Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
39279cc3 | 33 | #include <linux/compat.h> |
9ebefb18 | 34 | #include <linux/bit_spinlock.h> |
5103e947 | 35 | #include <linux/xattr.h> |
33268eaf | 36 | #include <linux/posix_acl.h> |
d899e052 | 37 | #include <linux/falloc.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
7a36ddec | 39 | #include <linux/ratelimit.h> |
22c44fe6 | 40 | #include <linux/mount.h> |
55e301fd | 41 | #include <linux/btrfs.h> |
53b381b3 | 42 | #include <linux/blkdev.h> |
f23b5a59 | 43 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 44 | #include <linux/uio.h> |
f6edc45e | 45 | #include <asm/unaligned.h> |
39279cc3 CM |
46 | #include "ctree.h" |
47 | #include "disk-io.h" | |
48 | #include "transaction.h" | |
49 | #include "btrfs_inode.h" | |
39279cc3 | 50 | #include "print-tree.h" |
e6dcd2dc | 51 | #include "ordered-data.h" |
95819c05 | 52 | #include "xattr.h" |
e02119d5 | 53 | #include "tree-log.h" |
4a54c8c1 | 54 | #include "volumes.h" |
c8b97818 | 55 | #include "compression.h" |
b4ce94de | 56 | #include "locking.h" |
dc89e982 | 57 | #include "free-space-cache.h" |
581bb050 | 58 | #include "inode-map.h" |
38c227d8 | 59 | #include "backref.h" |
f23b5a59 | 60 | #include "hash.h" |
63541927 | 61 | #include "props.h" |
31193213 | 62 | #include "qgroup.h" |
dda3245e | 63 | #include "dedupe.h" |
39279cc3 CM |
64 | |
65 | struct btrfs_iget_args { | |
90d3e592 | 66 | struct btrfs_key *location; |
39279cc3 CM |
67 | struct btrfs_root *root; |
68 | }; | |
69 | ||
f28a4928 FM |
70 | struct btrfs_dio_data { |
71 | u64 outstanding_extents; | |
72 | u64 reserve; | |
73 | u64 unsubmitted_oe_range_start; | |
74 | u64 unsubmitted_oe_range_end; | |
4aaedfb0 | 75 | int overwrite; |
f28a4928 FM |
76 | }; |
77 | ||
6e1d5dcc AD |
78 | static const struct inode_operations btrfs_dir_inode_operations; |
79 | static const struct inode_operations btrfs_symlink_inode_operations; | |
80 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
81 | static const struct inode_operations btrfs_special_inode_operations; | |
82 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
83 | static const struct address_space_operations btrfs_aops; |
84 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 85 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 86 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
87 | |
88 | static struct kmem_cache *btrfs_inode_cachep; | |
89 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 90 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 91 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
92 | |
93 | #define S_SHIFT 12 | |
4d4ab6d6 | 94 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
95 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
96 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
97 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
98 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
99 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
100 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
101 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
102 | }; | |
103 | ||
3972f260 | 104 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 105 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 106 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
107 | static noinline int cow_file_range(struct inode *inode, |
108 | struct page *locked_page, | |
dda3245e WX |
109 | u64 start, u64 end, u64 delalloc_end, |
110 | int *page_started, unsigned long *nr_written, | |
111 | int unlock, struct btrfs_dedupe_hash *hash); | |
6f9994db LB |
112 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
113 | u64 orig_start, u64 block_start, | |
114 | u64 block_len, u64 orig_block_len, | |
115 | u64 ram_bytes, int compress_type, | |
116 | int type); | |
7b128766 | 117 | |
52427260 QW |
118 | static void __endio_write_update_ordered(struct inode *inode, |
119 | const u64 offset, const u64 bytes, | |
120 | const bool uptodate); | |
121 | ||
122 | /* | |
123 | * Cleanup all submitted ordered extents in specified range to handle errors | |
124 | * from the fill_dellaloc() callback. | |
125 | * | |
126 | * NOTE: caller must ensure that when an error happens, it can not call | |
127 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
128 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
129 | * to be released, which we want to happen only when finishing the ordered | |
130 | * extent (btrfs_finish_ordered_io()). Also note that the caller of the | |
131 | * fill_delalloc() callback already does proper cleanup for the first page of | |
132 | * the range, that is, it invokes the callback writepage_end_io_hook() for the | |
133 | * range of the first page. | |
134 | */ | |
135 | static inline void btrfs_cleanup_ordered_extents(struct inode *inode, | |
136 | const u64 offset, | |
137 | const u64 bytes) | |
138 | { | |
63d71450 NA |
139 | unsigned long index = offset >> PAGE_SHIFT; |
140 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
141 | struct page *page; | |
142 | ||
143 | while (index <= end_index) { | |
144 | page = find_get_page(inode->i_mapping, index); | |
145 | index++; | |
146 | if (!page) | |
147 | continue; | |
148 | ClearPagePrivate2(page); | |
149 | put_page(page); | |
150 | } | |
52427260 QW |
151 | return __endio_write_update_ordered(inode, offset + PAGE_SIZE, |
152 | bytes - PAGE_SIZE, false); | |
153 | } | |
154 | ||
48a3b636 | 155 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 156 | |
6a3891c5 JB |
157 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
158 | void btrfs_test_inode_set_ops(struct inode *inode) | |
159 | { | |
160 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
161 | } | |
162 | #endif | |
163 | ||
f34f57a3 | 164 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
165 | struct inode *inode, struct inode *dir, |
166 | const struct qstr *qstr) | |
0279b4cd JO |
167 | { |
168 | int err; | |
169 | ||
f34f57a3 | 170 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 171 | if (!err) |
2a7dba39 | 172 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
173 | return err; |
174 | } | |
175 | ||
c8b97818 CM |
176 | /* |
177 | * this does all the hard work for inserting an inline extent into | |
178 | * the btree. The caller should have done a btrfs_drop_extents so that | |
179 | * no overlapping inline items exist in the btree | |
180 | */ | |
40f76580 | 181 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 182 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
183 | struct btrfs_root *root, struct inode *inode, |
184 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 185 | int compress_type, |
c8b97818 CM |
186 | struct page **compressed_pages) |
187 | { | |
c8b97818 CM |
188 | struct extent_buffer *leaf; |
189 | struct page *page = NULL; | |
190 | char *kaddr; | |
191 | unsigned long ptr; | |
192 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
193 | int ret; |
194 | size_t cur_size = size; | |
c8b97818 | 195 | unsigned long offset; |
c8b97818 | 196 | |
fe3f566c | 197 | if (compressed_size && compressed_pages) |
c8b97818 | 198 | cur_size = compressed_size; |
c8b97818 | 199 | |
1acae57b | 200 | inode_add_bytes(inode, size); |
c8b97818 | 201 | |
1acae57b FDBM |
202 | if (!extent_inserted) { |
203 | struct btrfs_key key; | |
204 | size_t datasize; | |
c8b97818 | 205 | |
4a0cc7ca | 206 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 207 | key.offset = start; |
962a298f | 208 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 209 | |
1acae57b FDBM |
210 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
211 | path->leave_spinning = 1; | |
212 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
213 | datasize); | |
79b4f4c6 | 214 | if (ret) |
1acae57b | 215 | goto fail; |
c8b97818 CM |
216 | } |
217 | leaf = path->nodes[0]; | |
218 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
219 | struct btrfs_file_extent_item); | |
220 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
221 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
222 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
223 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
224 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
225 | ptr = btrfs_file_extent_inline_start(ei); | |
226 | ||
261507a0 | 227 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
228 | struct page *cpage; |
229 | int i = 0; | |
d397712b | 230 | while (compressed_size > 0) { |
c8b97818 | 231 | cpage = compressed_pages[i]; |
5b050f04 | 232 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 233 | PAGE_SIZE); |
c8b97818 | 234 | |
7ac687d9 | 235 | kaddr = kmap_atomic(cpage); |
c8b97818 | 236 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 237 | kunmap_atomic(kaddr); |
c8b97818 CM |
238 | |
239 | i++; | |
240 | ptr += cur_size; | |
241 | compressed_size -= cur_size; | |
242 | } | |
243 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 244 | compress_type); |
c8b97818 CM |
245 | } else { |
246 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 247 | start >> PAGE_SHIFT); |
c8b97818 | 248 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 249 | kaddr = kmap_atomic(page); |
09cbfeaf | 250 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 251 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 252 | kunmap_atomic(kaddr); |
09cbfeaf | 253 | put_page(page); |
c8b97818 CM |
254 | } |
255 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 256 | btrfs_release_path(path); |
c8b97818 | 257 | |
c2167754 YZ |
258 | /* |
259 | * we're an inline extent, so nobody can | |
260 | * extend the file past i_size without locking | |
261 | * a page we already have locked. | |
262 | * | |
263 | * We must do any isize and inode updates | |
264 | * before we unlock the pages. Otherwise we | |
265 | * could end up racing with unlink. | |
266 | */ | |
c8b97818 | 267 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 268 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 269 | |
c8b97818 | 270 | fail: |
79b4f4c6 | 271 | return ret; |
c8b97818 CM |
272 | } |
273 | ||
274 | ||
275 | /* | |
276 | * conditionally insert an inline extent into the file. This | |
277 | * does the checks required to make sure the data is small enough | |
278 | * to fit as an inline extent. | |
279 | */ | |
00361589 JB |
280 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
281 | struct inode *inode, u64 start, | |
282 | u64 end, size_t compressed_size, | |
283 | int compress_type, | |
284 | struct page **compressed_pages) | |
c8b97818 | 285 | { |
0b246afa | 286 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 287 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
288 | u64 isize = i_size_read(inode); |
289 | u64 actual_end = min(end + 1, isize); | |
290 | u64 inline_len = actual_end - start; | |
0b246afa | 291 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
292 | u64 data_len = inline_len; |
293 | int ret; | |
1acae57b FDBM |
294 | struct btrfs_path *path; |
295 | int extent_inserted = 0; | |
296 | u32 extent_item_size; | |
c8b97818 CM |
297 | |
298 | if (compressed_size) | |
299 | data_len = compressed_size; | |
300 | ||
301 | if (start > 0 || | |
0b246afa JM |
302 | actual_end > fs_info->sectorsize || |
303 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 304 | (!compressed_size && |
0b246afa | 305 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 306 | end + 1 < isize || |
0b246afa | 307 | data_len > fs_info->max_inline) { |
c8b97818 CM |
308 | return 1; |
309 | } | |
310 | ||
1acae57b FDBM |
311 | path = btrfs_alloc_path(); |
312 | if (!path) | |
313 | return -ENOMEM; | |
314 | ||
00361589 | 315 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
316 | if (IS_ERR(trans)) { |
317 | btrfs_free_path(path); | |
00361589 | 318 | return PTR_ERR(trans); |
1acae57b | 319 | } |
0b246afa | 320 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
00361589 | 321 | |
1acae57b FDBM |
322 | if (compressed_size && compressed_pages) |
323 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
324 | compressed_size); | |
325 | else | |
326 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
327 | inline_len); | |
328 | ||
329 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
330 | start, aligned_end, NULL, | |
331 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 332 | if (ret) { |
66642832 | 333 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
334 | goto out; |
335 | } | |
c8b97818 CM |
336 | |
337 | if (isize > actual_end) | |
338 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
339 | ret = insert_inline_extent(trans, path, extent_inserted, |
340 | root, inode, start, | |
c8b97818 | 341 | inline_len, compressed_size, |
fe3f566c | 342 | compress_type, compressed_pages); |
2adcac1a | 343 | if (ret && ret != -ENOSPC) { |
66642832 | 344 | btrfs_abort_transaction(trans, ret); |
00361589 | 345 | goto out; |
2adcac1a | 346 | } else if (ret == -ENOSPC) { |
00361589 JB |
347 | ret = 1; |
348 | goto out; | |
79787eaa | 349 | } |
2adcac1a | 350 | |
bdc20e67 | 351 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
691fa059 | 352 | btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start); |
dcdbc059 | 353 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 354 | out: |
94ed938a QW |
355 | /* |
356 | * Don't forget to free the reserved space, as for inlined extent | |
357 | * it won't count as data extent, free them directly here. | |
358 | * And at reserve time, it's always aligned to page size, so | |
359 | * just free one page here. | |
360 | */ | |
bc42bda2 | 361 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 362 | btrfs_free_path(path); |
3a45bb20 | 363 | btrfs_end_transaction(trans); |
00361589 | 364 | return ret; |
c8b97818 CM |
365 | } |
366 | ||
771ed689 CM |
367 | struct async_extent { |
368 | u64 start; | |
369 | u64 ram_size; | |
370 | u64 compressed_size; | |
371 | struct page **pages; | |
372 | unsigned long nr_pages; | |
261507a0 | 373 | int compress_type; |
771ed689 CM |
374 | struct list_head list; |
375 | }; | |
376 | ||
377 | struct async_cow { | |
378 | struct inode *inode; | |
379 | struct btrfs_root *root; | |
380 | struct page *locked_page; | |
381 | u64 start; | |
382 | u64 end; | |
383 | struct list_head extents; | |
384 | struct btrfs_work work; | |
385 | }; | |
386 | ||
387 | static noinline int add_async_extent(struct async_cow *cow, | |
388 | u64 start, u64 ram_size, | |
389 | u64 compressed_size, | |
390 | struct page **pages, | |
261507a0 LZ |
391 | unsigned long nr_pages, |
392 | int compress_type) | |
771ed689 CM |
393 | { |
394 | struct async_extent *async_extent; | |
395 | ||
396 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 397 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
398 | async_extent->start = start; |
399 | async_extent->ram_size = ram_size; | |
400 | async_extent->compressed_size = compressed_size; | |
401 | async_extent->pages = pages; | |
402 | async_extent->nr_pages = nr_pages; | |
261507a0 | 403 | async_extent->compress_type = compress_type; |
771ed689 CM |
404 | list_add_tail(&async_extent->list, &cow->extents); |
405 | return 0; | |
406 | } | |
407 | ||
c2fcdcdf | 408 | static inline int inode_need_compress(struct inode *inode, u64 start, u64 end) |
f79707b0 | 409 | { |
0b246afa | 410 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
411 | |
412 | /* force compress */ | |
0b246afa | 413 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 414 | return 1; |
eec63c65 DS |
415 | /* defrag ioctl */ |
416 | if (BTRFS_I(inode)->defrag_compress) | |
417 | return 1; | |
f79707b0 WS |
418 | /* bad compression ratios */ |
419 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
420 | return 0; | |
0b246afa | 421 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 | 422 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
b52aa8c9 | 423 | BTRFS_I(inode)->prop_compress) |
c2fcdcdf | 424 | return btrfs_compress_heuristic(inode, start, end); |
f79707b0 WS |
425 | return 0; |
426 | } | |
427 | ||
6158e1ce | 428 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
429 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
430 | { | |
431 | /* If this is a small write inside eof, kick off a defrag */ | |
432 | if (num_bytes < small_write && | |
6158e1ce | 433 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
434 | btrfs_add_inode_defrag(NULL, inode); |
435 | } | |
436 | ||
d352ac68 | 437 | /* |
771ed689 CM |
438 | * we create compressed extents in two phases. The first |
439 | * phase compresses a range of pages that have already been | |
440 | * locked (both pages and state bits are locked). | |
c8b97818 | 441 | * |
771ed689 CM |
442 | * This is done inside an ordered work queue, and the compression |
443 | * is spread across many cpus. The actual IO submission is step | |
444 | * two, and the ordered work queue takes care of making sure that | |
445 | * happens in the same order things were put onto the queue by | |
446 | * writepages and friends. | |
c8b97818 | 447 | * |
771ed689 CM |
448 | * If this code finds it can't get good compression, it puts an |
449 | * entry onto the work queue to write the uncompressed bytes. This | |
450 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
451 | * are written in the same order that the flusher thread sent them |
452 | * down. | |
d352ac68 | 453 | */ |
c44f649e | 454 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
455 | struct page *locked_page, |
456 | u64 start, u64 end, | |
457 | struct async_cow *async_cow, | |
458 | int *num_added) | |
b888db2b | 459 | { |
0b246afa | 460 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
b888db2b | 461 | struct btrfs_root *root = BTRFS_I(inode)->root; |
db94535d | 462 | u64 num_bytes; |
0b246afa | 463 | u64 blocksize = fs_info->sectorsize; |
c8b97818 | 464 | u64 actual_end; |
42dc7bab | 465 | u64 isize = i_size_read(inode); |
e6dcd2dc | 466 | int ret = 0; |
c8b97818 CM |
467 | struct page **pages = NULL; |
468 | unsigned long nr_pages; | |
c8b97818 CM |
469 | unsigned long total_compressed = 0; |
470 | unsigned long total_in = 0; | |
c8b97818 CM |
471 | int i; |
472 | int will_compress; | |
0b246afa | 473 | int compress_type = fs_info->compress_type; |
4adaa611 | 474 | int redirty = 0; |
b888db2b | 475 | |
6158e1ce NB |
476 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
477 | SZ_16K); | |
4cb5300b | 478 | |
42dc7bab | 479 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
480 | again: |
481 | will_compress = 0; | |
09cbfeaf | 482 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
483 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
484 | nr_pages = min_t(unsigned long, nr_pages, | |
485 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 486 | |
f03d9301 CM |
487 | /* |
488 | * we don't want to send crud past the end of i_size through | |
489 | * compression, that's just a waste of CPU time. So, if the | |
490 | * end of the file is before the start of our current | |
491 | * requested range of bytes, we bail out to the uncompressed | |
492 | * cleanup code that can deal with all of this. | |
493 | * | |
494 | * It isn't really the fastest way to fix things, but this is a | |
495 | * very uncommon corner. | |
496 | */ | |
497 | if (actual_end <= start) | |
498 | goto cleanup_and_bail_uncompressed; | |
499 | ||
c8b97818 CM |
500 | total_compressed = actual_end - start; |
501 | ||
4bcbb332 SW |
502 | /* |
503 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 504 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
505 | */ |
506 | if (total_compressed <= blocksize && | |
507 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
508 | goto cleanup_and_bail_uncompressed; | |
509 | ||
069eac78 DS |
510 | total_compressed = min_t(unsigned long, total_compressed, |
511 | BTRFS_MAX_UNCOMPRESSED); | |
fda2832f | 512 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 513 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
514 | total_in = 0; |
515 | ret = 0; | |
db94535d | 516 | |
771ed689 CM |
517 | /* |
518 | * we do compression for mount -o compress and when the | |
519 | * inode has not been flagged as nocompress. This flag can | |
520 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 521 | */ |
c2fcdcdf | 522 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 523 | WARN_ON(pages); |
31e818fe | 524 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
525 | if (!pages) { |
526 | /* just bail out to the uncompressed code */ | |
527 | goto cont; | |
528 | } | |
c8b97818 | 529 | |
eec63c65 DS |
530 | if (BTRFS_I(inode)->defrag_compress) |
531 | compress_type = BTRFS_I(inode)->defrag_compress; | |
532 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 533 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 534 | |
4adaa611 CM |
535 | /* |
536 | * we need to call clear_page_dirty_for_io on each | |
537 | * page in the range. Otherwise applications with the file | |
538 | * mmap'd can wander in and change the page contents while | |
539 | * we are compressing them. | |
540 | * | |
541 | * If the compression fails for any reason, we set the pages | |
542 | * dirty again later on. | |
543 | */ | |
544 | extent_range_clear_dirty_for_io(inode, start, end); | |
545 | redirty = 1; | |
261507a0 LZ |
546 | ret = btrfs_compress_pages(compress_type, |
547 | inode->i_mapping, start, | |
38c31464 | 548 | pages, |
4d3a800e | 549 | &nr_pages, |
261507a0 | 550 | &total_in, |
e5d74902 | 551 | &total_compressed); |
c8b97818 CM |
552 | |
553 | if (!ret) { | |
554 | unsigned long offset = total_compressed & | |
09cbfeaf | 555 | (PAGE_SIZE - 1); |
4d3a800e | 556 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
557 | char *kaddr; |
558 | ||
559 | /* zero the tail end of the last page, we might be | |
560 | * sending it down to disk | |
561 | */ | |
562 | if (offset) { | |
7ac687d9 | 563 | kaddr = kmap_atomic(page); |
c8b97818 | 564 | memset(kaddr + offset, 0, |
09cbfeaf | 565 | PAGE_SIZE - offset); |
7ac687d9 | 566 | kunmap_atomic(kaddr); |
c8b97818 CM |
567 | } |
568 | will_compress = 1; | |
569 | } | |
570 | } | |
560f7d75 | 571 | cont: |
c8b97818 CM |
572 | if (start == 0) { |
573 | /* lets try to make an inline extent */ | |
771ed689 | 574 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 575 | /* we didn't compress the entire range, try |
771ed689 | 576 | * to make an uncompressed inline extent. |
c8b97818 | 577 | */ |
00361589 | 578 | ret = cow_file_range_inline(root, inode, start, end, |
f74670f7 | 579 | 0, BTRFS_COMPRESS_NONE, NULL); |
c8b97818 | 580 | } else { |
771ed689 | 581 | /* try making a compressed inline extent */ |
00361589 | 582 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
583 | total_compressed, |
584 | compress_type, pages); | |
c8b97818 | 585 | } |
79787eaa | 586 | if (ret <= 0) { |
151a41bc | 587 | unsigned long clear_flags = EXTENT_DELALLOC | |
a7e3b975 | 588 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG; |
e6eb4314 FM |
589 | unsigned long page_error_op; |
590 | ||
151a41bc | 591 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 592 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 593 | |
771ed689 | 594 | /* |
79787eaa JM |
595 | * inline extent creation worked or returned error, |
596 | * we don't need to create any more async work items. | |
597 | * Unlock and free up our temp pages. | |
771ed689 | 598 | */ |
ba8b04c1 QW |
599 | extent_clear_unlock_delalloc(inode, start, end, end, |
600 | NULL, clear_flags, | |
601 | PAGE_UNLOCK | | |
c2790a2e JB |
602 | PAGE_CLEAR_DIRTY | |
603 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 604 | page_error_op | |
c2790a2e | 605 | PAGE_END_WRITEBACK); |
1c81ba23 FM |
606 | if (ret == 0) |
607 | btrfs_free_reserved_data_space_noquota(inode, | |
608 | start, | |
609 | end - start + 1); | |
c8b97818 CM |
610 | goto free_pages_out; |
611 | } | |
612 | } | |
613 | ||
614 | if (will_compress) { | |
615 | /* | |
616 | * we aren't doing an inline extent round the compressed size | |
617 | * up to a block size boundary so the allocator does sane | |
618 | * things | |
619 | */ | |
fda2832f | 620 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
621 | |
622 | /* | |
623 | * one last check to make sure the compression is really a | |
170607eb TT |
624 | * win, compare the page count read with the blocks on disk, |
625 | * compression must free at least one sector size | |
c8b97818 | 626 | */ |
09cbfeaf | 627 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 628 | if (total_compressed + blocksize <= total_in) { |
c8b97818 | 629 | num_bytes = total_in; |
c8bb0c8b AS |
630 | *num_added += 1; |
631 | ||
632 | /* | |
633 | * The async work queues will take care of doing actual | |
634 | * allocation on disk for these compressed pages, and | |
635 | * will submit them to the elevator. | |
636 | */ | |
637 | add_async_extent(async_cow, start, num_bytes, | |
4d3a800e | 638 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
639 | compress_type); |
640 | ||
641 | if (start + num_bytes < end) { | |
642 | start += num_bytes; | |
643 | pages = NULL; | |
644 | cond_resched(); | |
645 | goto again; | |
646 | } | |
647 | return; | |
c8b97818 CM |
648 | } |
649 | } | |
c8bb0c8b | 650 | if (pages) { |
c8b97818 CM |
651 | /* |
652 | * the compression code ran but failed to make things smaller, | |
653 | * free any pages it allocated and our page pointer array | |
654 | */ | |
4d3a800e | 655 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 656 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 657 | put_page(pages[i]); |
c8b97818 CM |
658 | } |
659 | kfree(pages); | |
660 | pages = NULL; | |
661 | total_compressed = 0; | |
4d3a800e | 662 | nr_pages = 0; |
c8b97818 CM |
663 | |
664 | /* flag the file so we don't compress in the future */ | |
0b246afa | 665 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 666 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 667 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 668 | } |
c8b97818 | 669 | } |
f03d9301 | 670 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
671 | /* |
672 | * No compression, but we still need to write the pages in the file | |
673 | * we've been given so far. redirty the locked page if it corresponds | |
674 | * to our extent and set things up for the async work queue to run | |
675 | * cow_file_range to do the normal delalloc dance. | |
676 | */ | |
677 | if (page_offset(locked_page) >= start && | |
678 | page_offset(locked_page) <= end) | |
679 | __set_page_dirty_nobuffers(locked_page); | |
680 | /* unlocked later on in the async handlers */ | |
681 | ||
682 | if (redirty) | |
683 | extent_range_redirty_for_io(inode, start, end); | |
684 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0, | |
685 | BTRFS_COMPRESS_NONE); | |
686 | *num_added += 1; | |
3b951516 | 687 | |
c44f649e | 688 | return; |
771ed689 CM |
689 | |
690 | free_pages_out: | |
4d3a800e | 691 | for (i = 0; i < nr_pages; i++) { |
771ed689 | 692 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 693 | put_page(pages[i]); |
771ed689 | 694 | } |
d397712b | 695 | kfree(pages); |
771ed689 | 696 | } |
771ed689 | 697 | |
40ae837b FM |
698 | static void free_async_extent_pages(struct async_extent *async_extent) |
699 | { | |
700 | int i; | |
701 | ||
702 | if (!async_extent->pages) | |
703 | return; | |
704 | ||
705 | for (i = 0; i < async_extent->nr_pages; i++) { | |
706 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 707 | put_page(async_extent->pages[i]); |
40ae837b FM |
708 | } |
709 | kfree(async_extent->pages); | |
710 | async_extent->nr_pages = 0; | |
711 | async_extent->pages = NULL; | |
771ed689 CM |
712 | } |
713 | ||
714 | /* | |
715 | * phase two of compressed writeback. This is the ordered portion | |
716 | * of the code, which only gets called in the order the work was | |
717 | * queued. We walk all the async extents created by compress_file_range | |
718 | * and send them down to the disk. | |
719 | */ | |
dec8f175 | 720 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
721 | struct async_cow *async_cow) |
722 | { | |
0b246afa | 723 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
724 | struct async_extent *async_extent; |
725 | u64 alloc_hint = 0; | |
771ed689 CM |
726 | struct btrfs_key ins; |
727 | struct extent_map *em; | |
728 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
771ed689 | 729 | struct extent_io_tree *io_tree; |
f5a84ee3 | 730 | int ret = 0; |
771ed689 | 731 | |
3e04e7f1 | 732 | again: |
d397712b | 733 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
734 | async_extent = list_entry(async_cow->extents.next, |
735 | struct async_extent, list); | |
736 | list_del(&async_extent->list); | |
c8b97818 | 737 | |
771ed689 CM |
738 | io_tree = &BTRFS_I(inode)->io_tree; |
739 | ||
f5a84ee3 | 740 | retry: |
771ed689 CM |
741 | /* did the compression code fall back to uncompressed IO? */ |
742 | if (!async_extent->pages) { | |
743 | int page_started = 0; | |
744 | unsigned long nr_written = 0; | |
745 | ||
746 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 747 | async_extent->start + |
d0082371 | 748 | async_extent->ram_size - 1); |
771ed689 CM |
749 | |
750 | /* allocate blocks */ | |
f5a84ee3 JB |
751 | ret = cow_file_range(inode, async_cow->locked_page, |
752 | async_extent->start, | |
753 | async_extent->start + | |
754 | async_extent->ram_size - 1, | |
dda3245e WX |
755 | async_extent->start + |
756 | async_extent->ram_size - 1, | |
757 | &page_started, &nr_written, 0, | |
758 | NULL); | |
771ed689 | 759 | |
79787eaa JM |
760 | /* JDM XXX */ |
761 | ||
771ed689 CM |
762 | /* |
763 | * if page_started, cow_file_range inserted an | |
764 | * inline extent and took care of all the unlocking | |
765 | * and IO for us. Otherwise, we need to submit | |
766 | * all those pages down to the drive. | |
767 | */ | |
f5a84ee3 | 768 | if (!page_started && !ret) |
771ed689 CM |
769 | extent_write_locked_range(io_tree, |
770 | inode, async_extent->start, | |
d397712b | 771 | async_extent->start + |
771ed689 CM |
772 | async_extent->ram_size - 1, |
773 | btrfs_get_extent, | |
774 | WB_SYNC_ALL); | |
3e04e7f1 JB |
775 | else if (ret) |
776 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
777 | kfree(async_extent); |
778 | cond_resched(); | |
779 | continue; | |
780 | } | |
781 | ||
782 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 783 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 784 | |
18513091 | 785 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
786 | async_extent->compressed_size, |
787 | async_extent->compressed_size, | |
e570fd27 | 788 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 789 | if (ret) { |
40ae837b | 790 | free_async_extent_pages(async_extent); |
3e04e7f1 | 791 | |
fdf8e2ea JB |
792 | if (ret == -ENOSPC) { |
793 | unlock_extent(io_tree, async_extent->start, | |
794 | async_extent->start + | |
795 | async_extent->ram_size - 1); | |
ce62003f LB |
796 | |
797 | /* | |
798 | * we need to redirty the pages if we decide to | |
799 | * fallback to uncompressed IO, otherwise we | |
800 | * will not submit these pages down to lower | |
801 | * layers. | |
802 | */ | |
803 | extent_range_redirty_for_io(inode, | |
804 | async_extent->start, | |
805 | async_extent->start + | |
806 | async_extent->ram_size - 1); | |
807 | ||
79787eaa | 808 | goto retry; |
fdf8e2ea | 809 | } |
3e04e7f1 | 810 | goto out_free; |
f5a84ee3 | 811 | } |
c2167754 YZ |
812 | /* |
813 | * here we're doing allocation and writeback of the | |
814 | * compressed pages | |
815 | */ | |
6f9994db LB |
816 | em = create_io_em(inode, async_extent->start, |
817 | async_extent->ram_size, /* len */ | |
818 | async_extent->start, /* orig_start */ | |
819 | ins.objectid, /* block_start */ | |
820 | ins.offset, /* block_len */ | |
821 | ins.offset, /* orig_block_len */ | |
822 | async_extent->ram_size, /* ram_bytes */ | |
823 | async_extent->compress_type, | |
824 | BTRFS_ORDERED_COMPRESSED); | |
825 | if (IS_ERR(em)) | |
826 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 827 | goto out_free_reserve; |
6f9994db | 828 | free_extent_map(em); |
3e04e7f1 | 829 | |
261507a0 LZ |
830 | ret = btrfs_add_ordered_extent_compress(inode, |
831 | async_extent->start, | |
832 | ins.objectid, | |
833 | async_extent->ram_size, | |
834 | ins.offset, | |
835 | BTRFS_ORDERED_COMPRESSED, | |
836 | async_extent->compress_type); | |
d9f85963 | 837 | if (ret) { |
dcdbc059 NB |
838 | btrfs_drop_extent_cache(BTRFS_I(inode), |
839 | async_extent->start, | |
d9f85963 FM |
840 | async_extent->start + |
841 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 842 | goto out_free_reserve; |
d9f85963 | 843 | } |
0b246afa | 844 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 845 | |
771ed689 CM |
846 | /* |
847 | * clear dirty, set writeback and unlock the pages. | |
848 | */ | |
c2790a2e | 849 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
850 | async_extent->start + |
851 | async_extent->ram_size - 1, | |
a791e35e CM |
852 | async_extent->start + |
853 | async_extent->ram_size - 1, | |
151a41bc JB |
854 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
855 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 856 | PAGE_SET_WRITEBACK); |
4e4cbee9 | 857 | if (btrfs_submit_compressed_write(inode, |
d397712b CM |
858 | async_extent->start, |
859 | async_extent->ram_size, | |
860 | ins.objectid, | |
861 | ins.offset, async_extent->pages, | |
4e4cbee9 | 862 | async_extent->nr_pages)) { |
fce2a4e6 FM |
863 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
864 | struct page *p = async_extent->pages[0]; | |
865 | const u64 start = async_extent->start; | |
866 | const u64 end = start + async_extent->ram_size - 1; | |
867 | ||
868 | p->mapping = inode->i_mapping; | |
869 | tree->ops->writepage_end_io_hook(p, start, end, | |
870 | NULL, 0); | |
871 | p->mapping = NULL; | |
ba8b04c1 QW |
872 | extent_clear_unlock_delalloc(inode, start, end, end, |
873 | NULL, 0, | |
fce2a4e6 FM |
874 | PAGE_END_WRITEBACK | |
875 | PAGE_SET_ERROR); | |
40ae837b | 876 | free_async_extent_pages(async_extent); |
fce2a4e6 | 877 | } |
771ed689 CM |
878 | alloc_hint = ins.objectid + ins.offset; |
879 | kfree(async_extent); | |
880 | cond_resched(); | |
881 | } | |
dec8f175 | 882 | return; |
3e04e7f1 | 883 | out_free_reserve: |
0b246afa | 884 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 885 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 886 | out_free: |
c2790a2e | 887 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
888 | async_extent->start + |
889 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
890 | async_extent->start + |
891 | async_extent->ram_size - 1, | |
c2790a2e | 892 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 893 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
894 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
895 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
896 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
897 | PAGE_SET_ERROR); | |
40ae837b | 898 | free_async_extent_pages(async_extent); |
79787eaa | 899 | kfree(async_extent); |
3e04e7f1 | 900 | goto again; |
771ed689 CM |
901 | } |
902 | ||
4b46fce2 JB |
903 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
904 | u64 num_bytes) | |
905 | { | |
906 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
907 | struct extent_map *em; | |
908 | u64 alloc_hint = 0; | |
909 | ||
910 | read_lock(&em_tree->lock); | |
911 | em = search_extent_mapping(em_tree, start, num_bytes); | |
912 | if (em) { | |
913 | /* | |
914 | * if block start isn't an actual block number then find the | |
915 | * first block in this inode and use that as a hint. If that | |
916 | * block is also bogus then just don't worry about it. | |
917 | */ | |
918 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
919 | free_extent_map(em); | |
920 | em = search_extent_mapping(em_tree, 0, 0); | |
921 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
922 | alloc_hint = em->block_start; | |
923 | if (em) | |
924 | free_extent_map(em); | |
925 | } else { | |
926 | alloc_hint = em->block_start; | |
927 | free_extent_map(em); | |
928 | } | |
929 | } | |
930 | read_unlock(&em_tree->lock); | |
931 | ||
932 | return alloc_hint; | |
933 | } | |
934 | ||
771ed689 CM |
935 | /* |
936 | * when extent_io.c finds a delayed allocation range in the file, | |
937 | * the call backs end up in this code. The basic idea is to | |
938 | * allocate extents on disk for the range, and create ordered data structs | |
939 | * in ram to track those extents. | |
940 | * | |
941 | * locked_page is the page that writepage had locked already. We use | |
942 | * it to make sure we don't do extra locks or unlocks. | |
943 | * | |
944 | * *page_started is set to one if we unlock locked_page and do everything | |
945 | * required to start IO on it. It may be clean and already done with | |
946 | * IO when we return. | |
947 | */ | |
00361589 JB |
948 | static noinline int cow_file_range(struct inode *inode, |
949 | struct page *locked_page, | |
dda3245e WX |
950 | u64 start, u64 end, u64 delalloc_end, |
951 | int *page_started, unsigned long *nr_written, | |
952 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 953 | { |
0b246afa | 954 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 955 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
956 | u64 alloc_hint = 0; |
957 | u64 num_bytes; | |
958 | unsigned long ram_size; | |
959 | u64 disk_num_bytes; | |
a315e68f | 960 | u64 cur_alloc_size = 0; |
0b246afa | 961 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
962 | struct btrfs_key ins; |
963 | struct extent_map *em; | |
a315e68f FM |
964 | unsigned clear_bits; |
965 | unsigned long page_ops; | |
966 | bool extent_reserved = false; | |
771ed689 CM |
967 | int ret = 0; |
968 | ||
70ddc553 | 969 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 970 | WARN_ON_ONCE(1); |
29bce2f3 JB |
971 | ret = -EINVAL; |
972 | goto out_unlock; | |
02ecd2c2 | 973 | } |
771ed689 | 974 | |
fda2832f | 975 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
976 | num_bytes = max(blocksize, num_bytes); |
977 | disk_num_bytes = num_bytes; | |
771ed689 | 978 | |
6158e1ce | 979 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 980 | |
771ed689 CM |
981 | if (start == 0) { |
982 | /* lets try to make an inline extent */ | |
f74670f7 AJ |
983 | ret = cow_file_range_inline(root, inode, start, end, 0, |
984 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 985 | if (ret == 0) { |
ba8b04c1 QW |
986 | extent_clear_unlock_delalloc(inode, start, end, |
987 | delalloc_end, NULL, | |
c2790a2e | 988 | EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 989 | EXTENT_DELALLOC_NEW | |
151a41bc | 990 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
991 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
992 | PAGE_END_WRITEBACK); | |
18513091 WX |
993 | btrfs_free_reserved_data_space_noquota(inode, start, |
994 | end - start + 1); | |
771ed689 | 995 | *nr_written = *nr_written + |
09cbfeaf | 996 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 997 | *page_started = 1; |
771ed689 | 998 | goto out; |
79787eaa | 999 | } else if (ret < 0) { |
79787eaa | 1000 | goto out_unlock; |
771ed689 CM |
1001 | } |
1002 | } | |
1003 | ||
1004 | BUG_ON(disk_num_bytes > | |
0b246afa | 1005 | btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1006 | |
4b46fce2 | 1007 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
1008 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1009 | start + num_bytes - 1, 0); | |
771ed689 | 1010 | |
d397712b | 1011 | while (disk_num_bytes > 0) { |
287a0ab9 | 1012 | cur_alloc_size = disk_num_bytes; |
18513091 | 1013 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 1014 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 1015 | &ins, 1, 1); |
00361589 | 1016 | if (ret < 0) |
79787eaa | 1017 | goto out_unlock; |
a315e68f FM |
1018 | cur_alloc_size = ins.offset; |
1019 | extent_reserved = true; | |
d397712b | 1020 | |
771ed689 | 1021 | ram_size = ins.offset; |
6f9994db LB |
1022 | em = create_io_em(inode, start, ins.offset, /* len */ |
1023 | start, /* orig_start */ | |
1024 | ins.objectid, /* block_start */ | |
1025 | ins.offset, /* block_len */ | |
1026 | ins.offset, /* orig_block_len */ | |
1027 | ram_size, /* ram_bytes */ | |
1028 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1029 | BTRFS_ORDERED_REGULAR /* type */); |
4cb89420 SY |
1030 | if (IS_ERR(em)) { |
1031 | ret = PTR_ERR(em); | |
ace68bac | 1032 | goto out_reserve; |
4cb89420 | 1033 | } |
6f9994db | 1034 | free_extent_map(em); |
e6dcd2dc | 1035 | |
e6dcd2dc | 1036 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1037 | ram_size, cur_alloc_size, 0); |
ace68bac | 1038 | if (ret) |
d9f85963 | 1039 | goto out_drop_extent_cache; |
c8b97818 | 1040 | |
17d217fe YZ |
1041 | if (root->root_key.objectid == |
1042 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1043 | ret = btrfs_reloc_clone_csums(inode, start, | |
1044 | cur_alloc_size); | |
4dbd80fb QW |
1045 | /* |
1046 | * Only drop cache here, and process as normal. | |
1047 | * | |
1048 | * We must not allow extent_clear_unlock_delalloc() | |
1049 | * at out_unlock label to free meta of this ordered | |
1050 | * extent, as its meta should be freed by | |
1051 | * btrfs_finish_ordered_io(). | |
1052 | * | |
1053 | * So we must continue until @start is increased to | |
1054 | * skip current ordered extent. | |
1055 | */ | |
00361589 | 1056 | if (ret) |
4dbd80fb QW |
1057 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1058 | start + ram_size - 1, 0); | |
17d217fe YZ |
1059 | } |
1060 | ||
0b246afa | 1061 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1062 | |
c8b97818 CM |
1063 | /* we're not doing compressed IO, don't unlock the first |
1064 | * page (which the caller expects to stay locked), don't | |
1065 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1066 | * |
1067 | * Do set the Private2 bit so we know this page was properly | |
1068 | * setup for writepage | |
c8b97818 | 1069 | */ |
a315e68f FM |
1070 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1071 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1072 | |
c2790a2e | 1073 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1074 | start + ram_size - 1, |
1075 | delalloc_end, locked_page, | |
c2790a2e | 1076 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1077 | page_ops); |
4dbd80fb QW |
1078 | if (disk_num_bytes < cur_alloc_size) |
1079 | disk_num_bytes = 0; | |
1080 | else | |
1081 | disk_num_bytes -= cur_alloc_size; | |
c59f8951 CM |
1082 | num_bytes -= cur_alloc_size; |
1083 | alloc_hint = ins.objectid + ins.offset; | |
1084 | start += cur_alloc_size; | |
a315e68f | 1085 | extent_reserved = false; |
4dbd80fb QW |
1086 | |
1087 | /* | |
1088 | * btrfs_reloc_clone_csums() error, since start is increased | |
1089 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1090 | * free metadata of current ordered extent, we're OK to exit. | |
1091 | */ | |
1092 | if (ret) | |
1093 | goto out_unlock; | |
b888db2b | 1094 | } |
79787eaa | 1095 | out: |
be20aa9d | 1096 | return ret; |
b7d5b0a8 | 1097 | |
d9f85963 | 1098 | out_drop_extent_cache: |
dcdbc059 | 1099 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1100 | out_reserve: |
0b246afa | 1101 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1102 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1103 | out_unlock: |
a7e3b975 FM |
1104 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1105 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1106 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1107 | PAGE_END_WRITEBACK; | |
1108 | /* | |
1109 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1110 | * failed to create the respective ordered extent, then it means that | |
1111 | * when we reserved the extent we decremented the extent's size from | |
1112 | * the data space_info's bytes_may_use counter and incremented the | |
1113 | * space_info's bytes_reserved counter by the same amount. We must make | |
1114 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1115 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1116 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1117 | */ | |
1118 | if (extent_reserved) { | |
1119 | extent_clear_unlock_delalloc(inode, start, | |
1120 | start + cur_alloc_size, | |
1121 | start + cur_alloc_size, | |
1122 | locked_page, | |
1123 | clear_bits, | |
1124 | page_ops); | |
1125 | start += cur_alloc_size; | |
1126 | if (start >= end) | |
1127 | goto out; | |
1128 | } | |
ba8b04c1 QW |
1129 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1130 | locked_page, | |
a315e68f FM |
1131 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1132 | page_ops); | |
79787eaa | 1133 | goto out; |
771ed689 | 1134 | } |
c8b97818 | 1135 | |
771ed689 CM |
1136 | /* |
1137 | * work queue call back to started compression on a file and pages | |
1138 | */ | |
1139 | static noinline void async_cow_start(struct btrfs_work *work) | |
1140 | { | |
1141 | struct async_cow *async_cow; | |
1142 | int num_added = 0; | |
1143 | async_cow = container_of(work, struct async_cow, work); | |
1144 | ||
1145 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1146 | async_cow->start, async_cow->end, async_cow, | |
1147 | &num_added); | |
8180ef88 | 1148 | if (num_added == 0) { |
cb77fcd8 | 1149 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1150 | async_cow->inode = NULL; |
8180ef88 | 1151 | } |
771ed689 CM |
1152 | } |
1153 | ||
1154 | /* | |
1155 | * work queue call back to submit previously compressed pages | |
1156 | */ | |
1157 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1158 | { | |
0b246afa | 1159 | struct btrfs_fs_info *fs_info; |
771ed689 CM |
1160 | struct async_cow *async_cow; |
1161 | struct btrfs_root *root; | |
1162 | unsigned long nr_pages; | |
1163 | ||
1164 | async_cow = container_of(work, struct async_cow, work); | |
1165 | ||
1166 | root = async_cow->root; | |
0b246afa | 1167 | fs_info = root->fs_info; |
09cbfeaf KS |
1168 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1169 | PAGE_SHIFT; | |
771ed689 | 1170 | |
ee863954 DS |
1171 | /* |
1172 | * atomic_sub_return implies a barrier for waitqueue_active | |
1173 | */ | |
0b246afa | 1174 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
ee22184b | 1175 | 5 * SZ_1M && |
0b246afa JM |
1176 | waitqueue_active(&fs_info->async_submit_wait)) |
1177 | wake_up(&fs_info->async_submit_wait); | |
771ed689 | 1178 | |
d397712b | 1179 | if (async_cow->inode) |
771ed689 | 1180 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1181 | } |
c8b97818 | 1182 | |
771ed689 CM |
1183 | static noinline void async_cow_free(struct btrfs_work *work) |
1184 | { | |
1185 | struct async_cow *async_cow; | |
1186 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1187 | if (async_cow->inode) |
cb77fcd8 | 1188 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1189 | kfree(async_cow); |
1190 | } | |
1191 | ||
1192 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1193 | u64 start, u64 end, int *page_started, | |
1194 | unsigned long *nr_written) | |
1195 | { | |
0b246afa | 1196 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
1197 | struct async_cow *async_cow; |
1198 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1199 | unsigned long nr_pages; | |
1200 | u64 cur_end; | |
771ed689 | 1201 | |
a3429ab7 CM |
1202 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1203 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1204 | while (start < end) { |
771ed689 | 1205 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1206 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1207 | async_cow->inode = igrab(inode); |
771ed689 CM |
1208 | async_cow->root = root; |
1209 | async_cow->locked_page = locked_page; | |
1210 | async_cow->start = start; | |
1211 | ||
f79707b0 | 1212 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
0b246afa | 1213 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
771ed689 CM |
1214 | cur_end = end; |
1215 | else | |
ee22184b | 1216 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1217 | |
1218 | async_cow->end = cur_end; | |
1219 | INIT_LIST_HEAD(&async_cow->extents); | |
1220 | ||
9e0af237 LB |
1221 | btrfs_init_work(&async_cow->work, |
1222 | btrfs_delalloc_helper, | |
1223 | async_cow_start, async_cow_submit, | |
1224 | async_cow_free); | |
771ed689 | 1225 | |
09cbfeaf KS |
1226 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1227 | PAGE_SHIFT; | |
0b246afa | 1228 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1229 | |
0b246afa | 1230 | btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work); |
771ed689 | 1231 | |
0b246afa JM |
1232 | while (atomic_read(&fs_info->async_submit_draining) && |
1233 | atomic_read(&fs_info->async_delalloc_pages)) { | |
1234 | wait_event(fs_info->async_submit_wait, | |
1235 | (atomic_read(&fs_info->async_delalloc_pages) == | |
1236 | 0)); | |
771ed689 CM |
1237 | } |
1238 | ||
1239 | *nr_written += nr_pages; | |
1240 | start = cur_end + 1; | |
1241 | } | |
1242 | *page_started = 1; | |
1243 | return 0; | |
be20aa9d CM |
1244 | } |
1245 | ||
2ff7e61e | 1246 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1247 | u64 bytenr, u64 num_bytes) |
1248 | { | |
1249 | int ret; | |
1250 | struct btrfs_ordered_sum *sums; | |
1251 | LIST_HEAD(list); | |
1252 | ||
0b246afa | 1253 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1254 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1255 | if (ret == 0 && list_empty(&list)) |
1256 | return 0; | |
1257 | ||
1258 | while (!list_empty(&list)) { | |
1259 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1260 | list_del(&sums->list); | |
1261 | kfree(sums); | |
1262 | } | |
4be89529 LB |
1263 | if (ret < 0) |
1264 | return ret; | |
17d217fe YZ |
1265 | return 1; |
1266 | } | |
1267 | ||
d352ac68 CM |
1268 | /* |
1269 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1270 | * of the extents that exist in the file, and COWs the file as required. | |
1271 | * | |
1272 | * If no cow copies or snapshots exist, we write directly to the existing | |
1273 | * blocks on disk | |
1274 | */ | |
7f366cfe CM |
1275 | static noinline int run_delalloc_nocow(struct inode *inode, |
1276 | struct page *locked_page, | |
771ed689 CM |
1277 | u64 start, u64 end, int *page_started, int force, |
1278 | unsigned long *nr_written) | |
be20aa9d | 1279 | { |
0b246afa | 1280 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d CM |
1281 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1282 | struct extent_buffer *leaf; | |
be20aa9d | 1283 | struct btrfs_path *path; |
80ff3856 | 1284 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1285 | struct btrfs_key found_key; |
6f9994db | 1286 | struct extent_map *em; |
80ff3856 YZ |
1287 | u64 cow_start; |
1288 | u64 cur_offset; | |
1289 | u64 extent_end; | |
5d4f98a2 | 1290 | u64 extent_offset; |
80ff3856 YZ |
1291 | u64 disk_bytenr; |
1292 | u64 num_bytes; | |
b4939680 | 1293 | u64 disk_num_bytes; |
cc95bef6 | 1294 | u64 ram_bytes; |
80ff3856 | 1295 | int extent_type; |
79787eaa | 1296 | int ret, err; |
d899e052 | 1297 | int type; |
80ff3856 YZ |
1298 | int nocow; |
1299 | int check_prev = 1; | |
82d5902d | 1300 | bool nolock; |
4a0cc7ca | 1301 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1302 | |
1303 | path = btrfs_alloc_path(); | |
17ca04af | 1304 | if (!path) { |
ba8b04c1 QW |
1305 | extent_clear_unlock_delalloc(inode, start, end, end, |
1306 | locked_page, | |
c2790a2e | 1307 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1308 | EXTENT_DO_ACCOUNTING | |
1309 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1310 | PAGE_CLEAR_DIRTY | |
1311 | PAGE_SET_WRITEBACK | | |
1312 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1313 | return -ENOMEM; |
17ca04af | 1314 | } |
82d5902d | 1315 | |
70ddc553 | 1316 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
82d5902d | 1317 | |
80ff3856 YZ |
1318 | cow_start = (u64)-1; |
1319 | cur_offset = start; | |
1320 | while (1) { | |
e4c3b2dc | 1321 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1322 | cur_offset, 0); |
d788a349 | 1323 | if (ret < 0) |
79787eaa | 1324 | goto error; |
80ff3856 YZ |
1325 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1326 | leaf = path->nodes[0]; | |
1327 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1328 | path->slots[0] - 1); | |
33345d01 | 1329 | if (found_key.objectid == ino && |
80ff3856 YZ |
1330 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1331 | path->slots[0]--; | |
1332 | } | |
1333 | check_prev = 0; | |
1334 | next_slot: | |
1335 | leaf = path->nodes[0]; | |
1336 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1337 | ret = btrfs_next_leaf(root, path); | |
c766cb48 LB |
1338 | if (ret < 0) { |
1339 | if (cow_start != (u64)-1) | |
1340 | cur_offset = cow_start; | |
79787eaa | 1341 | goto error; |
c766cb48 | 1342 | } |
80ff3856 YZ |
1343 | if (ret > 0) |
1344 | break; | |
1345 | leaf = path->nodes[0]; | |
1346 | } | |
be20aa9d | 1347 | |
80ff3856 YZ |
1348 | nocow = 0; |
1349 | disk_bytenr = 0; | |
17d217fe | 1350 | num_bytes = 0; |
80ff3856 YZ |
1351 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1352 | ||
1d512cb7 FM |
1353 | if (found_key.objectid > ino) |
1354 | break; | |
1355 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1356 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1357 | path->slots[0]++; | |
1358 | goto next_slot; | |
1359 | } | |
1360 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1361 | found_key.offset > end) |
1362 | break; | |
1363 | ||
1364 | if (found_key.offset > cur_offset) { | |
1365 | extent_end = found_key.offset; | |
e9061e21 | 1366 | extent_type = 0; |
80ff3856 YZ |
1367 | goto out_check; |
1368 | } | |
1369 | ||
1370 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1371 | struct btrfs_file_extent_item); | |
1372 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1373 | ||
cc95bef6 | 1374 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1375 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1376 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1377 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1378 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1379 | extent_end = found_key.offset + |
1380 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1381 | disk_num_bytes = |
1382 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1383 | if (extent_end <= start) { |
1384 | path->slots[0]++; | |
1385 | goto next_slot; | |
1386 | } | |
17d217fe YZ |
1387 | if (disk_bytenr == 0) |
1388 | goto out_check; | |
80ff3856 YZ |
1389 | if (btrfs_file_extent_compression(leaf, fi) || |
1390 | btrfs_file_extent_encryption(leaf, fi) || | |
1391 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1392 | goto out_check; | |
d899e052 YZ |
1393 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1394 | goto out_check; | |
2ff7e61e | 1395 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1396 | goto out_check; |
4be89529 LB |
1397 | ret = btrfs_cross_ref_exist(root, ino, |
1398 | found_key.offset - | |
1399 | extent_offset, disk_bytenr); | |
1400 | if (ret) { | |
1401 | /* | |
1402 | * ret could be -EIO if the above fails to read | |
1403 | * metadata. | |
1404 | */ | |
1405 | if (ret < 0) { | |
1406 | if (cow_start != (u64)-1) | |
1407 | cur_offset = cow_start; | |
1408 | goto error; | |
1409 | } | |
1410 | ||
1411 | WARN_ON_ONCE(nolock); | |
17d217fe | 1412 | goto out_check; |
4be89529 | 1413 | } |
5d4f98a2 | 1414 | disk_bytenr += extent_offset; |
17d217fe YZ |
1415 | disk_bytenr += cur_offset - found_key.offset; |
1416 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1417 | /* |
1418 | * if there are pending snapshots for this root, | |
1419 | * we fall into common COW way. | |
1420 | */ | |
1421 | if (!nolock) { | |
ea14b57f | 1422 | err = btrfs_start_write_no_snapshotting(root); |
e9894fd3 WS |
1423 | if (!err) |
1424 | goto out_check; | |
1425 | } | |
17d217fe YZ |
1426 | /* |
1427 | * force cow if csum exists in the range. | |
1428 | * this ensure that csum for a given extent are | |
1429 | * either valid or do not exist. | |
1430 | */ | |
4be89529 LB |
1431 | ret = csum_exist_in_range(fs_info, disk_bytenr, |
1432 | num_bytes); | |
1433 | if (ret) { | |
91e1f56a | 1434 | if (!nolock) |
ea14b57f | 1435 | btrfs_end_write_no_snapshotting(root); |
4be89529 LB |
1436 | |
1437 | /* | |
1438 | * ret could be -EIO if the above fails to read | |
1439 | * metadata. | |
1440 | */ | |
1441 | if (ret < 0) { | |
1442 | if (cow_start != (u64)-1) | |
1443 | cur_offset = cow_start; | |
1444 | goto error; | |
1445 | } | |
1446 | WARN_ON_ONCE(nolock); | |
17d217fe | 1447 | goto out_check; |
91e1f56a RK |
1448 | } |
1449 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) { | |
1450 | if (!nolock) | |
ea14b57f | 1451 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1452 | goto out_check; |
91e1f56a | 1453 | } |
80ff3856 YZ |
1454 | nocow = 1; |
1455 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1456 | extent_end = found_key.offset + | |
514ac8ad CM |
1457 | btrfs_file_extent_inline_len(leaf, |
1458 | path->slots[0], fi); | |
da17066c | 1459 | extent_end = ALIGN(extent_end, |
0b246afa | 1460 | fs_info->sectorsize); |
80ff3856 YZ |
1461 | } else { |
1462 | BUG_ON(1); | |
1463 | } | |
1464 | out_check: | |
1465 | if (extent_end <= start) { | |
1466 | path->slots[0]++; | |
e9894fd3 | 1467 | if (!nolock && nocow) |
ea14b57f | 1468 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1469 | if (nocow) |
0b246afa | 1470 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
80ff3856 YZ |
1471 | goto next_slot; |
1472 | } | |
1473 | if (!nocow) { | |
1474 | if (cow_start == (u64)-1) | |
1475 | cow_start = cur_offset; | |
1476 | cur_offset = extent_end; | |
1477 | if (cur_offset > end) | |
1478 | break; | |
1479 | path->slots[0]++; | |
1480 | goto next_slot; | |
7ea394f1 YZ |
1481 | } |
1482 | ||
b3b4aa74 | 1483 | btrfs_release_path(path); |
80ff3856 | 1484 | if (cow_start != (u64)-1) { |
00361589 JB |
1485 | ret = cow_file_range(inode, locked_page, |
1486 | cow_start, found_key.offset - 1, | |
dda3245e WX |
1487 | end, page_started, nr_written, 1, |
1488 | NULL); | |
e9894fd3 WS |
1489 | if (ret) { |
1490 | if (!nolock && nocow) | |
ea14b57f | 1491 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1492 | if (nocow) |
0b246afa | 1493 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1494 | disk_bytenr); |
79787eaa | 1495 | goto error; |
e9894fd3 | 1496 | } |
80ff3856 | 1497 | cow_start = (u64)-1; |
7ea394f1 | 1498 | } |
80ff3856 | 1499 | |
d899e052 | 1500 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db LB |
1501 | u64 orig_start = found_key.offset - extent_offset; |
1502 | ||
1503 | em = create_io_em(inode, cur_offset, num_bytes, | |
1504 | orig_start, | |
1505 | disk_bytenr, /* block_start */ | |
1506 | num_bytes, /* block_len */ | |
1507 | disk_num_bytes, /* orig_block_len */ | |
1508 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1509 | BTRFS_ORDERED_PREALLOC); | |
1510 | if (IS_ERR(em)) { | |
1511 | if (!nolock && nocow) | |
ea14b57f | 1512 | btrfs_end_write_no_snapshotting(root); |
6f9994db LB |
1513 | if (nocow) |
1514 | btrfs_dec_nocow_writers(fs_info, | |
1515 | disk_bytenr); | |
1516 | ret = PTR_ERR(em); | |
1517 | goto error; | |
d899e052 | 1518 | } |
6f9994db LB |
1519 | free_extent_map(em); |
1520 | } | |
1521 | ||
1522 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
d899e052 YZ |
1523 | type = BTRFS_ORDERED_PREALLOC; |
1524 | } else { | |
1525 | type = BTRFS_ORDERED_NOCOW; | |
1526 | } | |
80ff3856 YZ |
1527 | |
1528 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1529 | num_bytes, num_bytes, type); |
f78c436c | 1530 | if (nocow) |
0b246afa | 1531 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1532 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1533 | |
efa56464 | 1534 | if (root->root_key.objectid == |
4dbd80fb QW |
1535 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1536 | /* | |
1537 | * Error handled later, as we must prevent | |
1538 | * extent_clear_unlock_delalloc() in error handler | |
1539 | * from freeing metadata of created ordered extent. | |
1540 | */ | |
efa56464 YZ |
1541 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
1542 | num_bytes); | |
efa56464 | 1543 | |
c2790a2e | 1544 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1545 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1546 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1547 | EXTENT_DELALLOC | |
1548 | EXTENT_CLEAR_DATA_RESV, | |
1549 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1550 | ||
e9894fd3 | 1551 | if (!nolock && nocow) |
ea14b57f | 1552 | btrfs_end_write_no_snapshotting(root); |
80ff3856 | 1553 | cur_offset = extent_end; |
4dbd80fb QW |
1554 | |
1555 | /* | |
1556 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1557 | * handler, as metadata for created ordered extent will only | |
1558 | * be freed by btrfs_finish_ordered_io(). | |
1559 | */ | |
1560 | if (ret) | |
1561 | goto error; | |
80ff3856 YZ |
1562 | if (cur_offset > end) |
1563 | break; | |
be20aa9d | 1564 | } |
b3b4aa74 | 1565 | btrfs_release_path(path); |
80ff3856 | 1566 | |
17ca04af | 1567 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1568 | cow_start = cur_offset; |
17ca04af JB |
1569 | cur_offset = end; |
1570 | } | |
1571 | ||
80ff3856 | 1572 | if (cow_start != (u64)-1) { |
dda3245e WX |
1573 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1574 | page_started, nr_written, 1, NULL); | |
d788a349 | 1575 | if (ret) |
79787eaa | 1576 | goto error; |
80ff3856 YZ |
1577 | } |
1578 | ||
79787eaa | 1579 | error: |
17ca04af | 1580 | if (ret && cur_offset < end) |
ba8b04c1 | 1581 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1582 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1583 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1584 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1585 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1586 | PAGE_SET_WRITEBACK | |
1587 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1588 | btrfs_free_path(path); |
79787eaa | 1589 | return ret; |
be20aa9d CM |
1590 | } |
1591 | ||
47059d93 WS |
1592 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1593 | { | |
1594 | ||
1595 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1596 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1597 | return 0; | |
1598 | ||
1599 | /* | |
1600 | * @defrag_bytes is a hint value, no spinlock held here, | |
1601 | * if is not zero, it means the file is defragging. | |
1602 | * Force cow if given extent needs to be defragged. | |
1603 | */ | |
1604 | if (BTRFS_I(inode)->defrag_bytes && | |
1605 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1606 | EXTENT_DEFRAG, 0, NULL)) | |
1607 | return 1; | |
1608 | ||
1609 | return 0; | |
1610 | } | |
1611 | ||
d352ac68 CM |
1612 | /* |
1613 | * extent_io.c call back to do delayed allocation processing | |
1614 | */ | |
c6100a4b | 1615 | static int run_delalloc_range(void *private_data, struct page *locked_page, |
771ed689 CM |
1616 | u64 start, u64 end, int *page_started, |
1617 | unsigned long *nr_written) | |
be20aa9d | 1618 | { |
c6100a4b | 1619 | struct inode *inode = private_data; |
be20aa9d | 1620 | int ret; |
47059d93 | 1621 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1622 | |
47059d93 | 1623 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1624 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1625 | page_started, 1, nr_written); |
47059d93 | 1626 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1627 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1628 | page_started, 0, nr_written); |
c2fcdcdf | 1629 | } else if (!inode_need_compress(inode, start, end)) { |
dda3245e WX |
1630 | ret = cow_file_range(inode, locked_page, start, end, end, |
1631 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1632 | } else { |
1633 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1634 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1635 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1636 | page_started, nr_written); |
7ddf5a42 | 1637 | } |
52427260 QW |
1638 | if (ret) |
1639 | btrfs_cleanup_ordered_extents(inode, start, end - start + 1); | |
b888db2b CM |
1640 | return ret; |
1641 | } | |
1642 | ||
c6100a4b | 1643 | static void btrfs_split_extent_hook(void *private_data, |
1bf85046 | 1644 | struct extent_state *orig, u64 split) |
9ed74f2d | 1645 | { |
c6100a4b | 1646 | struct inode *inode = private_data; |
dcab6a3b JB |
1647 | u64 size; |
1648 | ||
0ca1f7ce | 1649 | /* not delalloc, ignore it */ |
9ed74f2d | 1650 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1651 | return; |
9ed74f2d | 1652 | |
dcab6a3b JB |
1653 | size = orig->end - orig->start + 1; |
1654 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1655 | u32 num_extents; |
dcab6a3b JB |
1656 | u64 new_size; |
1657 | ||
1658 | /* | |
ba117213 JB |
1659 | * See the explanation in btrfs_merge_extent_hook, the same |
1660 | * applies here, just in reverse. | |
dcab6a3b JB |
1661 | */ |
1662 | new_size = orig->end - split + 1; | |
823bb20a | 1663 | num_extents = count_max_extents(new_size); |
ba117213 | 1664 | new_size = split - orig->start; |
823bb20a DS |
1665 | num_extents += count_max_extents(new_size); |
1666 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1667 | return; |
1668 | } | |
1669 | ||
9e0baf60 JB |
1670 | spin_lock(&BTRFS_I(inode)->lock); |
1671 | BTRFS_I(inode)->outstanding_extents++; | |
1672 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1673 | } |
1674 | ||
1675 | /* | |
1676 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1677 | * extents so we can keep track of new extents that are just merged onto old | |
1678 | * extents, such as when we are doing sequential writes, so we can properly | |
1679 | * account for the metadata space we'll need. | |
1680 | */ | |
c6100a4b | 1681 | static void btrfs_merge_extent_hook(void *private_data, |
1bf85046 JM |
1682 | struct extent_state *new, |
1683 | struct extent_state *other) | |
9ed74f2d | 1684 | { |
c6100a4b | 1685 | struct inode *inode = private_data; |
dcab6a3b | 1686 | u64 new_size, old_size; |
823bb20a | 1687 | u32 num_extents; |
dcab6a3b | 1688 | |
9ed74f2d JB |
1689 | /* not delalloc, ignore it */ |
1690 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1691 | return; |
9ed74f2d | 1692 | |
8461a3de JB |
1693 | if (new->start > other->start) |
1694 | new_size = new->end - other->start + 1; | |
1695 | else | |
1696 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1697 | |
1698 | /* we're not bigger than the max, unreserve the space and go */ | |
1699 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1700 | spin_lock(&BTRFS_I(inode)->lock); | |
1701 | BTRFS_I(inode)->outstanding_extents--; | |
1702 | spin_unlock(&BTRFS_I(inode)->lock); | |
1703 | return; | |
1704 | } | |
1705 | ||
1706 | /* | |
ba117213 JB |
1707 | * We have to add up either side to figure out how many extents were |
1708 | * accounted for before we merged into one big extent. If the number of | |
1709 | * extents we accounted for is <= the amount we need for the new range | |
1710 | * then we can return, otherwise drop. Think of it like this | |
1711 | * | |
1712 | * [ 4k][MAX_SIZE] | |
1713 | * | |
1714 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1715 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1716 | * we have 1 so they are == and we can return. But in this case | |
1717 | * | |
1718 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1719 | * | |
1720 | * Each range on their own accounts for 2 extents, but merged together | |
1721 | * they are only 3 extents worth of accounting, so we need to drop in | |
1722 | * this case. | |
dcab6a3b | 1723 | */ |
ba117213 | 1724 | old_size = other->end - other->start + 1; |
823bb20a | 1725 | num_extents = count_max_extents(old_size); |
ba117213 | 1726 | old_size = new->end - new->start + 1; |
823bb20a DS |
1727 | num_extents += count_max_extents(old_size); |
1728 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1729 | return; |
1730 | ||
9e0baf60 JB |
1731 | spin_lock(&BTRFS_I(inode)->lock); |
1732 | BTRFS_I(inode)->outstanding_extents--; | |
1733 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1734 | } |
1735 | ||
eb73c1b7 MX |
1736 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1737 | struct inode *inode) | |
1738 | { | |
0b246afa JM |
1739 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1740 | ||
eb73c1b7 MX |
1741 | spin_lock(&root->delalloc_lock); |
1742 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1743 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1744 | &root->delalloc_inodes); | |
1745 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1746 | &BTRFS_I(inode)->runtime_flags); | |
1747 | root->nr_delalloc_inodes++; | |
1748 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1749 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1750 | BUG_ON(!list_empty(&root->delalloc_root)); |
1751 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1752 | &fs_info->delalloc_roots); |
1753 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1754 | } |
1755 | } | |
1756 | spin_unlock(&root->delalloc_lock); | |
1757 | } | |
1758 | ||
0d670384 NB |
1759 | |
1760 | void __btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1761 | struct btrfs_inode *inode) | |
eb73c1b7 | 1762 | { |
9e3e97f4 | 1763 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0b246afa | 1764 | |
9e3e97f4 NB |
1765 | if (!list_empty(&inode->delalloc_inodes)) { |
1766 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1767 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1768 | &inode->runtime_flags); |
eb73c1b7 MX |
1769 | root->nr_delalloc_inodes--; |
1770 | if (!root->nr_delalloc_inodes) { | |
0b246afa | 1771 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1772 | BUG_ON(list_empty(&root->delalloc_root)); |
1773 | list_del_init(&root->delalloc_root); | |
0b246afa | 1774 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1775 | } |
1776 | } | |
0d670384 NB |
1777 | } |
1778 | ||
1779 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1780 | struct btrfs_inode *inode) | |
1781 | { | |
1782 | spin_lock(&root->delalloc_lock); | |
1783 | __btrfs_del_delalloc_inode(root, inode); | |
eb73c1b7 MX |
1784 | spin_unlock(&root->delalloc_lock); |
1785 | } | |
1786 | ||
d352ac68 CM |
1787 | /* |
1788 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1789 | * bytes in this file, and to maintain the list of inodes that | |
1790 | * have pending delalloc work to be done. | |
1791 | */ | |
c6100a4b | 1792 | static void btrfs_set_bit_hook(void *private_data, |
9ee49a04 | 1793 | struct extent_state *state, unsigned *bits) |
291d673e | 1794 | { |
c6100a4b | 1795 | struct inode *inode = private_data; |
9ed74f2d | 1796 | |
0b246afa JM |
1797 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1798 | ||
47059d93 WS |
1799 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1800 | WARN_ON(1); | |
75eff68e CM |
1801 | /* |
1802 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1803 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1804 | * bit, which is only set or cleared with irqs on |
1805 | */ | |
0ca1f7ce | 1806 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1807 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1808 | u64 len = state->end + 1 - state->start; |
70ddc553 | 1809 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1810 | |
9e0baf60 | 1811 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1812 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1813 | } else { |
1814 | spin_lock(&BTRFS_I(inode)->lock); | |
1815 | BTRFS_I(inode)->outstanding_extents++; | |
1816 | spin_unlock(&BTRFS_I(inode)->lock); | |
1817 | } | |
287a0ab9 | 1818 | |
6a3891c5 | 1819 | /* For sanity tests */ |
0b246afa | 1820 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1821 | return; |
1822 | ||
104b4e51 NB |
1823 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
1824 | fs_info->delalloc_batch); | |
df0af1a5 | 1825 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1826 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1827 | if (*bits & EXTENT_DEFRAG) |
1828 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1829 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1830 | &BTRFS_I(inode)->runtime_flags)) |
1831 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1832 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1833 | } |
a7e3b975 FM |
1834 | |
1835 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
1836 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1837 | spin_lock(&BTRFS_I(inode)->lock); | |
1838 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
1839 | state->start; | |
1840 | spin_unlock(&BTRFS_I(inode)->lock); | |
1841 | } | |
291d673e CM |
1842 | } |
1843 | ||
d352ac68 CM |
1844 | /* |
1845 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1846 | */ | |
c6100a4b | 1847 | static void btrfs_clear_bit_hook(void *private_data, |
41074888 | 1848 | struct extent_state *state, |
9ee49a04 | 1849 | unsigned *bits) |
291d673e | 1850 | { |
c6100a4b | 1851 | struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data); |
6fc0ef68 | 1852 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
47059d93 | 1853 | u64 len = state->end + 1 - state->start; |
823bb20a | 1854 | u32 num_extents = count_max_extents(len); |
47059d93 | 1855 | |
4a4b964f FM |
1856 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
1857 | spin_lock(&inode->lock); | |
6fc0ef68 | 1858 | inode->defrag_bytes -= len; |
4a4b964f FM |
1859 | spin_unlock(&inode->lock); |
1860 | } | |
47059d93 | 1861 | |
75eff68e CM |
1862 | /* |
1863 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1864 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1865 | * bit, which is only set or cleared with irqs on |
1866 | */ | |
0ca1f7ce | 1867 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1868 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1869 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1870 | |
9e0baf60 | 1871 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1872 | *bits &= ~EXTENT_FIRST_DELALLOC; |
a315e68f | 1873 | } else if (!(*bits & EXTENT_CLEAR_META_RESV)) { |
6fc0ef68 NB |
1874 | spin_lock(&inode->lock); |
1875 | inode->outstanding_extents -= num_extents; | |
1876 | spin_unlock(&inode->lock); | |
9e0baf60 | 1877 | } |
0ca1f7ce | 1878 | |
b6d08f06 JB |
1879 | /* |
1880 | * We don't reserve metadata space for space cache inodes so we | |
1881 | * don't need to call dellalloc_release_metadata if there is an | |
1882 | * error. | |
1883 | */ | |
a315e68f | 1884 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 1885 | root != fs_info->tree_root) |
0ca1f7ce YZ |
1886 | btrfs_delalloc_release_metadata(inode, len); |
1887 | ||
6a3891c5 | 1888 | /* For sanity tests. */ |
0b246afa | 1889 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1890 | return; |
1891 | ||
a315e68f FM |
1892 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
1893 | do_list && !(state->state & EXTENT_NORESERVE) && | |
1894 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
6fc0ef68 NB |
1895 | btrfs_free_reserved_data_space_noquota( |
1896 | &inode->vfs_inode, | |
51773bec | 1897 | state->start, len); |
9ed74f2d | 1898 | |
104b4e51 NB |
1899 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
1900 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1901 | spin_lock(&inode->lock); |
1902 | inode->delalloc_bytes -= len; | |
1903 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1904 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1905 | &inode->runtime_flags)) |
eb73c1b7 | 1906 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1907 | spin_unlock(&inode->lock); |
291d673e | 1908 | } |
a7e3b975 FM |
1909 | |
1910 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
1911 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1912 | spin_lock(&inode->lock); | |
1913 | ASSERT(inode->new_delalloc_bytes >= len); | |
1914 | inode->new_delalloc_bytes -= len; | |
1915 | spin_unlock(&inode->lock); | |
1916 | } | |
291d673e CM |
1917 | } |
1918 | ||
d352ac68 CM |
1919 | /* |
1920 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1921 | * we don't create bios that span stripes or chunks | |
6f034ece LB |
1922 | * |
1923 | * return 1 if page cannot be merged to bio | |
1924 | * return 0 if page can be merged to bio | |
1925 | * return error otherwise | |
d352ac68 | 1926 | */ |
81a75f67 | 1927 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1928 | size_t size, struct bio *bio, |
1929 | unsigned long bio_flags) | |
239b14b3 | 1930 | { |
0b246afa JM |
1931 | struct inode *inode = page->mapping->host; |
1932 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1933 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1934 | u64 length = 0; |
1935 | u64 map_length; | |
239b14b3 CM |
1936 | int ret; |
1937 | ||
771ed689 CM |
1938 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1939 | return 0; | |
1940 | ||
4f024f37 | 1941 | length = bio->bi_iter.bi_size; |
239b14b3 | 1942 | map_length = length; |
0b246afa JM |
1943 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1944 | NULL, 0); | |
6f034ece LB |
1945 | if (ret < 0) |
1946 | return ret; | |
d397712b | 1947 | if (map_length < length + size) |
239b14b3 | 1948 | return 1; |
3444a972 | 1949 | return 0; |
239b14b3 CM |
1950 | } |
1951 | ||
d352ac68 CM |
1952 | /* |
1953 | * in order to insert checksums into the metadata in large chunks, | |
1954 | * we wait until bio submission time. All the pages in the bio are | |
1955 | * checksummed and sums are attached onto the ordered extent record. | |
1956 | * | |
1957 | * At IO completion time the cums attached on the ordered extent record | |
1958 | * are inserted into the btree | |
1959 | */ | |
8c27cb35 | 1960 | static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio, |
81a75f67 | 1961 | int mirror_num, unsigned long bio_flags, |
eaf25d93 | 1962 | u64 bio_offset) |
065631f6 | 1963 | { |
c6100a4b | 1964 | struct inode *inode = private_data; |
4e4cbee9 | 1965 | blk_status_t ret = 0; |
e015640f | 1966 | |
2ff7e61e | 1967 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1968 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1969 | return 0; |
1970 | } | |
e015640f | 1971 | |
4a69a410 CM |
1972 | /* |
1973 | * in order to insert checksums into the metadata in large chunks, | |
1974 | * we wait until bio submission time. All the pages in the bio are | |
1975 | * checksummed and sums are attached onto the ordered extent record. | |
1976 | * | |
1977 | * At IO completion time the cums attached on the ordered extent record | |
1978 | * are inserted into the btree | |
1979 | */ | |
8c27cb35 | 1980 | static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio, |
eaf25d93 CM |
1981 | int mirror_num, unsigned long bio_flags, |
1982 | u64 bio_offset) | |
4a69a410 | 1983 | { |
c6100a4b | 1984 | struct inode *inode = private_data; |
2ff7e61e | 1985 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4e4cbee9 | 1986 | blk_status_t ret; |
61891923 | 1987 | |
2ff7e61e | 1988 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 1); |
4246a0b6 | 1989 | if (ret) { |
4e4cbee9 | 1990 | bio->bi_status = ret; |
4246a0b6 CH |
1991 | bio_endio(bio); |
1992 | } | |
61891923 | 1993 | return ret; |
44b8bd7e CM |
1994 | } |
1995 | ||
d352ac68 | 1996 | /* |
cad321ad CM |
1997 | * extent_io.c submission hook. This does the right thing for csum calculation |
1998 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1999 | */ |
8c27cb35 | 2000 | static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio, |
c6100a4b JB |
2001 | int mirror_num, unsigned long bio_flags, |
2002 | u64 bio_offset) | |
44b8bd7e | 2003 | { |
c6100a4b | 2004 | struct inode *inode = private_data; |
0b246afa | 2005 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 2006 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 2007 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 2008 | blk_status_t ret = 0; |
19b9bdb0 | 2009 | int skip_sum; |
b812ce28 | 2010 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 2011 | |
6cbff00f | 2012 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 2013 | |
70ddc553 | 2014 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 2015 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 2016 | |
37226b21 | 2017 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 2018 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 2019 | if (ret) |
61891923 | 2020 | goto out; |
5fd02043 | 2021 | |
d20f7043 | 2022 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
2023 | ret = btrfs_submit_compressed_read(inode, bio, |
2024 | mirror_num, | |
2025 | bio_flags); | |
2026 | goto out; | |
c2db1073 | 2027 | } else if (!skip_sum) { |
2ff7e61e | 2028 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 2029 | if (ret) |
61891923 | 2030 | goto out; |
c2db1073 | 2031 | } |
4d1b5fb4 | 2032 | goto mapit; |
b812ce28 | 2033 | } else if (async && !skip_sum) { |
17d217fe YZ |
2034 | /* csum items have already been cloned */ |
2035 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2036 | goto mapit; | |
19b9bdb0 | 2037 | /* we're doing a write, do the async checksumming */ |
c6100a4b JB |
2038 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
2039 | bio_offset, inode, | |
0b246afa JM |
2040 | __btrfs_submit_bio_start, |
2041 | __btrfs_submit_bio_done); | |
61891923 | 2042 | goto out; |
b812ce28 | 2043 | } else if (!skip_sum) { |
2ff7e61e | 2044 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2045 | if (ret) |
2046 | goto out; | |
19b9bdb0 CM |
2047 | } |
2048 | ||
0b86a832 | 2049 | mapit: |
2ff7e61e | 2050 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2051 | |
2052 | out: | |
4e4cbee9 CH |
2053 | if (ret) { |
2054 | bio->bi_status = ret; | |
4246a0b6 CH |
2055 | bio_endio(bio); |
2056 | } | |
61891923 | 2057 | return ret; |
065631f6 | 2058 | } |
6885f308 | 2059 | |
d352ac68 CM |
2060 | /* |
2061 | * given a list of ordered sums record them in the inode. This happens | |
2062 | * at IO completion time based on sums calculated at bio submission time. | |
2063 | */ | |
ba1da2f4 | 2064 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2065 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2066 | { |
e6dcd2dc CM |
2067 | struct btrfs_ordered_sum *sum; |
2068 | ||
c6e30871 | 2069 | list_for_each_entry(sum, list, list) { |
39847c4d | 2070 | trans->adding_csums = 1; |
d20f7043 CM |
2071 | btrfs_csum_file_blocks(trans, |
2072 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 2073 | trans->adding_csums = 0; |
e6dcd2dc CM |
2074 | } |
2075 | return 0; | |
2076 | } | |
2077 | ||
2ac55d41 | 2078 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
ba8b04c1 | 2079 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2080 | { |
09cbfeaf | 2081 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 2082 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
7cd8c752 | 2083 | cached_state); |
ea8c2819 CM |
2084 | } |
2085 | ||
d352ac68 | 2086 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2087 | struct btrfs_writepage_fixup { |
2088 | struct page *page; | |
2089 | struct btrfs_work work; | |
2090 | }; | |
2091 | ||
b2950863 | 2092 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2093 | { |
2094 | struct btrfs_writepage_fixup *fixup; | |
2095 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2096 | struct extent_state *cached_state = NULL; |
364ecf36 | 2097 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2098 | struct page *page; |
2099 | struct inode *inode; | |
2100 | u64 page_start; | |
2101 | u64 page_end; | |
87826df0 | 2102 | int ret; |
247e743c CM |
2103 | |
2104 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2105 | page = fixup->page; | |
4a096752 | 2106 | again: |
247e743c CM |
2107 | lock_page(page); |
2108 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2109 | ClearPageChecked(page); | |
2110 | goto out_page; | |
2111 | } | |
2112 | ||
2113 | inode = page->mapping->host; | |
2114 | page_start = page_offset(page); | |
09cbfeaf | 2115 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2116 | |
ff13db41 | 2117 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2118 | &cached_state); |
4a096752 CM |
2119 | |
2120 | /* already ordered? We're done */ | |
8b62b72b | 2121 | if (PagePrivate2(page)) |
247e743c | 2122 | goto out; |
4a096752 | 2123 | |
a776c6fa | 2124 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2125 | PAGE_SIZE); |
4a096752 | 2126 | if (ordered) { |
2ac55d41 JB |
2127 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2128 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2129 | unlock_page(page); |
2130 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2131 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2132 | goto again; |
2133 | } | |
247e743c | 2134 | |
364ecf36 | 2135 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2136 | PAGE_SIZE); |
87826df0 JM |
2137 | if (ret) { |
2138 | mapping_set_error(page->mapping, ret); | |
2139 | end_extent_writepage(page, ret, page_start, page_end); | |
2140 | ClearPageChecked(page); | |
2141 | goto out; | |
2142 | } | |
2143 | ||
4063cd56 NB |
2144 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
2145 | &cached_state, 0); | |
2146 | if (ret) { | |
2147 | mapping_set_error(page->mapping, ret); | |
2148 | end_extent_writepage(page, ret, page_start, page_end); | |
2149 | ClearPageChecked(page); | |
2150 | goto out; | |
2151 | } | |
2152 | ||
247e743c | 2153 | ClearPageChecked(page); |
87826df0 | 2154 | set_page_dirty(page); |
247e743c | 2155 | out: |
2ac55d41 JB |
2156 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2157 | &cached_state, GFP_NOFS); | |
247e743c CM |
2158 | out_page: |
2159 | unlock_page(page); | |
09cbfeaf | 2160 | put_page(page); |
b897abec | 2161 | kfree(fixup); |
364ecf36 | 2162 | extent_changeset_free(data_reserved); |
247e743c CM |
2163 | } |
2164 | ||
2165 | /* | |
2166 | * There are a few paths in the higher layers of the kernel that directly | |
2167 | * set the page dirty bit without asking the filesystem if it is a | |
2168 | * good idea. This causes problems because we want to make sure COW | |
2169 | * properly happens and the data=ordered rules are followed. | |
2170 | * | |
c8b97818 | 2171 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2172 | * hasn't been properly setup for IO. We kick off an async process |
2173 | * to fix it up. The async helper will wait for ordered extents, set | |
2174 | * the delalloc bit and make it safe to write the page. | |
2175 | */ | |
b2950863 | 2176 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2177 | { |
2178 | struct inode *inode = page->mapping->host; | |
0b246afa | 2179 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2180 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2181 | |
8b62b72b CM |
2182 | /* this page is properly in the ordered list */ |
2183 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2184 | return 0; |
2185 | ||
2186 | if (PageChecked(page)) | |
2187 | return -EAGAIN; | |
2188 | ||
2189 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2190 | if (!fixup) | |
2191 | return -EAGAIN; | |
f421950f | 2192 | |
247e743c | 2193 | SetPageChecked(page); |
09cbfeaf | 2194 | get_page(page); |
9e0af237 LB |
2195 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2196 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2197 | fixup->page = page; |
0b246afa | 2198 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2199 | return -EBUSY; |
247e743c CM |
2200 | } |
2201 | ||
d899e052 YZ |
2202 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2203 | struct inode *inode, u64 file_pos, | |
2204 | u64 disk_bytenr, u64 disk_num_bytes, | |
2205 | u64 num_bytes, u64 ram_bytes, | |
2206 | u8 compression, u8 encryption, | |
2207 | u16 other_encoding, int extent_type) | |
2208 | { | |
2209 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2210 | struct btrfs_file_extent_item *fi; | |
2211 | struct btrfs_path *path; | |
2212 | struct extent_buffer *leaf; | |
2213 | struct btrfs_key ins; | |
a12b877b | 2214 | u64 qg_released; |
1acae57b | 2215 | int extent_inserted = 0; |
d899e052 YZ |
2216 | int ret; |
2217 | ||
2218 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2219 | if (!path) |
2220 | return -ENOMEM; | |
d899e052 | 2221 | |
a1ed835e CM |
2222 | /* |
2223 | * we may be replacing one extent in the tree with another. | |
2224 | * The new extent is pinned in the extent map, and we don't want | |
2225 | * to drop it from the cache until it is completely in the btree. | |
2226 | * | |
2227 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2228 | * the caller is expected to unpin it and allow it to be merged | |
2229 | * with the others. | |
2230 | */ | |
1acae57b FDBM |
2231 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2232 | file_pos + num_bytes, NULL, 0, | |
2233 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2234 | if (ret) |
2235 | goto out; | |
d899e052 | 2236 | |
1acae57b | 2237 | if (!extent_inserted) { |
4a0cc7ca | 2238 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2239 | ins.offset = file_pos; |
2240 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2241 | ||
2242 | path->leave_spinning = 1; | |
2243 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2244 | sizeof(*fi)); | |
2245 | if (ret) | |
2246 | goto out; | |
2247 | } | |
d899e052 YZ |
2248 | leaf = path->nodes[0]; |
2249 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2250 | struct btrfs_file_extent_item); | |
2251 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2252 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2253 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2254 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2255 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2256 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2257 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2258 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2259 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2260 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2261 | |
d899e052 | 2262 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2263 | btrfs_release_path(path); |
d899e052 YZ |
2264 | |
2265 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2266 | |
2267 | ins.objectid = disk_bytenr; | |
2268 | ins.offset = disk_num_bytes; | |
2269 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2270 | |
297d750b | 2271 | /* |
5846a3c2 QW |
2272 | * Release the reserved range from inode dirty range map, as it is |
2273 | * already moved into delayed_ref_head | |
297d750b | 2274 | */ |
a12b877b QW |
2275 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2276 | if (ret < 0) | |
2277 | goto out; | |
2278 | qg_released = ret; | |
2279 | ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid, | |
2280 | btrfs_ino(BTRFS_I(inode)), file_pos, qg_released, &ins); | |
79787eaa | 2281 | out: |
d899e052 | 2282 | btrfs_free_path(path); |
b9473439 | 2283 | |
79787eaa | 2284 | return ret; |
d899e052 YZ |
2285 | } |
2286 | ||
38c227d8 LB |
2287 | /* snapshot-aware defrag */ |
2288 | struct sa_defrag_extent_backref { | |
2289 | struct rb_node node; | |
2290 | struct old_sa_defrag_extent *old; | |
2291 | u64 root_id; | |
2292 | u64 inum; | |
2293 | u64 file_pos; | |
2294 | u64 extent_offset; | |
2295 | u64 num_bytes; | |
2296 | u64 generation; | |
2297 | }; | |
2298 | ||
2299 | struct old_sa_defrag_extent { | |
2300 | struct list_head list; | |
2301 | struct new_sa_defrag_extent *new; | |
2302 | ||
2303 | u64 extent_offset; | |
2304 | u64 bytenr; | |
2305 | u64 offset; | |
2306 | u64 len; | |
2307 | int count; | |
2308 | }; | |
2309 | ||
2310 | struct new_sa_defrag_extent { | |
2311 | struct rb_root root; | |
2312 | struct list_head head; | |
2313 | struct btrfs_path *path; | |
2314 | struct inode *inode; | |
2315 | u64 file_pos; | |
2316 | u64 len; | |
2317 | u64 bytenr; | |
2318 | u64 disk_len; | |
2319 | u8 compress_type; | |
2320 | }; | |
2321 | ||
2322 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2323 | struct sa_defrag_extent_backref *b2) | |
2324 | { | |
2325 | if (b1->root_id < b2->root_id) | |
2326 | return -1; | |
2327 | else if (b1->root_id > b2->root_id) | |
2328 | return 1; | |
2329 | ||
2330 | if (b1->inum < b2->inum) | |
2331 | return -1; | |
2332 | else if (b1->inum > b2->inum) | |
2333 | return 1; | |
2334 | ||
2335 | if (b1->file_pos < b2->file_pos) | |
2336 | return -1; | |
2337 | else if (b1->file_pos > b2->file_pos) | |
2338 | return 1; | |
2339 | ||
2340 | /* | |
2341 | * [------------------------------] ===> (a range of space) | |
2342 | * |<--->| |<---->| =============> (fs/file tree A) | |
2343 | * |<---------------------------->| ===> (fs/file tree B) | |
2344 | * | |
2345 | * A range of space can refer to two file extents in one tree while | |
2346 | * refer to only one file extent in another tree. | |
2347 | * | |
2348 | * So we may process a disk offset more than one time(two extents in A) | |
2349 | * and locate at the same extent(one extent in B), then insert two same | |
2350 | * backrefs(both refer to the extent in B). | |
2351 | */ | |
2352 | return 0; | |
2353 | } | |
2354 | ||
2355 | static void backref_insert(struct rb_root *root, | |
2356 | struct sa_defrag_extent_backref *backref) | |
2357 | { | |
2358 | struct rb_node **p = &root->rb_node; | |
2359 | struct rb_node *parent = NULL; | |
2360 | struct sa_defrag_extent_backref *entry; | |
2361 | int ret; | |
2362 | ||
2363 | while (*p) { | |
2364 | parent = *p; | |
2365 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2366 | ||
2367 | ret = backref_comp(backref, entry); | |
2368 | if (ret < 0) | |
2369 | p = &(*p)->rb_left; | |
2370 | else | |
2371 | p = &(*p)->rb_right; | |
2372 | } | |
2373 | ||
2374 | rb_link_node(&backref->node, parent, p); | |
2375 | rb_insert_color(&backref->node, root); | |
2376 | } | |
2377 | ||
2378 | /* | |
2379 | * Note the backref might has changed, and in this case we just return 0. | |
2380 | */ | |
2381 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2382 | void *ctx) | |
2383 | { | |
2384 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2385 | struct old_sa_defrag_extent *old = ctx; |
2386 | struct new_sa_defrag_extent *new = old->new; | |
2387 | struct btrfs_path *path = new->path; | |
2388 | struct btrfs_key key; | |
2389 | struct btrfs_root *root; | |
2390 | struct sa_defrag_extent_backref *backref; | |
2391 | struct extent_buffer *leaf; | |
2392 | struct inode *inode = new->inode; | |
0b246afa | 2393 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2394 | int slot; |
2395 | int ret; | |
2396 | u64 extent_offset; | |
2397 | u64 num_bytes; | |
2398 | ||
2399 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2400 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2401 | return 0; |
2402 | ||
2403 | key.objectid = root_id; | |
2404 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2405 | key.offset = (u64)-1; | |
2406 | ||
38c227d8 LB |
2407 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2408 | if (IS_ERR(root)) { | |
2409 | if (PTR_ERR(root) == -ENOENT) | |
2410 | return 0; | |
2411 | WARN_ON(1); | |
ab8d0fc4 | 2412 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2413 | inum, offset, root_id); |
2414 | return PTR_ERR(root); | |
2415 | } | |
2416 | ||
2417 | key.objectid = inum; | |
2418 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2419 | if (offset > (u64)-1 << 32) | |
2420 | key.offset = 0; | |
2421 | else | |
2422 | key.offset = offset; | |
2423 | ||
2424 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2425 | if (WARN_ON(ret < 0)) |
38c227d8 | 2426 | return ret; |
50f1319c | 2427 | ret = 0; |
38c227d8 LB |
2428 | |
2429 | while (1) { | |
2430 | cond_resched(); | |
2431 | ||
2432 | leaf = path->nodes[0]; | |
2433 | slot = path->slots[0]; | |
2434 | ||
2435 | if (slot >= btrfs_header_nritems(leaf)) { | |
2436 | ret = btrfs_next_leaf(root, path); | |
2437 | if (ret < 0) { | |
2438 | goto out; | |
2439 | } else if (ret > 0) { | |
2440 | ret = 0; | |
2441 | goto out; | |
2442 | } | |
2443 | continue; | |
2444 | } | |
2445 | ||
2446 | path->slots[0]++; | |
2447 | ||
2448 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2449 | ||
2450 | if (key.objectid > inum) | |
2451 | goto out; | |
2452 | ||
2453 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2454 | continue; | |
2455 | ||
2456 | extent = btrfs_item_ptr(leaf, slot, | |
2457 | struct btrfs_file_extent_item); | |
2458 | ||
2459 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2460 | continue; | |
2461 | ||
e68afa49 LB |
2462 | /* |
2463 | * 'offset' refers to the exact key.offset, | |
2464 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2465 | * (key.offset - extent_offset). | |
2466 | */ | |
2467 | if (key.offset != offset) | |
38c227d8 LB |
2468 | continue; |
2469 | ||
e68afa49 | 2470 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2471 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2472 | |
38c227d8 LB |
2473 | if (extent_offset >= old->extent_offset + old->offset + |
2474 | old->len || extent_offset + num_bytes <= | |
2475 | old->extent_offset + old->offset) | |
2476 | continue; | |
38c227d8 LB |
2477 | break; |
2478 | } | |
2479 | ||
2480 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2481 | if (!backref) { | |
2482 | ret = -ENOENT; | |
2483 | goto out; | |
2484 | } | |
2485 | ||
2486 | backref->root_id = root_id; | |
2487 | backref->inum = inum; | |
e68afa49 | 2488 | backref->file_pos = offset; |
38c227d8 LB |
2489 | backref->num_bytes = num_bytes; |
2490 | backref->extent_offset = extent_offset; | |
2491 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2492 | backref->old = old; | |
2493 | backref_insert(&new->root, backref); | |
2494 | old->count++; | |
2495 | out: | |
2496 | btrfs_release_path(path); | |
2497 | WARN_ON(ret); | |
2498 | return ret; | |
2499 | } | |
2500 | ||
2501 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2502 | struct new_sa_defrag_extent *new) | |
2503 | { | |
0b246afa | 2504 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2505 | struct old_sa_defrag_extent *old, *tmp; |
2506 | int ret; | |
2507 | ||
2508 | new->path = path; | |
2509 | ||
2510 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2511 | ret = iterate_inodes_from_logical(old->bytenr + |
2512 | old->extent_offset, fs_info, | |
38c227d8 LB |
2513 | path, record_one_backref, |
2514 | old); | |
4724b106 JB |
2515 | if (ret < 0 && ret != -ENOENT) |
2516 | return false; | |
38c227d8 LB |
2517 | |
2518 | /* no backref to be processed for this extent */ | |
2519 | if (!old->count) { | |
2520 | list_del(&old->list); | |
2521 | kfree(old); | |
2522 | } | |
2523 | } | |
2524 | ||
2525 | if (list_empty(&new->head)) | |
2526 | return false; | |
2527 | ||
2528 | return true; | |
2529 | } | |
2530 | ||
2531 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2532 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2533 | struct new_sa_defrag_extent *new) |
38c227d8 | 2534 | { |
116e0024 | 2535 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2536 | return 0; |
2537 | ||
2538 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2539 | return 0; | |
2540 | ||
116e0024 LB |
2541 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2542 | return 0; | |
2543 | ||
2544 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2545 | btrfs_file_extent_other_encoding(leaf, fi)) |
2546 | return 0; | |
2547 | ||
2548 | return 1; | |
2549 | } | |
2550 | ||
2551 | /* | |
2552 | * Note the backref might has changed, and in this case we just return 0. | |
2553 | */ | |
2554 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2555 | struct sa_defrag_extent_backref *prev, | |
2556 | struct sa_defrag_extent_backref *backref) | |
2557 | { | |
2558 | struct btrfs_file_extent_item *extent; | |
2559 | struct btrfs_file_extent_item *item; | |
2560 | struct btrfs_ordered_extent *ordered; | |
2561 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2562 | struct btrfs_root *root; |
2563 | struct btrfs_key key; | |
2564 | struct extent_buffer *leaf; | |
2565 | struct old_sa_defrag_extent *old = backref->old; | |
2566 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2567 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2568 | struct inode *inode; |
2569 | struct extent_state *cached = NULL; | |
2570 | int ret = 0; | |
2571 | u64 start; | |
2572 | u64 len; | |
2573 | u64 lock_start; | |
2574 | u64 lock_end; | |
2575 | bool merge = false; | |
2576 | int index; | |
2577 | ||
2578 | if (prev && prev->root_id == backref->root_id && | |
2579 | prev->inum == backref->inum && | |
2580 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2581 | merge = true; | |
2582 | ||
2583 | /* step 1: get root */ | |
2584 | key.objectid = backref->root_id; | |
2585 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2586 | key.offset = (u64)-1; | |
2587 | ||
38c227d8 LB |
2588 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2589 | ||
2590 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2591 | if (IS_ERR(root)) { | |
2592 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2593 | if (PTR_ERR(root) == -ENOENT) | |
2594 | return 0; | |
2595 | return PTR_ERR(root); | |
2596 | } | |
38c227d8 | 2597 | |
bcbba5e6 WS |
2598 | if (btrfs_root_readonly(root)) { |
2599 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2600 | return 0; | |
2601 | } | |
2602 | ||
38c227d8 LB |
2603 | /* step 2: get inode */ |
2604 | key.objectid = backref->inum; | |
2605 | key.type = BTRFS_INODE_ITEM_KEY; | |
2606 | key.offset = 0; | |
2607 | ||
2608 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2609 | if (IS_ERR(inode)) { | |
2610 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2611 | return 0; | |
2612 | } | |
2613 | ||
2614 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2615 | ||
2616 | /* step 3: relink backref */ | |
2617 | lock_start = backref->file_pos; | |
2618 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2619 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2620 | &cached); |
38c227d8 LB |
2621 | |
2622 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2623 | if (ordered) { | |
2624 | btrfs_put_ordered_extent(ordered); | |
2625 | goto out_unlock; | |
2626 | } | |
2627 | ||
2628 | trans = btrfs_join_transaction(root); | |
2629 | if (IS_ERR(trans)) { | |
2630 | ret = PTR_ERR(trans); | |
2631 | goto out_unlock; | |
2632 | } | |
2633 | ||
2634 | key.objectid = backref->inum; | |
2635 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2636 | key.offset = backref->file_pos; | |
2637 | ||
2638 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2639 | if (ret < 0) { | |
2640 | goto out_free_path; | |
2641 | } else if (ret > 0) { | |
2642 | ret = 0; | |
2643 | goto out_free_path; | |
2644 | } | |
2645 | ||
2646 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2647 | struct btrfs_file_extent_item); | |
2648 | ||
2649 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2650 | backref->generation) | |
2651 | goto out_free_path; | |
2652 | ||
2653 | btrfs_release_path(path); | |
2654 | ||
2655 | start = backref->file_pos; | |
2656 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2657 | start += old->extent_offset + old->offset - | |
2658 | backref->extent_offset; | |
2659 | ||
2660 | len = min(backref->extent_offset + backref->num_bytes, | |
2661 | old->extent_offset + old->offset + old->len); | |
2662 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2663 | ||
2664 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2665 | start + len, 1); | |
2666 | if (ret) | |
2667 | goto out_free_path; | |
2668 | again: | |
4a0cc7ca | 2669 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2670 | key.type = BTRFS_EXTENT_DATA_KEY; |
2671 | key.offset = start; | |
2672 | ||
a09a0a70 | 2673 | path->leave_spinning = 1; |
38c227d8 LB |
2674 | if (merge) { |
2675 | struct btrfs_file_extent_item *fi; | |
2676 | u64 extent_len; | |
2677 | struct btrfs_key found_key; | |
2678 | ||
3c9665df | 2679 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2680 | if (ret < 0) |
2681 | goto out_free_path; | |
2682 | ||
2683 | path->slots[0]--; | |
2684 | leaf = path->nodes[0]; | |
2685 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2686 | ||
2687 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2688 | struct btrfs_file_extent_item); | |
2689 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2690 | ||
116e0024 LB |
2691 | if (extent_len + found_key.offset == start && |
2692 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2693 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2694 | extent_len + len); | |
2695 | btrfs_mark_buffer_dirty(leaf); | |
2696 | inode_add_bytes(inode, len); | |
2697 | ||
2698 | ret = 1; | |
2699 | goto out_free_path; | |
2700 | } else { | |
2701 | merge = false; | |
2702 | btrfs_release_path(path); | |
2703 | goto again; | |
2704 | } | |
2705 | } | |
2706 | ||
2707 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2708 | sizeof(*extent)); | |
2709 | if (ret) { | |
66642832 | 2710 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2711 | goto out_free_path; |
2712 | } | |
2713 | ||
2714 | leaf = path->nodes[0]; | |
2715 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2716 | struct btrfs_file_extent_item); | |
2717 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2718 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2719 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2720 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2721 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2722 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2723 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2724 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2725 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2726 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2727 | ||
2728 | btrfs_mark_buffer_dirty(leaf); | |
2729 | inode_add_bytes(inode, len); | |
a09a0a70 | 2730 | btrfs_release_path(path); |
38c227d8 | 2731 | |
2ff7e61e | 2732 | ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr, |
38c227d8 LB |
2733 | new->disk_len, 0, |
2734 | backref->root_id, backref->inum, | |
b06c4bf5 | 2735 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2736 | if (ret) { |
66642832 | 2737 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2738 | goto out_free_path; |
2739 | } | |
2740 | ||
2741 | ret = 1; | |
2742 | out_free_path: | |
2743 | btrfs_release_path(path); | |
a09a0a70 | 2744 | path->leave_spinning = 0; |
3a45bb20 | 2745 | btrfs_end_transaction(trans); |
38c227d8 LB |
2746 | out_unlock: |
2747 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2748 | &cached, GFP_NOFS); | |
2749 | iput(inode); | |
2750 | return ret; | |
2751 | } | |
2752 | ||
6f519564 LB |
2753 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2754 | { | |
2755 | struct old_sa_defrag_extent *old, *tmp; | |
2756 | ||
2757 | if (!new) | |
2758 | return; | |
2759 | ||
2760 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2761 | kfree(old); |
2762 | } | |
2763 | kfree(new); | |
2764 | } | |
2765 | ||
38c227d8 LB |
2766 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2767 | { | |
0b246afa | 2768 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2769 | struct btrfs_path *path; |
38c227d8 LB |
2770 | struct sa_defrag_extent_backref *backref; |
2771 | struct sa_defrag_extent_backref *prev = NULL; | |
2772 | struct inode *inode; | |
2773 | struct btrfs_root *root; | |
2774 | struct rb_node *node; | |
2775 | int ret; | |
2776 | ||
2777 | inode = new->inode; | |
2778 | root = BTRFS_I(inode)->root; | |
2779 | ||
2780 | path = btrfs_alloc_path(); | |
2781 | if (!path) | |
2782 | return; | |
2783 | ||
2784 | if (!record_extent_backrefs(path, new)) { | |
2785 | btrfs_free_path(path); | |
2786 | goto out; | |
2787 | } | |
2788 | btrfs_release_path(path); | |
2789 | ||
2790 | while (1) { | |
2791 | node = rb_first(&new->root); | |
2792 | if (!node) | |
2793 | break; | |
2794 | rb_erase(node, &new->root); | |
2795 | ||
2796 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2797 | ||
2798 | ret = relink_extent_backref(path, prev, backref); | |
2799 | WARN_ON(ret < 0); | |
2800 | ||
2801 | kfree(prev); | |
2802 | ||
2803 | if (ret == 1) | |
2804 | prev = backref; | |
2805 | else | |
2806 | prev = NULL; | |
2807 | cond_resched(); | |
2808 | } | |
2809 | kfree(prev); | |
2810 | ||
2811 | btrfs_free_path(path); | |
38c227d8 | 2812 | out: |
6f519564 LB |
2813 | free_sa_defrag_extent(new); |
2814 | ||
0b246afa JM |
2815 | atomic_dec(&fs_info->defrag_running); |
2816 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2817 | } |
2818 | ||
2819 | static struct new_sa_defrag_extent * | |
2820 | record_old_file_extents(struct inode *inode, | |
2821 | struct btrfs_ordered_extent *ordered) | |
2822 | { | |
0b246afa | 2823 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2824 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2825 | struct btrfs_path *path; | |
2826 | struct btrfs_key key; | |
6f519564 | 2827 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2828 | struct new_sa_defrag_extent *new; |
2829 | int ret; | |
2830 | ||
2831 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2832 | if (!new) | |
2833 | return NULL; | |
2834 | ||
2835 | new->inode = inode; | |
2836 | new->file_pos = ordered->file_offset; | |
2837 | new->len = ordered->len; | |
2838 | new->bytenr = ordered->start; | |
2839 | new->disk_len = ordered->disk_len; | |
2840 | new->compress_type = ordered->compress_type; | |
2841 | new->root = RB_ROOT; | |
2842 | INIT_LIST_HEAD(&new->head); | |
2843 | ||
2844 | path = btrfs_alloc_path(); | |
2845 | if (!path) | |
2846 | goto out_kfree; | |
2847 | ||
4a0cc7ca | 2848 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2849 | key.type = BTRFS_EXTENT_DATA_KEY; |
2850 | key.offset = new->file_pos; | |
2851 | ||
2852 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2853 | if (ret < 0) | |
2854 | goto out_free_path; | |
2855 | if (ret > 0 && path->slots[0] > 0) | |
2856 | path->slots[0]--; | |
2857 | ||
2858 | /* find out all the old extents for the file range */ | |
2859 | while (1) { | |
2860 | struct btrfs_file_extent_item *extent; | |
2861 | struct extent_buffer *l; | |
2862 | int slot; | |
2863 | u64 num_bytes; | |
2864 | u64 offset; | |
2865 | u64 end; | |
2866 | u64 disk_bytenr; | |
2867 | u64 extent_offset; | |
2868 | ||
2869 | l = path->nodes[0]; | |
2870 | slot = path->slots[0]; | |
2871 | ||
2872 | if (slot >= btrfs_header_nritems(l)) { | |
2873 | ret = btrfs_next_leaf(root, path); | |
2874 | if (ret < 0) | |
6f519564 | 2875 | goto out_free_path; |
38c227d8 LB |
2876 | else if (ret > 0) |
2877 | break; | |
2878 | continue; | |
2879 | } | |
2880 | ||
2881 | btrfs_item_key_to_cpu(l, &key, slot); | |
2882 | ||
4a0cc7ca | 2883 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2884 | break; |
2885 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2886 | break; | |
2887 | if (key.offset >= new->file_pos + new->len) | |
2888 | break; | |
2889 | ||
2890 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2891 | ||
2892 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2893 | if (key.offset + num_bytes < new->file_pos) | |
2894 | goto next; | |
2895 | ||
2896 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2897 | if (!disk_bytenr) | |
2898 | goto next; | |
2899 | ||
2900 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2901 | ||
2902 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2903 | if (!old) | |
6f519564 | 2904 | goto out_free_path; |
38c227d8 LB |
2905 | |
2906 | offset = max(new->file_pos, key.offset); | |
2907 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2908 | ||
2909 | old->bytenr = disk_bytenr; | |
2910 | old->extent_offset = extent_offset; | |
2911 | old->offset = offset - key.offset; | |
2912 | old->len = end - offset; | |
2913 | old->new = new; | |
2914 | old->count = 0; | |
2915 | list_add_tail(&old->list, &new->head); | |
2916 | next: | |
2917 | path->slots[0]++; | |
2918 | cond_resched(); | |
2919 | } | |
2920 | ||
2921 | btrfs_free_path(path); | |
0b246afa | 2922 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2923 | |
2924 | return new; | |
2925 | ||
38c227d8 LB |
2926 | out_free_path: |
2927 | btrfs_free_path(path); | |
2928 | out_kfree: | |
6f519564 | 2929 | free_sa_defrag_extent(new); |
38c227d8 LB |
2930 | return NULL; |
2931 | } | |
2932 | ||
2ff7e61e | 2933 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2934 | u64 start, u64 len) |
2935 | { | |
2936 | struct btrfs_block_group_cache *cache; | |
2937 | ||
0b246afa | 2938 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2939 | ASSERT(cache); |
2940 | ||
2941 | spin_lock(&cache->lock); | |
2942 | cache->delalloc_bytes -= len; | |
2943 | spin_unlock(&cache->lock); | |
2944 | ||
2945 | btrfs_put_block_group(cache); | |
2946 | } | |
2947 | ||
d352ac68 CM |
2948 | /* as ordered data IO finishes, this gets called so we can finish |
2949 | * an ordered extent if the range of bytes in the file it covers are | |
2950 | * fully written. | |
2951 | */ | |
5fd02043 | 2952 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2953 | { |
5fd02043 | 2954 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2955 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2956 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2957 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2958 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2959 | struct extent_state *cached_state = NULL; |
38c227d8 | 2960 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2961 | int compress_type = 0; |
77cef2ec JB |
2962 | int ret = 0; |
2963 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2964 | bool nolock; |
77cef2ec | 2965 | bool truncated = false; |
a7e3b975 FM |
2966 | bool range_locked = false; |
2967 | bool clear_new_delalloc_bytes = false; | |
2968 | ||
2969 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2970 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2971 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2972 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2973 | |
70ddc553 | 2974 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2975 | |
5fd02043 JB |
2976 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2977 | ret = -EIO; | |
2978 | goto out; | |
2979 | } | |
2980 | ||
7ab7956e NB |
2981 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2982 | ordered_extent->file_offset, | |
2983 | ordered_extent->file_offset + | |
2984 | ordered_extent->len - 1); | |
f612496b | 2985 | |
77cef2ec JB |
2986 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2987 | truncated = true; | |
2988 | logical_len = ordered_extent->truncated_len; | |
2989 | /* Truncated the entire extent, don't bother adding */ | |
2990 | if (!logical_len) | |
2991 | goto out; | |
2992 | } | |
2993 | ||
c2167754 | 2994 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2995 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2996 | |
2997 | /* | |
2998 | * For mwrite(mmap + memset to write) case, we still reserve | |
2999 | * space for NOCOW range. | |
3000 | * As NOCOW won't cause a new delayed ref, just free the space | |
3001 | */ | |
bc42bda2 | 3002 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 3003 | ordered_extent->len); |
6c760c07 JB |
3004 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3005 | if (nolock) | |
3006 | trans = btrfs_join_transaction_nolock(root); | |
3007 | else | |
3008 | trans = btrfs_join_transaction(root); | |
3009 | if (IS_ERR(trans)) { | |
3010 | ret = PTR_ERR(trans); | |
3011 | trans = NULL; | |
3012 | goto out; | |
c2167754 | 3013 | } |
0b246afa | 3014 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
6c760c07 JB |
3015 | ret = btrfs_update_inode_fallback(trans, root, inode); |
3016 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 3017 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
3018 | goto out; |
3019 | } | |
e6dcd2dc | 3020 | |
a7e3b975 | 3021 | range_locked = true; |
2ac55d41 JB |
3022 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
3023 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 3024 | &cached_state); |
e6dcd2dc | 3025 | |
38c227d8 LB |
3026 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
3027 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 3028 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
3029 | if (ret) { |
3030 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 3031 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
3032 | /* the inode is shared */ |
3033 | new = record_old_file_extents(inode, ordered_extent); | |
3034 | ||
3035 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
3036 | ordered_extent->file_offset + ordered_extent->len - 1, | |
3037 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
3038 | } | |
3039 | ||
0cb59c99 | 3040 | if (nolock) |
7a7eaa40 | 3041 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 3042 | else |
7a7eaa40 | 3043 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3044 | if (IS_ERR(trans)) { |
3045 | ret = PTR_ERR(trans); | |
3046 | trans = NULL; | |
a7e3b975 | 3047 | goto out; |
79787eaa | 3048 | } |
a79b7d4b | 3049 | |
0b246afa | 3050 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
c2167754 | 3051 | |
c8b97818 | 3052 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3053 | compress_type = ordered_extent->compress_type; |
d899e052 | 3054 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3055 | BUG_ON(compress_type); |
8edc5b97 JM |
3056 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3057 | ordered_extent->len); | |
7a6d7067 | 3058 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3059 | ordered_extent->file_offset, |
3060 | ordered_extent->file_offset + | |
77cef2ec | 3061 | logical_len); |
d899e052 | 3062 | } else { |
0b246afa | 3063 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3064 | ret = insert_reserved_file_extent(trans, inode, |
3065 | ordered_extent->file_offset, | |
3066 | ordered_extent->start, | |
3067 | ordered_extent->disk_len, | |
77cef2ec | 3068 | logical_len, logical_len, |
261507a0 | 3069 | compress_type, 0, 0, |
d899e052 | 3070 | BTRFS_FILE_EXTENT_REG); |
e570fd27 | 3071 | if (!ret) |
2ff7e61e | 3072 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3073 | ordered_extent->start, |
3074 | ordered_extent->disk_len); | |
d899e052 | 3075 | } |
5dc562c5 JB |
3076 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3077 | ordered_extent->file_offset, ordered_extent->len, | |
3078 | trans->transid); | |
79787eaa | 3079 | if (ret < 0) { |
66642832 | 3080 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3081 | goto out; |
79787eaa | 3082 | } |
2ac55d41 | 3083 | |
df9f628e | 3084 | add_pending_csums(trans, inode, &ordered_extent->list); |
e6dcd2dc | 3085 | |
6c760c07 JB |
3086 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3087 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3088 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3089 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3090 | goto out; |
1ef30be1 JB |
3091 | } |
3092 | ret = 0; | |
c2167754 | 3093 | out: |
a7e3b975 FM |
3094 | if (range_locked || clear_new_delalloc_bytes) { |
3095 | unsigned int clear_bits = 0; | |
3096 | ||
3097 | if (range_locked) | |
3098 | clear_bits |= EXTENT_LOCKED; | |
3099 | if (clear_new_delalloc_bytes) | |
3100 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3101 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3102 | ordered_extent->file_offset, | |
3103 | ordered_extent->file_offset + | |
3104 | ordered_extent->len - 1, | |
3105 | clear_bits, | |
3106 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
3107 | 0, &cached_state, GFP_NOFS); | |
3108 | } | |
3109 | ||
0b246afa | 3110 | if (root != fs_info->tree_root) |
691fa059 NB |
3111 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
3112 | ordered_extent->len); | |
a698d075 | 3113 | if (trans) |
3a45bb20 | 3114 | btrfs_end_transaction(trans); |
0cb59c99 | 3115 | |
77cef2ec JB |
3116 | if (ret || truncated) { |
3117 | u64 start, end; | |
3118 | ||
3119 | if (truncated) | |
3120 | start = ordered_extent->file_offset + logical_len; | |
3121 | else | |
3122 | start = ordered_extent->file_offset; | |
3123 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
3124 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
3125 | ||
3126 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3127 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3128 | |
0bec9ef5 JB |
3129 | /* |
3130 | * If the ordered extent had an IOERR or something else went | |
3131 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3132 | * back to the allocator. We only free the extent in the |
3133 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 3134 | */ |
77cef2ec JB |
3135 | if ((ret || !logical_len) && |
3136 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 | 3137 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3138 | btrfs_free_reserved_extent(fs_info, |
3139 | ordered_extent->start, | |
e570fd27 | 3140 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3141 | } |
3142 | ||
3143 | ||
5fd02043 | 3144 | /* |
8bad3c02 LB |
3145 | * This needs to be done to make sure anybody waiting knows we are done |
3146 | * updating everything for this ordered extent. | |
5fd02043 JB |
3147 | */ |
3148 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3149 | ||
38c227d8 | 3150 | /* for snapshot-aware defrag */ |
6f519564 LB |
3151 | if (new) { |
3152 | if (ret) { | |
3153 | free_sa_defrag_extent(new); | |
0b246afa | 3154 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3155 | } else { |
3156 | relink_file_extents(new); | |
3157 | } | |
3158 | } | |
38c227d8 | 3159 | |
e6dcd2dc CM |
3160 | /* once for us */ |
3161 | btrfs_put_ordered_extent(ordered_extent); | |
3162 | /* once for the tree */ | |
3163 | btrfs_put_ordered_extent(ordered_extent); | |
3164 | ||
59b837d5 EL |
3165 | /* Try to release some metadata so we don't get an OOM but don't wait */ |
3166 | btrfs_btree_balance_dirty_nodelay(fs_info); | |
3167 | ||
5fd02043 JB |
3168 | return ret; |
3169 | } | |
3170 | ||
3171 | static void finish_ordered_fn(struct btrfs_work *work) | |
3172 | { | |
3173 | struct btrfs_ordered_extent *ordered_extent; | |
3174 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3175 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3176 | } |
3177 | ||
c3988d63 | 3178 | static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3179 | struct extent_state *state, int uptodate) |
3180 | { | |
5fd02043 | 3181 | struct inode *inode = page->mapping->host; |
0b246afa | 3182 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3183 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3184 | struct btrfs_workqueue *wq; |
3185 | btrfs_work_func_t func; | |
5fd02043 | 3186 | |
1abe9b8a | 3187 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3188 | ||
8b62b72b | 3189 | ClearPagePrivate2(page); |
5fd02043 JB |
3190 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3191 | end - start + 1, uptodate)) | |
c3988d63 | 3192 | return; |
5fd02043 | 3193 | |
70ddc553 | 3194 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3195 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3196 | func = btrfs_freespace_write_helper; |
3197 | } else { | |
0b246afa | 3198 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3199 | func = btrfs_endio_write_helper; |
3200 | } | |
5fd02043 | 3201 | |
9e0af237 LB |
3202 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3203 | NULL); | |
3204 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3205 | } |
3206 | ||
dc380aea MX |
3207 | static int __readpage_endio_check(struct inode *inode, |
3208 | struct btrfs_io_bio *io_bio, | |
3209 | int icsum, struct page *page, | |
3210 | int pgoff, u64 start, size_t len) | |
3211 | { | |
3212 | char *kaddr; | |
3213 | u32 csum_expected; | |
3214 | u32 csum = ~(u32)0; | |
dc380aea MX |
3215 | |
3216 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3217 | ||
3218 | kaddr = kmap_atomic(page); | |
3219 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3220 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3221 | if (csum != csum_expected) |
3222 | goto zeroit; | |
3223 | ||
3224 | kunmap_atomic(kaddr); | |
3225 | return 0; | |
3226 | zeroit: | |
0970a22e | 3227 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3228 | io_bio->mirror_num); |
dc380aea MX |
3229 | memset(kaddr + pgoff, 1, len); |
3230 | flush_dcache_page(page); | |
3231 | kunmap_atomic(kaddr); | |
dc380aea MX |
3232 | return -EIO; |
3233 | } | |
3234 | ||
d352ac68 CM |
3235 | /* |
3236 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3237 | * if there's a match, we allow the bio to finish. If not, the code in |
3238 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3239 | */ |
facc8a22 MX |
3240 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3241 | u64 phy_offset, struct page *page, | |
3242 | u64 start, u64 end, int mirror) | |
07157aac | 3243 | { |
4eee4fa4 | 3244 | size_t offset = start - page_offset(page); |
07157aac | 3245 | struct inode *inode = page->mapping->host; |
d1310b2e | 3246 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3247 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3248 | |
d20f7043 CM |
3249 | if (PageChecked(page)) { |
3250 | ClearPageChecked(page); | |
dc380aea | 3251 | return 0; |
d20f7043 | 3252 | } |
6cbff00f CH |
3253 | |
3254 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3255 | return 0; |
17d217fe YZ |
3256 | |
3257 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3258 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3259 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3260 | return 0; |
17d217fe | 3261 | } |
d20f7043 | 3262 | |
facc8a22 | 3263 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3264 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3265 | start, (size_t)(end - start + 1)); | |
07157aac | 3266 | } |
b888db2b | 3267 | |
24bbcf04 YZ |
3268 | void btrfs_add_delayed_iput(struct inode *inode) |
3269 | { | |
0b246afa | 3270 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3271 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3272 | |
3273 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3274 | return; | |
3275 | ||
24bbcf04 | 3276 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3277 | if (binode->delayed_iput_count == 0) { |
3278 | ASSERT(list_empty(&binode->delayed_iput)); | |
3279 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3280 | } else { | |
3281 | binode->delayed_iput_count++; | |
3282 | } | |
24bbcf04 YZ |
3283 | spin_unlock(&fs_info->delayed_iput_lock); |
3284 | } | |
3285 | ||
2ff7e61e | 3286 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3287 | { |
24bbcf04 | 3288 | |
24bbcf04 | 3289 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3290 | while (!list_empty(&fs_info->delayed_iputs)) { |
3291 | struct btrfs_inode *inode; | |
3292 | ||
3293 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3294 | struct btrfs_inode, delayed_iput); | |
3295 | if (inode->delayed_iput_count) { | |
3296 | inode->delayed_iput_count--; | |
3297 | list_move_tail(&inode->delayed_iput, | |
3298 | &fs_info->delayed_iputs); | |
3299 | } else { | |
3300 | list_del_init(&inode->delayed_iput); | |
3301 | } | |
3302 | spin_unlock(&fs_info->delayed_iput_lock); | |
3303 | iput(&inode->vfs_inode); | |
3304 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3305 | } |
8089fe62 | 3306 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3307 | } |
3308 | ||
d68fc57b | 3309 | /* |
42b2aa86 | 3310 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3311 | * files in the subvolume, it removes orphan item and frees block_rsv |
3312 | * structure. | |
3313 | */ | |
3314 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3315 | struct btrfs_root *root) | |
3316 | { | |
0b246afa | 3317 | struct btrfs_fs_info *fs_info = root->fs_info; |
90290e19 | 3318 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3319 | int ret; |
3320 | ||
8a35d95f | 3321 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3322 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3323 | return; | |
3324 | ||
90290e19 | 3325 | spin_lock(&root->orphan_lock); |
8a35d95f | 3326 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3327 | spin_unlock(&root->orphan_lock); |
3328 | return; | |
3329 | } | |
3330 | ||
3331 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3332 | spin_unlock(&root->orphan_lock); | |
3333 | return; | |
3334 | } | |
3335 | ||
3336 | block_rsv = root->orphan_block_rsv; | |
3337 | root->orphan_block_rsv = NULL; | |
3338 | spin_unlock(&root->orphan_lock); | |
3339 | ||
27cdeb70 | 3340 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b | 3341 | btrfs_root_refs(&root->root_item) > 0) { |
0b246afa | 3342 | ret = btrfs_del_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3343 | root->root_key.objectid); |
4ef31a45 | 3344 | if (ret) |
66642832 | 3345 | btrfs_abort_transaction(trans, ret); |
4ef31a45 | 3346 | else |
27cdeb70 MX |
3347 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3348 | &root->state); | |
d68fc57b YZ |
3349 | } |
3350 | ||
90290e19 JB |
3351 | if (block_rsv) { |
3352 | WARN_ON(block_rsv->size > 0); | |
2ff7e61e | 3353 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b YZ |
3354 | } |
3355 | } | |
3356 | ||
7b128766 JB |
3357 | /* |
3358 | * This creates an orphan entry for the given inode in case something goes | |
3359 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3360 | * |
3361 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3362 | * this function. | |
7b128766 | 3363 | */ |
73f2e545 NB |
3364 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
3365 | struct btrfs_inode *inode) | |
7b128766 | 3366 | { |
73f2e545 NB |
3367 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
3368 | struct btrfs_root *root = inode->root; | |
d68fc57b YZ |
3369 | struct btrfs_block_rsv *block_rsv = NULL; |
3370 | int reserve = 0; | |
3371 | int insert = 0; | |
3372 | int ret; | |
7b128766 | 3373 | |
d68fc57b | 3374 | if (!root->orphan_block_rsv) { |
2ff7e61e JM |
3375 | block_rsv = btrfs_alloc_block_rsv(fs_info, |
3376 | BTRFS_BLOCK_RSV_TEMP); | |
b532402e TI |
3377 | if (!block_rsv) |
3378 | return -ENOMEM; | |
d68fc57b | 3379 | } |
7b128766 | 3380 | |
d68fc57b YZ |
3381 | spin_lock(&root->orphan_lock); |
3382 | if (!root->orphan_block_rsv) { | |
3383 | root->orphan_block_rsv = block_rsv; | |
3384 | } else if (block_rsv) { | |
2ff7e61e | 3385 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b | 3386 | block_rsv = NULL; |
7b128766 | 3387 | } |
7b128766 | 3388 | |
8a35d95f | 3389 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3390 | &inode->runtime_flags)) { |
d68fc57b YZ |
3391 | #if 0 |
3392 | /* | |
3393 | * For proper ENOSPC handling, we should do orphan | |
3394 | * cleanup when mounting. But this introduces backward | |
3395 | * compatibility issue. | |
3396 | */ | |
3397 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3398 | insert = 2; | |
3399 | else | |
3400 | insert = 1; | |
3401 | #endif | |
3402 | insert = 1; | |
321f0e70 | 3403 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3404 | } |
3405 | ||
72ac3c0d | 3406 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
73f2e545 | 3407 | &inode->runtime_flags)) |
d68fc57b | 3408 | reserve = 1; |
d68fc57b | 3409 | spin_unlock(&root->orphan_lock); |
7b128766 | 3410 | |
d68fc57b YZ |
3411 | /* grab metadata reservation from transaction handle */ |
3412 | if (reserve) { | |
3413 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3414 | ASSERT(!ret); |
3415 | if (ret) { | |
f30c7d95 LB |
3416 | /* |
3417 | * dec doesn't need spin_lock as ->orphan_block_rsv | |
3418 | * would be released only if ->orphan_inodes is | |
3419 | * zero. | |
3420 | */ | |
3b6571c1 JB |
3421 | atomic_dec(&root->orphan_inodes); |
3422 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3423 | &inode->runtime_flags); |
3b6571c1 JB |
3424 | if (insert) |
3425 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
73f2e545 | 3426 | &inode->runtime_flags); |
3b6571c1 JB |
3427 | return ret; |
3428 | } | |
d68fc57b | 3429 | } |
7b128766 | 3430 | |
d68fc57b YZ |
3431 | /* insert an orphan item to track this unlinked/truncated file */ |
3432 | if (insert >= 1) { | |
73f2e545 | 3433 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3434 | if (ret) { |
4ef31a45 JB |
3435 | if (reserve) { |
3436 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3437 | &inode->runtime_flags); |
4ef31a45 JB |
3438 | btrfs_orphan_release_metadata(inode); |
3439 | } | |
f30c7d95 LB |
3440 | /* |
3441 | * btrfs_orphan_commit_root may race with us and set | |
3442 | * ->orphan_block_rsv to zero, in order to avoid that, | |
3443 | * decrease ->orphan_inodes after everything is done. | |
3444 | */ | |
3445 | atomic_dec(&root->orphan_inodes); | |
4ef31a45 | 3446 | if (ret != -EEXIST) { |
e8e7cff6 | 3447 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3448 | &inode->runtime_flags); |
66642832 | 3449 | btrfs_abort_transaction(trans, ret); |
4ef31a45 JB |
3450 | return ret; |
3451 | } | |
79787eaa JM |
3452 | } |
3453 | ret = 0; | |
d68fc57b YZ |
3454 | } |
3455 | ||
3456 | /* insert an orphan item to track subvolume contains orphan files */ | |
3457 | if (insert >= 2) { | |
0b246afa | 3458 | ret = btrfs_insert_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3459 | root->root_key.objectid); |
79787eaa | 3460 | if (ret && ret != -EEXIST) { |
66642832 | 3461 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3462 | return ret; |
3463 | } | |
d68fc57b YZ |
3464 | } |
3465 | return 0; | |
7b128766 JB |
3466 | } |
3467 | ||
3468 | /* | |
3469 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3470 | * item for this particular inode. | |
3471 | */ | |
48a3b636 | 3472 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3473 | struct btrfs_inode *inode) |
7b128766 | 3474 | { |
3d6ae7bb | 3475 | struct btrfs_root *root = inode->root; |
d68fc57b | 3476 | int delete_item = 0; |
7b128766 JB |
3477 | int ret = 0; |
3478 | ||
8a35d95f | 3479 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3d6ae7bb | 3480 | &inode->runtime_flags)) |
d68fc57b | 3481 | delete_item = 1; |
7b128766 | 3482 | |
f30c7d95 LB |
3483 | if (delete_item && trans) |
3484 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); | |
3485 | ||
72ac3c0d | 3486 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3d6ae7bb | 3487 | &inode->runtime_flags)) |
f30c7d95 | 3488 | btrfs_orphan_release_metadata(inode); |
7b128766 | 3489 | |
f30c7d95 LB |
3490 | /* |
3491 | * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv | |
3492 | * to zero, in order to avoid that, decrease ->orphan_inodes after | |
3493 | * everything is done. | |
3494 | */ | |
3495 | if (delete_item) | |
8a35d95f | 3496 | atomic_dec(&root->orphan_inodes); |
703c88e0 | 3497 | |
4ef31a45 | 3498 | return ret; |
7b128766 JB |
3499 | } |
3500 | ||
3501 | /* | |
3502 | * this cleans up any orphans that may be left on the list from the last use | |
3503 | * of this root. | |
3504 | */ | |
66b4ffd1 | 3505 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3506 | { |
0b246afa | 3507 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3508 | struct btrfs_path *path; |
3509 | struct extent_buffer *leaf; | |
7b128766 JB |
3510 | struct btrfs_key key, found_key; |
3511 | struct btrfs_trans_handle *trans; | |
3512 | struct inode *inode; | |
8f6d7f4f | 3513 | u64 last_objectid = 0; |
7b128766 JB |
3514 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3515 | ||
d68fc57b | 3516 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3517 | return 0; |
c71bf099 YZ |
3518 | |
3519 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3520 | if (!path) { |
3521 | ret = -ENOMEM; | |
3522 | goto out; | |
3523 | } | |
e4058b54 | 3524 | path->reada = READA_BACK; |
7b128766 JB |
3525 | |
3526 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3527 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3528 | key.offset = (u64)-1; |
3529 | ||
7b128766 JB |
3530 | while (1) { |
3531 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3532 | if (ret < 0) |
3533 | goto out; | |
7b128766 JB |
3534 | |
3535 | /* | |
3536 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3537 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3538 | * find the key and see if we have stuff that matches |
3539 | */ | |
3540 | if (ret > 0) { | |
66b4ffd1 | 3541 | ret = 0; |
7b128766 JB |
3542 | if (path->slots[0] == 0) |
3543 | break; | |
3544 | path->slots[0]--; | |
3545 | } | |
3546 | ||
3547 | /* pull out the item */ | |
3548 | leaf = path->nodes[0]; | |
7b128766 JB |
3549 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3550 | ||
3551 | /* make sure the item matches what we want */ | |
3552 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3553 | break; | |
962a298f | 3554 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3555 | break; |
3556 | ||
3557 | /* release the path since we're done with it */ | |
b3b4aa74 | 3558 | btrfs_release_path(path); |
7b128766 JB |
3559 | |
3560 | /* | |
3561 | * this is where we are basically btrfs_lookup, without the | |
3562 | * crossing root thing. we store the inode number in the | |
3563 | * offset of the orphan item. | |
3564 | */ | |
8f6d7f4f JB |
3565 | |
3566 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3567 | btrfs_err(fs_info, |
3568 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3569 | ret = -EINVAL; |
3570 | goto out; | |
3571 | } | |
3572 | ||
3573 | last_objectid = found_key.offset; | |
3574 | ||
5d4f98a2 YZ |
3575 | found_key.objectid = found_key.offset; |
3576 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3577 | found_key.offset = 0; | |
0b246afa | 3578 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3579 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3580 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3581 | goto out; |
7b128766 | 3582 | |
0b246afa | 3583 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3584 | struct btrfs_root *dead_root; |
3585 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3586 | int is_dead_root = 0; | |
3587 | ||
3588 | /* | |
3589 | * this is an orphan in the tree root. Currently these | |
3590 | * could come from 2 sources: | |
3591 | * a) a snapshot deletion in progress | |
3592 | * b) a free space cache inode | |
3593 | * We need to distinguish those two, as the snapshot | |
3594 | * orphan must not get deleted. | |
3595 | * find_dead_roots already ran before us, so if this | |
3596 | * is a snapshot deletion, we should find the root | |
3597 | * in the dead_roots list | |
3598 | */ | |
3599 | spin_lock(&fs_info->trans_lock); | |
3600 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3601 | root_list) { | |
3602 | if (dead_root->root_key.objectid == | |
3603 | found_key.objectid) { | |
3604 | is_dead_root = 1; | |
3605 | break; | |
3606 | } | |
3607 | } | |
3608 | spin_unlock(&fs_info->trans_lock); | |
3609 | if (is_dead_root) { | |
3610 | /* prevent this orphan from being found again */ | |
3611 | key.offset = found_key.objectid - 1; | |
3612 | continue; | |
3613 | } | |
3614 | } | |
7b128766 | 3615 | /* |
a8c9e576 JB |
3616 | * Inode is already gone but the orphan item is still there, |
3617 | * kill the orphan item. | |
7b128766 | 3618 | */ |
67710892 | 3619 | if (ret == -ENOENT) { |
a8c9e576 | 3620 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3621 | if (IS_ERR(trans)) { |
3622 | ret = PTR_ERR(trans); | |
3623 | goto out; | |
3624 | } | |
0b246afa JM |
3625 | btrfs_debug(fs_info, "auto deleting %Lu", |
3626 | found_key.objectid); | |
a8c9e576 JB |
3627 | ret = btrfs_del_orphan_item(trans, root, |
3628 | found_key.objectid); | |
3a45bb20 | 3629 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3630 | if (ret) |
3631 | goto out; | |
7b128766 JB |
3632 | continue; |
3633 | } | |
3634 | ||
a8c9e576 JB |
3635 | /* |
3636 | * add this inode to the orphan list so btrfs_orphan_del does | |
3637 | * the proper thing when we hit it | |
3638 | */ | |
8a35d95f JB |
3639 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3640 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3641 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3642 | |
7b128766 JB |
3643 | /* if we have links, this was a truncate, lets do that */ |
3644 | if (inode->i_nlink) { | |
fae7f21c | 3645 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3646 | iput(inode); |
3647 | continue; | |
3648 | } | |
7b128766 | 3649 | nr_truncate++; |
f3fe820c JB |
3650 | |
3651 | /* 1 for the orphan item deletion. */ | |
3652 | trans = btrfs_start_transaction(root, 1); | |
3653 | if (IS_ERR(trans)) { | |
c69b26b0 | 3654 | iput(inode); |
f3fe820c JB |
3655 | ret = PTR_ERR(trans); |
3656 | goto out; | |
3657 | } | |
73f2e545 | 3658 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 3659 | btrfs_end_transaction(trans); |
c69b26b0 JB |
3660 | if (ret) { |
3661 | iput(inode); | |
f3fe820c | 3662 | goto out; |
c69b26b0 | 3663 | } |
f3fe820c | 3664 | |
66b4ffd1 | 3665 | ret = btrfs_truncate(inode); |
4a7d0f68 | 3666 | if (ret) |
3d6ae7bb | 3667 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7b128766 JB |
3668 | } else { |
3669 | nr_unlink++; | |
3670 | } | |
3671 | ||
3672 | /* this will do delete_inode and everything for us */ | |
3673 | iput(inode); | |
66b4ffd1 JB |
3674 | if (ret) |
3675 | goto out; | |
7b128766 | 3676 | } |
3254c876 MX |
3677 | /* release the path since we're done with it */ |
3678 | btrfs_release_path(path); | |
3679 | ||
d68fc57b YZ |
3680 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3681 | ||
3682 | if (root->orphan_block_rsv) | |
2ff7e61e | 3683 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, |
d68fc57b YZ |
3684 | (u64)-1); |
3685 | ||
27cdeb70 MX |
3686 | if (root->orphan_block_rsv || |
3687 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3688 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3689 | if (!IS_ERR(trans)) |
3a45bb20 | 3690 | btrfs_end_transaction(trans); |
d68fc57b | 3691 | } |
7b128766 JB |
3692 | |
3693 | if (nr_unlink) | |
0b246afa | 3694 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3695 | if (nr_truncate) |
0b246afa | 3696 | btrfs_debug(fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3697 | |
3698 | out: | |
3699 | if (ret) | |
0b246afa | 3700 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3701 | btrfs_free_path(path); |
3702 | return ret; | |
7b128766 JB |
3703 | } |
3704 | ||
46a53cca CM |
3705 | /* |
3706 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3707 | * don't find any xattrs, we know there can't be any acls. | |
3708 | * | |
3709 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3710 | */ | |
3711 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3712 | int slot, u64 objectid, |
3713 | int *first_xattr_slot) | |
46a53cca CM |
3714 | { |
3715 | u32 nritems = btrfs_header_nritems(leaf); | |
3716 | struct btrfs_key found_key; | |
f23b5a59 JB |
3717 | static u64 xattr_access = 0; |
3718 | static u64 xattr_default = 0; | |
46a53cca CM |
3719 | int scanned = 0; |
3720 | ||
f23b5a59 | 3721 | if (!xattr_access) { |
97d79299 AG |
3722 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3723 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3724 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3725 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3726 | } |
3727 | ||
46a53cca | 3728 | slot++; |
63541927 | 3729 | *first_xattr_slot = -1; |
46a53cca CM |
3730 | while (slot < nritems) { |
3731 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3732 | ||
3733 | /* we found a different objectid, there must not be acls */ | |
3734 | if (found_key.objectid != objectid) | |
3735 | return 0; | |
3736 | ||
3737 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3738 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3739 | if (*first_xattr_slot == -1) |
3740 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3741 | if (found_key.offset == xattr_access || |
3742 | found_key.offset == xattr_default) | |
3743 | return 1; | |
3744 | } | |
46a53cca CM |
3745 | |
3746 | /* | |
3747 | * we found a key greater than an xattr key, there can't | |
3748 | * be any acls later on | |
3749 | */ | |
3750 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3751 | return 0; | |
3752 | ||
3753 | slot++; | |
3754 | scanned++; | |
3755 | ||
3756 | /* | |
3757 | * it goes inode, inode backrefs, xattrs, extents, | |
3758 | * so if there are a ton of hard links to an inode there can | |
3759 | * be a lot of backrefs. Don't waste time searching too hard, | |
3760 | * this is just an optimization | |
3761 | */ | |
3762 | if (scanned >= 8) | |
3763 | break; | |
3764 | } | |
3765 | /* we hit the end of the leaf before we found an xattr or | |
3766 | * something larger than an xattr. We have to assume the inode | |
3767 | * has acls | |
3768 | */ | |
63541927 FDBM |
3769 | if (*first_xattr_slot == -1) |
3770 | *first_xattr_slot = slot; | |
46a53cca CM |
3771 | return 1; |
3772 | } | |
3773 | ||
d352ac68 CM |
3774 | /* |
3775 | * read an inode from the btree into the in-memory inode | |
3776 | */ | |
67710892 | 3777 | static int btrfs_read_locked_inode(struct inode *inode) |
39279cc3 | 3778 | { |
0b246afa | 3779 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 3780 | struct btrfs_path *path; |
5f39d397 | 3781 | struct extent_buffer *leaf; |
39279cc3 CM |
3782 | struct btrfs_inode_item *inode_item; |
3783 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3784 | struct btrfs_key location; | |
67de1176 | 3785 | unsigned long ptr; |
46a53cca | 3786 | int maybe_acls; |
618e21d5 | 3787 | u32 rdev; |
39279cc3 | 3788 | int ret; |
2f7e33d4 | 3789 | bool filled = false; |
63541927 | 3790 | int first_xattr_slot; |
2f7e33d4 MX |
3791 | |
3792 | ret = btrfs_fill_inode(inode, &rdev); | |
3793 | if (!ret) | |
3794 | filled = true; | |
39279cc3 CM |
3795 | |
3796 | path = btrfs_alloc_path(); | |
67710892 FM |
3797 | if (!path) { |
3798 | ret = -ENOMEM; | |
1748f843 | 3799 | goto make_bad; |
67710892 | 3800 | } |
1748f843 | 3801 | |
39279cc3 | 3802 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3803 | |
39279cc3 | 3804 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 FM |
3805 | if (ret) { |
3806 | if (ret > 0) | |
3807 | ret = -ENOENT; | |
39279cc3 | 3808 | goto make_bad; |
67710892 | 3809 | } |
39279cc3 | 3810 | |
5f39d397 | 3811 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3812 | |
3813 | if (filled) | |
67de1176 | 3814 | goto cache_index; |
2f7e33d4 | 3815 | |
5f39d397 CM |
3816 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3817 | struct btrfs_inode_item); | |
5f39d397 | 3818 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3819 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3820 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3821 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3822 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3823 | |
a937b979 DS |
3824 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3825 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3826 | |
a937b979 DS |
3827 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3828 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3829 | |
a937b979 DS |
3830 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3831 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3832 | |
9cc97d64 | 3833 | BTRFS_I(inode)->i_otime.tv_sec = |
3834 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3835 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3836 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3837 | |
a76a3cd4 | 3838 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3839 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3840 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3841 | ||
6e17d30b YD |
3842 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3843 | inode->i_generation = BTRFS_I(inode)->generation; | |
3844 | inode->i_rdev = 0; | |
3845 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3846 | ||
3847 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3848 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3849 | ||
3850 | cache_index: | |
5dc562c5 JB |
3851 | /* |
3852 | * If we were modified in the current generation and evicted from memory | |
3853 | * and then re-read we need to do a full sync since we don't have any | |
3854 | * idea about which extents were modified before we were evicted from | |
3855 | * cache. | |
6e17d30b YD |
3856 | * |
3857 | * This is required for both inode re-read from disk and delayed inode | |
3858 | * in delayed_nodes_tree. | |
5dc562c5 | 3859 | */ |
0b246afa | 3860 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3861 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3862 | &BTRFS_I(inode)->runtime_flags); | |
3863 | ||
bde6c242 FM |
3864 | /* |
3865 | * We don't persist the id of the transaction where an unlink operation | |
3866 | * against the inode was last made. So here we assume the inode might | |
3867 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3868 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3869 | * between the inode and its parent if the inode is fsync'ed and the log | |
3870 | * replayed. For example, in the scenario: | |
3871 | * | |
3872 | * touch mydir/foo | |
3873 | * ln mydir/foo mydir/bar | |
3874 | * sync | |
3875 | * unlink mydir/bar | |
3876 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3877 | * xfs_io -c fsync mydir/foo | |
3878 | * <power failure> | |
3879 | * mount fs, triggers fsync log replay | |
3880 | * | |
3881 | * We must make sure that when we fsync our inode foo we also log its | |
3882 | * parent inode, otherwise after log replay the parent still has the | |
3883 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3884 | * and doesn't have an inode ref with the name "bar" anymore. | |
3885 | * | |
3886 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3887 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3888 | * transaction commits on fsync if our inode is a directory, or if our |
3889 | * inode is not a directory, logging its parent unnecessarily. | |
3890 | */ | |
3891 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3892 | ||
67de1176 MX |
3893 | path->slots[0]++; |
3894 | if (inode->i_nlink != 1 || | |
3895 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3896 | goto cache_acl; | |
3897 | ||
3898 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3899 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3900 | goto cache_acl; |
3901 | ||
3902 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3903 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3904 | struct btrfs_inode_ref *ref; | |
3905 | ||
3906 | ref = (struct btrfs_inode_ref *)ptr; | |
3907 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3908 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3909 | struct btrfs_inode_extref *extref; | |
3910 | ||
3911 | extref = (struct btrfs_inode_extref *)ptr; | |
3912 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3913 | extref); | |
3914 | } | |
2f7e33d4 | 3915 | cache_acl: |
46a53cca CM |
3916 | /* |
3917 | * try to precache a NULL acl entry for files that don't have | |
3918 | * any xattrs or acls | |
3919 | */ | |
33345d01 | 3920 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3921 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3922 | if (first_xattr_slot != -1) { |
3923 | path->slots[0] = first_xattr_slot; | |
3924 | ret = btrfs_load_inode_props(inode, path); | |
3925 | if (ret) | |
0b246afa | 3926 | btrfs_err(fs_info, |
351fd353 | 3927 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3928 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3929 | root->root_key.objectid, ret); |
3930 | } | |
3931 | btrfs_free_path(path); | |
3932 | ||
72c04902 AV |
3933 | if (!maybe_acls) |
3934 | cache_no_acl(inode); | |
46a53cca | 3935 | |
39279cc3 | 3936 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3937 | case S_IFREG: |
3938 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3939 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3940 | inode->i_fop = &btrfs_file_operations; |
3941 | inode->i_op = &btrfs_file_inode_operations; | |
3942 | break; | |
3943 | case S_IFDIR: | |
3944 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3945 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3946 | break; |
3947 | case S_IFLNK: | |
3948 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3949 | inode_nohighmem(inode); |
39279cc3 CM |
3950 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3951 | break; | |
618e21d5 | 3952 | default: |
0279b4cd | 3953 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3954 | init_special_inode(inode, inode->i_mode, rdev); |
3955 | break; | |
39279cc3 | 3956 | } |
6cbff00f CH |
3957 | |
3958 | btrfs_update_iflags(inode); | |
67710892 | 3959 | return 0; |
39279cc3 CM |
3960 | |
3961 | make_bad: | |
39279cc3 | 3962 | btrfs_free_path(path); |
39279cc3 | 3963 | make_bad_inode(inode); |
67710892 | 3964 | return ret; |
39279cc3 CM |
3965 | } |
3966 | ||
d352ac68 CM |
3967 | /* |
3968 | * given a leaf and an inode, copy the inode fields into the leaf | |
3969 | */ | |
e02119d5 CM |
3970 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3971 | struct extent_buffer *leaf, | |
5f39d397 | 3972 | struct btrfs_inode_item *item, |
39279cc3 CM |
3973 | struct inode *inode) |
3974 | { | |
51fab693 LB |
3975 | struct btrfs_map_token token; |
3976 | ||
3977 | btrfs_init_map_token(&token); | |
5f39d397 | 3978 | |
51fab693 LB |
3979 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3980 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3981 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3982 | &token); | |
3983 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3984 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3985 | |
a937b979 | 3986 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3987 | inode->i_atime.tv_sec, &token); |
a937b979 | 3988 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3989 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3990 | |
a937b979 | 3991 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3992 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3993 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3994 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3995 | |
a937b979 | 3996 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3997 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3998 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3999 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 4000 | |
9cc97d64 | 4001 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
4002 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
4003 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
4004 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
4005 | ||
51fab693 LB |
4006 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
4007 | &token); | |
4008 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
4009 | &token); | |
4010 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
4011 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
4012 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
4013 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
4014 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
4015 | } |
4016 | ||
d352ac68 CM |
4017 | /* |
4018 | * copy everything in the in-memory inode into the btree. | |
4019 | */ | |
2115133f | 4020 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 4021 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
4022 | { |
4023 | struct btrfs_inode_item *inode_item; | |
4024 | struct btrfs_path *path; | |
5f39d397 | 4025 | struct extent_buffer *leaf; |
39279cc3 CM |
4026 | int ret; |
4027 | ||
4028 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
4029 | if (!path) |
4030 | return -ENOMEM; | |
4031 | ||
b9473439 | 4032 | path->leave_spinning = 1; |
16cdcec7 MX |
4033 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
4034 | 1); | |
39279cc3 CM |
4035 | if (ret) { |
4036 | if (ret > 0) | |
4037 | ret = -ENOENT; | |
4038 | goto failed; | |
4039 | } | |
4040 | ||
5f39d397 CM |
4041 | leaf = path->nodes[0]; |
4042 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 4043 | struct btrfs_inode_item); |
39279cc3 | 4044 | |
e02119d5 | 4045 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 4046 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 4047 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
4048 | ret = 0; |
4049 | failed: | |
39279cc3 CM |
4050 | btrfs_free_path(path); |
4051 | return ret; | |
4052 | } | |
4053 | ||
2115133f CM |
4054 | /* |
4055 | * copy everything in the in-memory inode into the btree. | |
4056 | */ | |
4057 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
4058 | struct btrfs_root *root, struct inode *inode) | |
4059 | { | |
0b246afa | 4060 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4061 | int ret; |
4062 | ||
4063 | /* | |
4064 | * If the inode is a free space inode, we can deadlock during commit | |
4065 | * if we put it into the delayed code. | |
4066 | * | |
4067 | * The data relocation inode should also be directly updated | |
4068 | * without delay | |
4069 | */ | |
70ddc553 | 4070 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 4071 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 4072 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4073 | btrfs_update_root_times(trans, root); |
4074 | ||
2115133f CM |
4075 | ret = btrfs_delayed_update_inode(trans, root, inode); |
4076 | if (!ret) | |
4077 | btrfs_set_inode_last_trans(trans, inode); | |
4078 | return ret; | |
4079 | } | |
4080 | ||
4081 | return btrfs_update_inode_item(trans, root, inode); | |
4082 | } | |
4083 | ||
be6aef60 JB |
4084 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4085 | struct btrfs_root *root, | |
4086 | struct inode *inode) | |
2115133f CM |
4087 | { |
4088 | int ret; | |
4089 | ||
4090 | ret = btrfs_update_inode(trans, root, inode); | |
4091 | if (ret == -ENOSPC) | |
4092 | return btrfs_update_inode_item(trans, root, inode); | |
4093 | return ret; | |
4094 | } | |
4095 | ||
d352ac68 CM |
4096 | /* |
4097 | * unlink helper that gets used here in inode.c and in the tree logging | |
4098 | * recovery code. It remove a link in a directory with a given name, and | |
4099 | * also drops the back refs in the inode to the directory | |
4100 | */ | |
92986796 AV |
4101 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4102 | struct btrfs_root *root, | |
4ec5934e NB |
4103 | struct btrfs_inode *dir, |
4104 | struct btrfs_inode *inode, | |
92986796 | 4105 | const char *name, int name_len) |
39279cc3 | 4106 | { |
0b246afa | 4107 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4108 | struct btrfs_path *path; |
39279cc3 | 4109 | int ret = 0; |
5f39d397 | 4110 | struct extent_buffer *leaf; |
39279cc3 | 4111 | struct btrfs_dir_item *di; |
5f39d397 | 4112 | struct btrfs_key key; |
aec7477b | 4113 | u64 index; |
33345d01 LZ |
4114 | u64 ino = btrfs_ino(inode); |
4115 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4116 | |
4117 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4118 | if (!path) { |
4119 | ret = -ENOMEM; | |
554233a6 | 4120 | goto out; |
54aa1f4d CM |
4121 | } |
4122 | ||
b9473439 | 4123 | path->leave_spinning = 1; |
33345d01 | 4124 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
4125 | name, name_len, -1); |
4126 | if (IS_ERR(di)) { | |
4127 | ret = PTR_ERR(di); | |
4128 | goto err; | |
4129 | } | |
4130 | if (!di) { | |
4131 | ret = -ENOENT; | |
4132 | goto err; | |
4133 | } | |
5f39d397 CM |
4134 | leaf = path->nodes[0]; |
4135 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 4136 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4137 | if (ret) |
4138 | goto err; | |
b3b4aa74 | 4139 | btrfs_release_path(path); |
39279cc3 | 4140 | |
67de1176 MX |
4141 | /* |
4142 | * If we don't have dir index, we have to get it by looking up | |
4143 | * the inode ref, since we get the inode ref, remove it directly, | |
4144 | * it is unnecessary to do delayed deletion. | |
4145 | * | |
4146 | * But if we have dir index, needn't search inode ref to get it. | |
4147 | * Since the inode ref is close to the inode item, it is better | |
4148 | * that we delay to delete it, and just do this deletion when | |
4149 | * we update the inode item. | |
4150 | */ | |
4ec5934e | 4151 | if (inode->dir_index) { |
67de1176 MX |
4152 | ret = btrfs_delayed_delete_inode_ref(inode); |
4153 | if (!ret) { | |
4ec5934e | 4154 | index = inode->dir_index; |
67de1176 MX |
4155 | goto skip_backref; |
4156 | } | |
4157 | } | |
4158 | ||
33345d01 LZ |
4159 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4160 | dir_ino, &index); | |
aec7477b | 4161 | if (ret) { |
0b246afa | 4162 | btrfs_info(fs_info, |
c2cf52eb | 4163 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4164 | name_len, name, ino, dir_ino); |
66642832 | 4165 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4166 | goto err; |
4167 | } | |
67de1176 | 4168 | skip_backref: |
2ff7e61e | 4169 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index); |
79787eaa | 4170 | if (ret) { |
66642832 | 4171 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4172 | goto err; |
79787eaa | 4173 | } |
39279cc3 | 4174 | |
4ec5934e NB |
4175 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4176 | dir_ino); | |
79787eaa | 4177 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4178 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4179 | goto err; |
4180 | } | |
e02119d5 | 4181 | |
4ec5934e NB |
4182 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4183 | index); | |
6418c961 CM |
4184 | if (ret == -ENOENT) |
4185 | ret = 0; | |
d4e3991b | 4186 | else if (ret) |
66642832 | 4187 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
4188 | err: |
4189 | btrfs_free_path(path); | |
e02119d5 CM |
4190 | if (ret) |
4191 | goto out; | |
4192 | ||
6ef06d27 | 4193 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4194 | inode_inc_iversion(&inode->vfs_inode); |
4195 | inode_inc_iversion(&dir->vfs_inode); | |
4196 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4197 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4198 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4199 | out: |
39279cc3 CM |
4200 | return ret; |
4201 | } | |
4202 | ||
92986796 AV |
4203 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4204 | struct btrfs_root *root, | |
4ec5934e | 4205 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4206 | const char *name, int name_len) |
4207 | { | |
4208 | int ret; | |
4209 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4210 | if (!ret) { | |
4ec5934e NB |
4211 | drop_nlink(&inode->vfs_inode); |
4212 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4213 | } |
4214 | return ret; | |
4215 | } | |
39279cc3 | 4216 | |
a22285a6 YZ |
4217 | /* |
4218 | * helper to start transaction for unlink and rmdir. | |
4219 | * | |
d52be818 JB |
4220 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4221 | * if we cannot make our reservations the normal way try and see if there is | |
4222 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4223 | * allow the unlink to occur. | |
a22285a6 | 4224 | */ |
d52be818 | 4225 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4226 | { |
a22285a6 | 4227 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4228 | |
e70bea5f JB |
4229 | /* |
4230 | * 1 for the possible orphan item | |
4231 | * 1 for the dir item | |
4232 | * 1 for the dir index | |
4233 | * 1 for the inode ref | |
e70bea5f JB |
4234 | * 1 for the inode |
4235 | */ | |
8eab77ff | 4236 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4237 | } |
4238 | ||
4239 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4240 | { | |
4241 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4242 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4243 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4244 | int ret; |
a22285a6 | 4245 | |
d52be818 | 4246 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4247 | if (IS_ERR(trans)) |
4248 | return PTR_ERR(trans); | |
5f39d397 | 4249 | |
4ec5934e NB |
4250 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4251 | 0); | |
12fcfd22 | 4252 | |
4ec5934e NB |
4253 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4254 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4255 | dentry->d_name.len); | |
b532402e TI |
4256 | if (ret) |
4257 | goto out; | |
7b128766 | 4258 | |
a22285a6 | 4259 | if (inode->i_nlink == 0) { |
73f2e545 | 4260 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4261 | if (ret) |
4262 | goto out; | |
a22285a6 | 4263 | } |
7b128766 | 4264 | |
b532402e | 4265 | out: |
3a45bb20 | 4266 | btrfs_end_transaction(trans); |
2ff7e61e | 4267 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4268 | return ret; |
4269 | } | |
4270 | ||
4df27c4d YZ |
4271 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4272 | struct btrfs_root *root, | |
4273 | struct inode *dir, u64 objectid, | |
4274 | const char *name, int name_len) | |
4275 | { | |
0b246afa | 4276 | struct btrfs_fs_info *fs_info = root->fs_info; |
4df27c4d YZ |
4277 | struct btrfs_path *path; |
4278 | struct extent_buffer *leaf; | |
4279 | struct btrfs_dir_item *di; | |
4280 | struct btrfs_key key; | |
4281 | u64 index; | |
4282 | int ret; | |
4a0cc7ca | 4283 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4284 | |
4285 | path = btrfs_alloc_path(); | |
4286 | if (!path) | |
4287 | return -ENOMEM; | |
4288 | ||
33345d01 | 4289 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4290 | name, name_len, -1); |
79787eaa JM |
4291 | if (IS_ERR_OR_NULL(di)) { |
4292 | if (!di) | |
4293 | ret = -ENOENT; | |
4294 | else | |
4295 | ret = PTR_ERR(di); | |
4296 | goto out; | |
4297 | } | |
4df27c4d YZ |
4298 | |
4299 | leaf = path->nodes[0]; | |
4300 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4301 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4302 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4303 | if (ret) { |
66642832 | 4304 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4305 | goto out; |
4306 | } | |
b3b4aa74 | 4307 | btrfs_release_path(path); |
4df27c4d | 4308 | |
0b246afa JM |
4309 | ret = btrfs_del_root_ref(trans, fs_info, objectid, |
4310 | root->root_key.objectid, dir_ino, | |
4311 | &index, name, name_len); | |
4df27c4d | 4312 | if (ret < 0) { |
79787eaa | 4313 | if (ret != -ENOENT) { |
66642832 | 4314 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4315 | goto out; |
4316 | } | |
33345d01 | 4317 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4318 | name, name_len); |
79787eaa JM |
4319 | if (IS_ERR_OR_NULL(di)) { |
4320 | if (!di) | |
4321 | ret = -ENOENT; | |
4322 | else | |
4323 | ret = PTR_ERR(di); | |
66642832 | 4324 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4325 | goto out; |
4326 | } | |
4df27c4d YZ |
4327 | |
4328 | leaf = path->nodes[0]; | |
4329 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4330 | btrfs_release_path(path); |
4df27c4d YZ |
4331 | index = key.offset; |
4332 | } | |
945d8962 | 4333 | btrfs_release_path(path); |
4df27c4d | 4334 | |
e67bbbb9 | 4335 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index); |
79787eaa | 4336 | if (ret) { |
66642832 | 4337 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4338 | goto out; |
4339 | } | |
4df27c4d | 4340 | |
6ef06d27 | 4341 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4342 | inode_inc_iversion(dir); |
c2050a45 | 4343 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4344 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4345 | if (ret) |
66642832 | 4346 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4347 | out: |
71d7aed0 | 4348 | btrfs_free_path(path); |
79787eaa | 4349 | return ret; |
4df27c4d YZ |
4350 | } |
4351 | ||
39279cc3 CM |
4352 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4353 | { | |
2b0143b5 | 4354 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4355 | int err = 0; |
39279cc3 | 4356 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4357 | struct btrfs_trans_handle *trans; |
44f714da | 4358 | u64 last_unlink_trans; |
39279cc3 | 4359 | |
b3ae244e | 4360 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4361 | return -ENOTEMPTY; |
4a0cc7ca | 4362 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
b3ae244e | 4363 | return -EPERM; |
134d4512 | 4364 | |
d52be818 | 4365 | trans = __unlink_start_trans(dir); |
a22285a6 | 4366 | if (IS_ERR(trans)) |
5df6a9f6 | 4367 | return PTR_ERR(trans); |
5df6a9f6 | 4368 | |
4a0cc7ca | 4369 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4370 | err = btrfs_unlink_subvol(trans, root, dir, |
4371 | BTRFS_I(inode)->location.objectid, | |
4372 | dentry->d_name.name, | |
4373 | dentry->d_name.len); | |
4374 | goto out; | |
4375 | } | |
4376 | ||
73f2e545 | 4377 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4378 | if (err) |
4df27c4d | 4379 | goto out; |
7b128766 | 4380 | |
44f714da FM |
4381 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4382 | ||
39279cc3 | 4383 | /* now the directory is empty */ |
4ec5934e NB |
4384 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4385 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4386 | dentry->d_name.len); | |
44f714da | 4387 | if (!err) { |
6ef06d27 | 4388 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4389 | /* |
4390 | * Propagate the last_unlink_trans value of the deleted dir to | |
4391 | * its parent directory. This is to prevent an unrecoverable | |
4392 | * log tree in the case we do something like this: | |
4393 | * 1) create dir foo | |
4394 | * 2) create snapshot under dir foo | |
4395 | * 3) delete the snapshot | |
4396 | * 4) rmdir foo | |
4397 | * 5) mkdir foo | |
4398 | * 6) fsync foo or some file inside foo | |
4399 | */ | |
4400 | if (last_unlink_trans >= trans->transid) | |
4401 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4402 | } | |
4df27c4d | 4403 | out: |
3a45bb20 | 4404 | btrfs_end_transaction(trans); |
2ff7e61e | 4405 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4406 | |
39279cc3 CM |
4407 | return err; |
4408 | } | |
4409 | ||
28f75a0e CM |
4410 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4411 | struct btrfs_root *root, | |
4412 | u64 bytes_deleted) | |
4413 | { | |
0b246afa | 4414 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4415 | int ret; |
4416 | ||
dc95f7bf JB |
4417 | /* |
4418 | * This is only used to apply pressure to the enospc system, we don't | |
4419 | * intend to use this reservation at all. | |
4420 | */ | |
2ff7e61e | 4421 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4422 | bytes_deleted *= fs_info->nodesize; |
4423 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4424 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4425 | if (!ret) { |
0b246afa | 4426 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4427 | trans->transid, |
4428 | bytes_deleted, 1); | |
28f75a0e | 4429 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4430 | } |
28f75a0e CM |
4431 | return ret; |
4432 | ||
4433 | } | |
4434 | ||
0305cd5f FM |
4435 | static int truncate_inline_extent(struct inode *inode, |
4436 | struct btrfs_path *path, | |
4437 | struct btrfs_key *found_key, | |
4438 | const u64 item_end, | |
4439 | const u64 new_size) | |
4440 | { | |
4441 | struct extent_buffer *leaf = path->nodes[0]; | |
4442 | int slot = path->slots[0]; | |
4443 | struct btrfs_file_extent_item *fi; | |
4444 | u32 size = (u32)(new_size - found_key->offset); | |
4445 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4446 | ||
4447 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4448 | ||
4449 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4450 | loff_t offset = new_size; | |
09cbfeaf | 4451 | loff_t page_end = ALIGN(offset, PAGE_SIZE); |
0305cd5f FM |
4452 | |
4453 | /* | |
4454 | * Zero out the remaining of the last page of our inline extent, | |
4455 | * instead of directly truncating our inline extent here - that | |
4456 | * would be much more complex (decompressing all the data, then | |
4457 | * compressing the truncated data, which might be bigger than | |
4458 | * the size of the inline extent, resize the extent, etc). | |
4459 | * We release the path because to get the page we might need to | |
4460 | * read the extent item from disk (data not in the page cache). | |
4461 | */ | |
4462 | btrfs_release_path(path); | |
9703fefe CR |
4463 | return btrfs_truncate_block(inode, offset, page_end - offset, |
4464 | 0); | |
0305cd5f FM |
4465 | } |
4466 | ||
4467 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4468 | size = btrfs_file_extent_calc_inline_size(size); | |
2ff7e61e | 4469 | btrfs_truncate_item(root->fs_info, path, size, 1); |
0305cd5f FM |
4470 | |
4471 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4472 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4473 | ||
4474 | return 0; | |
4475 | } | |
4476 | ||
39279cc3 CM |
4477 | /* |
4478 | * this can truncate away extent items, csum items and directory items. | |
4479 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4480 | * any higher than new_size |
39279cc3 CM |
4481 | * |
4482 | * csum items that cross the new i_size are truncated to the new size | |
4483 | * as well. | |
7b128766 JB |
4484 | * |
4485 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4486 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4487 | */ |
8082510e YZ |
4488 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4489 | struct btrfs_root *root, | |
4490 | struct inode *inode, | |
4491 | u64 new_size, u32 min_type) | |
39279cc3 | 4492 | { |
0b246afa | 4493 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4494 | struct btrfs_path *path; |
5f39d397 | 4495 | struct extent_buffer *leaf; |
39279cc3 | 4496 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4497 | struct btrfs_key key; |
4498 | struct btrfs_key found_key; | |
39279cc3 | 4499 | u64 extent_start = 0; |
db94535d | 4500 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4501 | u64 extent_offset = 0; |
39279cc3 | 4502 | u64 item_end = 0; |
c1aa4575 | 4503 | u64 last_size = new_size; |
8082510e | 4504 | u32 found_type = (u8)-1; |
39279cc3 CM |
4505 | int found_extent; |
4506 | int del_item; | |
85e21bac CM |
4507 | int pending_del_nr = 0; |
4508 | int pending_del_slot = 0; | |
179e29e4 | 4509 | int extent_type = -1; |
8082510e YZ |
4510 | int ret; |
4511 | int err = 0; | |
4a0cc7ca | 4512 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4513 | u64 bytes_deleted = 0; |
1262133b JB |
4514 | bool be_nice = 0; |
4515 | bool should_throttle = 0; | |
28f75a0e | 4516 | bool should_end = 0; |
8082510e YZ |
4517 | |
4518 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4519 | |
28ed1345 CM |
4520 | /* |
4521 | * for non-free space inodes and ref cows, we want to back off from | |
4522 | * time to time | |
4523 | */ | |
70ddc553 | 4524 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 CM |
4525 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
4526 | be_nice = 1; | |
4527 | ||
0eb0e19c MF |
4528 | path = btrfs_alloc_path(); |
4529 | if (!path) | |
4530 | return -ENOMEM; | |
e4058b54 | 4531 | path->reada = READA_BACK; |
0eb0e19c | 4532 | |
5dc562c5 JB |
4533 | /* |
4534 | * We want to drop from the next block forward in case this new size is | |
4535 | * not block aligned since we will be keeping the last block of the | |
4536 | * extent just the way it is. | |
4537 | */ | |
27cdeb70 | 4538 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4539 | root == fs_info->tree_root) |
dcdbc059 | 4540 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4541 | fs_info->sectorsize), |
da17066c | 4542 | (u64)-1, 0); |
8082510e | 4543 | |
16cdcec7 MX |
4544 | /* |
4545 | * This function is also used to drop the items in the log tree before | |
4546 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4547 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4548 | * items. | |
4549 | */ | |
4550 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4551 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4552 | |
33345d01 | 4553 | key.objectid = ino; |
39279cc3 | 4554 | key.offset = (u64)-1; |
5f39d397 CM |
4555 | key.type = (u8)-1; |
4556 | ||
85e21bac | 4557 | search_again: |
28ed1345 CM |
4558 | /* |
4559 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4560 | * up a huge file in a single leaf. Most of the time that | |
4561 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4562 | */ | |
ee22184b | 4563 | if (be_nice && bytes_deleted > SZ_32M) { |
3a45bb20 | 4564 | if (btrfs_should_end_transaction(trans)) { |
28ed1345 CM |
4565 | err = -EAGAIN; |
4566 | goto error; | |
4567 | } | |
4568 | } | |
4569 | ||
4570 | ||
b9473439 | 4571 | path->leave_spinning = 1; |
85e21bac | 4572 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4573 | if (ret < 0) { |
4574 | err = ret; | |
4575 | goto out; | |
4576 | } | |
d397712b | 4577 | |
85e21bac | 4578 | if (ret > 0) { |
e02119d5 CM |
4579 | /* there are no items in the tree for us to truncate, we're |
4580 | * done | |
4581 | */ | |
8082510e YZ |
4582 | if (path->slots[0] == 0) |
4583 | goto out; | |
85e21bac CM |
4584 | path->slots[0]--; |
4585 | } | |
4586 | ||
d397712b | 4587 | while (1) { |
39279cc3 | 4588 | fi = NULL; |
5f39d397 CM |
4589 | leaf = path->nodes[0]; |
4590 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4591 | found_type = found_key.type; |
39279cc3 | 4592 | |
33345d01 | 4593 | if (found_key.objectid != ino) |
39279cc3 | 4594 | break; |
5f39d397 | 4595 | |
85e21bac | 4596 | if (found_type < min_type) |
39279cc3 CM |
4597 | break; |
4598 | ||
5f39d397 | 4599 | item_end = found_key.offset; |
39279cc3 | 4600 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4601 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4602 | struct btrfs_file_extent_item); |
179e29e4 CM |
4603 | extent_type = btrfs_file_extent_type(leaf, fi); |
4604 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4605 | item_end += |
db94535d | 4606 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4607 | |
4608 | trace_btrfs_truncate_show_fi_regular( | |
4609 | BTRFS_I(inode), leaf, fi, | |
4610 | found_key.offset); | |
179e29e4 | 4611 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4612 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4613 | path->slots[0], fi); |
09ed2f16 LB |
4614 | |
4615 | trace_btrfs_truncate_show_fi_inline( | |
4616 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4617 | found_key.offset); | |
39279cc3 | 4618 | } |
008630c1 | 4619 | item_end--; |
39279cc3 | 4620 | } |
8082510e YZ |
4621 | if (found_type > min_type) { |
4622 | del_item = 1; | |
4623 | } else { | |
76b42abb | 4624 | if (item_end < new_size) |
b888db2b | 4625 | break; |
8082510e YZ |
4626 | if (found_key.offset >= new_size) |
4627 | del_item = 1; | |
4628 | else | |
4629 | del_item = 0; | |
39279cc3 | 4630 | } |
39279cc3 | 4631 | found_extent = 0; |
39279cc3 | 4632 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4633 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4634 | goto delete; | |
4635 | ||
7f4f6e0a JB |
4636 | if (del_item) |
4637 | last_size = found_key.offset; | |
4638 | else | |
4639 | last_size = new_size; | |
4640 | ||
179e29e4 | 4641 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4642 | u64 num_dec; |
db94535d | 4643 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4644 | if (!del_item) { |
db94535d CM |
4645 | u64 orig_num_bytes = |
4646 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4647 | extent_num_bytes = ALIGN(new_size - |
4648 | found_key.offset, | |
0b246afa | 4649 | fs_info->sectorsize); |
db94535d CM |
4650 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4651 | extent_num_bytes); | |
4652 | num_dec = (orig_num_bytes - | |
9069218d | 4653 | extent_num_bytes); |
27cdeb70 MX |
4654 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4655 | &root->state) && | |
4656 | extent_start != 0) | |
a76a3cd4 | 4657 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4658 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4659 | } else { |
db94535d CM |
4660 | extent_num_bytes = |
4661 | btrfs_file_extent_disk_num_bytes(leaf, | |
4662 | fi); | |
5d4f98a2 YZ |
4663 | extent_offset = found_key.offset - |
4664 | btrfs_file_extent_offset(leaf, fi); | |
4665 | ||
39279cc3 | 4666 | /* FIXME blocksize != 4096 */ |
9069218d | 4667 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4668 | if (extent_start != 0) { |
4669 | found_extent = 1; | |
27cdeb70 MX |
4670 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4671 | &root->state)) | |
a76a3cd4 | 4672 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4673 | } |
39279cc3 | 4674 | } |
9069218d | 4675 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4676 | /* |
4677 | * we can't truncate inline items that have had | |
4678 | * special encodings | |
4679 | */ | |
4680 | if (!del_item && | |
c8b97818 CM |
4681 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4682 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4683 | |
4684 | /* | |
0305cd5f FM |
4685 | * Need to release path in order to truncate a |
4686 | * compressed extent. So delete any accumulated | |
4687 | * extent items so far. | |
514ac8ad | 4688 | */ |
0305cd5f FM |
4689 | if (btrfs_file_extent_compression(leaf, fi) != |
4690 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4691 | err = btrfs_del_items(trans, root, path, | |
4692 | pending_del_slot, | |
4693 | pending_del_nr); | |
4694 | if (err) { | |
4695 | btrfs_abort_transaction(trans, | |
0305cd5f FM |
4696 | err); |
4697 | goto error; | |
4698 | } | |
4699 | pending_del_nr = 0; | |
4700 | } | |
4701 | ||
4702 | err = truncate_inline_extent(inode, path, | |
4703 | &found_key, | |
4704 | item_end, | |
4705 | new_size); | |
4706 | if (err) { | |
66642832 | 4707 | btrfs_abort_transaction(trans, err); |
0305cd5f FM |
4708 | goto error; |
4709 | } | |
27cdeb70 MX |
4710 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4711 | &root->state)) { | |
0305cd5f | 4712 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4713 | } |
39279cc3 | 4714 | } |
179e29e4 | 4715 | delete: |
39279cc3 | 4716 | if (del_item) { |
85e21bac CM |
4717 | if (!pending_del_nr) { |
4718 | /* no pending yet, add ourselves */ | |
4719 | pending_del_slot = path->slots[0]; | |
4720 | pending_del_nr = 1; | |
4721 | } else if (pending_del_nr && | |
4722 | path->slots[0] + 1 == pending_del_slot) { | |
4723 | /* hop on the pending chunk */ | |
4724 | pending_del_nr++; | |
4725 | pending_del_slot = path->slots[0]; | |
4726 | } else { | |
d397712b | 4727 | BUG(); |
85e21bac | 4728 | } |
39279cc3 CM |
4729 | } else { |
4730 | break; | |
4731 | } | |
28f75a0e CM |
4732 | should_throttle = 0; |
4733 | ||
27cdeb70 MX |
4734 | if (found_extent && |
4735 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4736 | root == fs_info->tree_root)) { |
b9473439 | 4737 | btrfs_set_path_blocking(path); |
28ed1345 | 4738 | bytes_deleted += extent_num_bytes; |
2ff7e61e | 4739 | ret = btrfs_free_extent(trans, fs_info, extent_start, |
5d4f98a2 YZ |
4740 | extent_num_bytes, 0, |
4741 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4742 | ino, extent_offset); |
3bf16538 OS |
4743 | if (ret) { |
4744 | btrfs_abort_transaction(trans, ret); | |
4745 | break; | |
4746 | } | |
2ff7e61e JM |
4747 | if (btrfs_should_throttle_delayed_refs(trans, fs_info)) |
4748 | btrfs_async_run_delayed_refs(fs_info, | |
dd4b857a WX |
4749 | trans->delayed_ref_updates * 2, |
4750 | trans->transid, 0); | |
28f75a0e CM |
4751 | if (be_nice) { |
4752 | if (truncate_space_check(trans, root, | |
4753 | extent_num_bytes)) { | |
4754 | should_end = 1; | |
4755 | } | |
4756 | if (btrfs_should_throttle_delayed_refs(trans, | |
2ff7e61e | 4757 | fs_info)) |
28f75a0e | 4758 | should_throttle = 1; |
28f75a0e | 4759 | } |
39279cc3 | 4760 | } |
85e21bac | 4761 | |
8082510e YZ |
4762 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4763 | break; | |
4764 | ||
4765 | if (path->slots[0] == 0 || | |
1262133b | 4766 | path->slots[0] != pending_del_slot || |
28f75a0e | 4767 | should_throttle || should_end) { |
8082510e YZ |
4768 | if (pending_del_nr) { |
4769 | ret = btrfs_del_items(trans, root, path, | |
4770 | pending_del_slot, | |
4771 | pending_del_nr); | |
79787eaa | 4772 | if (ret) { |
66642832 | 4773 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4774 | goto error; |
4775 | } | |
8082510e YZ |
4776 | pending_del_nr = 0; |
4777 | } | |
b3b4aa74 | 4778 | btrfs_release_path(path); |
28f75a0e | 4779 | if (should_throttle) { |
1262133b JB |
4780 | unsigned long updates = trans->delayed_ref_updates; |
4781 | if (updates) { | |
4782 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4783 | ret = btrfs_run_delayed_refs(trans, |
4784 | fs_info, | |
4785 | updates * 2); | |
1262133b JB |
4786 | if (ret && !err) |
4787 | err = ret; | |
4788 | } | |
4789 | } | |
28f75a0e CM |
4790 | /* |
4791 | * if we failed to refill our space rsv, bail out | |
4792 | * and let the transaction restart | |
4793 | */ | |
4794 | if (should_end) { | |
4795 | err = -EAGAIN; | |
4796 | goto error; | |
4797 | } | |
85e21bac | 4798 | goto search_again; |
8082510e YZ |
4799 | } else { |
4800 | path->slots[0]--; | |
85e21bac | 4801 | } |
39279cc3 | 4802 | } |
8082510e | 4803 | out: |
85e21bac CM |
4804 | if (pending_del_nr) { |
4805 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4806 | pending_del_nr); | |
79787eaa | 4807 | if (ret) |
66642832 | 4808 | btrfs_abort_transaction(trans, ret); |
85e21bac | 4809 | } |
79787eaa | 4810 | error: |
76b42abb FM |
4811 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4812 | ASSERT(last_size >= new_size); | |
4813 | if (!err && last_size > new_size) | |
4814 | last_size = new_size; | |
7f4f6e0a | 4815 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4816 | } |
28ed1345 | 4817 | |
39279cc3 | 4818 | btrfs_free_path(path); |
28ed1345 | 4819 | |
ee22184b | 4820 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4821 | unsigned long updates = trans->delayed_ref_updates; |
4822 | if (updates) { | |
4823 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4824 | ret = btrfs_run_delayed_refs(trans, fs_info, |
4825 | updates * 2); | |
28ed1345 CM |
4826 | if (ret && !err) |
4827 | err = ret; | |
4828 | } | |
4829 | } | |
8082510e | 4830 | return err; |
39279cc3 CM |
4831 | } |
4832 | ||
4833 | /* | |
9703fefe | 4834 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4835 | * @inode - inode that we're zeroing |
4836 | * @from - the offset to start zeroing | |
4837 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4838 | * offset | |
4839 | * @front - zero up to the offset instead of from the offset on | |
4840 | * | |
9703fefe | 4841 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4842 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4843 | */ |
9703fefe | 4844 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4845 | int front) |
39279cc3 | 4846 | { |
0b246afa | 4847 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4848 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4849 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4850 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4851 | struct extent_state *cached_state = NULL; |
364ecf36 | 4852 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4853 | char *kaddr; |
0b246afa | 4854 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4855 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4856 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4857 | struct page *page; |
3b16a4e3 | 4858 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4859 | int ret = 0; |
9703fefe CR |
4860 | u64 block_start; |
4861 | u64 block_end; | |
39279cc3 | 4862 | |
2aaa6655 JB |
4863 | if ((offset & (blocksize - 1)) == 0 && |
4864 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4865 | goto out; |
9703fefe | 4866 | |
364ecf36 | 4867 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
9703fefe | 4868 | round_down(from, blocksize), blocksize); |
5d5e103a JB |
4869 | if (ret) |
4870 | goto out; | |
39279cc3 | 4871 | |
211c17f5 | 4872 | again: |
3b16a4e3 | 4873 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4874 | if (!page) { |
bc42bda2 | 4875 | btrfs_delalloc_release_space(inode, data_reserved, |
9703fefe CR |
4876 | round_down(from, blocksize), |
4877 | blocksize); | |
ac6a2b36 | 4878 | ret = -ENOMEM; |
39279cc3 | 4879 | goto out; |
5d5e103a | 4880 | } |
e6dcd2dc | 4881 | |
9703fefe CR |
4882 | block_start = round_down(from, blocksize); |
4883 | block_end = block_start + blocksize - 1; | |
e6dcd2dc | 4884 | |
39279cc3 | 4885 | if (!PageUptodate(page)) { |
9ebefb18 | 4886 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4887 | lock_page(page); |
211c17f5 CM |
4888 | if (page->mapping != mapping) { |
4889 | unlock_page(page); | |
09cbfeaf | 4890 | put_page(page); |
211c17f5 CM |
4891 | goto again; |
4892 | } | |
39279cc3 CM |
4893 | if (!PageUptodate(page)) { |
4894 | ret = -EIO; | |
89642229 | 4895 | goto out_unlock; |
39279cc3 CM |
4896 | } |
4897 | } | |
211c17f5 | 4898 | wait_on_page_writeback(page); |
e6dcd2dc | 4899 | |
9703fefe | 4900 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4901 | set_page_extent_mapped(page); |
4902 | ||
9703fefe | 4903 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4904 | if (ordered) { |
9703fefe | 4905 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4906 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4907 | unlock_page(page); |
09cbfeaf | 4908 | put_page(page); |
eb84ae03 | 4909 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4910 | btrfs_put_ordered_extent(ordered); |
4911 | goto again; | |
4912 | } | |
4913 | ||
9703fefe | 4914 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4915 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4916 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4917 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4918 | |
9703fefe | 4919 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, |
ba8b04c1 | 4920 | &cached_state, 0); |
9ed74f2d | 4921 | if (ret) { |
9703fefe | 4922 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4923 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4924 | goto out_unlock; |
4925 | } | |
4926 | ||
9703fefe | 4927 | if (offset != blocksize) { |
2aaa6655 | 4928 | if (!len) |
9703fefe | 4929 | len = blocksize - offset; |
e6dcd2dc | 4930 | kaddr = kmap(page); |
2aaa6655 | 4931 | if (front) |
9703fefe CR |
4932 | memset(kaddr + (block_start - page_offset(page)), |
4933 | 0, offset); | |
2aaa6655 | 4934 | else |
9703fefe CR |
4935 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4936 | 0, len); | |
e6dcd2dc CM |
4937 | flush_dcache_page(page); |
4938 | kunmap(page); | |
4939 | } | |
247e743c | 4940 | ClearPageChecked(page); |
e6dcd2dc | 4941 | set_page_dirty(page); |
9703fefe | 4942 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4943 | GFP_NOFS); |
39279cc3 | 4944 | |
89642229 | 4945 | out_unlock: |
5d5e103a | 4946 | if (ret) |
bc42bda2 | 4947 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
9703fefe | 4948 | blocksize); |
39279cc3 | 4949 | unlock_page(page); |
09cbfeaf | 4950 | put_page(page); |
39279cc3 | 4951 | out: |
364ecf36 | 4952 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4953 | return ret; |
4954 | } | |
4955 | ||
16e7549f JB |
4956 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4957 | u64 offset, u64 len) | |
4958 | { | |
0b246afa | 4959 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4960 | struct btrfs_trans_handle *trans; |
4961 | int ret; | |
4962 | ||
4963 | /* | |
4964 | * Still need to make sure the inode looks like it's been updated so | |
4965 | * that any holes get logged if we fsync. | |
4966 | */ | |
0b246afa JM |
4967 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4968 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4969 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4970 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4971 | return 0; | |
4972 | } | |
4973 | ||
4974 | /* | |
4975 | * 1 - for the one we're dropping | |
4976 | * 1 - for the one we're adding | |
4977 | * 1 - for updating the inode. | |
4978 | */ | |
4979 | trans = btrfs_start_transaction(root, 3); | |
4980 | if (IS_ERR(trans)) | |
4981 | return PTR_ERR(trans); | |
4982 | ||
4983 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4984 | if (ret) { | |
66642832 | 4985 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4986 | btrfs_end_transaction(trans); |
16e7549f JB |
4987 | return ret; |
4988 | } | |
4989 | ||
f85b7379 DS |
4990 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4991 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4992 | if (ret) |
66642832 | 4993 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4994 | else |
4995 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4996 | btrfs_end_transaction(trans); |
16e7549f JB |
4997 | return ret; |
4998 | } | |
4999 | ||
695a0d0d JB |
5000 | /* |
5001 | * This function puts in dummy file extents for the area we're creating a hole | |
5002 | * for. So if we are truncating this file to a larger size we need to insert | |
5003 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
5004 | * the range between oldsize and size | |
5005 | */ | |
a41ad394 | 5006 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 5007 | { |
0b246afa | 5008 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
5009 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5010 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 5011 | struct extent_map *em = NULL; |
2ac55d41 | 5012 | struct extent_state *cached_state = NULL; |
5dc562c5 | 5013 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
5014 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
5015 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
5016 | u64 last_byte; |
5017 | u64 cur_offset; | |
5018 | u64 hole_size; | |
9ed74f2d | 5019 | int err = 0; |
39279cc3 | 5020 | |
a71754fc | 5021 | /* |
9703fefe CR |
5022 | * If our size started in the middle of a block we need to zero out the |
5023 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
5024 | * expose stale data. |
5025 | */ | |
9703fefe | 5026 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
5027 | if (err) |
5028 | return err; | |
5029 | ||
9036c102 YZ |
5030 | if (size <= hole_start) |
5031 | return 0; | |
5032 | ||
9036c102 YZ |
5033 | while (1) { |
5034 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 5035 | |
ff13db41 | 5036 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 5037 | &cached_state); |
a776c6fa | 5038 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start, |
fa7c1494 | 5039 | block_end - hole_start); |
9036c102 YZ |
5040 | if (!ordered) |
5041 | break; | |
2ac55d41 JB |
5042 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
5043 | &cached_state, GFP_NOFS); | |
fa7c1494 | 5044 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
5045 | btrfs_put_ordered_extent(ordered); |
5046 | } | |
39279cc3 | 5047 | |
9036c102 YZ |
5048 | cur_offset = hole_start; |
5049 | while (1) { | |
fc4f21b1 | 5050 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 5051 | block_end - cur_offset, 0); |
79787eaa JM |
5052 | if (IS_ERR(em)) { |
5053 | err = PTR_ERR(em); | |
f2767956 | 5054 | em = NULL; |
79787eaa JM |
5055 | break; |
5056 | } | |
9036c102 | 5057 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 5058 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 5059 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 5060 | struct extent_map *hole_em; |
9036c102 | 5061 | hole_size = last_byte - cur_offset; |
9ed74f2d | 5062 | |
16e7549f JB |
5063 | err = maybe_insert_hole(root, inode, cur_offset, |
5064 | hole_size); | |
5065 | if (err) | |
3893e33b | 5066 | break; |
dcdbc059 | 5067 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
5068 | cur_offset + hole_size - 1, 0); |
5069 | hole_em = alloc_extent_map(); | |
5070 | if (!hole_em) { | |
5071 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
5072 | &BTRFS_I(inode)->runtime_flags); | |
5073 | goto next; | |
5074 | } | |
5075 | hole_em->start = cur_offset; | |
5076 | hole_em->len = hole_size; | |
5077 | hole_em->orig_start = cur_offset; | |
8082510e | 5078 | |
5dc562c5 JB |
5079 | hole_em->block_start = EXTENT_MAP_HOLE; |
5080 | hole_em->block_len = 0; | |
b4939680 | 5081 | hole_em->orig_block_len = 0; |
cc95bef6 | 5082 | hole_em->ram_bytes = hole_size; |
0b246afa | 5083 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 5084 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 5085 | hole_em->generation = fs_info->generation; |
8082510e | 5086 | |
5dc562c5 JB |
5087 | while (1) { |
5088 | write_lock(&em_tree->lock); | |
09a2a8f9 | 5089 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
5090 | write_unlock(&em_tree->lock); |
5091 | if (err != -EEXIST) | |
5092 | break; | |
dcdbc059 NB |
5093 | btrfs_drop_extent_cache(BTRFS_I(inode), |
5094 | cur_offset, | |
5dc562c5 JB |
5095 | cur_offset + |
5096 | hole_size - 1, 0); | |
5097 | } | |
5098 | free_extent_map(hole_em); | |
9036c102 | 5099 | } |
16e7549f | 5100 | next: |
9036c102 | 5101 | free_extent_map(em); |
a22285a6 | 5102 | em = NULL; |
9036c102 | 5103 | cur_offset = last_byte; |
8082510e | 5104 | if (cur_offset >= block_end) |
9036c102 YZ |
5105 | break; |
5106 | } | |
a22285a6 | 5107 | free_extent_map(em); |
2ac55d41 JB |
5108 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
5109 | GFP_NOFS); | |
9036c102 YZ |
5110 | return err; |
5111 | } | |
39279cc3 | 5112 | |
3972f260 | 5113 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5114 | { |
f4a2f4c5 MX |
5115 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5116 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5117 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5118 | loff_t newsize = attr->ia_size; |
5119 | int mask = attr->ia_valid; | |
8082510e YZ |
5120 | int ret; |
5121 | ||
3972f260 ES |
5122 | /* |
5123 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5124 | * special case where we need to update the times despite not having | |
5125 | * these flags set. For all other operations the VFS set these flags | |
5126 | * explicitly if it wants a timestamp update. | |
5127 | */ | |
dff6efc3 CH |
5128 | if (newsize != oldsize) { |
5129 | inode_inc_iversion(inode); | |
5130 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5131 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5132 | current_time(inode); |
dff6efc3 | 5133 | } |
3972f260 | 5134 | |
a41ad394 | 5135 | if (newsize > oldsize) { |
9ea24bbe | 5136 | /* |
ea14b57f | 5137 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5138 | * This is to ensure the snapshot captures a fully consistent |
5139 | * state of this file - if the snapshot captures this expanding | |
5140 | * truncation, it must capture all writes that happened before | |
5141 | * this truncation. | |
5142 | */ | |
0bc19f90 | 5143 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5144 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5145 | if (ret) { |
ea14b57f | 5146 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5147 | return ret; |
9ea24bbe | 5148 | } |
8082510e | 5149 | |
f4a2f4c5 | 5150 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5151 | if (IS_ERR(trans)) { |
ea14b57f | 5152 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5153 | return PTR_ERR(trans); |
9ea24bbe | 5154 | } |
f4a2f4c5 MX |
5155 | |
5156 | i_size_write(inode, newsize); | |
5157 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5158 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5159 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5160 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5161 | btrfs_end_transaction(trans); |
a41ad394 | 5162 | } else { |
8082510e | 5163 | |
a41ad394 JB |
5164 | /* |
5165 | * We're truncating a file that used to have good data down to | |
5166 | * zero. Make sure it gets into the ordered flush list so that | |
5167 | * any new writes get down to disk quickly. | |
5168 | */ | |
5169 | if (newsize == 0) | |
72ac3c0d JB |
5170 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5171 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5172 | |
f3fe820c JB |
5173 | /* |
5174 | * 1 for the orphan item we're going to add | |
5175 | * 1 for the orphan item deletion. | |
5176 | */ | |
5177 | trans = btrfs_start_transaction(root, 2); | |
5178 | if (IS_ERR(trans)) | |
5179 | return PTR_ERR(trans); | |
5180 | ||
5181 | /* | |
5182 | * We need to do this in case we fail at _any_ point during the | |
5183 | * actual truncate. Once we do the truncate_setsize we could | |
5184 | * invalidate pages which forces any outstanding ordered io to | |
5185 | * be instantly completed which will give us extents that need | |
5186 | * to be truncated. If we fail to get an orphan inode down we | |
5187 | * could have left over extents that were never meant to live, | |
01327610 | 5188 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
5189 | * will be consistent. |
5190 | */ | |
73f2e545 | 5191 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 5192 | btrfs_end_transaction(trans); |
f3fe820c JB |
5193 | if (ret) |
5194 | return ret; | |
5195 | ||
a41ad394 JB |
5196 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
5197 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
5198 | |
5199 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
abcefb1e | 5200 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5201 | inode_dio_wait(inode); |
0b581701 | 5202 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5203 | |
a41ad394 | 5204 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
5205 | if (ret && inode->i_nlink) { |
5206 | int err; | |
5207 | ||
19fd2df5 LB |
5208 | /* To get a stable disk_i_size */ |
5209 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5210 | if (err) { | |
3d6ae7bb | 5211 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
19fd2df5 LB |
5212 | return err; |
5213 | } | |
5214 | ||
7f4f6e0a JB |
5215 | /* |
5216 | * failed to truncate, disk_i_size is only adjusted down | |
5217 | * as we remove extents, so it should represent the true | |
5218 | * size of the inode, so reset the in memory size and | |
5219 | * delete our orphan entry. | |
5220 | */ | |
5221 | trans = btrfs_join_transaction(root); | |
5222 | if (IS_ERR(trans)) { | |
3d6ae7bb | 5223 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7f4f6e0a JB |
5224 | return ret; |
5225 | } | |
5226 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
3d6ae7bb | 5227 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
7f4f6e0a | 5228 | if (err) |
66642832 | 5229 | btrfs_abort_transaction(trans, err); |
3a45bb20 | 5230 | btrfs_end_transaction(trans); |
7f4f6e0a | 5231 | } |
8082510e YZ |
5232 | } |
5233 | ||
a41ad394 | 5234 | return ret; |
8082510e YZ |
5235 | } |
5236 | ||
9036c102 YZ |
5237 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5238 | { | |
2b0143b5 | 5239 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5240 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5241 | int err; |
39279cc3 | 5242 | |
b83cc969 LZ |
5243 | if (btrfs_root_readonly(root)) |
5244 | return -EROFS; | |
5245 | ||
31051c85 | 5246 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5247 | if (err) |
5248 | return err; | |
2bf5a725 | 5249 | |
5a3f23d5 | 5250 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5251 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5252 | if (err) |
5253 | return err; | |
39279cc3 | 5254 | } |
9036c102 | 5255 | |
1025774c CH |
5256 | if (attr->ia_valid) { |
5257 | setattr_copy(inode, attr); | |
0c4d2d95 | 5258 | inode_inc_iversion(inode); |
22c44fe6 | 5259 | err = btrfs_dirty_inode(inode); |
1025774c | 5260 | |
22c44fe6 | 5261 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5262 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5263 | } |
33268eaf | 5264 | |
39279cc3 CM |
5265 | return err; |
5266 | } | |
61295eb8 | 5267 | |
131e404a FDBM |
5268 | /* |
5269 | * While truncating the inode pages during eviction, we get the VFS calling | |
5270 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5271 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5272 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5273 | * extent_state structures over and over, wasting lots of time. | |
5274 | * | |
5275 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5276 | * those expensive operations on a per page basis and do only the ordered io | |
5277 | * finishing, while we release here the extent_map and extent_state structures, | |
5278 | * without the excessive merging and splitting. | |
5279 | */ | |
5280 | static void evict_inode_truncate_pages(struct inode *inode) | |
5281 | { | |
5282 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5283 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5284 | struct rb_node *node; | |
5285 | ||
5286 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5287 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5288 | |
5289 | write_lock(&map_tree->lock); | |
5290 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5291 | struct extent_map *em; | |
5292 | ||
5293 | node = rb_first(&map_tree->map); | |
5294 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5295 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5296 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5297 | remove_extent_mapping(map_tree, em); |
5298 | free_extent_map(em); | |
7064dd5c FM |
5299 | if (need_resched()) { |
5300 | write_unlock(&map_tree->lock); | |
5301 | cond_resched(); | |
5302 | write_lock(&map_tree->lock); | |
5303 | } | |
131e404a FDBM |
5304 | } |
5305 | write_unlock(&map_tree->lock); | |
5306 | ||
6ca07097 FM |
5307 | /* |
5308 | * Keep looping until we have no more ranges in the io tree. | |
5309 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5310 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5311 | * still in progress (unlocked the pages in the bio but did not yet | |
5312 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5313 | * ranges can still be locked and eviction started because before |
5314 | * submitting those bios, which are executed by a separate task (work | |
5315 | * queue kthread), inode references (inode->i_count) were not taken | |
5316 | * (which would be dropped in the end io callback of each bio). | |
5317 | * Therefore here we effectively end up waiting for those bios and | |
5318 | * anyone else holding locked ranges without having bumped the inode's | |
5319 | * reference count - if we don't do it, when they access the inode's | |
5320 | * io_tree to unlock a range it may be too late, leading to an | |
5321 | * use-after-free issue. | |
5322 | */ | |
131e404a FDBM |
5323 | spin_lock(&io_tree->lock); |
5324 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5325 | struct extent_state *state; | |
5326 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5327 | u64 start; |
5328 | u64 end; | |
131e404a FDBM |
5329 | |
5330 | node = rb_first(&io_tree->state); | |
5331 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5332 | start = state->start; |
5333 | end = state->end; | |
131e404a FDBM |
5334 | spin_unlock(&io_tree->lock); |
5335 | ||
ff13db41 | 5336 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5337 | |
5338 | /* | |
5339 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5340 | * and its reserved space won't be freed by delayed_ref. | |
5341 | * So we need to free its reserved space here. | |
5342 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5343 | * | |
5344 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5345 | */ | |
5346 | if (state->state & EXTENT_DELALLOC) | |
bc42bda2 | 5347 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5348 | |
6ca07097 | 5349 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5350 | EXTENT_LOCKED | EXTENT_DIRTY | |
5351 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5352 | EXTENT_DEFRAG, 1, 1, | |
5353 | &cached_state, GFP_NOFS); | |
131e404a | 5354 | |
7064dd5c | 5355 | cond_resched(); |
131e404a FDBM |
5356 | spin_lock(&io_tree->lock); |
5357 | } | |
5358 | spin_unlock(&io_tree->lock); | |
5359 | } | |
5360 | ||
bd555975 | 5361 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5362 | { |
0b246afa | 5363 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5364 | struct btrfs_trans_handle *trans; |
5365 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5366 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5367 | int steal_from_global = 0; |
3d48d981 | 5368 | u64 min_size; |
39279cc3 CM |
5369 | int ret; |
5370 | ||
1abe9b8a | 5371 | trace_btrfs_inode_evict(inode); |
5372 | ||
3d48d981 | 5373 | if (!root) { |
1371798b | 5374 | clear_inode(inode); |
3d48d981 NB |
5375 | return; |
5376 | } | |
5377 | ||
0b246afa | 5378 | min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
3d48d981 | 5379 | |
131e404a FDBM |
5380 | evict_inode_truncate_pages(inode); |
5381 | ||
69e9c6c6 SB |
5382 | if (inode->i_nlink && |
5383 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5384 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5385 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5386 | goto no_delete; |
5387 | ||
39279cc3 | 5388 | if (is_bad_inode(inode)) { |
3d6ae7bb | 5389 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
39279cc3 CM |
5390 | goto no_delete; |
5391 | } | |
bd555975 | 5392 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5393 | if (!special_file(inode->i_mode)) |
5394 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5395 | |
7ab7956e | 5396 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5397 | |
0b246afa | 5398 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
6bf02314 | 5399 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5400 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5401 | goto no_delete; |
5402 | } | |
5403 | ||
76dda93c | 5404 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5405 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5406 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5407 | goto no_delete; |
5408 | } | |
5409 | ||
aa79021f | 5410 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
0e8c36a9 | 5411 | if (ret) { |
3d6ae7bb | 5412 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
0e8c36a9 MX |
5413 | goto no_delete; |
5414 | } | |
5415 | ||
2ff7e61e | 5416 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
4289a667 | 5417 | if (!rsv) { |
3d6ae7bb | 5418 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
4289a667 JB |
5419 | goto no_delete; |
5420 | } | |
4a338542 | 5421 | rsv->size = min_size; |
ca7e70f5 | 5422 | rsv->failfast = 1; |
0b246afa | 5423 | global_rsv = &fs_info->global_block_rsv; |
4289a667 | 5424 | |
6ef06d27 | 5425 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5426 | |
4289a667 | 5427 | /* |
8407aa46 MX |
5428 | * This is a bit simpler than btrfs_truncate since we've already |
5429 | * reserved our space for our orphan item in the unlink, so we just | |
5430 | * need to reserve some slack space in case we add bytes and update | |
5431 | * inode item when doing the truncate. | |
4289a667 | 5432 | */ |
8082510e | 5433 | while (1) { |
08e007d2 MX |
5434 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5435 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5436 | |
5437 | /* | |
5438 | * Try and steal from the global reserve since we will | |
5439 | * likely not use this space anyway, we want to try as | |
5440 | * hard as possible to get this to work. | |
5441 | */ | |
5442 | if (ret) | |
3bce876f JB |
5443 | steal_from_global++; |
5444 | else | |
5445 | steal_from_global = 0; | |
5446 | ret = 0; | |
d68fc57b | 5447 | |
3bce876f JB |
5448 | /* |
5449 | * steal_from_global == 0: we reserved stuff, hooray! | |
5450 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5451 | * steal_from_global == 2: we've committed, still not a lot of | |
5452 | * room but maybe we'll have room in the global reserve this | |
5453 | * time. | |
5454 | * steal_from_global == 3: abandon all hope! | |
5455 | */ | |
5456 | if (steal_from_global > 2) { | |
0b246afa JM |
5457 | btrfs_warn(fs_info, |
5458 | "Could not get space for a delete, will truncate on mount %d", | |
5459 | ret); | |
3d6ae7bb | 5460 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5461 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5462 | goto no_delete; |
d68fc57b | 5463 | } |
7b128766 | 5464 | |
0e8c36a9 | 5465 | trans = btrfs_join_transaction(root); |
4289a667 | 5466 | if (IS_ERR(trans)) { |
3d6ae7bb | 5467 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5468 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5469 | goto no_delete; |
d68fc57b | 5470 | } |
7b128766 | 5471 | |
3bce876f | 5472 | /* |
01327610 | 5473 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5474 | * sure there is room to do it, if not we need to commit and try |
5475 | * again. | |
5476 | */ | |
5477 | if (steal_from_global) { | |
2ff7e61e | 5478 | if (!btrfs_check_space_for_delayed_refs(trans, fs_info)) |
3bce876f | 5479 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, |
25d609f8 | 5480 | min_size, 0); |
3bce876f JB |
5481 | else |
5482 | ret = -ENOSPC; | |
5483 | } | |
5484 | ||
5485 | /* | |
5486 | * Couldn't steal from the global reserve, we have too much | |
5487 | * pending stuff built up, commit the transaction and try it | |
5488 | * again. | |
5489 | */ | |
5490 | if (ret) { | |
3a45bb20 | 5491 | ret = btrfs_commit_transaction(trans); |
3bce876f | 5492 | if (ret) { |
3d6ae7bb | 5493 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5494 | btrfs_free_block_rsv(fs_info, rsv); |
3bce876f JB |
5495 | goto no_delete; |
5496 | } | |
5497 | continue; | |
5498 | } else { | |
5499 | steal_from_global = 0; | |
5500 | } | |
5501 | ||
4289a667 JB |
5502 | trans->block_rsv = rsv; |
5503 | ||
d68fc57b | 5504 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
ef61d940 OS |
5505 | if (ret) { |
5506 | trans->block_rsv = &fs_info->trans_block_rsv; | |
5507 | btrfs_end_transaction(trans); | |
5508 | btrfs_btree_balance_dirty(fs_info); | |
5509 | if (ret != -ENOSPC && ret != -EAGAIN) { | |
5510 | btrfs_orphan_del(NULL, BTRFS_I(inode)); | |
5511 | btrfs_free_block_rsv(fs_info, rsv); | |
5512 | goto no_delete; | |
5513 | } | |
5514 | } else { | |
8082510e | 5515 | break; |
ef61d940 | 5516 | } |
8082510e | 5517 | } |
5f39d397 | 5518 | |
2ff7e61e | 5519 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5520 | |
4ef31a45 JB |
5521 | /* |
5522 | * Errors here aren't a big deal, it just means we leave orphan items | |
5523 | * in the tree. They will be cleaned up on the next mount. | |
5524 | */ | |
ef61d940 OS |
5525 | trans->block_rsv = root->orphan_block_rsv; |
5526 | btrfs_orphan_del(trans, BTRFS_I(inode)); | |
54aa1f4d | 5527 | |
0b246afa JM |
5528 | trans->block_rsv = &fs_info->trans_block_rsv; |
5529 | if (!(root == fs_info->tree_root || | |
581bb050 | 5530 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5531 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5532 | |
3a45bb20 | 5533 | btrfs_end_transaction(trans); |
2ff7e61e | 5534 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 5535 | no_delete: |
f48d1cf5 | 5536 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5537 | clear_inode(inode); |
39279cc3 CM |
5538 | } |
5539 | ||
5540 | /* | |
5541 | * this returns the key found in the dir entry in the location pointer. | |
5542 | * If no dir entries were found, location->objectid is 0. | |
5543 | */ | |
5544 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5545 | struct btrfs_key *location) | |
5546 | { | |
5547 | const char *name = dentry->d_name.name; | |
5548 | int namelen = dentry->d_name.len; | |
5549 | struct btrfs_dir_item *di; | |
5550 | struct btrfs_path *path; | |
5551 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5552 | int ret = 0; |
39279cc3 CM |
5553 | |
5554 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5555 | if (!path) |
5556 | return -ENOMEM; | |
3954401f | 5557 | |
f85b7379 DS |
5558 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5559 | name, namelen, 0); | |
0d9f7f3e Y |
5560 | if (IS_ERR(di)) |
5561 | ret = PTR_ERR(di); | |
d397712b | 5562 | |
c704005d | 5563 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5564 | goto out_err; |
d397712b | 5565 | |
5f39d397 | 5566 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
71341a8a LB |
5567 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5568 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
5569 | btrfs_warn(root->fs_info, | |
5570 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5571 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5572 | location->objectid, location->type, location->offset); | |
5573 | goto out_err; | |
5574 | } | |
39279cc3 | 5575 | out: |
39279cc3 CM |
5576 | btrfs_free_path(path); |
5577 | return ret; | |
3954401f CM |
5578 | out_err: |
5579 | location->objectid = 0; | |
5580 | goto out; | |
39279cc3 CM |
5581 | } |
5582 | ||
5583 | /* | |
5584 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5585 | * needs to be changed to reflect the root directory of the tree root. This | |
5586 | * is kind of like crossing a mount point. | |
5587 | */ | |
2ff7e61e | 5588 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5589 | struct inode *dir, |
5590 | struct dentry *dentry, | |
5591 | struct btrfs_key *location, | |
5592 | struct btrfs_root **sub_root) | |
39279cc3 | 5593 | { |
4df27c4d YZ |
5594 | struct btrfs_path *path; |
5595 | struct btrfs_root *new_root; | |
5596 | struct btrfs_root_ref *ref; | |
5597 | struct extent_buffer *leaf; | |
1d4c08e0 | 5598 | struct btrfs_key key; |
4df27c4d YZ |
5599 | int ret; |
5600 | int err = 0; | |
39279cc3 | 5601 | |
4df27c4d YZ |
5602 | path = btrfs_alloc_path(); |
5603 | if (!path) { | |
5604 | err = -ENOMEM; | |
5605 | goto out; | |
5606 | } | |
39279cc3 | 5607 | |
4df27c4d | 5608 | err = -ENOENT; |
1d4c08e0 DS |
5609 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5610 | key.type = BTRFS_ROOT_REF_KEY; | |
5611 | key.offset = location->objectid; | |
5612 | ||
0b246afa | 5613 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5614 | if (ret) { |
5615 | if (ret < 0) | |
5616 | err = ret; | |
5617 | goto out; | |
5618 | } | |
39279cc3 | 5619 | |
4df27c4d YZ |
5620 | leaf = path->nodes[0]; |
5621 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5622 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5623 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5624 | goto out; | |
39279cc3 | 5625 | |
4df27c4d YZ |
5626 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5627 | (unsigned long)(ref + 1), | |
5628 | dentry->d_name.len); | |
5629 | if (ret) | |
5630 | goto out; | |
5631 | ||
b3b4aa74 | 5632 | btrfs_release_path(path); |
4df27c4d | 5633 | |
0b246afa | 5634 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5635 | if (IS_ERR(new_root)) { |
5636 | err = PTR_ERR(new_root); | |
5637 | goto out; | |
5638 | } | |
5639 | ||
4df27c4d YZ |
5640 | *sub_root = new_root; |
5641 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5642 | location->type = BTRFS_INODE_ITEM_KEY; | |
5643 | location->offset = 0; | |
5644 | err = 0; | |
5645 | out: | |
5646 | btrfs_free_path(path); | |
5647 | return err; | |
39279cc3 CM |
5648 | } |
5649 | ||
5d4f98a2 YZ |
5650 | static void inode_tree_add(struct inode *inode) |
5651 | { | |
5652 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5653 | struct btrfs_inode *entry; | |
03e860bd FNP |
5654 | struct rb_node **p; |
5655 | struct rb_node *parent; | |
cef21937 | 5656 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5657 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5658 | |
1d3382cb | 5659 | if (inode_unhashed(inode)) |
76dda93c | 5660 | return; |
e1409cef | 5661 | parent = NULL; |
5d4f98a2 | 5662 | spin_lock(&root->inode_lock); |
e1409cef | 5663 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5664 | while (*p) { |
5665 | parent = *p; | |
5666 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5667 | ||
4a0cc7ca | 5668 | if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5669 | p = &parent->rb_left; |
4a0cc7ca | 5670 | else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5671 | p = &parent->rb_right; |
5d4f98a2 YZ |
5672 | else { |
5673 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5674 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5675 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5676 | RB_CLEAR_NODE(parent); |
5677 | spin_unlock(&root->inode_lock); | |
cef21937 | 5678 | return; |
5d4f98a2 YZ |
5679 | } |
5680 | } | |
cef21937 FDBM |
5681 | rb_link_node(new, parent, p); |
5682 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5683 | spin_unlock(&root->inode_lock); |
5684 | } | |
5685 | ||
5686 | static void inode_tree_del(struct inode *inode) | |
5687 | { | |
0b246afa | 5688 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5689 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5690 | int empty = 0; |
5d4f98a2 | 5691 | |
03e860bd | 5692 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5693 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5694 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5695 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5696 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5697 | } |
03e860bd | 5698 | spin_unlock(&root->inode_lock); |
76dda93c | 5699 | |
69e9c6c6 | 5700 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5701 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5702 | spin_lock(&root->inode_lock); |
5703 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5704 | spin_unlock(&root->inode_lock); | |
5705 | if (empty) | |
5706 | btrfs_add_dead_root(root); | |
5707 | } | |
5708 | } | |
5709 | ||
143bede5 | 5710 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c | 5711 | { |
0b246afa | 5712 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c YZ |
5713 | struct rb_node *node; |
5714 | struct rb_node *prev; | |
5715 | struct btrfs_inode *entry; | |
5716 | struct inode *inode; | |
5717 | u64 objectid = 0; | |
5718 | ||
0b246afa | 5719 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
7813b3db | 5720 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
76dda93c YZ |
5721 | |
5722 | spin_lock(&root->inode_lock); | |
5723 | again: | |
5724 | node = root->inode_tree.rb_node; | |
5725 | prev = NULL; | |
5726 | while (node) { | |
5727 | prev = node; | |
5728 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5729 | ||
4a0cc7ca | 5730 | if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c | 5731 | node = node->rb_left; |
4a0cc7ca | 5732 | else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c YZ |
5733 | node = node->rb_right; |
5734 | else | |
5735 | break; | |
5736 | } | |
5737 | if (!node) { | |
5738 | while (prev) { | |
5739 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5740 | if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) { |
76dda93c YZ |
5741 | node = prev; |
5742 | break; | |
5743 | } | |
5744 | prev = rb_next(prev); | |
5745 | } | |
5746 | } | |
5747 | while (node) { | |
5748 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5749 | objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1; |
76dda93c YZ |
5750 | inode = igrab(&entry->vfs_inode); |
5751 | if (inode) { | |
5752 | spin_unlock(&root->inode_lock); | |
5753 | if (atomic_read(&inode->i_count) > 1) | |
5754 | d_prune_aliases(inode); | |
5755 | /* | |
45321ac5 | 5756 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5757 | * the inode cache when its usage count |
5758 | * hits zero. | |
5759 | */ | |
5760 | iput(inode); | |
5761 | cond_resched(); | |
5762 | spin_lock(&root->inode_lock); | |
5763 | goto again; | |
5764 | } | |
5765 | ||
5766 | if (cond_resched_lock(&root->inode_lock)) | |
5767 | goto again; | |
5768 | ||
5769 | node = rb_next(node); | |
5770 | } | |
5771 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5772 | } |
5773 | ||
e02119d5 CM |
5774 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5775 | { | |
5776 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5777 | inode->i_ino = args->location->objectid; |
5778 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5779 | sizeof(*args->location)); | |
e02119d5 | 5780 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5781 | return 0; |
5782 | } | |
5783 | ||
5784 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5785 | { | |
5786 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5787 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5788 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5789 | } |
5790 | ||
5d4f98a2 | 5791 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5792 | struct btrfs_key *location, |
5d4f98a2 | 5793 | struct btrfs_root *root) |
39279cc3 CM |
5794 | { |
5795 | struct inode *inode; | |
5796 | struct btrfs_iget_args args; | |
90d3e592 | 5797 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5798 | |
90d3e592 | 5799 | args.location = location; |
39279cc3 CM |
5800 | args.root = root; |
5801 | ||
778ba82b | 5802 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5803 | btrfs_init_locked_inode, |
5804 | (void *)&args); | |
5805 | return inode; | |
5806 | } | |
5807 | ||
1a54ef8c BR |
5808 | /* Get an inode object given its location and corresponding root. |
5809 | * Returns in *is_new if the inode was read from disk | |
5810 | */ | |
5811 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5812 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5813 | { |
5814 | struct inode *inode; | |
5815 | ||
90d3e592 | 5816 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5817 | if (!inode) |
5d4f98a2 | 5818 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5819 | |
5820 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5821 | int ret; |
5822 | ||
5823 | ret = btrfs_read_locked_inode(inode); | |
1748f843 MF |
5824 | if (!is_bad_inode(inode)) { |
5825 | inode_tree_add(inode); | |
5826 | unlock_new_inode(inode); | |
5827 | if (new) | |
5828 | *new = 1; | |
5829 | } else { | |
e0b6d65b ST |
5830 | unlock_new_inode(inode); |
5831 | iput(inode); | |
67710892 FM |
5832 | ASSERT(ret < 0); |
5833 | inode = ERR_PTR(ret < 0 ? ret : -ESTALE); | |
1748f843 MF |
5834 | } |
5835 | } | |
5836 | ||
1a54ef8c BR |
5837 | return inode; |
5838 | } | |
5839 | ||
4df27c4d YZ |
5840 | static struct inode *new_simple_dir(struct super_block *s, |
5841 | struct btrfs_key *key, | |
5842 | struct btrfs_root *root) | |
5843 | { | |
5844 | struct inode *inode = new_inode(s); | |
5845 | ||
5846 | if (!inode) | |
5847 | return ERR_PTR(-ENOMEM); | |
5848 | ||
4df27c4d YZ |
5849 | BTRFS_I(inode)->root = root; |
5850 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5851 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5852 | |
5853 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5854 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5855 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5856 | inode->i_fop = &simple_dir_operations; |
5857 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5858 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5859 | inode->i_atime = inode->i_mtime; |
5860 | inode->i_ctime = inode->i_mtime; | |
5861 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5862 | |
5863 | return inode; | |
5864 | } | |
5865 | ||
3de4586c | 5866 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5867 | { |
0b246afa | 5868 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5869 | struct inode *inode; |
4df27c4d | 5870 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5871 | struct btrfs_root *sub_root = root; |
5872 | struct btrfs_key location; | |
76dda93c | 5873 | int index; |
b4aff1f8 | 5874 | int ret = 0; |
39279cc3 CM |
5875 | |
5876 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5877 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5878 | |
39e3c955 | 5879 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5880 | if (ret < 0) |
5881 | return ERR_PTR(ret); | |
5f39d397 | 5882 | |
4df27c4d | 5883 | if (location.objectid == 0) |
5662344b | 5884 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5885 | |
5886 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5887 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5888 | return inode; |
5889 | } | |
5890 | ||
0b246afa | 5891 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5892 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5893 | &location, &sub_root); |
5894 | if (ret < 0) { | |
5895 | if (ret != -ENOENT) | |
5896 | inode = ERR_PTR(ret); | |
5897 | else | |
5898 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5899 | } else { | |
73f73415 | 5900 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5901 | } |
0b246afa | 5902 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5903 | |
34d19bad | 5904 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5905 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5906 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5907 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5908 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5909 | if (ret) { |
5910 | iput(inode); | |
66b4ffd1 | 5911 | inode = ERR_PTR(ret); |
01cd3367 | 5912 | } |
c71bf099 YZ |
5913 | } |
5914 | ||
3de4586c CM |
5915 | return inode; |
5916 | } | |
5917 | ||
fe15ce44 | 5918 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5919 | { |
5920 | struct btrfs_root *root; | |
2b0143b5 | 5921 | struct inode *inode = d_inode(dentry); |
76dda93c | 5922 | |
848cce0d | 5923 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5924 | inode = d_inode(dentry->d_parent); |
76dda93c | 5925 | |
848cce0d LZ |
5926 | if (inode) { |
5927 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5928 | if (btrfs_root_refs(&root->root_item) == 0) |
5929 | return 1; | |
848cce0d | 5930 | |
4a0cc7ca | 5931 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5932 | return 1; |
efefb143 | 5933 | } |
76dda93c YZ |
5934 | return 0; |
5935 | } | |
5936 | ||
b4aff1f8 JB |
5937 | static void btrfs_dentry_release(struct dentry *dentry) |
5938 | { | |
944a4515 | 5939 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5940 | } |
5941 | ||
3de4586c | 5942 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5943 | unsigned int flags) |
3de4586c | 5944 | { |
5662344b | 5945 | struct inode *inode; |
a66e7cc6 | 5946 | |
5662344b TI |
5947 | inode = btrfs_lookup_dentry(dir, dentry); |
5948 | if (IS_ERR(inode)) { | |
5949 | if (PTR_ERR(inode) == -ENOENT) | |
5950 | inode = NULL; | |
5951 | else | |
5952 | return ERR_CAST(inode); | |
5953 | } | |
5954 | ||
41d28bca | 5955 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5956 | } |
5957 | ||
16cdcec7 | 5958 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5959 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5960 | }; | |
5961 | ||
23b5ec74 JB |
5962 | /* |
5963 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5964 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5965 | * our information into that, and then dir_emit from the buffer. This is | |
5966 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5967 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5968 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5969 | * tree lock. | |
5970 | */ | |
5971 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5972 | { | |
5973 | struct btrfs_file_private *private; | |
5974 | ||
5975 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5976 | if (!private) | |
5977 | return -ENOMEM; | |
5978 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5979 | if (!private->filldir_buf) { | |
5980 | kfree(private); | |
5981 | return -ENOMEM; | |
5982 | } | |
5983 | file->private_data = private; | |
5984 | return 0; | |
5985 | } | |
5986 | ||
5987 | struct dir_entry { | |
5988 | u64 ino; | |
5989 | u64 offset; | |
5990 | unsigned type; | |
5991 | int name_len; | |
5992 | }; | |
5993 | ||
5994 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5995 | { | |
5996 | while (entries--) { | |
5997 | struct dir_entry *entry = addr; | |
5998 | char *name = (char *)(entry + 1); | |
5999 | ||
f6edc45e DS |
6000 | ctx->pos = get_unaligned(&entry->offset); |
6001 | if (!dir_emit(ctx, name, get_unaligned(&entry->name_len), | |
6002 | get_unaligned(&entry->ino), | |
6003 | get_unaligned(&entry->type))) | |
23b5ec74 | 6004 | return 1; |
f6edc45e DS |
6005 | addr += sizeof(struct dir_entry) + |
6006 | get_unaligned(&entry->name_len); | |
23b5ec74 JB |
6007 | ctx->pos++; |
6008 | } | |
6009 | return 0; | |
6010 | } | |
6011 | ||
9cdda8d3 | 6012 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 6013 | { |
9cdda8d3 | 6014 | struct inode *inode = file_inode(file); |
2ff7e61e | 6015 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 6016 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 6017 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
6018 | struct btrfs_dir_item *di; |
6019 | struct btrfs_key key; | |
5f39d397 | 6020 | struct btrfs_key found_key; |
39279cc3 | 6021 | struct btrfs_path *path; |
23b5ec74 | 6022 | void *addr; |
16cdcec7 MX |
6023 | struct list_head ins_list; |
6024 | struct list_head del_list; | |
39279cc3 | 6025 | int ret; |
5f39d397 | 6026 | struct extent_buffer *leaf; |
39279cc3 | 6027 | int slot; |
5f39d397 CM |
6028 | char *name_ptr; |
6029 | int name_len; | |
23b5ec74 JB |
6030 | int entries = 0; |
6031 | int total_len = 0; | |
02dbfc99 | 6032 | bool put = false; |
c2951f32 | 6033 | struct btrfs_key location; |
5f39d397 | 6034 | |
9cdda8d3 AV |
6035 | if (!dir_emit_dots(file, ctx)) |
6036 | return 0; | |
6037 | ||
49593bfa | 6038 | path = btrfs_alloc_path(); |
16cdcec7 MX |
6039 | if (!path) |
6040 | return -ENOMEM; | |
ff5714cc | 6041 | |
23b5ec74 | 6042 | addr = private->filldir_buf; |
e4058b54 | 6043 | path->reada = READA_FORWARD; |
49593bfa | 6044 | |
c2951f32 JM |
6045 | INIT_LIST_HEAD(&ins_list); |
6046 | INIT_LIST_HEAD(&del_list); | |
6047 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 6048 | |
23b5ec74 | 6049 | again: |
c2951f32 | 6050 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 6051 | key.offset = ctx->pos; |
4a0cc7ca | 6052 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 6053 | |
39279cc3 CM |
6054 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6055 | if (ret < 0) | |
6056 | goto err; | |
49593bfa DW |
6057 | |
6058 | while (1) { | |
23b5ec74 JB |
6059 | struct dir_entry *entry; |
6060 | ||
5f39d397 | 6061 | leaf = path->nodes[0]; |
39279cc3 | 6062 | slot = path->slots[0]; |
b9e03af0 LZ |
6063 | if (slot >= btrfs_header_nritems(leaf)) { |
6064 | ret = btrfs_next_leaf(root, path); | |
6065 | if (ret < 0) | |
6066 | goto err; | |
6067 | else if (ret > 0) | |
6068 | break; | |
6069 | continue; | |
39279cc3 | 6070 | } |
3de4586c | 6071 | |
5f39d397 CM |
6072 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
6073 | ||
6074 | if (found_key.objectid != key.objectid) | |
39279cc3 | 6075 | break; |
c2951f32 | 6076 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 6077 | break; |
9cdda8d3 | 6078 | if (found_key.offset < ctx->pos) |
b9e03af0 | 6079 | goto next; |
c2951f32 | 6080 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 6081 | goto next; |
39279cc3 | 6082 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
e79a3327 | 6083 | if (verify_dir_item(fs_info, leaf, slot, di)) |
c2951f32 | 6084 | goto next; |
22a94d44 | 6085 | |
c2951f32 | 6086 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
6087 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
6088 | PAGE_SIZE) { | |
6089 | btrfs_release_path(path); | |
6090 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6091 | if (ret) | |
6092 | goto nopos; | |
6093 | addr = private->filldir_buf; | |
6094 | entries = 0; | |
6095 | total_len = 0; | |
6096 | goto again; | |
c2951f32 | 6097 | } |
23b5ec74 JB |
6098 | |
6099 | entry = addr; | |
f6edc45e | 6100 | put_unaligned(name_len, &entry->name_len); |
23b5ec74 | 6101 | name_ptr = (char *)(entry + 1); |
c2951f32 JM |
6102 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
6103 | name_len); | |
f6edc45e DS |
6104 | put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)], |
6105 | &entry->type); | |
c2951f32 | 6106 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
f6edc45e DS |
6107 | put_unaligned(location.objectid, &entry->ino); |
6108 | put_unaligned(found_key.offset, &entry->offset); | |
23b5ec74 JB |
6109 | entries++; |
6110 | addr += sizeof(struct dir_entry) + name_len; | |
6111 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
6112 | next: |
6113 | path->slots[0]++; | |
39279cc3 | 6114 | } |
23b5ec74 JB |
6115 | btrfs_release_path(path); |
6116 | ||
6117 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6118 | if (ret) | |
6119 | goto nopos; | |
49593bfa | 6120 | |
d2fbb2b5 | 6121 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6122 | if (ret) |
bc4ef759 DS |
6123 | goto nopos; |
6124 | ||
db62efbb ZB |
6125 | /* |
6126 | * Stop new entries from being returned after we return the last | |
6127 | * entry. | |
6128 | * | |
6129 | * New directory entries are assigned a strictly increasing | |
6130 | * offset. This means that new entries created during readdir | |
6131 | * are *guaranteed* to be seen in the future by that readdir. | |
6132 | * This has broken buggy programs which operate on names as | |
6133 | * they're returned by readdir. Until we re-use freed offsets | |
6134 | * we have this hack to stop new entries from being returned | |
6135 | * under the assumption that they'll never reach this huge | |
6136 | * offset. | |
6137 | * | |
6138 | * This is being careful not to overflow 32bit loff_t unless the | |
6139 | * last entry requires it because doing so has broken 32bit apps | |
6140 | * in the past. | |
6141 | */ | |
c2951f32 JM |
6142 | if (ctx->pos >= INT_MAX) |
6143 | ctx->pos = LLONG_MAX; | |
6144 | else | |
6145 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6146 | nopos: |
6147 | ret = 0; | |
6148 | err: | |
02dbfc99 OS |
6149 | if (put) |
6150 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6151 | btrfs_free_path(path); |
39279cc3 CM |
6152 | return ret; |
6153 | } | |
6154 | ||
39279cc3 | 6155 | /* |
54aa1f4d | 6156 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6157 | * inode changes. But, it is most likely to find the inode in cache. |
6158 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6159 | * to keep or drop this code. | |
6160 | */ | |
48a3b636 | 6161 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6162 | { |
2ff7e61e | 6163 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6164 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6165 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6166 | int ret; |
6167 | ||
72ac3c0d | 6168 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6169 | return 0; |
39279cc3 | 6170 | |
7a7eaa40 | 6171 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6172 | if (IS_ERR(trans)) |
6173 | return PTR_ERR(trans); | |
8929ecfa YZ |
6174 | |
6175 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6176 | if (ret && ret == -ENOSPC) { |
6177 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6178 | btrfs_end_transaction(trans); |
94b60442 | 6179 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6180 | if (IS_ERR(trans)) |
6181 | return PTR_ERR(trans); | |
8929ecfa | 6182 | |
94b60442 | 6183 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6184 | } |
3a45bb20 | 6185 | btrfs_end_transaction(trans); |
16cdcec7 | 6186 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6187 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6188 | |
6189 | return ret; | |
6190 | } | |
6191 | ||
6192 | /* | |
6193 | * This is a copy of file_update_time. We need this so we can return error on | |
6194 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6195 | */ | |
e41f941a JB |
6196 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
6197 | int flags) | |
22c44fe6 | 6198 | { |
2bc55652 AB |
6199 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6200 | ||
6201 | if (btrfs_root_readonly(root)) | |
6202 | return -EROFS; | |
6203 | ||
e41f941a | 6204 | if (flags & S_VERSION) |
22c44fe6 | 6205 | inode_inc_iversion(inode); |
e41f941a JB |
6206 | if (flags & S_CTIME) |
6207 | inode->i_ctime = *now; | |
6208 | if (flags & S_MTIME) | |
6209 | inode->i_mtime = *now; | |
6210 | if (flags & S_ATIME) | |
6211 | inode->i_atime = *now; | |
6212 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
6213 | } |
6214 | ||
d352ac68 CM |
6215 | /* |
6216 | * find the highest existing sequence number in a directory | |
6217 | * and then set the in-memory index_cnt variable to reflect | |
6218 | * free sequence numbers | |
6219 | */ | |
4c570655 | 6220 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6221 | { |
4c570655 | 6222 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6223 | struct btrfs_key key, found_key; |
6224 | struct btrfs_path *path; | |
6225 | struct extent_buffer *leaf; | |
6226 | int ret; | |
6227 | ||
4c570655 | 6228 | key.objectid = btrfs_ino(inode); |
962a298f | 6229 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6230 | key.offset = (u64)-1; |
6231 | ||
6232 | path = btrfs_alloc_path(); | |
6233 | if (!path) | |
6234 | return -ENOMEM; | |
6235 | ||
6236 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6237 | if (ret < 0) | |
6238 | goto out; | |
6239 | /* FIXME: we should be able to handle this */ | |
6240 | if (ret == 0) | |
6241 | goto out; | |
6242 | ret = 0; | |
6243 | ||
6244 | /* | |
6245 | * MAGIC NUMBER EXPLANATION: | |
6246 | * since we search a directory based on f_pos we have to start at 2 | |
6247 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6248 | * else has to start at 2 | |
6249 | */ | |
6250 | if (path->slots[0] == 0) { | |
4c570655 | 6251 | inode->index_cnt = 2; |
aec7477b JB |
6252 | goto out; |
6253 | } | |
6254 | ||
6255 | path->slots[0]--; | |
6256 | ||
6257 | leaf = path->nodes[0]; | |
6258 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6259 | ||
4c570655 | 6260 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6261 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6262 | inode->index_cnt = 2; |
aec7477b JB |
6263 | goto out; |
6264 | } | |
6265 | ||
4c570655 | 6266 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6267 | out: |
6268 | btrfs_free_path(path); | |
6269 | return ret; | |
6270 | } | |
6271 | ||
d352ac68 CM |
6272 | /* |
6273 | * helper to find a free sequence number in a given directory. This current | |
6274 | * code is very simple, later versions will do smarter things in the btree | |
6275 | */ | |
877574e2 | 6276 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6277 | { |
6278 | int ret = 0; | |
6279 | ||
877574e2 NB |
6280 | if (dir->index_cnt == (u64)-1) { |
6281 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6282 | if (ret) { |
6283 | ret = btrfs_set_inode_index_count(dir); | |
6284 | if (ret) | |
6285 | return ret; | |
6286 | } | |
aec7477b JB |
6287 | } |
6288 | ||
877574e2 NB |
6289 | *index = dir->index_cnt; |
6290 | dir->index_cnt++; | |
aec7477b JB |
6291 | |
6292 | return ret; | |
6293 | } | |
6294 | ||
b0d5d10f CM |
6295 | static int btrfs_insert_inode_locked(struct inode *inode) |
6296 | { | |
6297 | struct btrfs_iget_args args; | |
6298 | args.location = &BTRFS_I(inode)->location; | |
6299 | args.root = BTRFS_I(inode)->root; | |
6300 | ||
6301 | return insert_inode_locked4(inode, | |
6302 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6303 | btrfs_find_actor, &args); | |
6304 | } | |
6305 | ||
19aee8de AJ |
6306 | /* |
6307 | * Inherit flags from the parent inode. | |
6308 | * | |
6309 | * Currently only the compression flags and the cow flags are inherited. | |
6310 | */ | |
6311 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6312 | { | |
6313 | unsigned int flags; | |
6314 | ||
6315 | if (!dir) | |
6316 | return; | |
6317 | ||
6318 | flags = BTRFS_I(dir)->flags; | |
6319 | ||
6320 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6321 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6322 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6323 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6324 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6325 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6326 | } | |
6327 | ||
6328 | if (flags & BTRFS_INODE_NODATACOW) { | |
6329 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6330 | if (S_ISREG(inode->i_mode)) | |
6331 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6332 | } | |
6333 | ||
6334 | btrfs_update_iflags(inode); | |
6335 | } | |
6336 | ||
39279cc3 CM |
6337 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6338 | struct btrfs_root *root, | |
aec7477b | 6339 | struct inode *dir, |
9c58309d | 6340 | const char *name, int name_len, |
175a4eb7 AV |
6341 | u64 ref_objectid, u64 objectid, |
6342 | umode_t mode, u64 *index) | |
39279cc3 | 6343 | { |
0b246afa | 6344 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6345 | struct inode *inode; |
5f39d397 | 6346 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6347 | struct btrfs_key *location; |
5f39d397 | 6348 | struct btrfs_path *path; |
9c58309d CM |
6349 | struct btrfs_inode_ref *ref; |
6350 | struct btrfs_key key[2]; | |
6351 | u32 sizes[2]; | |
ef3b9af5 | 6352 | int nitems = name ? 2 : 1; |
9c58309d | 6353 | unsigned long ptr; |
39279cc3 | 6354 | int ret; |
39279cc3 | 6355 | |
5f39d397 | 6356 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6357 | if (!path) |
6358 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6359 | |
0b246afa | 6360 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6361 | if (!inode) { |
6362 | btrfs_free_path(path); | |
39279cc3 | 6363 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6364 | } |
39279cc3 | 6365 | |
5762b5c9 FM |
6366 | /* |
6367 | * O_TMPFILE, set link count to 0, so that after this point, | |
6368 | * we fill in an inode item with the correct link count. | |
6369 | */ | |
6370 | if (!name) | |
6371 | set_nlink(inode, 0); | |
6372 | ||
581bb050 LZ |
6373 | /* |
6374 | * we have to initialize this early, so we can reclaim the inode | |
6375 | * number if we fail afterwards in this function. | |
6376 | */ | |
6377 | inode->i_ino = objectid; | |
6378 | ||
ef3b9af5 | 6379 | if (dir && name) { |
1abe9b8a | 6380 | trace_btrfs_inode_request(dir); |
6381 | ||
877574e2 | 6382 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6383 | if (ret) { |
8fb27640 | 6384 | btrfs_free_path(path); |
09771430 | 6385 | iput(inode); |
aec7477b | 6386 | return ERR_PTR(ret); |
09771430 | 6387 | } |
ef3b9af5 FM |
6388 | } else if (dir) { |
6389 | *index = 0; | |
aec7477b JB |
6390 | } |
6391 | /* | |
6392 | * index_cnt is ignored for everything but a dir, | |
6393 | * btrfs_get_inode_index_count has an explanation for the magic | |
6394 | * number | |
6395 | */ | |
6396 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6397 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6398 | BTRFS_I(inode)->root = root; |
e02119d5 | 6399 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6400 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6401 | |
5dc562c5 JB |
6402 | /* |
6403 | * We could have gotten an inode number from somebody who was fsynced | |
6404 | * and then removed in this same transaction, so let's just set full | |
6405 | * sync since it will be a full sync anyway and this will blow away the | |
6406 | * old info in the log. | |
6407 | */ | |
6408 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6409 | ||
9c58309d | 6410 | key[0].objectid = objectid; |
962a298f | 6411 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6412 | key[0].offset = 0; |
6413 | ||
9c58309d | 6414 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6415 | |
6416 | if (name) { | |
6417 | /* | |
6418 | * Start new inodes with an inode_ref. This is slightly more | |
6419 | * efficient for small numbers of hard links since they will | |
6420 | * be packed into one item. Extended refs will kick in if we | |
6421 | * add more hard links than can fit in the ref item. | |
6422 | */ | |
6423 | key[1].objectid = objectid; | |
962a298f | 6424 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6425 | key[1].offset = ref_objectid; |
6426 | ||
6427 | sizes[1] = name_len + sizeof(*ref); | |
6428 | } | |
9c58309d | 6429 | |
b0d5d10f CM |
6430 | location = &BTRFS_I(inode)->location; |
6431 | location->objectid = objectid; | |
6432 | location->offset = 0; | |
962a298f | 6433 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6434 | |
6435 | ret = btrfs_insert_inode_locked(inode); | |
6436 | if (ret < 0) | |
6437 | goto fail; | |
6438 | ||
b9473439 | 6439 | path->leave_spinning = 1; |
ef3b9af5 | 6440 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6441 | if (ret != 0) |
b0d5d10f | 6442 | goto fail_unlock; |
5f39d397 | 6443 | |
ecc11fab | 6444 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6445 | inode_set_bytes(inode, 0); |
9cc97d64 | 6446 | |
c2050a45 | 6447 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6448 | inode->i_atime = inode->i_mtime; |
6449 | inode->i_ctime = inode->i_mtime; | |
6450 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6451 | ||
5f39d397 CM |
6452 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6453 | struct btrfs_inode_item); | |
b159fa28 | 6454 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6455 | sizeof(*inode_item)); |
e02119d5 | 6456 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6457 | |
ef3b9af5 FM |
6458 | if (name) { |
6459 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6460 | struct btrfs_inode_ref); | |
6461 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6462 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6463 | ptr = (unsigned long)(ref + 1); | |
6464 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6465 | } | |
9c58309d | 6466 | |
5f39d397 CM |
6467 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6468 | btrfs_free_path(path); | |
6469 | ||
6cbff00f CH |
6470 | btrfs_inherit_iflags(inode, dir); |
6471 | ||
569254b0 | 6472 | if (S_ISREG(mode)) { |
0b246afa | 6473 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6474 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6475 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6476 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6477 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6478 | } |
6479 | ||
5d4f98a2 | 6480 | inode_tree_add(inode); |
1abe9b8a | 6481 | |
6482 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6483 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6484 | |
8ea05e3a AB |
6485 | btrfs_update_root_times(trans, root); |
6486 | ||
63541927 FDBM |
6487 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6488 | if (ret) | |
0b246afa | 6489 | btrfs_err(fs_info, |
63541927 | 6490 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6491 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6492 | |
39279cc3 | 6493 | return inode; |
b0d5d10f CM |
6494 | |
6495 | fail_unlock: | |
6496 | unlock_new_inode(inode); | |
5f39d397 | 6497 | fail: |
ef3b9af5 | 6498 | if (dir && name) |
aec7477b | 6499 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6500 | btrfs_free_path(path); |
09771430 | 6501 | iput(inode); |
5f39d397 | 6502 | return ERR_PTR(ret); |
39279cc3 CM |
6503 | } |
6504 | ||
6505 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6506 | { | |
6507 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6508 | } | |
6509 | ||
d352ac68 CM |
6510 | /* |
6511 | * utility function to add 'inode' into 'parent_inode' with | |
6512 | * a give name and a given sequence number. | |
6513 | * if 'add_backref' is true, also insert a backref from the | |
6514 | * inode to the parent directory. | |
6515 | */ | |
e02119d5 | 6516 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6517 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6518 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6519 | { |
db0a669f | 6520 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4df27c4d | 6521 | int ret = 0; |
39279cc3 | 6522 | struct btrfs_key key; |
db0a669f NB |
6523 | struct btrfs_root *root = parent_inode->root; |
6524 | u64 ino = btrfs_ino(inode); | |
6525 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6526 | |
33345d01 | 6527 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6528 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6529 | } else { |
33345d01 | 6530 | key.objectid = ino; |
962a298f | 6531 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6532 | key.offset = 0; |
6533 | } | |
6534 | ||
33345d01 | 6535 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
0b246afa JM |
6536 | ret = btrfs_add_root_ref(trans, fs_info, key.objectid, |
6537 | root->root_key.objectid, parent_ino, | |
6538 | index, name, name_len); | |
4df27c4d | 6539 | } else if (add_backref) { |
33345d01 LZ |
6540 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6541 | parent_ino, index); | |
4df27c4d | 6542 | } |
39279cc3 | 6543 | |
79787eaa JM |
6544 | /* Nothing to clean up yet */ |
6545 | if (ret) | |
6546 | return ret; | |
4df27c4d | 6547 | |
79787eaa JM |
6548 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6549 | parent_inode, &key, | |
db0a669f | 6550 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6551 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6552 | goto fail_dir_item; |
6553 | else if (ret) { | |
66642832 | 6554 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6555 | return ret; |
39279cc3 | 6556 | } |
79787eaa | 6557 | |
db0a669f | 6558 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6559 | name_len * 2); |
db0a669f NB |
6560 | inode_inc_iversion(&parent_inode->vfs_inode); |
6561 | parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime = | |
6562 | current_time(&parent_inode->vfs_inode); | |
6563 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); | |
79787eaa | 6564 | if (ret) |
66642832 | 6565 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6566 | return ret; |
fe66a05a CM |
6567 | |
6568 | fail_dir_item: | |
6569 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6570 | u64 local_index; | |
6571 | int err; | |
0b246afa JM |
6572 | err = btrfs_del_root_ref(trans, fs_info, key.objectid, |
6573 | root->root_key.objectid, parent_ino, | |
6574 | &local_index, name, name_len); | |
fe66a05a CM |
6575 | |
6576 | } else if (add_backref) { | |
6577 | u64 local_index; | |
6578 | int err; | |
6579 | ||
6580 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6581 | ino, parent_ino, &local_index); | |
6582 | } | |
6583 | return ret; | |
39279cc3 CM |
6584 | } |
6585 | ||
6586 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6587 | struct btrfs_inode *dir, struct dentry *dentry, |
6588 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6589 | { |
a1b075d2 JB |
6590 | int err = btrfs_add_link(trans, dir, inode, |
6591 | dentry->d_name.name, dentry->d_name.len, | |
6592 | backref, index); | |
39279cc3 CM |
6593 | if (err > 0) |
6594 | err = -EEXIST; | |
6595 | return err; | |
6596 | } | |
6597 | ||
618e21d5 | 6598 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6599 | umode_t mode, dev_t rdev) |
618e21d5 | 6600 | { |
2ff7e61e | 6601 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6602 | struct btrfs_trans_handle *trans; |
6603 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6604 | struct inode *inode = NULL; |
618e21d5 JB |
6605 | int err; |
6606 | int drop_inode = 0; | |
6607 | u64 objectid; | |
00e4e6b3 | 6608 | u64 index = 0; |
618e21d5 | 6609 | |
9ed74f2d JB |
6610 | /* |
6611 | * 2 for inode item and ref | |
6612 | * 2 for dir items | |
6613 | * 1 for xattr if selinux is on | |
6614 | */ | |
a22285a6 YZ |
6615 | trans = btrfs_start_transaction(root, 5); |
6616 | if (IS_ERR(trans)) | |
6617 | return PTR_ERR(trans); | |
1832a6d5 | 6618 | |
581bb050 LZ |
6619 | err = btrfs_find_free_ino(root, &objectid); |
6620 | if (err) | |
6621 | goto out_unlock; | |
6622 | ||
aec7477b | 6623 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6624 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6625 | mode, &index); | |
7cf96da3 TI |
6626 | if (IS_ERR(inode)) { |
6627 | err = PTR_ERR(inode); | |
618e21d5 | 6628 | goto out_unlock; |
7cf96da3 | 6629 | } |
618e21d5 | 6630 | |
ad19db71 CS |
6631 | /* |
6632 | * If the active LSM wants to access the inode during | |
6633 | * d_instantiate it needs these. Smack checks to see | |
6634 | * if the filesystem supports xattrs by looking at the | |
6635 | * ops vector. | |
6636 | */ | |
ad19db71 | 6637 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6638 | init_special_inode(inode, inode->i_mode, rdev); |
6639 | ||
6640 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6641 | if (err) |
b0d5d10f CM |
6642 | goto out_unlock_inode; |
6643 | ||
cef415af NB |
6644 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6645 | 0, index); | |
b0d5d10f CM |
6646 | if (err) { |
6647 | goto out_unlock_inode; | |
6648 | } else { | |
1b4ab1bb | 6649 | btrfs_update_inode(trans, root, inode); |
f440ea85 | 6650 | d_instantiate_new(dentry, inode); |
618e21d5 | 6651 | } |
b0d5d10f | 6652 | |
618e21d5 | 6653 | out_unlock: |
3a45bb20 | 6654 | btrfs_end_transaction(trans); |
2ff7e61e JM |
6655 | btrfs_balance_delayed_items(fs_info); |
6656 | btrfs_btree_balance_dirty(fs_info); | |
618e21d5 JB |
6657 | if (drop_inode) { |
6658 | inode_dec_link_count(inode); | |
6659 | iput(inode); | |
6660 | } | |
618e21d5 | 6661 | return err; |
b0d5d10f CM |
6662 | |
6663 | out_unlock_inode: | |
6664 | drop_inode = 1; | |
6665 | unlock_new_inode(inode); | |
6666 | goto out_unlock; | |
6667 | ||
618e21d5 JB |
6668 | } |
6669 | ||
39279cc3 | 6670 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6671 | umode_t mode, bool excl) |
39279cc3 | 6672 | { |
2ff7e61e | 6673 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6674 | struct btrfs_trans_handle *trans; |
6675 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6676 | struct inode *inode = NULL; |
43baa579 | 6677 | int drop_inode_on_err = 0; |
a22285a6 | 6678 | int err; |
39279cc3 | 6679 | u64 objectid; |
00e4e6b3 | 6680 | u64 index = 0; |
39279cc3 | 6681 | |
9ed74f2d JB |
6682 | /* |
6683 | * 2 for inode item and ref | |
6684 | * 2 for dir items | |
6685 | * 1 for xattr if selinux is on | |
6686 | */ | |
a22285a6 YZ |
6687 | trans = btrfs_start_transaction(root, 5); |
6688 | if (IS_ERR(trans)) | |
6689 | return PTR_ERR(trans); | |
9ed74f2d | 6690 | |
581bb050 LZ |
6691 | err = btrfs_find_free_ino(root, &objectid); |
6692 | if (err) | |
6693 | goto out_unlock; | |
6694 | ||
aec7477b | 6695 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6696 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6697 | mode, &index); | |
7cf96da3 TI |
6698 | if (IS_ERR(inode)) { |
6699 | err = PTR_ERR(inode); | |
39279cc3 | 6700 | goto out_unlock; |
7cf96da3 | 6701 | } |
43baa579 | 6702 | drop_inode_on_err = 1; |
ad19db71 CS |
6703 | /* |
6704 | * If the active LSM wants to access the inode during | |
6705 | * d_instantiate it needs these. Smack checks to see | |
6706 | * if the filesystem supports xattrs by looking at the | |
6707 | * ops vector. | |
6708 | */ | |
6709 | inode->i_fop = &btrfs_file_operations; | |
6710 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6711 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6712 | |
6713 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6714 | if (err) | |
6715 | goto out_unlock_inode; | |
6716 | ||
6717 | err = btrfs_update_inode(trans, root, inode); | |
6718 | if (err) | |
6719 | goto out_unlock_inode; | |
ad19db71 | 6720 | |
cef415af NB |
6721 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6722 | 0, index); | |
39279cc3 | 6723 | if (err) |
b0d5d10f | 6724 | goto out_unlock_inode; |
43baa579 | 6725 | |
43baa579 | 6726 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
f440ea85 | 6727 | d_instantiate_new(dentry, inode); |
43baa579 | 6728 | |
39279cc3 | 6729 | out_unlock: |
3a45bb20 | 6730 | btrfs_end_transaction(trans); |
43baa579 | 6731 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6732 | inode_dec_link_count(inode); |
6733 | iput(inode); | |
6734 | } | |
2ff7e61e JM |
6735 | btrfs_balance_delayed_items(fs_info); |
6736 | btrfs_btree_balance_dirty(fs_info); | |
39279cc3 | 6737 | return err; |
b0d5d10f CM |
6738 | |
6739 | out_unlock_inode: | |
6740 | unlock_new_inode(inode); | |
6741 | goto out_unlock; | |
6742 | ||
39279cc3 CM |
6743 | } |
6744 | ||
6745 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6746 | struct dentry *dentry) | |
6747 | { | |
271dba45 | 6748 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6749 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6750 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6751 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6752 | u64 index; |
39279cc3 CM |
6753 | int err; |
6754 | int drop_inode = 0; | |
6755 | ||
4a8be425 TH |
6756 | /* do not allow sys_link's with other subvols of the same device */ |
6757 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6758 | return -EXDEV; |
4a8be425 | 6759 | |
f186373f | 6760 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6761 | return -EMLINK; |
4a8be425 | 6762 | |
877574e2 | 6763 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6764 | if (err) |
6765 | goto fail; | |
6766 | ||
a22285a6 | 6767 | /* |
7e6b6465 | 6768 | * 2 items for inode and inode ref |
a22285a6 | 6769 | * 2 items for dir items |
7e6b6465 | 6770 | * 1 item for parent inode |
a22285a6 | 6771 | */ |
7e6b6465 | 6772 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6773 | if (IS_ERR(trans)) { |
6774 | err = PTR_ERR(trans); | |
271dba45 | 6775 | trans = NULL; |
a22285a6 YZ |
6776 | goto fail; |
6777 | } | |
5f39d397 | 6778 | |
67de1176 MX |
6779 | /* There are several dir indexes for this inode, clear the cache. */ |
6780 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6781 | inc_nlink(inode); |
0c4d2d95 | 6782 | inode_inc_iversion(inode); |
c2050a45 | 6783 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6784 | ihold(inode); |
e9976151 | 6785 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6786 | |
cef415af NB |
6787 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6788 | 1, index); | |
5f39d397 | 6789 | |
a5719521 | 6790 | if (err) { |
54aa1f4d | 6791 | drop_inode = 1; |
a5719521 | 6792 | } else { |
10d9f309 | 6793 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6794 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6795 | if (err) |
6796 | goto fail; | |
ef3b9af5 FM |
6797 | if (inode->i_nlink == 1) { |
6798 | /* | |
6799 | * If new hard link count is 1, it's a file created | |
6800 | * with open(2) O_TMPFILE flag. | |
6801 | */ | |
3d6ae7bb | 6802 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6803 | if (err) |
6804 | goto fail; | |
6805 | } | |
08c422c2 | 6806 | d_instantiate(dentry, inode); |
9ca5fbfb | 6807 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6808 | } |
39279cc3 | 6809 | |
2ff7e61e | 6810 | btrfs_balance_delayed_items(fs_info); |
1832a6d5 | 6811 | fail: |
271dba45 | 6812 | if (trans) |
3a45bb20 | 6813 | btrfs_end_transaction(trans); |
39279cc3 CM |
6814 | if (drop_inode) { |
6815 | inode_dec_link_count(inode); | |
6816 | iput(inode); | |
6817 | } | |
2ff7e61e | 6818 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6819 | return err; |
6820 | } | |
6821 | ||
18bb1db3 | 6822 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6823 | { |
2ff7e61e | 6824 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6825 | struct inode *inode = NULL; |
39279cc3 CM |
6826 | struct btrfs_trans_handle *trans; |
6827 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6828 | int err = 0; | |
6829 | int drop_on_err = 0; | |
b9d86667 | 6830 | u64 objectid = 0; |
00e4e6b3 | 6831 | u64 index = 0; |
39279cc3 | 6832 | |
9ed74f2d JB |
6833 | /* |
6834 | * 2 items for inode and ref | |
6835 | * 2 items for dir items | |
6836 | * 1 for xattr if selinux is on | |
6837 | */ | |
a22285a6 YZ |
6838 | trans = btrfs_start_transaction(root, 5); |
6839 | if (IS_ERR(trans)) | |
6840 | return PTR_ERR(trans); | |
39279cc3 | 6841 | |
581bb050 LZ |
6842 | err = btrfs_find_free_ino(root, &objectid); |
6843 | if (err) | |
6844 | goto out_fail; | |
6845 | ||
aec7477b | 6846 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6847 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6848 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6849 | if (IS_ERR(inode)) { |
6850 | err = PTR_ERR(inode); | |
6851 | goto out_fail; | |
6852 | } | |
5f39d397 | 6853 | |
39279cc3 | 6854 | drop_on_err = 1; |
b0d5d10f CM |
6855 | /* these must be set before we unlock the inode */ |
6856 | inode->i_op = &btrfs_dir_inode_operations; | |
6857 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6858 | |
2a7dba39 | 6859 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6860 | if (err) |
b0d5d10f | 6861 | goto out_fail_inode; |
39279cc3 | 6862 | |
6ef06d27 | 6863 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6864 | err = btrfs_update_inode(trans, root, inode); |
6865 | if (err) | |
b0d5d10f | 6866 | goto out_fail_inode; |
5f39d397 | 6867 | |
db0a669f NB |
6868 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6869 | dentry->d_name.name, | |
6870 | dentry->d_name.len, 0, index); | |
39279cc3 | 6871 | if (err) |
b0d5d10f | 6872 | goto out_fail_inode; |
5f39d397 | 6873 | |
f440ea85 | 6874 | d_instantiate_new(dentry, inode); |
39279cc3 | 6875 | drop_on_err = 0; |
39279cc3 CM |
6876 | |
6877 | out_fail: | |
3a45bb20 | 6878 | btrfs_end_transaction(trans); |
c7cfb8a5 WS |
6879 | if (drop_on_err) { |
6880 | inode_dec_link_count(inode); | |
39279cc3 | 6881 | iput(inode); |
c7cfb8a5 | 6882 | } |
2ff7e61e JM |
6883 | btrfs_balance_delayed_items(fs_info); |
6884 | btrfs_btree_balance_dirty(fs_info); | |
39279cc3 | 6885 | return err; |
b0d5d10f CM |
6886 | |
6887 | out_fail_inode: | |
6888 | unlock_new_inode(inode); | |
6889 | goto out_fail; | |
39279cc3 CM |
6890 | } |
6891 | ||
e6c4efd8 QW |
6892 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6893 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6894 | { | |
6895 | struct rb_node *next; | |
6896 | ||
6897 | next = rb_next(&em->rb_node); | |
6898 | if (!next) | |
6899 | return NULL; | |
6900 | return container_of(next, struct extent_map, rb_node); | |
6901 | } | |
6902 | ||
6903 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6904 | { | |
6905 | struct rb_node *prev; | |
6906 | ||
6907 | prev = rb_prev(&em->rb_node); | |
6908 | if (!prev) | |
6909 | return NULL; | |
6910 | return container_of(prev, struct extent_map, rb_node); | |
6911 | } | |
6912 | ||
d352ac68 | 6913 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6914 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6915 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6916 | * the best fitted new extent into the tree. |
d352ac68 | 6917 | */ |
3b951516 CM |
6918 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6919 | struct extent_map *existing, | |
e6dcd2dc | 6920 | struct extent_map *em, |
51f395ad | 6921 | u64 map_start) |
3b951516 | 6922 | { |
e6c4efd8 QW |
6923 | struct extent_map *prev; |
6924 | struct extent_map *next; | |
6925 | u64 start; | |
6926 | u64 end; | |
3b951516 | 6927 | u64 start_diff; |
3b951516 | 6928 | |
e6dcd2dc | 6929 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6930 | |
6931 | if (existing->start > map_start) { | |
6932 | next = existing; | |
6933 | prev = prev_extent_map(next); | |
6934 | } else { | |
6935 | prev = existing; | |
6936 | next = next_extent_map(prev); | |
6937 | } | |
6938 | ||
6939 | start = prev ? extent_map_end(prev) : em->start; | |
6940 | start = max_t(u64, start, em->start); | |
6941 | end = next ? next->start : extent_map_end(em); | |
6942 | end = min_t(u64, end, extent_map_end(em)); | |
6943 | start_diff = start - em->start; | |
6944 | em->start = start; | |
6945 | em->len = end - start; | |
c8b97818 CM |
6946 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6947 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6948 | em->block_start += start_diff; |
c8b97818 CM |
6949 | em->block_len -= start_diff; |
6950 | } | |
09a2a8f9 | 6951 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6952 | } |
6953 | ||
c8b97818 | 6954 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6955 | struct page *page, |
c8b97818 CM |
6956 | size_t pg_offset, u64 extent_offset, |
6957 | struct btrfs_file_extent_item *item) | |
6958 | { | |
6959 | int ret; | |
6960 | struct extent_buffer *leaf = path->nodes[0]; | |
6961 | char *tmp; | |
6962 | size_t max_size; | |
6963 | unsigned long inline_size; | |
6964 | unsigned long ptr; | |
261507a0 | 6965 | int compress_type; |
c8b97818 CM |
6966 | |
6967 | WARN_ON(pg_offset != 0); | |
261507a0 | 6968 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6969 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6970 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6971 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6972 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6973 | if (!tmp) |
6974 | return -ENOMEM; | |
c8b97818 CM |
6975 | ptr = btrfs_file_extent_inline_start(item); |
6976 | ||
6977 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6978 | ||
09cbfeaf | 6979 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6980 | ret = btrfs_decompress(compress_type, tmp, page, |
6981 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6982 | |
6983 | /* | |
6984 | * decompression code contains a memset to fill in any space between the end | |
6985 | * of the uncompressed data and the end of max_size in case the decompressed | |
6986 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6987 | * the end of an inline extent and the beginning of the next block, so we | |
6988 | * cover that region here. | |
6989 | */ | |
6990 | ||
6991 | if (max_size + pg_offset < PAGE_SIZE) { | |
6992 | char *map = kmap(page); | |
6993 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6994 | kunmap(page); | |
6995 | } | |
c8b97818 | 6996 | kfree(tmp); |
166ae5a4 | 6997 | return ret; |
c8b97818 CM |
6998 | } |
6999 | ||
d352ac68 CM |
7000 | /* |
7001 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
7002 | * the ugly parts come from merging extents from the disk with the in-ram |
7003 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
7004 | * where the in-ram extents might be locked pending data=ordered completion. |
7005 | * | |
7006 | * This also copies inline extents directly into the page. | |
7007 | */ | |
fc4f21b1 NB |
7008 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
7009 | struct page *page, | |
7010 | size_t pg_offset, u64 start, u64 len, | |
7011 | int create) | |
a52d9a80 | 7012 | { |
fc4f21b1 | 7013 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
a52d9a80 CM |
7014 | int ret; |
7015 | int err = 0; | |
a52d9a80 CM |
7016 | u64 extent_start = 0; |
7017 | u64 extent_end = 0; | |
fc4f21b1 | 7018 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 7019 | u32 found_type; |
f421950f | 7020 | struct btrfs_path *path = NULL; |
fc4f21b1 | 7021 | struct btrfs_root *root = inode->root; |
a52d9a80 | 7022 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
7023 | struct extent_buffer *leaf; |
7024 | struct btrfs_key found_key; | |
a52d9a80 | 7025 | struct extent_map *em = NULL; |
fc4f21b1 NB |
7026 | struct extent_map_tree *em_tree = &inode->extent_tree; |
7027 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 7028 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 7029 | const bool new_inline = !page || create; |
a52d9a80 | 7030 | |
a52d9a80 | 7031 | again: |
890871be | 7032 | read_lock(&em_tree->lock); |
d1310b2e | 7033 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 7034 | if (em) |
0b246afa | 7035 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 7036 | read_unlock(&em_tree->lock); |
d1310b2e | 7037 | |
a52d9a80 | 7038 | if (em) { |
e1c4b745 CM |
7039 | if (em->start > start || em->start + em->len <= start) |
7040 | free_extent_map(em); | |
7041 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
7042 | free_extent_map(em); |
7043 | else | |
7044 | goto out; | |
a52d9a80 | 7045 | } |
172ddd60 | 7046 | em = alloc_extent_map(); |
a52d9a80 | 7047 | if (!em) { |
d1310b2e CM |
7048 | err = -ENOMEM; |
7049 | goto out; | |
a52d9a80 | 7050 | } |
0b246afa | 7051 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 7052 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 7053 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 7054 | em->len = (u64)-1; |
c8b97818 | 7055 | em->block_len = (u64)-1; |
f421950f CM |
7056 | |
7057 | if (!path) { | |
7058 | path = btrfs_alloc_path(); | |
026fd317 JB |
7059 | if (!path) { |
7060 | err = -ENOMEM; | |
7061 | goto out; | |
7062 | } | |
7063 | /* | |
7064 | * Chances are we'll be called again, so go ahead and do | |
7065 | * readahead | |
7066 | */ | |
e4058b54 | 7067 | path->reada = READA_FORWARD; |
f421950f CM |
7068 | } |
7069 | ||
179e29e4 CM |
7070 | ret = btrfs_lookup_file_extent(trans, root, path, |
7071 | objectid, start, trans != NULL); | |
a52d9a80 CM |
7072 | if (ret < 0) { |
7073 | err = ret; | |
7074 | goto out; | |
7075 | } | |
7076 | ||
7077 | if (ret != 0) { | |
7078 | if (path->slots[0] == 0) | |
7079 | goto not_found; | |
7080 | path->slots[0]--; | |
7081 | } | |
7082 | ||
5f39d397 CM |
7083 | leaf = path->nodes[0]; |
7084 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 7085 | struct btrfs_file_extent_item); |
a52d9a80 | 7086 | /* are we inside the extent that was found? */ |
5f39d397 | 7087 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 7088 | found_type = found_key.type; |
5f39d397 | 7089 | if (found_key.objectid != objectid || |
a52d9a80 | 7090 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
7091 | /* |
7092 | * If we backup past the first extent we want to move forward | |
7093 | * and see if there is an extent in front of us, otherwise we'll | |
7094 | * say there is a hole for our whole search range which can | |
7095 | * cause problems. | |
7096 | */ | |
7097 | extent_end = start; | |
7098 | goto next; | |
a52d9a80 CM |
7099 | } |
7100 | ||
5f39d397 CM |
7101 | found_type = btrfs_file_extent_type(leaf, item); |
7102 | extent_start = found_key.offset; | |
d899e052 YZ |
7103 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7104 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7105 | extent_end = extent_start + |
db94535d | 7106 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
7107 | |
7108 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
7109 | extent_start); | |
9036c102 YZ |
7110 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
7111 | size_t size; | |
514ac8ad | 7112 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
da17066c | 7113 | extent_end = ALIGN(extent_start + size, |
0b246afa | 7114 | fs_info->sectorsize); |
09ed2f16 LB |
7115 | |
7116 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
7117 | path->slots[0], | |
7118 | extent_start); | |
9036c102 | 7119 | } |
25a50341 | 7120 | next: |
9036c102 YZ |
7121 | if (start >= extent_end) { |
7122 | path->slots[0]++; | |
7123 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
7124 | ret = btrfs_next_leaf(root, path); | |
7125 | if (ret < 0) { | |
7126 | err = ret; | |
7127 | goto out; | |
a52d9a80 | 7128 | } |
9036c102 YZ |
7129 | if (ret > 0) |
7130 | goto not_found; | |
7131 | leaf = path->nodes[0]; | |
a52d9a80 | 7132 | } |
9036c102 YZ |
7133 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
7134 | if (found_key.objectid != objectid || | |
7135 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
7136 | goto not_found; | |
7137 | if (start + len <= found_key.offset) | |
7138 | goto not_found; | |
e2eca69d WS |
7139 | if (start > found_key.offset) |
7140 | goto next; | |
9036c102 | 7141 | em->start = start; |
70c8a91c | 7142 | em->orig_start = start; |
9036c102 YZ |
7143 | em->len = found_key.offset - start; |
7144 | goto not_found_em; | |
7145 | } | |
7146 | ||
fc4f21b1 | 7147 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 7148 | new_inline, em); |
7ffbb598 | 7149 | |
d899e052 YZ |
7150 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7151 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
7152 | goto insert; |
7153 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 7154 | unsigned long ptr; |
a52d9a80 | 7155 | char *map; |
3326d1b0 CM |
7156 | size_t size; |
7157 | size_t extent_offset; | |
7158 | size_t copy_size; | |
a52d9a80 | 7159 | |
7ffbb598 | 7160 | if (new_inline) |
689f9346 | 7161 | goto out; |
5f39d397 | 7162 | |
514ac8ad | 7163 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 7164 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7165 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7166 | size - extent_offset); | |
3326d1b0 | 7167 | em->start = extent_start + extent_offset; |
0b246afa | 7168 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7169 | em->orig_block_len = em->len; |
70c8a91c | 7170 | em->orig_start = em->start; |
689f9346 | 7171 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 7172 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
7173 | if (btrfs_file_extent_compression(leaf, item) != |
7174 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7175 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7176 | extent_offset, item); |
166ae5a4 ZB |
7177 | if (ret) { |
7178 | err = ret; | |
7179 | goto out; | |
7180 | } | |
c8b97818 CM |
7181 | } else { |
7182 | map = kmap(page); | |
7183 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
7184 | copy_size); | |
09cbfeaf | 7185 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7186 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7187 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7188 | copy_size); |
7189 | } | |
c8b97818 CM |
7190 | kunmap(page); |
7191 | } | |
179e29e4 CM |
7192 | flush_dcache_page(page); |
7193 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 7194 | BUG(); |
179e29e4 CM |
7195 | if (!trans) { |
7196 | kunmap(page); | |
7197 | free_extent_map(em); | |
7198 | em = NULL; | |
ff5714cc | 7199 | |
b3b4aa74 | 7200 | btrfs_release_path(path); |
7a7eaa40 | 7201 | trans = btrfs_join_transaction(root); |
ff5714cc | 7202 | |
3612b495 TI |
7203 | if (IS_ERR(trans)) |
7204 | return ERR_CAST(trans); | |
179e29e4 CM |
7205 | goto again; |
7206 | } | |
c8b97818 | 7207 | map = kmap(page); |
70dec807 | 7208 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 7209 | copy_size); |
c8b97818 | 7210 | kunmap(page); |
179e29e4 | 7211 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 7212 | } |
d1310b2e | 7213 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7214 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7215 | goto insert; |
a52d9a80 CM |
7216 | } |
7217 | not_found: | |
7218 | em->start = start; | |
70c8a91c | 7219 | em->orig_start = start; |
d1310b2e | 7220 | em->len = len; |
a52d9a80 | 7221 | not_found_em: |
5f39d397 | 7222 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 7223 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 7224 | insert: |
b3b4aa74 | 7225 | btrfs_release_path(path); |
d1310b2e | 7226 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7227 | btrfs_err(fs_info, |
5d163e0e JM |
7228 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7229 | em->start, em->len, start, len); | |
a52d9a80 CM |
7230 | err = -EIO; |
7231 | goto out; | |
7232 | } | |
d1310b2e CM |
7233 | |
7234 | err = 0; | |
890871be | 7235 | write_lock(&em_tree->lock); |
09a2a8f9 | 7236 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
7237 | /* it is possible that someone inserted the extent into the tree |
7238 | * while we had the lock dropped. It is also possible that | |
7239 | * an overlapping map exists in the tree | |
7240 | */ | |
a52d9a80 | 7241 | if (ret == -EEXIST) { |
3b951516 | 7242 | struct extent_map *existing; |
e6dcd2dc CM |
7243 | |
7244 | ret = 0; | |
7245 | ||
e6c4efd8 QW |
7246 | existing = search_extent_mapping(em_tree, start, len); |
7247 | /* | |
7248 | * existing will always be non-NULL, since there must be | |
7249 | * extent causing the -EEXIST. | |
7250 | */ | |
ebe06440 LB |
7251 | if (start >= existing->start && |
7252 | start < extent_map_end(existing)) { | |
8dff9c85 CM |
7253 | free_extent_map(em); |
7254 | em = existing; | |
7255 | err = 0; | |
ebe06440 | 7256 | } else { |
e6c4efd8 QW |
7257 | /* |
7258 | * The existing extent map is the one nearest to | |
7259 | * the [start, start + len) range which overlaps | |
7260 | */ | |
7261 | err = merge_extent_mapping(em_tree, existing, | |
7262 | em, start); | |
e1c4b745 | 7263 | free_extent_map(existing); |
e6c4efd8 | 7264 | if (err) { |
3b951516 CM |
7265 | free_extent_map(em); |
7266 | em = NULL; | |
7267 | } | |
a52d9a80 | 7268 | } |
a52d9a80 | 7269 | } |
890871be | 7270 | write_unlock(&em_tree->lock); |
a52d9a80 | 7271 | out: |
1abe9b8a | 7272 | |
fc4f21b1 | 7273 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7274 | |
527afb44 | 7275 | btrfs_free_path(path); |
a52d9a80 | 7276 | if (trans) { |
3a45bb20 | 7277 | ret = btrfs_end_transaction(trans); |
d397712b | 7278 | if (!err) |
a52d9a80 CM |
7279 | err = ret; |
7280 | } | |
a52d9a80 CM |
7281 | if (err) { |
7282 | free_extent_map(em); | |
a52d9a80 CM |
7283 | return ERR_PTR(err); |
7284 | } | |
79787eaa | 7285 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7286 | return em; |
7287 | } | |
7288 | ||
fc4f21b1 NB |
7289 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
7290 | struct page *page, | |
7291 | size_t pg_offset, u64 start, u64 len, | |
7292 | int create) | |
ec29ed5b CM |
7293 | { |
7294 | struct extent_map *em; | |
7295 | struct extent_map *hole_em = NULL; | |
7296 | u64 range_start = start; | |
7297 | u64 end; | |
7298 | u64 found; | |
7299 | u64 found_end; | |
7300 | int err = 0; | |
7301 | ||
7302 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7303 | if (IS_ERR(em)) | |
7304 | return em; | |
9986277e DC |
7305 | /* |
7306 | * If our em maps to: | |
7307 | * - a hole or | |
7308 | * - a pre-alloc extent, | |
7309 | * there might actually be delalloc bytes behind it. | |
7310 | */ | |
7311 | if (em->block_start != EXTENT_MAP_HOLE && | |
7312 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7313 | return em; | |
7314 | else | |
7315 | hole_em = em; | |
ec29ed5b CM |
7316 | |
7317 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7318 | end = start + len; | |
7319 | if (end < start) | |
7320 | end = (u64)-1; | |
7321 | else | |
7322 | end -= 1; | |
7323 | ||
7324 | em = NULL; | |
7325 | ||
7326 | /* ok, we didn't find anything, lets look for delalloc */ | |
fc4f21b1 | 7327 | found = count_range_bits(&inode->io_tree, &range_start, |
ec29ed5b CM |
7328 | end, len, EXTENT_DELALLOC, 1); |
7329 | found_end = range_start + found; | |
7330 | if (found_end < range_start) | |
7331 | found_end = (u64)-1; | |
7332 | ||
7333 | /* | |
7334 | * we didn't find anything useful, return | |
7335 | * the original results from get_extent() | |
7336 | */ | |
7337 | if (range_start > end || found_end <= start) { | |
7338 | em = hole_em; | |
7339 | hole_em = NULL; | |
7340 | goto out; | |
7341 | } | |
7342 | ||
7343 | /* adjust the range_start to make sure it doesn't | |
7344 | * go backwards from the start they passed in | |
7345 | */ | |
67871254 | 7346 | range_start = max(start, range_start); |
ec29ed5b CM |
7347 | found = found_end - range_start; |
7348 | ||
7349 | if (found > 0) { | |
7350 | u64 hole_start = start; | |
7351 | u64 hole_len = len; | |
7352 | ||
172ddd60 | 7353 | em = alloc_extent_map(); |
ec29ed5b CM |
7354 | if (!em) { |
7355 | err = -ENOMEM; | |
7356 | goto out; | |
7357 | } | |
7358 | /* | |
7359 | * when btrfs_get_extent can't find anything it | |
7360 | * returns one huge hole | |
7361 | * | |
7362 | * make sure what it found really fits our range, and | |
7363 | * adjust to make sure it is based on the start from | |
7364 | * the caller | |
7365 | */ | |
7366 | if (hole_em) { | |
7367 | u64 calc_end = extent_map_end(hole_em); | |
7368 | ||
7369 | if (calc_end <= start || (hole_em->start > end)) { | |
7370 | free_extent_map(hole_em); | |
7371 | hole_em = NULL; | |
7372 | } else { | |
7373 | hole_start = max(hole_em->start, start); | |
7374 | hole_len = calc_end - hole_start; | |
7375 | } | |
7376 | } | |
7377 | em->bdev = NULL; | |
7378 | if (hole_em && range_start > hole_start) { | |
7379 | /* our hole starts before our delalloc, so we | |
7380 | * have to return just the parts of the hole | |
7381 | * that go until the delalloc starts | |
7382 | */ | |
7383 | em->len = min(hole_len, | |
7384 | range_start - hole_start); | |
7385 | em->start = hole_start; | |
7386 | em->orig_start = hole_start; | |
7387 | /* | |
7388 | * don't adjust block start at all, | |
7389 | * it is fixed at EXTENT_MAP_HOLE | |
7390 | */ | |
7391 | em->block_start = hole_em->block_start; | |
7392 | em->block_len = hole_len; | |
f9e4fb53 LB |
7393 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7394 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7395 | } else { |
7396 | em->start = range_start; | |
7397 | em->len = found; | |
7398 | em->orig_start = range_start; | |
7399 | em->block_start = EXTENT_MAP_DELALLOC; | |
7400 | em->block_len = found; | |
7401 | } | |
7402 | } else if (hole_em) { | |
7403 | return hole_em; | |
7404 | } | |
7405 | out: | |
7406 | ||
7407 | free_extent_map(hole_em); | |
7408 | if (err) { | |
7409 | free_extent_map(em); | |
7410 | return ERR_PTR(err); | |
7411 | } | |
7412 | return em; | |
7413 | } | |
7414 | ||
5f9a8a51 FM |
7415 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7416 | const u64 start, | |
7417 | const u64 len, | |
7418 | const u64 orig_start, | |
7419 | const u64 block_start, | |
7420 | const u64 block_len, | |
7421 | const u64 orig_block_len, | |
7422 | const u64 ram_bytes, | |
7423 | const int type) | |
7424 | { | |
7425 | struct extent_map *em = NULL; | |
7426 | int ret; | |
7427 | ||
5f9a8a51 | 7428 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7429 | em = create_io_em(inode, start, len, orig_start, |
7430 | block_start, block_len, orig_block_len, | |
7431 | ram_bytes, | |
7432 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7433 | type); | |
5f9a8a51 FM |
7434 | if (IS_ERR(em)) |
7435 | goto out; | |
7436 | } | |
7437 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7438 | len, block_len, type); | |
7439 | if (ret) { | |
7440 | if (em) { | |
7441 | free_extent_map(em); | |
dcdbc059 | 7442 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7443 | start + len - 1, 0); |
7444 | } | |
7445 | em = ERR_PTR(ret); | |
7446 | } | |
7447 | out: | |
5f9a8a51 FM |
7448 | |
7449 | return em; | |
7450 | } | |
7451 | ||
4b46fce2 JB |
7452 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7453 | u64 start, u64 len) | |
7454 | { | |
0b246afa | 7455 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7456 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7457 | struct extent_map *em; |
4b46fce2 JB |
7458 | struct btrfs_key ins; |
7459 | u64 alloc_hint; | |
7460 | int ret; | |
4b46fce2 | 7461 | |
4b46fce2 | 7462 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7463 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7464 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7465 | if (ret) |
7466 | return ERR_PTR(ret); | |
4b46fce2 | 7467 | |
5f9a8a51 FM |
7468 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7469 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7470 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7471 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7472 | if (IS_ERR(em)) |
2ff7e61e JM |
7473 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7474 | ins.offset, 1); | |
de0ee0ed | 7475 | |
4b46fce2 JB |
7476 | return em; |
7477 | } | |
7478 | ||
46bfbb5c CM |
7479 | /* |
7480 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7481 | * block must be cow'd | |
7482 | */ | |
00361589 | 7483 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7484 | u64 *orig_start, u64 *orig_block_len, |
7485 | u64 *ram_bytes) | |
46bfbb5c | 7486 | { |
2ff7e61e | 7487 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7488 | struct btrfs_path *path; |
7489 | int ret; | |
7490 | struct extent_buffer *leaf; | |
7491 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7492 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7493 | struct btrfs_file_extent_item *fi; |
7494 | struct btrfs_key key; | |
7495 | u64 disk_bytenr; | |
7496 | u64 backref_offset; | |
7497 | u64 extent_end; | |
7498 | u64 num_bytes; | |
7499 | int slot; | |
7500 | int found_type; | |
7ee9e440 | 7501 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7502 | |
46bfbb5c CM |
7503 | path = btrfs_alloc_path(); |
7504 | if (!path) | |
7505 | return -ENOMEM; | |
7506 | ||
f85b7379 DS |
7507 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7508 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7509 | if (ret < 0) |
7510 | goto out; | |
7511 | ||
7512 | slot = path->slots[0]; | |
7513 | if (ret == 1) { | |
7514 | if (slot == 0) { | |
7515 | /* can't find the item, must cow */ | |
7516 | ret = 0; | |
7517 | goto out; | |
7518 | } | |
7519 | slot--; | |
7520 | } | |
7521 | ret = 0; | |
7522 | leaf = path->nodes[0]; | |
7523 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7524 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7525 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7526 | /* not our file or wrong item type, must cow */ | |
7527 | goto out; | |
7528 | } | |
7529 | ||
7530 | if (key.offset > offset) { | |
7531 | /* Wrong offset, must cow */ | |
7532 | goto out; | |
7533 | } | |
7534 | ||
7535 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7536 | found_type = btrfs_file_extent_type(leaf, fi); | |
7537 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7538 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7539 | /* not a regular extent, must cow */ | |
7540 | goto out; | |
7541 | } | |
7ee9e440 JB |
7542 | |
7543 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7544 | goto out; | |
7545 | ||
e77751aa MX |
7546 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7547 | if (extent_end <= offset) | |
7548 | goto out; | |
7549 | ||
46bfbb5c | 7550 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7551 | if (disk_bytenr == 0) |
7552 | goto out; | |
7553 | ||
7554 | if (btrfs_file_extent_compression(leaf, fi) || | |
7555 | btrfs_file_extent_encryption(leaf, fi) || | |
7556 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7557 | goto out; | |
7558 | ||
46bfbb5c CM |
7559 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7560 | ||
7ee9e440 JB |
7561 | if (orig_start) { |
7562 | *orig_start = key.offset - backref_offset; | |
7563 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7564 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7565 | } | |
eb384b55 | 7566 | |
2ff7e61e | 7567 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7568 | goto out; |
7b2b7085 MX |
7569 | |
7570 | num_bytes = min(offset + *len, extent_end) - offset; | |
7571 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7572 | u64 range_end; | |
7573 | ||
da17066c JM |
7574 | range_end = round_up(offset + num_bytes, |
7575 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7576 | ret = test_range_bit(io_tree, offset, range_end, |
7577 | EXTENT_DELALLOC, 0, NULL); | |
7578 | if (ret) { | |
7579 | ret = -EAGAIN; | |
7580 | goto out; | |
7581 | } | |
7582 | } | |
7583 | ||
1bda19eb | 7584 | btrfs_release_path(path); |
46bfbb5c CM |
7585 | |
7586 | /* | |
7587 | * look for other files referencing this extent, if we | |
7588 | * find any we must cow | |
7589 | */ | |
00361589 | 7590 | |
e4c3b2dc | 7591 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7592 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7593 | if (ret) { |
7594 | ret = 0; | |
7595 | goto out; | |
7596 | } | |
46bfbb5c CM |
7597 | |
7598 | /* | |
7599 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7600 | * in this extent we are about to write. If there | |
7601 | * are any csums in that range we have to cow in order | |
7602 | * to keep the csums correct | |
7603 | */ | |
7604 | disk_bytenr += backref_offset; | |
7605 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7606 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7607 | goto out; | |
46bfbb5c CM |
7608 | /* |
7609 | * all of the above have passed, it is safe to overwrite this extent | |
7610 | * without cow | |
7611 | */ | |
eb384b55 | 7612 | *len = num_bytes; |
46bfbb5c CM |
7613 | ret = 1; |
7614 | out: | |
7615 | btrfs_free_path(path); | |
7616 | return ret; | |
7617 | } | |
7618 | ||
fc4adbff AG |
7619 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7620 | { | |
7621 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
e03733da | 7622 | bool found = false; |
fc4adbff AG |
7623 | void **pagep = NULL; |
7624 | struct page *page = NULL; | |
cc2b702c DS |
7625 | unsigned long start_idx; |
7626 | unsigned long end_idx; | |
fc4adbff | 7627 | |
09cbfeaf | 7628 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7629 | |
7630 | /* | |
7631 | * end is the last byte in the last page. end == start is legal | |
7632 | */ | |
09cbfeaf | 7633 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7634 | |
7635 | rcu_read_lock(); | |
7636 | ||
7637 | /* Most of the code in this while loop is lifted from | |
7638 | * find_get_page. It's been modified to begin searching from a | |
7639 | * page and return just the first page found in that range. If the | |
7640 | * found idx is less than or equal to the end idx then we know that | |
7641 | * a page exists. If no pages are found or if those pages are | |
7642 | * outside of the range then we're fine (yay!) */ | |
7643 | while (page == NULL && | |
7644 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7645 | page = radix_tree_deref_slot(pagep); | |
7646 | if (unlikely(!page)) | |
7647 | break; | |
7648 | ||
7649 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7650 | if (radix_tree_deref_retry(page)) { |
7651 | page = NULL; | |
fc4adbff | 7652 | continue; |
809f9016 | 7653 | } |
fc4adbff AG |
7654 | /* |
7655 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7656 | * here as an exceptional entry: so return it without | |
7657 | * attempting to raise page count. | |
7658 | */ | |
6fdef6d4 | 7659 | page = NULL; |
fc4adbff AG |
7660 | break; /* TODO: Is this relevant for this use case? */ |
7661 | } | |
7662 | ||
91405151 FM |
7663 | if (!page_cache_get_speculative(page)) { |
7664 | page = NULL; | |
fc4adbff | 7665 | continue; |
91405151 | 7666 | } |
fc4adbff AG |
7667 | |
7668 | /* | |
7669 | * Has the page moved? | |
7670 | * This is part of the lockless pagecache protocol. See | |
7671 | * include/linux/pagemap.h for details. | |
7672 | */ | |
7673 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7674 | put_page(page); |
fc4adbff AG |
7675 | page = NULL; |
7676 | } | |
7677 | } | |
7678 | ||
7679 | if (page) { | |
7680 | if (page->index <= end_idx) | |
7681 | found = true; | |
09cbfeaf | 7682 | put_page(page); |
fc4adbff AG |
7683 | } |
7684 | ||
7685 | rcu_read_unlock(); | |
7686 | return found; | |
7687 | } | |
7688 | ||
eb838e73 JB |
7689 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7690 | struct extent_state **cached_state, int writing) | |
7691 | { | |
7692 | struct btrfs_ordered_extent *ordered; | |
7693 | int ret = 0; | |
7694 | ||
7695 | while (1) { | |
7696 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7697 | cached_state); |
eb838e73 JB |
7698 | /* |
7699 | * We're concerned with the entire range that we're going to be | |
01327610 | 7700 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7701 | * extents in this range. |
7702 | */ | |
a776c6fa | 7703 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7704 | lockend - lockstart + 1); |
7705 | ||
7706 | /* | |
7707 | * We need to make sure there are no buffered pages in this | |
7708 | * range either, we could have raced between the invalidate in | |
7709 | * generic_file_direct_write and locking the extent. The | |
7710 | * invalidate needs to happen so that reads after a write do not | |
7711 | * get stale data. | |
7712 | */ | |
fc4adbff AG |
7713 | if (!ordered && |
7714 | (!writing || | |
7715 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7716 | break; |
7717 | ||
7718 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7719 | cached_state, GFP_NOFS); | |
7720 | ||
7721 | if (ordered) { | |
ade77029 FM |
7722 | /* |
7723 | * If we are doing a DIO read and the ordered extent we | |
7724 | * found is for a buffered write, we can not wait for it | |
7725 | * to complete and retry, because if we do so we can | |
7726 | * deadlock with concurrent buffered writes on page | |
7727 | * locks. This happens only if our DIO read covers more | |
7728 | * than one extent map, if at this point has already | |
7729 | * created an ordered extent for a previous extent map | |
7730 | * and locked its range in the inode's io tree, and a | |
7731 | * concurrent write against that previous extent map's | |
7732 | * range and this range started (we unlock the ranges | |
7733 | * in the io tree only when the bios complete and | |
7734 | * buffered writes always lock pages before attempting | |
7735 | * to lock range in the io tree). | |
7736 | */ | |
7737 | if (writing || | |
7738 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7739 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7740 | else | |
7741 | ret = -ENOTBLK; | |
eb838e73 JB |
7742 | btrfs_put_ordered_extent(ordered); |
7743 | } else { | |
eb838e73 | 7744 | /* |
b850ae14 FM |
7745 | * We could trigger writeback for this range (and wait |
7746 | * for it to complete) and then invalidate the pages for | |
7747 | * this range (through invalidate_inode_pages2_range()), | |
7748 | * but that can lead us to a deadlock with a concurrent | |
7749 | * call to readpages() (a buffered read or a defrag call | |
7750 | * triggered a readahead) on a page lock due to an | |
7751 | * ordered dio extent we created before but did not have | |
7752 | * yet a corresponding bio submitted (whence it can not | |
7753 | * complete), which makes readpages() wait for that | |
7754 | * ordered extent to complete while holding a lock on | |
7755 | * that page. | |
eb838e73 | 7756 | */ |
b850ae14 | 7757 | ret = -ENOTBLK; |
eb838e73 JB |
7758 | } |
7759 | ||
ade77029 FM |
7760 | if (ret) |
7761 | break; | |
7762 | ||
eb838e73 JB |
7763 | cond_resched(); |
7764 | } | |
7765 | ||
7766 | return ret; | |
7767 | } | |
7768 | ||
6f9994db LB |
7769 | /* The callers of this must take lock_extent() */ |
7770 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7771 | u64 orig_start, u64 block_start, | |
7772 | u64 block_len, u64 orig_block_len, | |
7773 | u64 ram_bytes, int compress_type, | |
7774 | int type) | |
69ffb543 JB |
7775 | { |
7776 | struct extent_map_tree *em_tree; | |
7777 | struct extent_map *em; | |
7778 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7779 | int ret; | |
7780 | ||
6f9994db LB |
7781 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7782 | type == BTRFS_ORDERED_COMPRESSED || | |
7783 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7784 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7785 | |
69ffb543 JB |
7786 | em_tree = &BTRFS_I(inode)->extent_tree; |
7787 | em = alloc_extent_map(); | |
7788 | if (!em) | |
7789 | return ERR_PTR(-ENOMEM); | |
7790 | ||
7791 | em->start = start; | |
7792 | em->orig_start = orig_start; | |
7793 | em->len = len; | |
7794 | em->block_len = block_len; | |
7795 | em->block_start = block_start; | |
7796 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7797 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7798 | em->ram_bytes = ram_bytes; |
70c8a91c | 7799 | em->generation = -1; |
69ffb543 | 7800 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7801 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7802 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7803 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7804 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7805 | em->compress_type = compress_type; | |
7806 | } | |
69ffb543 JB |
7807 | |
7808 | do { | |
dcdbc059 | 7809 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7810 | em->start + em->len - 1, 0); |
7811 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7812 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7813 | write_unlock(&em_tree->lock); |
6f9994db LB |
7814 | /* |
7815 | * The caller has taken lock_extent(), who could race with us | |
7816 | * to add em? | |
7817 | */ | |
69ffb543 JB |
7818 | } while (ret == -EEXIST); |
7819 | ||
7820 | if (ret) { | |
7821 | free_extent_map(em); | |
7822 | return ERR_PTR(ret); | |
7823 | } | |
7824 | ||
6f9994db | 7825 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7826 | return em; |
7827 | } | |
7828 | ||
9c9464cc FM |
7829 | static void adjust_dio_outstanding_extents(struct inode *inode, |
7830 | struct btrfs_dio_data *dio_data, | |
7831 | const u64 len) | |
7832 | { | |
823bb20a | 7833 | unsigned num_extents = count_max_extents(len); |
9c9464cc | 7834 | |
9c9464cc FM |
7835 | /* |
7836 | * If we have an outstanding_extents count still set then we're | |
7837 | * within our reservation, otherwise we need to adjust our inode | |
7838 | * counter appropriately. | |
7839 | */ | |
c2931667 | 7840 | if (dio_data->outstanding_extents >= num_extents) { |
9c9464cc FM |
7841 | dio_data->outstanding_extents -= num_extents; |
7842 | } else { | |
c2931667 LB |
7843 | /* |
7844 | * If dio write length has been split due to no large enough | |
7845 | * contiguous space, we need to compensate our inode counter | |
7846 | * appropriately. | |
7847 | */ | |
7848 | u64 num_needed = num_extents - dio_data->outstanding_extents; | |
7849 | ||
9c9464cc | 7850 | spin_lock(&BTRFS_I(inode)->lock); |
c2931667 | 7851 | BTRFS_I(inode)->outstanding_extents += num_needed; |
9c9464cc FM |
7852 | spin_unlock(&BTRFS_I(inode)->lock); |
7853 | } | |
7854 | } | |
7855 | ||
4b46fce2 JB |
7856 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7857 | struct buffer_head *bh_result, int create) | |
7858 | { | |
0b246afa | 7859 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7860 | struct extent_map *em; |
eb838e73 | 7861 | struct extent_state *cached_state = NULL; |
50745b0a | 7862 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7863 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7864 | u64 lockstart, lockend; |
4b46fce2 | 7865 | u64 len = bh_result->b_size; |
eb838e73 | 7866 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7867 | int ret = 0; |
eb838e73 | 7868 | |
172a5049 | 7869 | if (create) |
3266789f | 7870 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7871 | else |
0b246afa | 7872 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7873 | |
c329861d JB |
7874 | lockstart = start; |
7875 | lockend = start + len - 1; | |
7876 | ||
e1cbbfa5 JB |
7877 | if (current->journal_info) { |
7878 | /* | |
7879 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7880 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7881 | * confused. |
7882 | */ | |
50745b0a | 7883 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7884 | current->journal_info = NULL; |
7885 | } | |
7886 | ||
eb838e73 JB |
7887 | /* |
7888 | * If this errors out it's because we couldn't invalidate pagecache for | |
7889 | * this range and we need to fallback to buffered. | |
7890 | */ | |
9c9464cc FM |
7891 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7892 | create)) { | |
7893 | ret = -ENOTBLK; | |
7894 | goto err; | |
7895 | } | |
eb838e73 | 7896 | |
fc4f21b1 | 7897 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7898 | if (IS_ERR(em)) { |
7899 | ret = PTR_ERR(em); | |
7900 | goto unlock_err; | |
7901 | } | |
4b46fce2 JB |
7902 | |
7903 | /* | |
7904 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7905 | * io. INLINE is special, and we could probably kludge it in here, but | |
7906 | * it's still buffered so for safety lets just fall back to the generic | |
7907 | * buffered path. | |
7908 | * | |
7909 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7910 | * decompress it, so there will be buffering required no matter what we | |
7911 | * do, so go ahead and fallback to buffered. | |
7912 | * | |
01327610 | 7913 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7914 | * to buffered IO. Don't blame me, this is the price we pay for using |
7915 | * the generic code. | |
7916 | */ | |
7917 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7918 | em->block_start == EXTENT_MAP_INLINE) { | |
7919 | free_extent_map(em); | |
eb838e73 JB |
7920 | ret = -ENOTBLK; |
7921 | goto unlock_err; | |
4b46fce2 JB |
7922 | } |
7923 | ||
7924 | /* Just a good old fashioned hole, return */ | |
7925 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7926 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7927 | free_extent_map(em); | |
eb838e73 | 7928 | goto unlock_err; |
4b46fce2 JB |
7929 | } |
7930 | ||
7931 | /* | |
7932 | * We don't allocate a new extent in the following cases | |
7933 | * | |
7934 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7935 | * existing extent. | |
7936 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7937 | * just use the extent. | |
7938 | * | |
7939 | */ | |
46bfbb5c | 7940 | if (!create) { |
eb838e73 JB |
7941 | len = min(len, em->len - (start - em->start)); |
7942 | lockstart = start + len; | |
7943 | goto unlock; | |
46bfbb5c | 7944 | } |
4b46fce2 JB |
7945 | |
7946 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7947 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7948 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7949 | int type; |
eb384b55 | 7950 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7951 | |
7952 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7953 | type = BTRFS_ORDERED_PREALLOC; | |
7954 | else | |
7955 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7956 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7957 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7958 | |
00361589 | 7959 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c | 7960 | &orig_block_len, &ram_bytes) == 1 && |
0b246afa | 7961 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
5f9a8a51 | 7962 | struct extent_map *em2; |
0b901916 | 7963 | |
5f9a8a51 FM |
7964 | em2 = btrfs_create_dio_extent(inode, start, len, |
7965 | orig_start, block_start, | |
7966 | len, orig_block_len, | |
7967 | ram_bytes, type); | |
0b246afa | 7968 | btrfs_dec_nocow_writers(fs_info, block_start); |
69ffb543 JB |
7969 | if (type == BTRFS_ORDERED_PREALLOC) { |
7970 | free_extent_map(em); | |
5f9a8a51 | 7971 | em = em2; |
69ffb543 | 7972 | } |
5f9a8a51 FM |
7973 | if (em2 && IS_ERR(em2)) { |
7974 | ret = PTR_ERR(em2); | |
eb838e73 | 7975 | goto unlock_err; |
46bfbb5c | 7976 | } |
18513091 WX |
7977 | /* |
7978 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7979 | * use the existing or preallocated extent, so does not | |
7980 | * need to adjust btrfs_space_info's bytes_may_use. | |
7981 | */ | |
7982 | btrfs_free_reserved_data_space_noquota(inode, | |
7983 | start, len); | |
46bfbb5c | 7984 | goto unlock; |
4b46fce2 | 7985 | } |
4b46fce2 | 7986 | } |
00361589 | 7987 | |
46bfbb5c CM |
7988 | /* |
7989 | * this will cow the extent, reset the len in case we changed | |
7990 | * it above | |
7991 | */ | |
7992 | len = bh_result->b_size; | |
70c8a91c JB |
7993 | free_extent_map(em); |
7994 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7995 | if (IS_ERR(em)) { |
7996 | ret = PTR_ERR(em); | |
7997 | goto unlock_err; | |
7998 | } | |
46bfbb5c CM |
7999 | len = min(len, em->len - (start - em->start)); |
8000 | unlock: | |
4b46fce2 JB |
8001 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
8002 | inode->i_blkbits; | |
46bfbb5c | 8003 | bh_result->b_size = len; |
4b46fce2 JB |
8004 | bh_result->b_bdev = em->bdev; |
8005 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
8006 | if (create) { |
8007 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
8008 | set_buffer_new(bh_result); | |
8009 | ||
8010 | /* | |
8011 | * Need to update the i_size under the extent lock so buffered | |
8012 | * readers will get the updated i_size when we unlock. | |
8013 | */ | |
4aaedfb0 | 8014 | if (!dio_data->overwrite && start + len > i_size_read(inode)) |
c3473e83 | 8015 | i_size_write(inode, start + len); |
0934856d | 8016 | |
9c9464cc | 8017 | adjust_dio_outstanding_extents(inode, dio_data, len); |
50745b0a | 8018 | WARN_ON(dio_data->reserve < len); |
8019 | dio_data->reserve -= len; | |
f28a4928 | 8020 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 8021 | current->journal_info = dio_data; |
c3473e83 | 8022 | } |
4b46fce2 | 8023 | |
eb838e73 JB |
8024 | /* |
8025 | * In the case of write we need to clear and unlock the entire range, | |
8026 | * in the case of read we need to unlock only the end area that we | |
8027 | * aren't using if there is any left over space. | |
8028 | */ | |
24c03fa5 | 8029 | if (lockstart < lockend) { |
0934856d MX |
8030 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
8031 | lockend, unlock_bits, 1, 0, | |
8032 | &cached_state, GFP_NOFS); | |
24c03fa5 | 8033 | } else { |
eb838e73 | 8034 | free_extent_state(cached_state); |
24c03fa5 | 8035 | } |
eb838e73 | 8036 | |
4b46fce2 JB |
8037 | free_extent_map(em); |
8038 | ||
8039 | return 0; | |
eb838e73 JB |
8040 | |
8041 | unlock_err: | |
eb838e73 JB |
8042 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
8043 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 8044 | err: |
50745b0a | 8045 | if (dio_data) |
8046 | current->journal_info = dio_data; | |
9c9464cc FM |
8047 | /* |
8048 | * Compensate the delalloc release we do in btrfs_direct_IO() when we | |
8049 | * write less data then expected, so that we don't underflow our inode's | |
8050 | * outstanding extents counter. | |
8051 | */ | |
8052 | if (create && dio_data) | |
8053 | adjust_dio_outstanding_extents(inode, dio_data, len); | |
8054 | ||
eb838e73 | 8055 | return ret; |
4b46fce2 JB |
8056 | } |
8057 | ||
58efbc9f OS |
8058 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
8059 | struct bio *bio, | |
8060 | int mirror_num) | |
8b110e39 | 8061 | { |
2ff7e61e | 8062 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 8063 | blk_status_t ret; |
8b110e39 | 8064 | |
37226b21 | 8065 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 MX |
8066 | |
8067 | bio_get(bio); | |
8068 | ||
2ff7e61e | 8069 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 MX |
8070 | if (ret) |
8071 | goto err; | |
8072 | ||
2ff7e61e | 8073 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
8b110e39 MX |
8074 | err: |
8075 | bio_put(bio); | |
8076 | return ret; | |
8077 | } | |
8078 | ||
8079 | static int btrfs_check_dio_repairable(struct inode *inode, | |
8080 | struct bio *failed_bio, | |
8081 | struct io_failure_record *failrec, | |
8082 | int failed_mirror) | |
8083 | { | |
ab8d0fc4 | 8084 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
8085 | int num_copies; |
8086 | ||
ab8d0fc4 | 8087 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
8088 | if (num_copies == 1) { |
8089 | /* | |
8090 | * we only have a single copy of the data, so don't bother with | |
8091 | * all the retry and error correction code that follows. no | |
8092 | * matter what the error is, it is very likely to persist. | |
8093 | */ | |
ab8d0fc4 JM |
8094 | btrfs_debug(fs_info, |
8095 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
8096 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
8097 | return 0; |
8098 | } | |
8099 | ||
8100 | failrec->failed_mirror = failed_mirror; | |
8101 | failrec->this_mirror++; | |
8102 | if (failrec->this_mirror == failed_mirror) | |
8103 | failrec->this_mirror++; | |
8104 | ||
8105 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
8106 | btrfs_debug(fs_info, |
8107 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
8108 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
8109 | return 0; |
8110 | } | |
8111 | ||
8112 | return 1; | |
8113 | } | |
8114 | ||
58efbc9f OS |
8115 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
8116 | struct page *page, unsigned int pgoff, | |
8117 | u64 start, u64 end, int failed_mirror, | |
8118 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
8119 | { |
8120 | struct io_failure_record *failrec; | |
7870d082 JB |
8121 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8122 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
8123 | struct bio *bio; |
8124 | int isector; | |
f1c77c55 | 8125 | unsigned int read_mode = 0; |
17347cec | 8126 | int segs; |
8b110e39 | 8127 | int ret; |
58efbc9f | 8128 | blk_status_t status; |
8b110e39 | 8129 | |
37226b21 | 8130 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
8131 | |
8132 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
8133 | if (ret) | |
58efbc9f | 8134 | return errno_to_blk_status(ret); |
8b110e39 MX |
8135 | |
8136 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
8137 | failed_mirror); | |
8138 | if (!ret) { | |
7870d082 | 8139 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 8140 | return BLK_STS_IOERR; |
8b110e39 MX |
8141 | } |
8142 | ||
17347cec LB |
8143 | segs = bio_segments(failed_bio); |
8144 | if (segs > 1 || | |
8145 | (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode))) | |
70fd7614 | 8146 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
8147 | |
8148 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
8149 | isector >>= inode->i_sb->s_blocksize_bits; | |
8150 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 8151 | pgoff, isector, repair_endio, repair_arg); |
37226b21 | 8152 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
8b110e39 MX |
8153 | |
8154 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 8155 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
8156 | read_mode, failrec->this_mirror, failrec->in_validation); |
8157 | ||
58efbc9f OS |
8158 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
8159 | if (status) { | |
7870d082 | 8160 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
8161 | bio_put(bio); |
8162 | } | |
8163 | ||
58efbc9f | 8164 | return status; |
8b110e39 MX |
8165 | } |
8166 | ||
8167 | struct btrfs_retry_complete { | |
8168 | struct completion done; | |
8169 | struct inode *inode; | |
8170 | u64 start; | |
8171 | int uptodate; | |
8172 | }; | |
8173 | ||
4246a0b6 | 8174 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
8175 | { |
8176 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 8177 | struct inode *inode = done->inode; |
8b110e39 | 8178 | struct bio_vec *bvec; |
7870d082 | 8179 | struct extent_io_tree *io_tree, *failure_tree; |
8b110e39 MX |
8180 | int i; |
8181 | ||
4e4cbee9 | 8182 | if (bio->bi_status) |
8b110e39 MX |
8183 | goto end; |
8184 | ||
2dabb324 | 8185 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
8186 | io_tree = &BTRFS_I(inode)->io_tree; |
8187 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8188 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); | |
2dabb324 | 8189 | |
8b110e39 | 8190 | done->uptodate = 1; |
c09abff8 | 8191 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8192 | bio_for_each_segment_all(bvec, bio, i) |
7870d082 JB |
8193 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
8194 | io_tree, done->start, bvec->bv_page, | |
8195 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
8196 | end: |
8197 | complete(&done->done); | |
8198 | bio_put(bio); | |
8199 | } | |
8200 | ||
58efbc9f OS |
8201 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
8202 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 8203 | { |
2dabb324 | 8204 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8205 | struct bio_vec bvec; |
8206 | struct bvec_iter iter; | |
8b110e39 | 8207 | struct btrfs_retry_complete done; |
4b46fce2 | 8208 | u64 start; |
2dabb324 CR |
8209 | unsigned int pgoff; |
8210 | u32 sectorsize; | |
8211 | int nr_sectors; | |
58efbc9f OS |
8212 | blk_status_t ret; |
8213 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 8214 | |
2dabb324 | 8215 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8216 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8217 | |
8b110e39 MX |
8218 | start = io_bio->logical; |
8219 | done.inode = inode; | |
17347cec | 8220 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8221 | |
17347cec LB |
8222 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8223 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
8224 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
8225 | |
8226 | next_block_or_try_again: | |
8b110e39 MX |
8227 | done.uptodate = 0; |
8228 | done.start = start; | |
8229 | init_completion(&done.done); | |
8230 | ||
17347cec | 8231 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
8232 | pgoff, start, start + sectorsize - 1, |
8233 | io_bio->mirror_num, | |
8234 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
8235 | if (ret) { |
8236 | err = ret; | |
8237 | goto next; | |
8238 | } | |
8b110e39 | 8239 | |
9c17f6cd | 8240 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8241 | |
8242 | if (!done.uptodate) { | |
8243 | /* We might have another mirror, so try again */ | |
2dabb324 | 8244 | goto next_block_or_try_again; |
8b110e39 MX |
8245 | } |
8246 | ||
629ebf4f | 8247 | next: |
2dabb324 CR |
8248 | start += sectorsize; |
8249 | ||
97bf5a55 LB |
8250 | nr_sectors--; |
8251 | if (nr_sectors) { | |
2dabb324 | 8252 | pgoff += sectorsize; |
97bf5a55 | 8253 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8254 | goto next_block_or_try_again; |
8255 | } | |
8b110e39 MX |
8256 | } |
8257 | ||
629ebf4f | 8258 | return err; |
8b110e39 MX |
8259 | } |
8260 | ||
4246a0b6 | 8261 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
8262 | { |
8263 | struct btrfs_retry_complete *done = bio->bi_private; | |
8264 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
8265 | struct extent_io_tree *io_tree, *failure_tree; |
8266 | struct inode *inode = done->inode; | |
8b110e39 MX |
8267 | struct bio_vec *bvec; |
8268 | int uptodate; | |
8269 | int ret; | |
8270 | int i; | |
8271 | ||
4e4cbee9 | 8272 | if (bio->bi_status) |
8b110e39 MX |
8273 | goto end; |
8274 | ||
8275 | uptodate = 1; | |
2dabb324 | 8276 | |
2dabb324 | 8277 | ASSERT(bio->bi_vcnt == 1); |
2e949b0a | 8278 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 8279 | |
7870d082 JB |
8280 | io_tree = &BTRFS_I(inode)->io_tree; |
8281 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8282 | ||
c09abff8 | 8283 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8284 | bio_for_each_segment_all(bvec, bio, i) { |
7870d082 JB |
8285 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
8286 | bvec->bv_offset, done->start, | |
8287 | bvec->bv_len); | |
8b110e39 | 8288 | if (!ret) |
7870d082 JB |
8289 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
8290 | failure_tree, io_tree, done->start, | |
8291 | bvec->bv_page, | |
8292 | btrfs_ino(BTRFS_I(inode)), | |
8293 | bvec->bv_offset); | |
8b110e39 MX |
8294 | else |
8295 | uptodate = 0; | |
8296 | } | |
8297 | ||
8298 | done->uptodate = uptodate; | |
8299 | end: | |
8300 | complete(&done->done); | |
8301 | bio_put(bio); | |
8302 | } | |
8303 | ||
4e4cbee9 CH |
8304 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
8305 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 8306 | { |
2dabb324 | 8307 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8308 | struct bio_vec bvec; |
8309 | struct bvec_iter iter; | |
8b110e39 MX |
8310 | struct btrfs_retry_complete done; |
8311 | u64 start; | |
8312 | u64 offset = 0; | |
2dabb324 CR |
8313 | u32 sectorsize; |
8314 | int nr_sectors; | |
8315 | unsigned int pgoff; | |
8316 | int csum_pos; | |
ef7cdac1 | 8317 | bool uptodate = (err == 0); |
8b110e39 | 8318 | int ret; |
58efbc9f | 8319 | blk_status_t status; |
dc380aea | 8320 | |
2dabb324 | 8321 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8322 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8323 | |
58efbc9f | 8324 | err = BLK_STS_OK; |
c1dc0896 | 8325 | start = io_bio->logical; |
8b110e39 | 8326 | done.inode = inode; |
17347cec | 8327 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8328 | |
17347cec LB |
8329 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8330 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 8331 | |
17347cec | 8332 | pgoff = bvec.bv_offset; |
2dabb324 | 8333 | next_block: |
ef7cdac1 LB |
8334 | if (uptodate) { |
8335 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8336 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8337 | bvec.bv_page, pgoff, start, sectorsize); | |
8338 | if (likely(!ret)) | |
8339 | goto next; | |
8340 | } | |
8b110e39 MX |
8341 | try_again: |
8342 | done.uptodate = 0; | |
8343 | done.start = start; | |
8344 | init_completion(&done.done); | |
8345 | ||
58efbc9f OS |
8346 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
8347 | pgoff, start, start + sectorsize - 1, | |
8348 | io_bio->mirror_num, btrfs_retry_endio, | |
8349 | &done); | |
8350 | if (status) { | |
8351 | err = status; | |
8b110e39 MX |
8352 | goto next; |
8353 | } | |
8354 | ||
9c17f6cd | 8355 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8356 | |
8357 | if (!done.uptodate) { | |
8358 | /* We might have another mirror, so try again */ | |
8359 | goto try_again; | |
8360 | } | |
8361 | next: | |
2dabb324 CR |
8362 | offset += sectorsize; |
8363 | start += sectorsize; | |
8364 | ||
8365 | ASSERT(nr_sectors); | |
8366 | ||
97bf5a55 LB |
8367 | nr_sectors--; |
8368 | if (nr_sectors) { | |
2dabb324 | 8369 | pgoff += sectorsize; |
97bf5a55 | 8370 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8371 | goto next_block; |
8372 | } | |
2c30c71b | 8373 | } |
c1dc0896 MX |
8374 | |
8375 | return err; | |
8376 | } | |
8377 | ||
4e4cbee9 CH |
8378 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8379 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8380 | { |
8381 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8382 | ||
8383 | if (skip_csum) { | |
8384 | if (unlikely(err)) | |
8385 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8386 | else | |
58efbc9f | 8387 | return BLK_STS_OK; |
8b110e39 MX |
8388 | } else { |
8389 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8390 | } | |
8391 | } | |
8392 | ||
4246a0b6 | 8393 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8394 | { |
8395 | struct btrfs_dio_private *dip = bio->bi_private; | |
8396 | struct inode *inode = dip->inode; | |
8397 | struct bio *dio_bio; | |
8398 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8399 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8400 | |
99c4e3b9 | 8401 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8402 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8403 | |
4b46fce2 | 8404 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8405 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8406 | dio_bio = dip->dio_bio; |
4b46fce2 | 8407 | |
4b46fce2 | 8408 | kfree(dip); |
c0da7aa1 | 8409 | |
99c4e3b9 | 8410 | dio_bio->bi_status = err; |
4055351c | 8411 | dio_end_io(dio_bio); |
23ea8e5a MX |
8412 | |
8413 | if (io_bio->end_io) | |
4e4cbee9 | 8414 | io_bio->end_io(io_bio, blk_status_to_errno(err)); |
9be3395b | 8415 | bio_put(bio); |
4b46fce2 JB |
8416 | } |
8417 | ||
52427260 QW |
8418 | static void __endio_write_update_ordered(struct inode *inode, |
8419 | const u64 offset, const u64 bytes, | |
8420 | const bool uptodate) | |
4b46fce2 | 8421 | { |
0b246afa | 8422 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8423 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8424 | struct btrfs_workqueue *wq; |
8425 | btrfs_work_func_t func; | |
14543774 FM |
8426 | u64 ordered_offset = offset; |
8427 | u64 ordered_bytes = bytes; | |
67c003f9 | 8428 | u64 last_offset; |
4b46fce2 JB |
8429 | int ret; |
8430 | ||
52427260 QW |
8431 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8432 | wq = fs_info->endio_freespace_worker; | |
8433 | func = btrfs_freespace_write_helper; | |
8434 | } else { | |
8435 | wq = fs_info->endio_write_workers; | |
8436 | func = btrfs_endio_write_helper; | |
8437 | } | |
8438 | ||
163cf09c | 8439 | again: |
67c003f9 | 8440 | last_offset = ordered_offset; |
163cf09c CM |
8441 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, |
8442 | &ordered_offset, | |
4246a0b6 | 8443 | ordered_bytes, |
14543774 | 8444 | uptodate); |
4b46fce2 | 8445 | if (!ret) |
163cf09c | 8446 | goto out_test; |
4b46fce2 | 8447 | |
52427260 QW |
8448 | btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL); |
8449 | btrfs_queue_work(wq, &ordered->work); | |
163cf09c | 8450 | out_test: |
67c003f9 NA |
8451 | /* |
8452 | * If btrfs_dec_test_ordered_pending does not find any ordered extent | |
8453 | * in the range, we can exit. | |
8454 | */ | |
8455 | if (ordered_offset == last_offset) | |
8456 | return; | |
163cf09c CM |
8457 | /* |
8458 | * our bio might span multiple ordered extents. If we haven't | |
8459 | * completed the accounting for the whole dio, go back and try again | |
8460 | */ | |
14543774 FM |
8461 | if (ordered_offset < offset + bytes) { |
8462 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8463 | ordered = NULL; |
163cf09c CM |
8464 | goto again; |
8465 | } | |
14543774 FM |
8466 | } |
8467 | ||
8468 | static void btrfs_endio_direct_write(struct bio *bio) | |
8469 | { | |
8470 | struct btrfs_dio_private *dip = bio->bi_private; | |
8471 | struct bio *dio_bio = dip->dio_bio; | |
8472 | ||
52427260 | 8473 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8474 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8475 | |
4b46fce2 | 8476 | kfree(dip); |
c0da7aa1 | 8477 | |
4e4cbee9 | 8478 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8479 | dio_end_io(dio_bio); |
9be3395b | 8480 | bio_put(bio); |
4b46fce2 JB |
8481 | } |
8482 | ||
8c27cb35 | 8483 | static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data, |
eaf25d93 CM |
8484 | struct bio *bio, int mirror_num, |
8485 | unsigned long bio_flags, u64 offset) | |
8486 | { | |
c6100a4b | 8487 | struct inode *inode = private_data; |
4e4cbee9 | 8488 | blk_status_t ret; |
2ff7e61e | 8489 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8490 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8491 | return 0; |
8492 | } | |
8493 | ||
4246a0b6 | 8494 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8495 | { |
8496 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8497 | blk_status_t err = bio->bi_status; |
e65e1535 | 8498 | |
8b110e39 MX |
8499 | if (err) |
8500 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8501 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8502 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8503 | bio->bi_opf, | |
8b110e39 MX |
8504 | (unsigned long long)bio->bi_iter.bi_sector, |
8505 | bio->bi_iter.bi_size, err); | |
8506 | ||
8507 | if (dip->subio_endio) | |
8508 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8509 | |
8510 | if (err) { | |
e65e1535 MX |
8511 | dip->errors = 1; |
8512 | ||
8513 | /* | |
8514 | * before atomic variable goto zero, we must make sure | |
8515 | * dip->errors is perceived to be set. | |
8516 | */ | |
4e857c58 | 8517 | smp_mb__before_atomic(); |
e65e1535 MX |
8518 | } |
8519 | ||
8520 | /* if there are more bios still pending for this dio, just exit */ | |
8521 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8522 | goto out; | |
8523 | ||
9be3395b | 8524 | if (dip->errors) { |
e65e1535 | 8525 | bio_io_error(dip->orig_bio); |
9be3395b | 8526 | } else { |
4e4cbee9 | 8527 | dip->dio_bio->bi_status = 0; |
4246a0b6 | 8528 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8529 | } |
8530 | out: | |
8531 | bio_put(bio); | |
8532 | } | |
8533 | ||
4e4cbee9 | 8534 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8535 | struct btrfs_dio_private *dip, |
8536 | struct bio *bio, | |
8537 | u64 file_offset) | |
8538 | { | |
8539 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8540 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8541 | blk_status_t ret; |
c1dc0896 MX |
8542 | |
8543 | /* | |
8544 | * We load all the csum data we need when we submit | |
8545 | * the first bio to reduce the csum tree search and | |
8546 | * contention. | |
8547 | */ | |
8548 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8549 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8550 | file_offset); |
8551 | if (ret) | |
8552 | return ret; | |
8553 | } | |
8554 | ||
8555 | if (bio == dip->orig_bio) | |
8556 | return 0; | |
8557 | ||
8558 | file_offset -= dip->logical_offset; | |
8559 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8560 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8561 | ||
8562 | return 0; | |
8563 | } | |
8564 | ||
58efbc9f OS |
8565 | static inline blk_status_t |
8566 | __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset, | |
66ba772e | 8567 | int async_submit) |
e65e1535 | 8568 | { |
0b246afa | 8569 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8570 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8571 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8572 | blk_status_t ret; |
e65e1535 | 8573 | |
b812ce28 JB |
8574 | if (async_submit) |
8575 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8576 | ||
e65e1535 | 8577 | bio_get(bio); |
5fd02043 JB |
8578 | |
8579 | if (!write) { | |
0b246afa | 8580 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8581 | if (ret) |
8582 | goto err; | |
8583 | } | |
e65e1535 | 8584 | |
e6961cac | 8585 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8586 | goto map; |
8587 | ||
8588 | if (write && async_submit) { | |
c6100a4b JB |
8589 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8590 | file_offset, inode, | |
0b246afa JM |
8591 | __btrfs_submit_bio_start_direct_io, |
8592 | __btrfs_submit_bio_done); | |
e65e1535 | 8593 | goto err; |
1ae39938 JB |
8594 | } else if (write) { |
8595 | /* | |
8596 | * If we aren't doing async submit, calculate the csum of the | |
8597 | * bio now. | |
8598 | */ | |
2ff7e61e | 8599 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8600 | if (ret) |
8601 | goto err; | |
23ea8e5a | 8602 | } else { |
2ff7e61e | 8603 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8604 | file_offset); |
c2db1073 TI |
8605 | if (ret) |
8606 | goto err; | |
8607 | } | |
1ae39938 | 8608 | map: |
2ff7e61e | 8609 | ret = btrfs_map_bio(fs_info, bio, 0, async_submit); |
e65e1535 MX |
8610 | err: |
8611 | bio_put(bio); | |
8612 | return ret; | |
8613 | } | |
8614 | ||
e6961cac | 8615 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8616 | { |
8617 | struct inode *inode = dip->inode; | |
0b246afa | 8618 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8619 | struct bio *bio; |
8620 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8621 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8622 | u64 file_offset = dip->logical_offset; |
e65e1535 | 8623 | u64 map_length; |
1ae39938 | 8624 | int async_submit = 0; |
725130ba LB |
8625 | u64 submit_len; |
8626 | int clone_offset = 0; | |
8627 | int clone_len; | |
5f4dc8fc | 8628 | int ret; |
58efbc9f | 8629 | blk_status_t status; |
e65e1535 | 8630 | |
4f024f37 | 8631 | map_length = orig_bio->bi_iter.bi_size; |
725130ba | 8632 | submit_len = map_length; |
0b246afa JM |
8633 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8634 | &map_length, NULL, 0); | |
7a5c3c9b | 8635 | if (ret) |
e65e1535 | 8636 | return -EIO; |
facc8a22 | 8637 | |
725130ba | 8638 | if (map_length >= submit_len) { |
02f57c7a | 8639 | bio = orig_bio; |
c1dc0896 | 8640 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8641 | goto submit; |
8642 | } | |
8643 | ||
53b381b3 | 8644 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8645 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8646 | async_submit = 0; |
8647 | else | |
8648 | async_submit = 1; | |
8649 | ||
725130ba LB |
8650 | /* bio split */ |
8651 | ASSERT(map_length <= INT_MAX); | |
02f57c7a | 8652 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8653 | do { |
725130ba | 8654 | clone_len = min_t(int, submit_len, map_length); |
02f57c7a | 8655 | |
725130ba LB |
8656 | /* |
8657 | * This will never fail as it's passing GPF_NOFS and | |
8658 | * the allocation is backed by btrfs_bioset. | |
8659 | */ | |
e477094f | 8660 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8661 | clone_len); |
8662 | bio->bi_private = dip; | |
8663 | bio->bi_end_io = btrfs_end_dio_bio; | |
8664 | btrfs_io_bio(bio)->logical = file_offset; | |
8665 | ||
8666 | ASSERT(submit_len >= clone_len); | |
8667 | submit_len -= clone_len; | |
8668 | if (submit_len == 0) | |
8669 | break; | |
e65e1535 | 8670 | |
725130ba LB |
8671 | /* |
8672 | * Increase the count before we submit the bio so we know | |
8673 | * the end IO handler won't happen before we increase the | |
8674 | * count. Otherwise, the dip might get freed before we're | |
8675 | * done setting it up. | |
8676 | */ | |
8677 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8678 | |
66ba772e | 8679 | status = __btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8680 | async_submit); |
8681 | if (status) { | |
725130ba LB |
8682 | bio_put(bio); |
8683 | atomic_dec(&dip->pending_bios); | |
8684 | goto out_err; | |
8685 | } | |
e65e1535 | 8686 | |
725130ba LB |
8687 | clone_offset += clone_len; |
8688 | start_sector += clone_len >> 9; | |
8689 | file_offset += clone_len; | |
5f4dc8fc | 8690 | |
725130ba LB |
8691 | map_length = submit_len; |
8692 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), | |
8693 | start_sector << 9, &map_length, NULL, 0); | |
8694 | if (ret) | |
8695 | goto out_err; | |
3c91ee69 | 8696 | } while (submit_len > 0); |
e65e1535 | 8697 | |
02f57c7a | 8698 | submit: |
66ba772e | 8699 | status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8700 | if (!status) |
e65e1535 MX |
8701 | return 0; |
8702 | ||
8703 | bio_put(bio); | |
8704 | out_err: | |
8705 | dip->errors = 1; | |
8706 | /* | |
8707 | * before atomic variable goto zero, we must | |
8708 | * make sure dip->errors is perceived to be set. | |
8709 | */ | |
4e857c58 | 8710 | smp_mb__before_atomic(); |
e65e1535 MX |
8711 | if (atomic_dec_and_test(&dip->pending_bios)) |
8712 | bio_io_error(dip->orig_bio); | |
8713 | ||
8714 | /* bio_end_io() will handle error, so we needn't return it */ | |
8715 | return 0; | |
8716 | } | |
8717 | ||
8a4c1e42 MC |
8718 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8719 | loff_t file_offset) | |
4b46fce2 | 8720 | { |
61de718f | 8721 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8722 | struct bio *bio = NULL; |
8723 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8724 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8725 | int ret = 0; |
8726 | ||
8b6c1d56 | 8727 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8728 | |
c1dc0896 | 8729 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8730 | if (!dip) { |
8731 | ret = -ENOMEM; | |
61de718f | 8732 | goto free_ordered; |
4b46fce2 | 8733 | } |
4b46fce2 | 8734 | |
9be3395b | 8735 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8736 | dip->inode = inode; |
8737 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8738 | dip->bytes = dio_bio->bi_iter.bi_size; |
8739 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8740 | bio->bi_private = dip; |
8741 | dip->orig_bio = bio; | |
9be3395b | 8742 | dip->dio_bio = dio_bio; |
e65e1535 | 8743 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8744 | io_bio = btrfs_io_bio(bio); |
8745 | io_bio->logical = file_offset; | |
4b46fce2 | 8746 | |
c1dc0896 | 8747 | if (write) { |
3892ac90 | 8748 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8749 | } else { |
3892ac90 | 8750 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8751 | dip->subio_endio = btrfs_subio_endio_read; |
8752 | } | |
4b46fce2 | 8753 | |
f28a4928 FM |
8754 | /* |
8755 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8756 | * even if we fail to submit a bio, because in such case we do the | |
8757 | * corresponding error handling below and it must not be done a second | |
8758 | * time by btrfs_direct_IO(). | |
8759 | */ | |
8760 | if (write) { | |
8761 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8762 | ||
8763 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8764 | dip->bytes; | |
8765 | dio_data->unsubmitted_oe_range_start = | |
8766 | dio_data->unsubmitted_oe_range_end; | |
8767 | } | |
8768 | ||
e6961cac | 8769 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8770 | if (!ret) |
eaf25d93 | 8771 | return; |
9be3395b | 8772 | |
3892ac90 LB |
8773 | if (io_bio->end_io) |
8774 | io_bio->end_io(io_bio, ret); | |
9be3395b | 8775 | |
4b46fce2 JB |
8776 | free_ordered: |
8777 | /* | |
61de718f FM |
8778 | * If we arrived here it means either we failed to submit the dip |
8779 | * or we either failed to clone the dio_bio or failed to allocate the | |
8780 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8781 | * call bio_endio against our io_bio so that we get proper resource | |
8782 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8783 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8784 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8785 | */ |
3892ac90 | 8786 | if (bio && dip) { |
054ec2f6 | 8787 | bio_io_error(bio); |
61de718f | 8788 | /* |
3892ac90 | 8789 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8790 | * and all the cleanup and final put for dio_bio (through |
8791 | * dio_end_io()). | |
8792 | */ | |
8793 | dip = NULL; | |
3892ac90 | 8794 | bio = NULL; |
61de718f | 8795 | } else { |
14543774 | 8796 | if (write) |
52427260 | 8797 | __endio_write_update_ordered(inode, |
14543774 FM |
8798 | file_offset, |
8799 | dio_bio->bi_iter.bi_size, | |
52427260 | 8800 | false); |
14543774 | 8801 | else |
61de718f FM |
8802 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8803 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8804 | |
4e4cbee9 | 8805 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8806 | /* |
8807 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8808 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8809 | */ | |
4055351c | 8810 | dio_end_io(dio_bio); |
4b46fce2 | 8811 | } |
3892ac90 LB |
8812 | if (bio) |
8813 | bio_put(bio); | |
61de718f | 8814 | kfree(dip); |
4b46fce2 JB |
8815 | } |
8816 | ||
2ff7e61e JM |
8817 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
8818 | struct kiocb *iocb, | |
8819 | const struct iov_iter *iter, loff_t offset) | |
5a5f79b5 CM |
8820 | { |
8821 | int seg; | |
a1b75f7d | 8822 | int i; |
0b246afa | 8823 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8824 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8825 | |
8826 | if (offset & blocksize_mask) | |
8827 | goto out; | |
8828 | ||
28060d5d AV |
8829 | if (iov_iter_alignment(iter) & blocksize_mask) |
8830 | goto out; | |
a1b75f7d | 8831 | |
28060d5d | 8832 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8833 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8834 | return 0; |
8835 | /* | |
8836 | * Check to make sure we don't have duplicate iov_base's in this | |
8837 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8838 | * when reading back. | |
8839 | */ | |
8840 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8841 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8842 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8843 | goto out; |
8844 | } | |
5a5f79b5 CM |
8845 | } |
8846 | retval = 0; | |
8847 | out: | |
8848 | return retval; | |
8849 | } | |
eb838e73 | 8850 | |
c8b8e32d | 8851 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8852 | { |
4b46fce2 JB |
8853 | struct file *file = iocb->ki_filp; |
8854 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8855 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8856 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8857 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8858 | loff_t offset = iocb->ki_pos; |
0934856d | 8859 | size_t count = 0; |
2e60a51e | 8860 | int flags = 0; |
38851cc1 MX |
8861 | bool wakeup = true; |
8862 | bool relock = false; | |
0934856d | 8863 | ssize_t ret; |
4b46fce2 | 8864 | |
2ff7e61e | 8865 | if (check_direct_IO(fs_info, iocb, iter, offset)) |
5a5f79b5 | 8866 | return 0; |
3f7c579c | 8867 | |
fe0f07d0 | 8868 | inode_dio_begin(inode); |
38851cc1 | 8869 | |
0e267c44 | 8870 | /* |
41bd9ca4 MX |
8871 | * The generic stuff only does filemap_write_and_wait_range, which |
8872 | * isn't enough if we've written compressed pages to this area, so | |
8873 | * we need to flush the dirty pages again to make absolutely sure | |
8874 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8875 | */ |
a6cbcd4a | 8876 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8877 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8878 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8879 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8880 | offset + count - 1); | |
0e267c44 | 8881 | |
6f673763 | 8882 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8883 | /* |
8884 | * If the write DIO is beyond the EOF, we need update | |
8885 | * the isize, but it is protected by i_mutex. So we can | |
8886 | * not unlock the i_mutex at this case. | |
8887 | */ | |
8888 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8889 | dio_data.overwrite = 1; |
5955102c | 8890 | inode_unlock(inode); |
38851cc1 | 8891 | relock = true; |
edf064e7 GR |
8892 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8893 | ret = -EAGAIN; | |
8894 | goto out; | |
38851cc1 | 8895 | } |
364ecf36 QW |
8896 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8897 | offset, count); | |
0934856d | 8898 | if (ret) |
38851cc1 | 8899 | goto out; |
823bb20a | 8900 | dio_data.outstanding_extents = count_max_extents(count); |
e1cbbfa5 JB |
8901 | |
8902 | /* | |
8903 | * We need to know how many extents we reserved so that we can | |
8904 | * do the accounting properly if we go over the number we | |
8905 | * originally calculated. Abuse current->journal_info for this. | |
8906 | */ | |
da17066c | 8907 | dio_data.reserve = round_up(count, |
0b246afa | 8908 | fs_info->sectorsize); |
f28a4928 FM |
8909 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8910 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8911 | current->journal_info = &dio_data; |
97dcdea0 | 8912 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8913 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8914 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8915 | inode_dio_end(inode); |
38851cc1 MX |
8916 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8917 | wakeup = false; | |
0934856d MX |
8918 | } |
8919 | ||
17f8c842 | 8920 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8921 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8922 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8923 | btrfs_submit_direct, flags); |
6f673763 | 8924 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8925 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8926 | current->journal_info = NULL; |
ddba1bfc | 8927 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8928 | if (dio_data.reserve) |
bc42bda2 QW |
8929 | btrfs_delalloc_release_space(inode, data_reserved, |
8930 | offset, dio_data.reserve); | |
f28a4928 FM |
8931 | /* |
8932 | * On error we might have left some ordered extents | |
8933 | * without submitting corresponding bios for them, so | |
8934 | * cleanup them up to avoid other tasks getting them | |
8935 | * and waiting for them to complete forever. | |
8936 | */ | |
8937 | if (dio_data.unsubmitted_oe_range_start < | |
8938 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8939 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8940 | dio_data.unsubmitted_oe_range_start, |
8941 | dio_data.unsubmitted_oe_range_end - | |
8942 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8943 | false); |
ddba1bfc | 8944 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 QW |
8945 | btrfs_delalloc_release_space(inode, data_reserved, |
8946 | offset, count - (size_t)ret); | |
0934856d | 8947 | } |
38851cc1 | 8948 | out: |
2e60a51e | 8949 | if (wakeup) |
fe0f07d0 | 8950 | inode_dio_end(inode); |
38851cc1 | 8951 | if (relock) |
5955102c | 8952 | inode_lock(inode); |
0934856d | 8953 | |
364ecf36 | 8954 | extent_changeset_free(data_reserved); |
0934856d | 8955 | return ret; |
16432985 CM |
8956 | } |
8957 | ||
05dadc09 TI |
8958 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8959 | ||
1506fcc8 YS |
8960 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8961 | __u64 start, __u64 len) | |
8962 | { | |
05dadc09 TI |
8963 | int ret; |
8964 | ||
8965 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8966 | if (ret) | |
8967 | return ret; | |
8968 | ||
ec29ed5b | 8969 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8970 | } |
8971 | ||
a52d9a80 | 8972 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8973 | { |
d1310b2e CM |
8974 | struct extent_io_tree *tree; |
8975 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8976 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8977 | } |
1832a6d5 | 8978 | |
a52d9a80 | 8979 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8980 | { |
d1310b2e | 8981 | struct extent_io_tree *tree; |
be7bd730 JB |
8982 | struct inode *inode = page->mapping->host; |
8983 | int ret; | |
b888db2b CM |
8984 | |
8985 | if (current->flags & PF_MEMALLOC) { | |
8986 | redirty_page_for_writepage(wbc, page); | |
8987 | unlock_page(page); | |
8988 | return 0; | |
8989 | } | |
be7bd730 JB |
8990 | |
8991 | /* | |
8992 | * If we are under memory pressure we will call this directly from the | |
8993 | * VM, we need to make sure we have the inode referenced for the ordered | |
8994 | * extent. If not just return like we didn't do anything. | |
8995 | */ | |
8996 | if (!igrab(inode)) { | |
8997 | redirty_page_for_writepage(wbc, page); | |
8998 | return AOP_WRITEPAGE_ACTIVATE; | |
8999 | } | |
d1310b2e | 9000 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
9001 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9002 | btrfs_add_delayed_iput(inode); | |
9003 | return ret; | |
9ebefb18 CM |
9004 | } |
9005 | ||
48a3b636 ES |
9006 | static int btrfs_writepages(struct address_space *mapping, |
9007 | struct writeback_control *wbc) | |
b293f02e | 9008 | { |
d1310b2e | 9009 | struct extent_io_tree *tree; |
771ed689 | 9010 | |
d1310b2e | 9011 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
9012 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
9013 | } | |
9014 | ||
3ab2fb5a CM |
9015 | static int |
9016 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
9017 | struct list_head *pages, unsigned nr_pages) | |
9018 | { | |
d1310b2e CM |
9019 | struct extent_io_tree *tree; |
9020 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
9021 | return extent_readpages(tree, mapping, pages, nr_pages, |
9022 | btrfs_get_extent); | |
9023 | } | |
e6dcd2dc | 9024 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 9025 | { |
d1310b2e CM |
9026 | struct extent_io_tree *tree; |
9027 | struct extent_map_tree *map; | |
a52d9a80 | 9028 | int ret; |
8c2383c3 | 9029 | |
d1310b2e CM |
9030 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
9031 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 9032 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
9033 | if (ret == 1) { |
9034 | ClearPagePrivate(page); | |
9035 | set_page_private(page, 0); | |
09cbfeaf | 9036 | put_page(page); |
39279cc3 | 9037 | } |
a52d9a80 | 9038 | return ret; |
39279cc3 CM |
9039 | } |
9040 | ||
e6dcd2dc CM |
9041 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9042 | { | |
98509cfc CM |
9043 | if (PageWriteback(page) || PageDirty(page)) |
9044 | return 0; | |
3ba7ab22 | 9045 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
9046 | } |
9047 | ||
d47992f8 LC |
9048 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
9049 | unsigned int length) | |
39279cc3 | 9050 | { |
5fd02043 | 9051 | struct inode *inode = page->mapping->host; |
d1310b2e | 9052 | struct extent_io_tree *tree; |
e6dcd2dc | 9053 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 9054 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 9055 | u64 page_start = page_offset(page); |
09cbfeaf | 9056 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
9057 | u64 start; |
9058 | u64 end; | |
131e404a | 9059 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 9060 | |
8b62b72b CM |
9061 | /* |
9062 | * we have the page locked, so new writeback can't start, | |
9063 | * and the dirty bit won't be cleared while we are here. | |
9064 | * | |
9065 | * Wait for IO on this page so that we can safely clear | |
9066 | * the PagePrivate2 bit and do ordered accounting | |
9067 | */ | |
e6dcd2dc | 9068 | wait_on_page_writeback(page); |
8b62b72b | 9069 | |
5fd02043 | 9070 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
9071 | if (offset) { |
9072 | btrfs_releasepage(page, GFP_NOFS); | |
9073 | return; | |
9074 | } | |
131e404a FDBM |
9075 | |
9076 | if (!inode_evicting) | |
ff13db41 | 9077 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
9078 | again: |
9079 | start = page_start; | |
a776c6fa | 9080 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 9081 | page_end - start + 1); |
e6dcd2dc | 9082 | if (ordered) { |
dbfdb6d1 | 9083 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
9084 | /* |
9085 | * IO on this page will never be started, so we need | |
9086 | * to account for any ordered extents now | |
9087 | */ | |
131e404a | 9088 | if (!inode_evicting) |
dbfdb6d1 | 9089 | clear_extent_bit(tree, start, end, |
131e404a | 9090 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 9091 | EXTENT_DELALLOC_NEW | |
131e404a FDBM |
9092 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
9093 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
9094 | GFP_NOFS); | |
8b62b72b CM |
9095 | /* |
9096 | * whoever cleared the private bit is responsible | |
9097 | * for the finish_ordered_io | |
9098 | */ | |
77cef2ec JB |
9099 | if (TestClearPagePrivate2(page)) { |
9100 | struct btrfs_ordered_inode_tree *tree; | |
9101 | u64 new_len; | |
9102 | ||
9103 | tree = &BTRFS_I(inode)->ordered_tree; | |
9104 | ||
9105 | spin_lock_irq(&tree->lock); | |
9106 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 9107 | new_len = start - ordered->file_offset; |
77cef2ec JB |
9108 | if (new_len < ordered->truncated_len) |
9109 | ordered->truncated_len = new_len; | |
9110 | spin_unlock_irq(&tree->lock); | |
9111 | ||
9112 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
9113 | start, |
9114 | end - start + 1, 1)) | |
77cef2ec | 9115 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 9116 | } |
e6dcd2dc | 9117 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
9118 | if (!inode_evicting) { |
9119 | cached_state = NULL; | |
dbfdb6d1 | 9120 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
9121 | &cached_state); |
9122 | } | |
dbfdb6d1 CR |
9123 | |
9124 | start = end + 1; | |
9125 | if (start < page_end) | |
9126 | goto again; | |
131e404a FDBM |
9127 | } |
9128 | ||
b9d0b389 QW |
9129 | /* |
9130 | * Qgroup reserved space handler | |
9131 | * Page here will be either | |
9132 | * 1) Already written to disk | |
9133 | * In this case, its reserved space is released from data rsv map | |
9134 | * and will be freed by delayed_ref handler finally. | |
9135 | * So even we call qgroup_free_data(), it won't decrease reserved | |
9136 | * space. | |
9137 | * 2) Not written to disk | |
0b34c261 GR |
9138 | * This means the reserved space should be freed here. However, |
9139 | * if a truncate invalidates the page (by clearing PageDirty) | |
9140 | * and the page is accounted for while allocating extent | |
9141 | * in btrfs_check_data_free_space() we let delayed_ref to | |
9142 | * free the entire extent. | |
b9d0b389 | 9143 | */ |
0b34c261 | 9144 | if (PageDirty(page)) |
bc42bda2 | 9145 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
9146 | if (!inode_evicting) { |
9147 | clear_extent_bit(tree, page_start, page_end, | |
9148 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
9149 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
9150 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
131e404a FDBM |
9151 | &cached_state, GFP_NOFS); |
9152 | ||
9153 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 9154 | } |
e6dcd2dc | 9155 | |
4a096752 | 9156 | ClearPageChecked(page); |
9ad6b7bc | 9157 | if (PagePrivate(page)) { |
9ad6b7bc CM |
9158 | ClearPagePrivate(page); |
9159 | set_page_private(page, 0); | |
09cbfeaf | 9160 | put_page(page); |
9ad6b7bc | 9161 | } |
39279cc3 CM |
9162 | } |
9163 | ||
9ebefb18 CM |
9164 | /* |
9165 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
9166 | * called from a page fault handler when a page is first dirtied. Hence we must | |
9167 | * be careful to check for EOF conditions here. We set the page up correctly | |
9168 | * for a written page which means we get ENOSPC checking when writing into | |
9169 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
9170 | * support these features. | |
9171 | * | |
9172 | * We are not allowed to take the i_mutex here so we have to play games to | |
9173 | * protect against truncate races as the page could now be beyond EOF. Because | |
9174 | * vmtruncate() writes the inode size before removing pages, once we have the | |
9175 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
9176 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
9177 | * unlock the page. | |
9178 | */ | |
11bac800 | 9179 | int btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 9180 | { |
c2ec175c | 9181 | struct page *page = vmf->page; |
11bac800 | 9182 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 9183 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
9184 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
9185 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 9186 | struct extent_state *cached_state = NULL; |
364ecf36 | 9187 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
9188 | char *kaddr; |
9189 | unsigned long zero_start; | |
9ebefb18 | 9190 | loff_t size; |
1832a6d5 | 9191 | int ret; |
9998eb70 | 9192 | int reserved = 0; |
d0b7da88 | 9193 | u64 reserved_space; |
a52d9a80 | 9194 | u64 page_start; |
e6dcd2dc | 9195 | u64 page_end; |
d0b7da88 CR |
9196 | u64 end; |
9197 | ||
09cbfeaf | 9198 | reserved_space = PAGE_SIZE; |
9ebefb18 | 9199 | |
b2b5ef5c | 9200 | sb_start_pagefault(inode->i_sb); |
df480633 | 9201 | page_start = page_offset(page); |
09cbfeaf | 9202 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 9203 | end = page_end; |
df480633 | 9204 | |
d0b7da88 CR |
9205 | /* |
9206 | * Reserving delalloc space after obtaining the page lock can lead to | |
9207 | * deadlock. For example, if a dirty page is locked by this function | |
9208 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
9209 | * dirty page write out, then the btrfs_writepage() function could | |
9210 | * end up waiting indefinitely to get a lock on the page currently | |
9211 | * being processed by btrfs_page_mkwrite() function. | |
9212 | */ | |
364ecf36 | 9213 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 9214 | reserved_space); |
9998eb70 | 9215 | if (!ret) { |
11bac800 | 9216 | ret = file_update_time(vmf->vma->vm_file); |
9998eb70 CM |
9217 | reserved = 1; |
9218 | } | |
56a76f82 NP |
9219 | if (ret) { |
9220 | if (ret == -ENOMEM) | |
9221 | ret = VM_FAULT_OOM; | |
9222 | else /* -ENOSPC, -EIO, etc */ | |
9223 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
9224 | if (reserved) |
9225 | goto out; | |
9226 | goto out_noreserve; | |
56a76f82 | 9227 | } |
1832a6d5 | 9228 | |
56a76f82 | 9229 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 9230 | again: |
9ebefb18 | 9231 | lock_page(page); |
9ebefb18 | 9232 | size = i_size_read(inode); |
a52d9a80 | 9233 | |
9ebefb18 | 9234 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 9235 | (page_start >= size)) { |
9ebefb18 CM |
9236 | /* page got truncated out from underneath us */ |
9237 | goto out_unlock; | |
9238 | } | |
e6dcd2dc CM |
9239 | wait_on_page_writeback(page); |
9240 | ||
ff13db41 | 9241 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
9242 | set_page_extent_mapped(page); |
9243 | ||
eb84ae03 CM |
9244 | /* |
9245 | * we can't set the delalloc bits if there are pending ordered | |
9246 | * extents. Drop our locks and wait for them to finish | |
9247 | */ | |
a776c6fa NB |
9248 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
9249 | PAGE_SIZE); | |
e6dcd2dc | 9250 | if (ordered) { |
2ac55d41 JB |
9251 | unlock_extent_cached(io_tree, page_start, page_end, |
9252 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 9253 | unlock_page(page); |
eb84ae03 | 9254 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
9255 | btrfs_put_ordered_extent(ordered); |
9256 | goto again; | |
9257 | } | |
9258 | ||
09cbfeaf | 9259 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 9260 | reserved_space = round_up(size - page_start, |
0b246afa | 9261 | fs_info->sectorsize); |
09cbfeaf | 9262 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 CR |
9263 | end = page_start + reserved_space - 1; |
9264 | spin_lock(&BTRFS_I(inode)->lock); | |
9265 | BTRFS_I(inode)->outstanding_extents++; | |
9266 | spin_unlock(&BTRFS_I(inode)->lock); | |
bc42bda2 QW |
9267 | btrfs_delalloc_release_space(inode, data_reserved, |
9268 | page_start, PAGE_SIZE - reserved_space); | |
d0b7da88 CR |
9269 | } |
9270 | } | |
9271 | ||
fbf19087 | 9272 | /* |
5416034f LB |
9273 | * page_mkwrite gets called when the page is firstly dirtied after it's |
9274 | * faulted in, but write(2) could also dirty a page and set delalloc | |
9275 | * bits, thus in this case for space account reason, we still need to | |
9276 | * clear any delalloc bits within this page range since we have to | |
9277 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 9278 | */ |
d0b7da88 | 9279 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9280 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9281 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 9282 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 9283 | |
d0b7da88 | 9284 | ret = btrfs_set_extent_delalloc(inode, page_start, end, |
ba8b04c1 | 9285 | &cached_state, 0); |
9ed74f2d | 9286 | if (ret) { |
2ac55d41 JB |
9287 | unlock_extent_cached(io_tree, page_start, page_end, |
9288 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
9289 | ret = VM_FAULT_SIGBUS; |
9290 | goto out_unlock; | |
9291 | } | |
e6dcd2dc | 9292 | ret = 0; |
9ebefb18 CM |
9293 | |
9294 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9295 | if (page_start + PAGE_SIZE > size) |
9296 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9297 | else |
09cbfeaf | 9298 | zero_start = PAGE_SIZE; |
9ebefb18 | 9299 | |
09cbfeaf | 9300 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9301 | kaddr = kmap(page); |
09cbfeaf | 9302 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9303 | flush_dcache_page(page); |
9304 | kunmap(page); | |
9305 | } | |
247e743c | 9306 | ClearPageChecked(page); |
e6dcd2dc | 9307 | set_page_dirty(page); |
50a9b214 | 9308 | SetPageUptodate(page); |
5a3f23d5 | 9309 | |
0b246afa | 9310 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9311 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9312 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9313 | |
2ac55d41 | 9314 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9315 | |
9316 | out_unlock: | |
b2b5ef5c JK |
9317 | if (!ret) { |
9318 | sb_end_pagefault(inode->i_sb); | |
364ecf36 | 9319 | extent_changeset_free(data_reserved); |
50a9b214 | 9320 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9321 | } |
9ebefb18 | 9322 | unlock_page(page); |
1832a6d5 | 9323 | out: |
bc42bda2 QW |
9324 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
9325 | reserved_space); | |
9998eb70 | 9326 | out_noreserve: |
b2b5ef5c | 9327 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9328 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
9329 | return ret; |
9330 | } | |
9331 | ||
a41ad394 | 9332 | static int btrfs_truncate(struct inode *inode) |
39279cc3 | 9333 | { |
0b246afa | 9334 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9335 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9336 | struct btrfs_block_rsv *rsv; |
a71754fc | 9337 | int ret = 0; |
3893e33b | 9338 | int err = 0; |
39279cc3 | 9339 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9340 | u64 mask = fs_info->sectorsize - 1; |
9341 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9342 | |
0ef8b726 JB |
9343 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9344 | (u64)-1); | |
9345 | if (ret) | |
9346 | return ret; | |
39279cc3 | 9347 | |
fcb80c2a | 9348 | /* |
01327610 | 9349 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9350 | * 3 things going on here |
9351 | * | |
9352 | * 1) We need to reserve space for our orphan item and the space to | |
9353 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9354 | * orphan item because we didn't reserve space to remove it. | |
9355 | * | |
9356 | * 2) We need to reserve space to update our inode. | |
9357 | * | |
9358 | * 3) We need to have something to cache all the space that is going to | |
9359 | * be free'd up by the truncate operation, but also have some slack | |
9360 | * space reserved in case it uses space during the truncate (thank you | |
9361 | * very much snapshotting). | |
9362 | * | |
01327610 | 9363 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9364 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9365 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9366 | * doesn't end up using space reserved for updating the inode or |
9367 | * removing the orphan item. We also need to be able to stop the | |
9368 | * transaction and start a new one, which means we need to be able to | |
9369 | * update the inode several times, and we have no idea of knowing how | |
9370 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9371 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9372 | * Then there is the orphan item, which does indeed need to be held on |
9373 | * to for the whole operation, and we need nobody to touch this reserved | |
9374 | * space except the orphan code. | |
9375 | * | |
9376 | * So that leaves us with | |
9377 | * | |
9378 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9379 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9380 | * transaction reservation. | |
9381 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9382 | * updating the inode. | |
9383 | */ | |
2ff7e61e | 9384 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9385 | if (!rsv) |
9386 | return -ENOMEM; | |
4a338542 | 9387 | rsv->size = min_size; |
ca7e70f5 | 9388 | rsv->failfast = 1; |
f0cd846e | 9389 | |
907cbceb | 9390 | /* |
07127184 | 9391 | * 1 for the truncate slack space |
907cbceb JB |
9392 | * 1 for updating the inode. |
9393 | */ | |
f3fe820c | 9394 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9395 | if (IS_ERR(trans)) { |
9396 | err = PTR_ERR(trans); | |
9397 | goto out; | |
9398 | } | |
f0cd846e | 9399 | |
907cbceb | 9400 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9401 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 9402 | min_size, 0); |
fcb80c2a | 9403 | BUG_ON(ret); |
f0cd846e | 9404 | |
5dc562c5 JB |
9405 | /* |
9406 | * So if we truncate and then write and fsync we normally would just | |
9407 | * write the extents that changed, which is a problem if we need to | |
9408 | * first truncate that entire inode. So set this flag so we write out | |
9409 | * all of the extents in the inode to the sync log so we're completely | |
9410 | * safe. | |
9411 | */ | |
9412 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9413 | trans->block_rsv = rsv; |
907cbceb | 9414 | |
8082510e YZ |
9415 | while (1) { |
9416 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9417 | inode->i_size, | |
9418 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 9419 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9420 | err = ret; |
8082510e | 9421 | break; |
3893e33b | 9422 | } |
39279cc3 | 9423 | |
0b246afa | 9424 | trans->block_rsv = &fs_info->trans_block_rsv; |
8082510e | 9425 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9426 | if (ret) { |
9427 | err = ret; | |
9428 | break; | |
9429 | } | |
ca7e70f5 | 9430 | |
3a45bb20 | 9431 | btrfs_end_transaction(trans); |
2ff7e61e | 9432 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9433 | |
9434 | trans = btrfs_start_transaction(root, 2); | |
9435 | if (IS_ERR(trans)) { | |
9436 | ret = err = PTR_ERR(trans); | |
9437 | trans = NULL; | |
9438 | break; | |
9439 | } | |
9440 | ||
47b5d646 | 9441 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9442 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 9443 | rsv, min_size, 0); |
ca7e70f5 JB |
9444 | BUG_ON(ret); /* shouldn't happen */ |
9445 | trans->block_rsv = rsv; | |
8082510e YZ |
9446 | } |
9447 | ||
9448 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 9449 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 9450 | ret = btrfs_orphan_del(trans, BTRFS_I(inode)); |
3893e33b JB |
9451 | if (ret) |
9452 | err = ret; | |
8082510e YZ |
9453 | } |
9454 | ||
917c16b2 | 9455 | if (trans) { |
0b246afa | 9456 | trans->block_rsv = &fs_info->trans_block_rsv; |
917c16b2 CM |
9457 | ret = btrfs_update_inode(trans, root, inode); |
9458 | if (ret && !err) | |
9459 | err = ret; | |
7b128766 | 9460 | |
3a45bb20 | 9461 | ret = btrfs_end_transaction(trans); |
2ff7e61e | 9462 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9463 | } |
fcb80c2a | 9464 | out: |
2ff7e61e | 9465 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9466 | |
3893e33b JB |
9467 | if (ret && !err) |
9468 | err = ret; | |
a41ad394 | 9469 | |
3893e33b | 9470 | return err; |
39279cc3 CM |
9471 | } |
9472 | ||
d352ac68 CM |
9473 | /* |
9474 | * create a new subvolume directory/inode (helper for the ioctl). | |
9475 | */ | |
d2fb3437 | 9476 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9477 | struct btrfs_root *new_root, |
9478 | struct btrfs_root *parent_root, | |
9479 | u64 new_dirid) | |
39279cc3 | 9480 | { |
39279cc3 | 9481 | struct inode *inode; |
76dda93c | 9482 | int err; |
00e4e6b3 | 9483 | u64 index = 0; |
39279cc3 | 9484 | |
12fc9d09 FA |
9485 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9486 | new_dirid, new_dirid, | |
9487 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9488 | &index); | |
54aa1f4d | 9489 | if (IS_ERR(inode)) |
f46b5a66 | 9490 | return PTR_ERR(inode); |
39279cc3 CM |
9491 | inode->i_op = &btrfs_dir_inode_operations; |
9492 | inode->i_fop = &btrfs_dir_file_operations; | |
9493 | ||
bfe86848 | 9494 | set_nlink(inode, 1); |
6ef06d27 | 9495 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9496 | unlock_new_inode(inode); |
3b96362c | 9497 | |
63541927 FDBM |
9498 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9499 | if (err) | |
9500 | btrfs_err(new_root->fs_info, | |
351fd353 | 9501 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9502 | new_root->root_key.objectid, err); |
9503 | ||
76dda93c | 9504 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9505 | |
76dda93c | 9506 | iput(inode); |
ce598979 | 9507 | return err; |
39279cc3 CM |
9508 | } |
9509 | ||
39279cc3 CM |
9510 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9511 | { | |
9512 | struct btrfs_inode *ei; | |
2ead6ae7 | 9513 | struct inode *inode; |
39279cc3 CM |
9514 | |
9515 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9516 | if (!ei) | |
9517 | return NULL; | |
2ead6ae7 YZ |
9518 | |
9519 | ei->root = NULL; | |
2ead6ae7 | 9520 | ei->generation = 0; |
15ee9bc7 | 9521 | ei->last_trans = 0; |
257c62e1 | 9522 | ei->last_sub_trans = 0; |
e02119d5 | 9523 | ei->logged_trans = 0; |
2ead6ae7 | 9524 | ei->delalloc_bytes = 0; |
a7e3b975 | 9525 | ei->new_delalloc_bytes = 0; |
47059d93 | 9526 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9527 | ei->disk_i_size = 0; |
9528 | ei->flags = 0; | |
7709cde3 | 9529 | ei->csum_bytes = 0; |
2ead6ae7 | 9530 | ei->index_cnt = (u64)-1; |
67de1176 | 9531 | ei->dir_index = 0; |
2ead6ae7 | 9532 | ei->last_unlink_trans = 0; |
46d8bc34 | 9533 | ei->last_log_commit = 0; |
8089fe62 | 9534 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9535 | |
9e0baf60 JB |
9536 | spin_lock_init(&ei->lock); |
9537 | ei->outstanding_extents = 0; | |
9538 | ei->reserved_extents = 0; | |
2ead6ae7 | 9539 | |
72ac3c0d | 9540 | ei->runtime_flags = 0; |
b52aa8c9 | 9541 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9542 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9543 | |
16cdcec7 MX |
9544 | ei->delayed_node = NULL; |
9545 | ||
9cc97d64 | 9546 | ei->i_otime.tv_sec = 0; |
9547 | ei->i_otime.tv_nsec = 0; | |
9548 | ||
2ead6ae7 | 9549 | inode = &ei->vfs_inode; |
a8067e02 | 9550 | extent_map_tree_init(&ei->extent_tree); |
c6100a4b JB |
9551 | extent_io_tree_init(&ei->io_tree, inode); |
9552 | extent_io_tree_init(&ei->io_failure_tree, inode); | |
0b32f4bb JB |
9553 | ei->io_tree.track_uptodate = 1; |
9554 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9555 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9556 | mutex_init(&ei->log_mutex); |
f248679e | 9557 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9558 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9559 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9560 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9561 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9562 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9563 | |
9564 | return inode; | |
39279cc3 CM |
9565 | } |
9566 | ||
aaedb55b JB |
9567 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9568 | void btrfs_test_destroy_inode(struct inode *inode) | |
9569 | { | |
dcdbc059 | 9570 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9571 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9572 | } | |
9573 | #endif | |
9574 | ||
fa0d7e3d NP |
9575 | static void btrfs_i_callback(struct rcu_head *head) |
9576 | { | |
9577 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9578 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9579 | } | |
9580 | ||
39279cc3 CM |
9581 | void btrfs_destroy_inode(struct inode *inode) |
9582 | { | |
0b246afa | 9583 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9584 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9585 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9586 | ||
b3d9b7a3 | 9587 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9588 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9589 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9590 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 | 9591 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9592 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9593 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9594 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9595 | |
a6dbd429 JB |
9596 | /* |
9597 | * This can happen where we create an inode, but somebody else also | |
9598 | * created the same inode and we need to destroy the one we already | |
9599 | * created. | |
9600 | */ | |
9601 | if (!root) | |
9602 | goto free; | |
9603 | ||
8a35d95f JB |
9604 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9605 | &BTRFS_I(inode)->runtime_flags)) { | |
0b246afa | 9606 | btrfs_info(fs_info, "inode %llu still on the orphan list", |
4a0cc7ca | 9607 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 9608 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9609 | } |
7b128766 | 9610 | |
d397712b | 9611 | while (1) { |
e6dcd2dc CM |
9612 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9613 | if (!ordered) | |
9614 | break; | |
9615 | else { | |
0b246afa | 9616 | btrfs_err(fs_info, |
5d163e0e JM |
9617 | "found ordered extent %llu %llu on inode cleanup", |
9618 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9619 | btrfs_remove_ordered_extent(inode, ordered); |
9620 | btrfs_put_ordered_extent(ordered); | |
9621 | btrfs_put_ordered_extent(ordered); | |
9622 | } | |
9623 | } | |
56fa9d07 | 9624 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9625 | inode_tree_del(inode); |
dcdbc059 | 9626 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
a6dbd429 | 9627 | free: |
fa0d7e3d | 9628 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9629 | } |
9630 | ||
45321ac5 | 9631 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9632 | { |
9633 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9634 | |
6379ef9f NA |
9635 | if (root == NULL) |
9636 | return 1; | |
9637 | ||
fa6ac876 | 9638 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9639 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9640 | return 1; |
76dda93c | 9641 | else |
45321ac5 | 9642 | return generic_drop_inode(inode); |
76dda93c YZ |
9643 | } |
9644 | ||
0ee0fda0 | 9645 | static void init_once(void *foo) |
39279cc3 CM |
9646 | { |
9647 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9648 | ||
9649 | inode_init_once(&ei->vfs_inode); | |
9650 | } | |
9651 | ||
9652 | void btrfs_destroy_cachep(void) | |
9653 | { | |
8c0a8537 KS |
9654 | /* |
9655 | * Make sure all delayed rcu free inodes are flushed before we | |
9656 | * destroy cache. | |
9657 | */ | |
9658 | rcu_barrier(); | |
5598e900 KM |
9659 | kmem_cache_destroy(btrfs_inode_cachep); |
9660 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9661 | kmem_cache_destroy(btrfs_path_cachep); |
9662 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9663 | } |
9664 | ||
9665 | int btrfs_init_cachep(void) | |
9666 | { | |
837e1972 | 9667 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9668 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9669 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9670 | init_once); | |
39279cc3 CM |
9671 | if (!btrfs_inode_cachep) |
9672 | goto fail; | |
9601e3f6 | 9673 | |
837e1972 | 9674 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9675 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9676 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9677 | if (!btrfs_trans_handle_cachep) |
9678 | goto fail; | |
9601e3f6 | 9679 | |
837e1972 | 9680 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9681 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9682 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9683 | if (!btrfs_path_cachep) |
9684 | goto fail; | |
9601e3f6 | 9685 | |
837e1972 | 9686 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9687 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9688 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9689 | if (!btrfs_free_space_cachep) |
9690 | goto fail; | |
9691 | ||
39279cc3 CM |
9692 | return 0; |
9693 | fail: | |
9694 | btrfs_destroy_cachep(); | |
9695 | return -ENOMEM; | |
9696 | } | |
9697 | ||
a528d35e DH |
9698 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9699 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9700 | { |
df0af1a5 | 9701 | u64 delalloc_bytes; |
a528d35e | 9702 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9703 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9704 | u32 bi_flags = BTRFS_I(inode)->flags; |
9705 | ||
9706 | stat->result_mask |= STATX_BTIME; | |
9707 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9708 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9709 | if (bi_flags & BTRFS_INODE_APPEND) | |
9710 | stat->attributes |= STATX_ATTR_APPEND; | |
9711 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9712 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9713 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9714 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9715 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9716 | stat->attributes |= STATX_ATTR_NODUMP; | |
9717 | ||
9718 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9719 | STATX_ATTR_COMPRESSED | | |
9720 | STATX_ATTR_IMMUTABLE | | |
9721 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9722 | |
39279cc3 | 9723 | generic_fillattr(inode, stat); |
0ee5dc67 | 9724 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9725 | |
9726 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9727 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9728 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9729 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9730 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9731 | return 0; |
9732 | } | |
9733 | ||
cdd1fedf DF |
9734 | static int btrfs_rename_exchange(struct inode *old_dir, |
9735 | struct dentry *old_dentry, | |
9736 | struct inode *new_dir, | |
9737 | struct dentry *new_dentry) | |
9738 | { | |
0b246afa | 9739 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9740 | struct btrfs_trans_handle *trans; |
9741 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9742 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9743 | struct inode *new_inode = new_dentry->d_inode; | |
9744 | struct inode *old_inode = old_dentry->d_inode; | |
c2050a45 | 9745 | struct timespec ctime = current_time(old_inode); |
cdd1fedf | 9746 | struct dentry *parent; |
4a0cc7ca NB |
9747 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9748 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9749 | u64 old_idx = 0; |
9750 | u64 new_idx = 0; | |
9751 | u64 root_objectid; | |
9752 | int ret; | |
6101eea4 | 9753 | int ret2; |
86e8aa0e FM |
9754 | bool root_log_pinned = false; |
9755 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9756 | |
9757 | /* we only allow rename subvolume link between subvolumes */ | |
9758 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9759 | return -EXDEV; | |
9760 | ||
9761 | /* close the race window with snapshot create/destroy ioctl */ | |
9762 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9763 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9764 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9765 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9766 | |
9767 | /* | |
9768 | * We want to reserve the absolute worst case amount of items. So if | |
9769 | * both inodes are subvols and we need to unlink them then that would | |
9770 | * require 4 item modifications, but if they are both normal inodes it | |
9771 | * would require 5 item modifications, so we'll assume their normal | |
9772 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9773 | * should cover the worst case number of items we'll modify. | |
9774 | */ | |
9775 | trans = btrfs_start_transaction(root, 12); | |
9776 | if (IS_ERR(trans)) { | |
9777 | ret = PTR_ERR(trans); | |
9778 | goto out_notrans; | |
9779 | } | |
9780 | ||
9781 | /* | |
9782 | * We need to find a free sequence number both in the source and | |
9783 | * in the destination directory for the exchange. | |
9784 | */ | |
877574e2 | 9785 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9786 | if (ret) |
9787 | goto out_fail; | |
877574e2 | 9788 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9789 | if (ret) |
9790 | goto out_fail; | |
9791 | ||
9792 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9793 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9794 | ||
9795 | /* Reference for the source. */ | |
9796 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9797 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9798 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9799 | } else { |
376e5a57 FM |
9800 | btrfs_pin_log_trans(root); |
9801 | root_log_pinned = true; | |
cdd1fedf DF |
9802 | ret = btrfs_insert_inode_ref(trans, dest, |
9803 | new_dentry->d_name.name, | |
9804 | new_dentry->d_name.len, | |
9805 | old_ino, | |
f85b7379 DS |
9806 | btrfs_ino(BTRFS_I(new_dir)), |
9807 | old_idx); | |
cdd1fedf DF |
9808 | if (ret) |
9809 | goto out_fail; | |
cdd1fedf DF |
9810 | } |
9811 | ||
9812 | /* And now for the dest. */ | |
9813 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9814 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9815 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9816 | } else { |
376e5a57 FM |
9817 | btrfs_pin_log_trans(dest); |
9818 | dest_log_pinned = true; | |
cdd1fedf DF |
9819 | ret = btrfs_insert_inode_ref(trans, root, |
9820 | old_dentry->d_name.name, | |
9821 | old_dentry->d_name.len, | |
9822 | new_ino, | |
f85b7379 DS |
9823 | btrfs_ino(BTRFS_I(old_dir)), |
9824 | new_idx); | |
cdd1fedf DF |
9825 | if (ret) |
9826 | goto out_fail; | |
cdd1fedf DF |
9827 | } |
9828 | ||
9829 | /* Update inode version and ctime/mtime. */ | |
9830 | inode_inc_iversion(old_dir); | |
9831 | inode_inc_iversion(new_dir); | |
9832 | inode_inc_iversion(old_inode); | |
9833 | inode_inc_iversion(new_inode); | |
9834 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9835 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9836 | old_inode->i_ctime = ctime; | |
9837 | new_inode->i_ctime = ctime; | |
9838 | ||
9839 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9840 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9841 | BTRFS_I(old_inode), 1); | |
9842 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9843 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9844 | } |
9845 | ||
9846 | /* src is a subvolume */ | |
9847 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9848 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9849 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9850 | root_objectid, | |
9851 | old_dentry->d_name.name, | |
9852 | old_dentry->d_name.len); | |
9853 | } else { /* src is an inode */ | |
4ec5934e NB |
9854 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9855 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9856 | old_dentry->d_name.name, |
9857 | old_dentry->d_name.len); | |
9858 | if (!ret) | |
9859 | ret = btrfs_update_inode(trans, root, old_inode); | |
9860 | } | |
9861 | if (ret) { | |
66642832 | 9862 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9863 | goto out_fail; |
9864 | } | |
9865 | ||
9866 | /* dest is a subvolume */ | |
9867 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9868 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9869 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9870 | root_objectid, | |
9871 | new_dentry->d_name.name, | |
9872 | new_dentry->d_name.len); | |
9873 | } else { /* dest is an inode */ | |
4ec5934e NB |
9874 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9875 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9876 | new_dentry->d_name.name, |
9877 | new_dentry->d_name.len); | |
9878 | if (!ret) | |
9879 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9880 | } | |
9881 | if (ret) { | |
66642832 | 9882 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9883 | goto out_fail; |
9884 | } | |
9885 | ||
db0a669f | 9886 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9887 | new_dentry->d_name.name, |
9888 | new_dentry->d_name.len, 0, old_idx); | |
9889 | if (ret) { | |
66642832 | 9890 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9891 | goto out_fail; |
9892 | } | |
9893 | ||
db0a669f | 9894 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9895 | old_dentry->d_name.name, |
9896 | old_dentry->d_name.len, 0, new_idx); | |
9897 | if (ret) { | |
66642832 | 9898 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9899 | goto out_fail; |
9900 | } | |
9901 | ||
9902 | if (old_inode->i_nlink == 1) | |
9903 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9904 | if (new_inode->i_nlink == 1) | |
9905 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9906 | ||
86e8aa0e | 9907 | if (root_log_pinned) { |
cdd1fedf | 9908 | parent = new_dentry->d_parent; |
f85b7379 DS |
9909 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9910 | parent); | |
cdd1fedf | 9911 | btrfs_end_log_trans(root); |
86e8aa0e | 9912 | root_log_pinned = false; |
cdd1fedf | 9913 | } |
86e8aa0e | 9914 | if (dest_log_pinned) { |
cdd1fedf | 9915 | parent = old_dentry->d_parent; |
f85b7379 DS |
9916 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9917 | parent); | |
cdd1fedf | 9918 | btrfs_end_log_trans(dest); |
86e8aa0e | 9919 | dest_log_pinned = false; |
cdd1fedf DF |
9920 | } |
9921 | out_fail: | |
86e8aa0e FM |
9922 | /* |
9923 | * If we have pinned a log and an error happened, we unpin tasks | |
9924 | * trying to sync the log and force them to fallback to a transaction | |
9925 | * commit if the log currently contains any of the inodes involved in | |
9926 | * this rename operation (to ensure we do not persist a log with an | |
9927 | * inconsistent state for any of these inodes or leading to any | |
9928 | * inconsistencies when replayed). If the transaction was aborted, the | |
9929 | * abortion reason is propagated to userspace when attempting to commit | |
9930 | * the transaction. If the log does not contain any of these inodes, we | |
9931 | * allow the tasks to sync it. | |
9932 | */ | |
9933 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9934 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9935 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9936 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9937 | (new_inode && |
0f8939b8 | 9938 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9939 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9940 | |
9941 | if (root_log_pinned) { | |
9942 | btrfs_end_log_trans(root); | |
9943 | root_log_pinned = false; | |
9944 | } | |
9945 | if (dest_log_pinned) { | |
9946 | btrfs_end_log_trans(dest); | |
9947 | dest_log_pinned = false; | |
9948 | } | |
9949 | } | |
6101eea4 FM |
9950 | ret2 = btrfs_end_transaction(trans); |
9951 | ret = ret ? ret : ret2; | |
cdd1fedf DF |
9952 | out_notrans: |
9953 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9954 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9955 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9956 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9957 | |
9958 | return ret; | |
9959 | } | |
9960 | ||
9961 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9962 | struct btrfs_root *root, | |
9963 | struct inode *dir, | |
9964 | struct dentry *dentry) | |
9965 | { | |
9966 | int ret; | |
9967 | struct inode *inode; | |
9968 | u64 objectid; | |
9969 | u64 index; | |
9970 | ||
9971 | ret = btrfs_find_free_ino(root, &objectid); | |
9972 | if (ret) | |
9973 | return ret; | |
9974 | ||
9975 | inode = btrfs_new_inode(trans, root, dir, | |
9976 | dentry->d_name.name, | |
9977 | dentry->d_name.len, | |
4a0cc7ca | 9978 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9979 | objectid, |
9980 | S_IFCHR | WHITEOUT_MODE, | |
9981 | &index); | |
9982 | ||
9983 | if (IS_ERR(inode)) { | |
9984 | ret = PTR_ERR(inode); | |
9985 | return ret; | |
9986 | } | |
9987 | ||
9988 | inode->i_op = &btrfs_special_inode_operations; | |
9989 | init_special_inode(inode, inode->i_mode, | |
9990 | WHITEOUT_DEV); | |
9991 | ||
9992 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9993 | &dentry->d_name); | |
9994 | if (ret) | |
c9901618 | 9995 | goto out; |
cdd1fedf | 9996 | |
cef415af NB |
9997 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9998 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9999 | if (ret) |
c9901618 | 10000 | goto out; |
cdd1fedf DF |
10001 | |
10002 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 10003 | out: |
cdd1fedf | 10004 | unlock_new_inode(inode); |
c9901618 FM |
10005 | if (ret) |
10006 | inode_dec_link_count(inode); | |
cdd1fedf DF |
10007 | iput(inode); |
10008 | ||
c9901618 | 10009 | return ret; |
cdd1fedf DF |
10010 | } |
10011 | ||
d397712b | 10012 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
10013 | struct inode *new_dir, struct dentry *new_dentry, |
10014 | unsigned int flags) | |
39279cc3 | 10015 | { |
0b246afa | 10016 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 10017 | struct btrfs_trans_handle *trans; |
5062af35 | 10018 | unsigned int trans_num_items; |
39279cc3 | 10019 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 10020 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
10021 | struct inode *new_inode = d_inode(new_dentry); |
10022 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 10023 | u64 index = 0; |
4df27c4d | 10024 | u64 root_objectid; |
39279cc3 | 10025 | int ret; |
4a0cc7ca | 10026 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 10027 | bool log_pinned = false; |
39279cc3 | 10028 | |
4a0cc7ca | 10029 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
10030 | return -EPERM; |
10031 | ||
4df27c4d | 10032 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 10033 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
10034 | return -EXDEV; |
10035 | ||
33345d01 | 10036 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 10037 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 10038 | return -ENOTEMPTY; |
5f39d397 | 10039 | |
4df27c4d YZ |
10040 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
10041 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
10042 | return -ENOTEMPTY; | |
9c52057c CM |
10043 | |
10044 | ||
10045 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 10046 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
10047 | new_dentry->d_name.name, |
10048 | new_dentry->d_name.len); | |
10049 | ||
10050 | if (ret) { | |
10051 | if (ret == -EEXIST) { | |
10052 | /* we shouldn't get | |
10053 | * eexist without a new_inode */ | |
fae7f21c | 10054 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
10055 | return ret; |
10056 | } | |
10057 | } else { | |
10058 | /* maybe -EOVERFLOW */ | |
10059 | return ret; | |
10060 | } | |
10061 | } | |
10062 | ret = 0; | |
10063 | ||
5a3f23d5 | 10064 | /* |
8d875f95 CM |
10065 | * we're using rename to replace one file with another. Start IO on it |
10066 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 10067 | */ |
8d875f95 | 10068 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
10069 | filemap_flush(old_inode->i_mapping); |
10070 | ||
76dda93c | 10071 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 10072 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 10073 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
10074 | /* |
10075 | * We want to reserve the absolute worst case amount of items. So if | |
10076 | * both inodes are subvols and we need to unlink them then that would | |
10077 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 10078 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
10079 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
10080 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
10081 | * If our rename has the whiteout flag, we need more 5 units for the |
10082 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
10083 | * when selinux is enabled). | |
a22285a6 | 10084 | */ |
5062af35 FM |
10085 | trans_num_items = 11; |
10086 | if (flags & RENAME_WHITEOUT) | |
10087 | trans_num_items += 5; | |
10088 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 10089 | if (IS_ERR(trans)) { |
cdd1fedf DF |
10090 | ret = PTR_ERR(trans); |
10091 | goto out_notrans; | |
10092 | } | |
76dda93c | 10093 | |
4df27c4d YZ |
10094 | if (dest != root) |
10095 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 10096 | |
877574e2 | 10097 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
10098 | if (ret) |
10099 | goto out_fail; | |
5a3f23d5 | 10100 | |
67de1176 | 10101 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 10102 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 10103 | /* force full log commit if subvolume involved. */ |
0b246afa | 10104 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 10105 | } else { |
c4aba954 FM |
10106 | btrfs_pin_log_trans(root); |
10107 | log_pinned = true; | |
a5719521 YZ |
10108 | ret = btrfs_insert_inode_ref(trans, dest, |
10109 | new_dentry->d_name.name, | |
10110 | new_dentry->d_name.len, | |
33345d01 | 10111 | old_ino, |
4a0cc7ca | 10112 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
10113 | if (ret) |
10114 | goto out_fail; | |
4df27c4d | 10115 | } |
5a3f23d5 | 10116 | |
0c4d2d95 JB |
10117 | inode_inc_iversion(old_dir); |
10118 | inode_inc_iversion(new_dir); | |
10119 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
10120 | old_dir->i_ctime = old_dir->i_mtime = |
10121 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 10122 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 10123 | |
12fcfd22 | 10124 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
10125 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
10126 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 10127 | |
33345d01 | 10128 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
10129 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
10130 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
10131 | old_dentry->d_name.name, | |
10132 | old_dentry->d_name.len); | |
10133 | } else { | |
4ec5934e NB |
10134 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
10135 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
10136 | old_dentry->d_name.name, |
10137 | old_dentry->d_name.len); | |
10138 | if (!ret) | |
10139 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 10140 | } |
79787eaa | 10141 | if (ret) { |
66642832 | 10142 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10143 | goto out_fail; |
10144 | } | |
39279cc3 CM |
10145 | |
10146 | if (new_inode) { | |
0c4d2d95 | 10147 | inode_inc_iversion(new_inode); |
c2050a45 | 10148 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 10149 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
10150 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
10151 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
10152 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
10153 | root_objectid, | |
10154 | new_dentry->d_name.name, | |
10155 | new_dentry->d_name.len); | |
10156 | BUG_ON(new_inode->i_nlink == 0); | |
10157 | } else { | |
4ec5934e NB |
10158 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
10159 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
10160 | new_dentry->d_name.name, |
10161 | new_dentry->d_name.len); | |
10162 | } | |
4ef31a45 | 10163 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
10164 | ret = btrfs_orphan_add(trans, |
10165 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 10166 | if (ret) { |
66642832 | 10167 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10168 | goto out_fail; |
10169 | } | |
39279cc3 | 10170 | } |
aec7477b | 10171 | |
db0a669f | 10172 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 10173 | new_dentry->d_name.name, |
a5719521 | 10174 | new_dentry->d_name.len, 0, index); |
79787eaa | 10175 | if (ret) { |
66642832 | 10176 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10177 | goto out_fail; |
10178 | } | |
39279cc3 | 10179 | |
67de1176 MX |
10180 | if (old_inode->i_nlink == 1) |
10181 | BTRFS_I(old_inode)->dir_index = index; | |
10182 | ||
3dc9e8f7 | 10183 | if (log_pinned) { |
10d9f309 | 10184 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 10185 | |
f85b7379 DS |
10186 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
10187 | parent); | |
4df27c4d | 10188 | btrfs_end_log_trans(root); |
3dc9e8f7 | 10189 | log_pinned = false; |
4df27c4d | 10190 | } |
cdd1fedf DF |
10191 | |
10192 | if (flags & RENAME_WHITEOUT) { | |
10193 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
10194 | old_dentry); | |
10195 | ||
10196 | if (ret) { | |
66642832 | 10197 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
10198 | goto out_fail; |
10199 | } | |
4df27c4d | 10200 | } |
39279cc3 | 10201 | out_fail: |
3dc9e8f7 FM |
10202 | /* |
10203 | * If we have pinned the log and an error happened, we unpin tasks | |
10204 | * trying to sync the log and force them to fallback to a transaction | |
10205 | * commit if the log currently contains any of the inodes involved in | |
10206 | * this rename operation (to ensure we do not persist a log with an | |
10207 | * inconsistent state for any of these inodes or leading to any | |
10208 | * inconsistencies when replayed). If the transaction was aborted, the | |
10209 | * abortion reason is propagated to userspace when attempting to commit | |
10210 | * the transaction. If the log does not contain any of these inodes, we | |
10211 | * allow the tasks to sync it. | |
10212 | */ | |
10213 | if (ret && log_pinned) { | |
0f8939b8 NB |
10214 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
10215 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
10216 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 10217 | (new_inode && |
0f8939b8 | 10218 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 10219 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
10220 | |
10221 | btrfs_end_log_trans(root); | |
10222 | log_pinned = false; | |
10223 | } | |
3a45bb20 | 10224 | btrfs_end_transaction(trans); |
b44c59a8 | 10225 | out_notrans: |
33345d01 | 10226 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 10227 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 10228 | |
39279cc3 CM |
10229 | return ret; |
10230 | } | |
10231 | ||
80ace85c MS |
10232 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
10233 | struct inode *new_dir, struct dentry *new_dentry, | |
10234 | unsigned int flags) | |
10235 | { | |
cdd1fedf | 10236 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
10237 | return -EINVAL; |
10238 | ||
cdd1fedf DF |
10239 | if (flags & RENAME_EXCHANGE) |
10240 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
10241 | new_dentry); | |
10242 | ||
10243 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
10244 | } |
10245 | ||
8ccf6f19 MX |
10246 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
10247 | { | |
10248 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 10249 | struct inode *inode; |
8ccf6f19 MX |
10250 | |
10251 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
10252 | work); | |
9f23e289 | 10253 | inode = delalloc_work->inode; |
30424601 DS |
10254 | filemap_flush(inode->i_mapping); |
10255 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
10256 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 10257 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
10258 | |
10259 | if (delalloc_work->delay_iput) | |
9f23e289 | 10260 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 10261 | else |
9f23e289 | 10262 | iput(inode); |
8ccf6f19 MX |
10263 | complete(&delalloc_work->completion); |
10264 | } | |
10265 | ||
10266 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 10267 | int delay_iput) |
8ccf6f19 MX |
10268 | { |
10269 | struct btrfs_delalloc_work *work; | |
10270 | ||
100d5702 | 10271 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
10272 | if (!work) |
10273 | return NULL; | |
10274 | ||
10275 | init_completion(&work->completion); | |
10276 | INIT_LIST_HEAD(&work->list); | |
10277 | work->inode = inode; | |
8ccf6f19 | 10278 | work->delay_iput = delay_iput; |
9e0af237 LB |
10279 | WARN_ON_ONCE(!inode); |
10280 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
10281 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10282 | |
10283 | return work; | |
10284 | } | |
10285 | ||
10286 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
10287 | { | |
10288 | wait_for_completion(&work->completion); | |
100d5702 | 10289 | kfree(work); |
8ccf6f19 MX |
10290 | } |
10291 | ||
d352ac68 CM |
10292 | /* |
10293 | * some fairly slow code that needs optimization. This walks the list | |
10294 | * of all the inodes with pending delalloc and forces them to disk. | |
10295 | */ | |
6c255e67 MX |
10296 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
10297 | int nr) | |
ea8c2819 | 10298 | { |
ea8c2819 | 10299 | struct btrfs_inode *binode; |
5b21f2ed | 10300 | struct inode *inode; |
8ccf6f19 MX |
10301 | struct btrfs_delalloc_work *work, *next; |
10302 | struct list_head works; | |
1eafa6c7 | 10303 | struct list_head splice; |
8ccf6f19 | 10304 | int ret = 0; |
ea8c2819 | 10305 | |
8ccf6f19 | 10306 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10307 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10308 | |
573bfb72 | 10309 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10310 | spin_lock(&root->delalloc_lock); |
10311 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10312 | while (!list_empty(&splice)) { |
10313 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10314 | delalloc_inodes); |
1eafa6c7 | 10315 | |
eb73c1b7 MX |
10316 | list_move_tail(&binode->delalloc_inodes, |
10317 | &root->delalloc_inodes); | |
5b21f2ed | 10318 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10319 | if (!inode) { |
eb73c1b7 | 10320 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10321 | continue; |
df0af1a5 | 10322 | } |
eb73c1b7 | 10323 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10324 | |
651d494a | 10325 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10326 | if (!work) { |
f4ab9ea7 JB |
10327 | if (delay_iput) |
10328 | btrfs_add_delayed_iput(inode); | |
10329 | else | |
10330 | iput(inode); | |
1eafa6c7 | 10331 | ret = -ENOMEM; |
a1ecaabb | 10332 | goto out; |
5b21f2ed | 10333 | } |
1eafa6c7 | 10334 | list_add_tail(&work->list, &works); |
a44903ab QW |
10335 | btrfs_queue_work(root->fs_info->flush_workers, |
10336 | &work->work); | |
6c255e67 MX |
10337 | ret++; |
10338 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10339 | goto out; |
5b21f2ed | 10340 | cond_resched(); |
eb73c1b7 | 10341 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10342 | } |
eb73c1b7 | 10343 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10344 | |
a1ecaabb | 10345 | out: |
eb73c1b7 MX |
10346 | list_for_each_entry_safe(work, next, &works, list) { |
10347 | list_del_init(&work->list); | |
10348 | btrfs_wait_and_free_delalloc_work(work); | |
10349 | } | |
10350 | ||
10351 | if (!list_empty_careful(&splice)) { | |
10352 | spin_lock(&root->delalloc_lock); | |
10353 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10354 | spin_unlock(&root->delalloc_lock); | |
10355 | } | |
573bfb72 | 10356 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10357 | return ret; |
10358 | } | |
1eafa6c7 | 10359 | |
eb73c1b7 MX |
10360 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10361 | { | |
0b246afa | 10362 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10363 | int ret; |
1eafa6c7 | 10364 | |
0b246afa | 10365 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10366 | return -EROFS; |
10367 | ||
6c255e67 MX |
10368 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10369 | if (ret > 0) | |
10370 | ret = 0; | |
eb73c1b7 MX |
10371 | /* |
10372 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
10373 | * we have to make sure the IO is actually started and that |
10374 | * ordered extents get created before we return | |
10375 | */ | |
0b246afa JM |
10376 | atomic_inc(&fs_info->async_submit_draining); |
10377 | while (atomic_read(&fs_info->nr_async_submits) || | |
10378 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10379 | wait_event(fs_info->async_submit_wait, | |
10380 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10381 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10382 | } | |
10383 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 MX |
10384 | return ret; |
10385 | } | |
10386 | ||
6c255e67 MX |
10387 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10388 | int nr) | |
eb73c1b7 MX |
10389 | { |
10390 | struct btrfs_root *root; | |
10391 | struct list_head splice; | |
10392 | int ret; | |
10393 | ||
2c21b4d7 | 10394 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10395 | return -EROFS; |
10396 | ||
10397 | INIT_LIST_HEAD(&splice); | |
10398 | ||
573bfb72 | 10399 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10400 | spin_lock(&fs_info->delalloc_root_lock); |
10401 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10402 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10403 | root = list_first_entry(&splice, struct btrfs_root, |
10404 | delalloc_root); | |
10405 | root = btrfs_grab_fs_root(root); | |
10406 | BUG_ON(!root); | |
10407 | list_move_tail(&root->delalloc_root, | |
10408 | &fs_info->delalloc_roots); | |
10409 | spin_unlock(&fs_info->delalloc_root_lock); | |
10410 | ||
6c255e67 | 10411 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10412 | btrfs_put_fs_root(root); |
6c255e67 | 10413 | if (ret < 0) |
eb73c1b7 MX |
10414 | goto out; |
10415 | ||
6c255e67 MX |
10416 | if (nr != -1) { |
10417 | nr -= ret; | |
10418 | WARN_ON(nr < 0); | |
10419 | } | |
eb73c1b7 | 10420 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10421 | } |
eb73c1b7 | 10422 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10423 | |
6c255e67 | 10424 | ret = 0; |
eb73c1b7 MX |
10425 | atomic_inc(&fs_info->async_submit_draining); |
10426 | while (atomic_read(&fs_info->nr_async_submits) || | |
10427 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10428 | wait_event(fs_info->async_submit_wait, | |
10429 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10430 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10431 | } | |
10432 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 10433 | out: |
1eafa6c7 | 10434 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10435 | spin_lock(&fs_info->delalloc_root_lock); |
10436 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10437 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10438 | } |
573bfb72 | 10439 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10440 | return ret; |
ea8c2819 CM |
10441 | } |
10442 | ||
39279cc3 CM |
10443 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10444 | const char *symname) | |
10445 | { | |
0b246afa | 10446 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10447 | struct btrfs_trans_handle *trans; |
10448 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10449 | struct btrfs_path *path; | |
10450 | struct btrfs_key key; | |
1832a6d5 | 10451 | struct inode *inode = NULL; |
39279cc3 CM |
10452 | int err; |
10453 | int drop_inode = 0; | |
10454 | u64 objectid; | |
67871254 | 10455 | u64 index = 0; |
39279cc3 CM |
10456 | int name_len; |
10457 | int datasize; | |
5f39d397 | 10458 | unsigned long ptr; |
39279cc3 | 10459 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10460 | struct extent_buffer *leaf; |
39279cc3 | 10461 | |
f06becc4 | 10462 | name_len = strlen(symname); |
0b246afa | 10463 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10464 | return -ENAMETOOLONG; |
1832a6d5 | 10465 | |
9ed74f2d JB |
10466 | /* |
10467 | * 2 items for inode item and ref | |
10468 | * 2 items for dir items | |
9269d12b FM |
10469 | * 1 item for updating parent inode item |
10470 | * 1 item for the inline extent item | |
9ed74f2d JB |
10471 | * 1 item for xattr if selinux is on |
10472 | */ | |
9269d12b | 10473 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10474 | if (IS_ERR(trans)) |
10475 | return PTR_ERR(trans); | |
1832a6d5 | 10476 | |
581bb050 LZ |
10477 | err = btrfs_find_free_ino(root, &objectid); |
10478 | if (err) | |
10479 | goto out_unlock; | |
10480 | ||
aec7477b | 10481 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10482 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10483 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10484 | if (IS_ERR(inode)) { |
10485 | err = PTR_ERR(inode); | |
39279cc3 | 10486 | goto out_unlock; |
7cf96da3 | 10487 | } |
39279cc3 | 10488 | |
ad19db71 CS |
10489 | /* |
10490 | * If the active LSM wants to access the inode during | |
10491 | * d_instantiate it needs these. Smack checks to see | |
10492 | * if the filesystem supports xattrs by looking at the | |
10493 | * ops vector. | |
10494 | */ | |
10495 | inode->i_fop = &btrfs_file_operations; | |
10496 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10497 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10498 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10499 | ||
10500 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10501 | if (err) | |
10502 | goto out_unlock_inode; | |
ad19db71 | 10503 | |
39279cc3 | 10504 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10505 | if (!path) { |
10506 | err = -ENOMEM; | |
b0d5d10f | 10507 | goto out_unlock_inode; |
d8926bb3 | 10508 | } |
4a0cc7ca | 10509 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10510 | key.offset = 0; |
962a298f | 10511 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10512 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10513 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10514 | datasize); | |
54aa1f4d | 10515 | if (err) { |
b0839166 | 10516 | btrfs_free_path(path); |
b0d5d10f | 10517 | goto out_unlock_inode; |
54aa1f4d | 10518 | } |
5f39d397 CM |
10519 | leaf = path->nodes[0]; |
10520 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10521 | struct btrfs_file_extent_item); | |
10522 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10523 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10524 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10525 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10526 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10527 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10528 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10529 | ||
39279cc3 | 10530 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10531 | write_extent_buffer(leaf, symname, ptr, name_len); |
10532 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10533 | btrfs_free_path(path); |
5f39d397 | 10534 | |
39279cc3 | 10535 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10536 | inode_nohighmem(inode); |
39279cc3 | 10537 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10538 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10539 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10540 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10541 | /* |
10542 | * Last step, add directory indexes for our symlink inode. This is the | |
10543 | * last step to avoid extra cleanup of these indexes if an error happens | |
10544 | * elsewhere above. | |
10545 | */ | |
10546 | if (!err) | |
cef415af NB |
10547 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10548 | BTRFS_I(inode), 0, index); | |
b0d5d10f | 10549 | if (err) { |
54aa1f4d | 10550 | drop_inode = 1; |
b0d5d10f CM |
10551 | goto out_unlock_inode; |
10552 | } | |
10553 | ||
f440ea85 | 10554 | d_instantiate_new(dentry, inode); |
39279cc3 CM |
10555 | |
10556 | out_unlock: | |
3a45bb20 | 10557 | btrfs_end_transaction(trans); |
39279cc3 CM |
10558 | if (drop_inode) { |
10559 | inode_dec_link_count(inode); | |
10560 | iput(inode); | |
10561 | } | |
2ff7e61e | 10562 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 10563 | return err; |
b0d5d10f CM |
10564 | |
10565 | out_unlock_inode: | |
10566 | drop_inode = 1; | |
10567 | unlock_new_inode(inode); | |
10568 | goto out_unlock; | |
39279cc3 | 10569 | } |
16432985 | 10570 | |
0af3d00b JB |
10571 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10572 | u64 start, u64 num_bytes, u64 min_size, | |
10573 | loff_t actual_len, u64 *alloc_hint, | |
10574 | struct btrfs_trans_handle *trans) | |
d899e052 | 10575 | { |
0b246afa | 10576 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10577 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10578 | struct extent_map *em; | |
d899e052 YZ |
10579 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10580 | struct btrfs_key ins; | |
d899e052 | 10581 | u64 cur_offset = start; |
55a61d1d | 10582 | u64 i_size; |
154ea289 | 10583 | u64 cur_bytes; |
0b670dc4 | 10584 | u64 last_alloc = (u64)-1; |
d899e052 | 10585 | int ret = 0; |
0af3d00b | 10586 | bool own_trans = true; |
18513091 | 10587 | u64 end = start + num_bytes - 1; |
d899e052 | 10588 | |
0af3d00b JB |
10589 | if (trans) |
10590 | own_trans = false; | |
d899e052 | 10591 | while (num_bytes > 0) { |
0af3d00b JB |
10592 | if (own_trans) { |
10593 | trans = btrfs_start_transaction(root, 3); | |
10594 | if (IS_ERR(trans)) { | |
10595 | ret = PTR_ERR(trans); | |
10596 | break; | |
10597 | } | |
5a303d5d YZ |
10598 | } |
10599 | ||
ee22184b | 10600 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10601 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10602 | /* |
10603 | * If we are severely fragmented we could end up with really | |
10604 | * small allocations, so if the allocator is returning small | |
10605 | * chunks lets make its job easier by only searching for those | |
10606 | * sized chunks. | |
10607 | */ | |
10608 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10609 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10610 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10611 | if (ret) { |
0af3d00b | 10612 | if (own_trans) |
3a45bb20 | 10613 | btrfs_end_transaction(trans); |
a22285a6 | 10614 | break; |
d899e052 | 10615 | } |
0b246afa | 10616 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10617 | |
0b670dc4 | 10618 | last_alloc = ins.offset; |
d899e052 YZ |
10619 | ret = insert_reserved_file_extent(trans, inode, |
10620 | cur_offset, ins.objectid, | |
10621 | ins.offset, ins.offset, | |
920bbbfb | 10622 | ins.offset, 0, 0, 0, |
d899e052 | 10623 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10624 | if (ret) { |
2ff7e61e | 10625 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10626 | ins.offset, 0); |
66642832 | 10627 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10628 | if (own_trans) |
3a45bb20 | 10629 | btrfs_end_transaction(trans); |
79787eaa JM |
10630 | break; |
10631 | } | |
31193213 | 10632 | |
dcdbc059 | 10633 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10634 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10635 | |
5dc562c5 JB |
10636 | em = alloc_extent_map(); |
10637 | if (!em) { | |
10638 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10639 | &BTRFS_I(inode)->runtime_flags); | |
10640 | goto next; | |
10641 | } | |
10642 | ||
10643 | em->start = cur_offset; | |
10644 | em->orig_start = cur_offset; | |
10645 | em->len = ins.offset; | |
10646 | em->block_start = ins.objectid; | |
10647 | em->block_len = ins.offset; | |
b4939680 | 10648 | em->orig_block_len = ins.offset; |
cc95bef6 | 10649 | em->ram_bytes = ins.offset; |
0b246afa | 10650 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10651 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10652 | em->generation = trans->transid; | |
10653 | ||
10654 | while (1) { | |
10655 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10656 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10657 | write_unlock(&em_tree->lock); |
10658 | if (ret != -EEXIST) | |
10659 | break; | |
dcdbc059 | 10660 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10661 | cur_offset + ins.offset - 1, |
10662 | 0); | |
10663 | } | |
10664 | free_extent_map(em); | |
10665 | next: | |
d899e052 YZ |
10666 | num_bytes -= ins.offset; |
10667 | cur_offset += ins.offset; | |
efa56464 | 10668 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10669 | |
0c4d2d95 | 10670 | inode_inc_iversion(inode); |
c2050a45 | 10671 | inode->i_ctime = current_time(inode); |
6cbff00f | 10672 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10673 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10674 | (actual_len > inode->i_size) && |
10675 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10676 | if (cur_offset > actual_len) |
55a61d1d | 10677 | i_size = actual_len; |
d1ea6a61 | 10678 | else |
55a61d1d JB |
10679 | i_size = cur_offset; |
10680 | i_size_write(inode, i_size); | |
10681 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10682 | } |
10683 | ||
d899e052 | 10684 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10685 | |
10686 | if (ret) { | |
66642832 | 10687 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10688 | if (own_trans) |
3a45bb20 | 10689 | btrfs_end_transaction(trans); |
79787eaa JM |
10690 | break; |
10691 | } | |
d899e052 | 10692 | |
0af3d00b | 10693 | if (own_trans) |
3a45bb20 | 10694 | btrfs_end_transaction(trans); |
5a303d5d | 10695 | } |
18513091 | 10696 | if (cur_offset < end) |
bc42bda2 | 10697 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10698 | end - cur_offset + 1); |
d899e052 YZ |
10699 | return ret; |
10700 | } | |
10701 | ||
0af3d00b JB |
10702 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10703 | u64 start, u64 num_bytes, u64 min_size, | |
10704 | loff_t actual_len, u64 *alloc_hint) | |
10705 | { | |
10706 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10707 | min_size, actual_len, alloc_hint, | |
10708 | NULL); | |
10709 | } | |
10710 | ||
10711 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10712 | struct btrfs_trans_handle *trans, int mode, | |
10713 | u64 start, u64 num_bytes, u64 min_size, | |
10714 | loff_t actual_len, u64 *alloc_hint) | |
10715 | { | |
10716 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10717 | min_size, actual_len, alloc_hint, trans); | |
10718 | } | |
10719 | ||
e6dcd2dc CM |
10720 | static int btrfs_set_page_dirty(struct page *page) |
10721 | { | |
e6dcd2dc CM |
10722 | return __set_page_dirty_nobuffers(page); |
10723 | } | |
10724 | ||
10556cb2 | 10725 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10726 | { |
b83cc969 | 10727 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10728 | umode_t mode = inode->i_mode; |
b83cc969 | 10729 | |
cb6db4e5 JM |
10730 | if (mask & MAY_WRITE && |
10731 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10732 | if (btrfs_root_readonly(root)) | |
10733 | return -EROFS; | |
10734 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10735 | return -EACCES; | |
10736 | } | |
2830ba7f | 10737 | return generic_permission(inode, mask); |
fdebe2bd | 10738 | } |
39279cc3 | 10739 | |
ef3b9af5 FM |
10740 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10741 | { | |
2ff7e61e | 10742 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10743 | struct btrfs_trans_handle *trans; |
10744 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10745 | struct inode *inode = NULL; | |
10746 | u64 objectid; | |
10747 | u64 index; | |
10748 | int ret = 0; | |
10749 | ||
10750 | /* | |
10751 | * 5 units required for adding orphan entry | |
10752 | */ | |
10753 | trans = btrfs_start_transaction(root, 5); | |
10754 | if (IS_ERR(trans)) | |
10755 | return PTR_ERR(trans); | |
10756 | ||
10757 | ret = btrfs_find_free_ino(root, &objectid); | |
10758 | if (ret) | |
10759 | goto out; | |
10760 | ||
10761 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10762 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10763 | if (IS_ERR(inode)) { |
10764 | ret = PTR_ERR(inode); | |
10765 | inode = NULL; | |
10766 | goto out; | |
10767 | } | |
10768 | ||
ef3b9af5 FM |
10769 | inode->i_fop = &btrfs_file_operations; |
10770 | inode->i_op = &btrfs_file_inode_operations; | |
10771 | ||
10772 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10773 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10774 | ||
b0d5d10f CM |
10775 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10776 | if (ret) | |
10777 | goto out_inode; | |
10778 | ||
10779 | ret = btrfs_update_inode(trans, root, inode); | |
10780 | if (ret) | |
10781 | goto out_inode; | |
73f2e545 | 10782 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10783 | if (ret) |
b0d5d10f | 10784 | goto out_inode; |
ef3b9af5 | 10785 | |
5762b5c9 FM |
10786 | /* |
10787 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10788 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10789 | * through: | |
10790 | * | |
10791 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10792 | */ | |
10793 | set_nlink(inode, 1); | |
b0d5d10f | 10794 | unlock_new_inode(inode); |
ef3b9af5 FM |
10795 | d_tmpfile(dentry, inode); |
10796 | mark_inode_dirty(inode); | |
10797 | ||
10798 | out: | |
3a45bb20 | 10799 | btrfs_end_transaction(trans); |
ef3b9af5 FM |
10800 | if (ret) |
10801 | iput(inode); | |
2ff7e61e JM |
10802 | btrfs_balance_delayed_items(fs_info); |
10803 | btrfs_btree_balance_dirty(fs_info); | |
ef3b9af5 | 10804 | return ret; |
b0d5d10f CM |
10805 | |
10806 | out_inode: | |
10807 | unlock_new_inode(inode); | |
10808 | goto out; | |
10809 | ||
ef3b9af5 FM |
10810 | } |
10811 | ||
20a7db8a | 10812 | __attribute__((const)) |
9d0d1c8b | 10813 | static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror) |
20a7db8a | 10814 | { |
9d0d1c8b | 10815 | return -EAGAIN; |
20a7db8a DS |
10816 | } |
10817 | ||
c6100a4b JB |
10818 | static struct btrfs_fs_info *iotree_fs_info(void *private_data) |
10819 | { | |
10820 | struct inode *inode = private_data; | |
10821 | return btrfs_sb(inode->i_sb); | |
10822 | } | |
10823 | ||
10824 | static void btrfs_check_extent_io_range(void *private_data, const char *caller, | |
10825 | u64 start, u64 end) | |
10826 | { | |
10827 | struct inode *inode = private_data; | |
10828 | u64 isize; | |
10829 | ||
10830 | isize = i_size_read(inode); | |
10831 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
10832 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
10833 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
10834 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
10835 | } | |
10836 | } | |
10837 | ||
10838 | void btrfs_set_range_writeback(void *private_data, u64 start, u64 end) | |
10839 | { | |
10840 | struct inode *inode = private_data; | |
10841 | unsigned long index = start >> PAGE_SHIFT; | |
10842 | unsigned long end_index = end >> PAGE_SHIFT; | |
10843 | struct page *page; | |
10844 | ||
10845 | while (index <= end_index) { | |
10846 | page = find_get_page(inode->i_mapping, index); | |
10847 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10848 | set_page_writeback(page); | |
10849 | put_page(page); | |
10850 | index++; | |
10851 | } | |
10852 | } | |
10853 | ||
6e1d5dcc | 10854 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10855 | .getattr = btrfs_getattr, |
39279cc3 CM |
10856 | .lookup = btrfs_lookup, |
10857 | .create = btrfs_create, | |
10858 | .unlink = btrfs_unlink, | |
10859 | .link = btrfs_link, | |
10860 | .mkdir = btrfs_mkdir, | |
10861 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10862 | .rename = btrfs_rename2, |
39279cc3 CM |
10863 | .symlink = btrfs_symlink, |
10864 | .setattr = btrfs_setattr, | |
618e21d5 | 10865 | .mknod = btrfs_mknod, |
5103e947 | 10866 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10867 | .permission = btrfs_permission, |
4e34e719 | 10868 | .get_acl = btrfs_get_acl, |
996a710d | 10869 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10870 | .update_time = btrfs_update_time, |
ef3b9af5 | 10871 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10872 | }; |
6e1d5dcc | 10873 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10874 | .lookup = btrfs_lookup, |
fdebe2bd | 10875 | .permission = btrfs_permission, |
93fd63c2 | 10876 | .update_time = btrfs_update_time, |
39279cc3 | 10877 | }; |
76dda93c | 10878 | |
828c0950 | 10879 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10880 | .llseek = generic_file_llseek, |
10881 | .read = generic_read_dir, | |
02dbfc99 | 10882 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10883 | .open = btrfs_opendir, |
34287aa3 | 10884 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10885 | #ifdef CONFIG_COMPAT |
4c63c245 | 10886 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10887 | #endif |
6bf13c0c | 10888 | .release = btrfs_release_file, |
e02119d5 | 10889 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10890 | }; |
10891 | ||
20e5506b | 10892 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10893 | /* mandatory callbacks */ |
065631f6 | 10894 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac | 10895 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
4d53dddb | 10896 | .merge_bio_hook = btrfs_merge_bio_hook, |
9d0d1c8b | 10897 | .readpage_io_failed_hook = btrfs_readpage_io_failed_hook, |
c6100a4b JB |
10898 | .tree_fs_info = iotree_fs_info, |
10899 | .set_range_writeback = btrfs_set_range_writeback, | |
4d53dddb DS |
10900 | |
10901 | /* optional callbacks */ | |
10902 | .fill_delalloc = run_delalloc_range, | |
e6dcd2dc | 10903 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10904 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10905 | .set_bit_hook = btrfs_set_bit_hook, |
10906 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10907 | .merge_extent_hook = btrfs_merge_extent_hook, |
10908 | .split_extent_hook = btrfs_split_extent_hook, | |
c6100a4b | 10909 | .check_extent_io_range = btrfs_check_extent_io_range, |
07157aac CM |
10910 | }; |
10911 | ||
35054394 CM |
10912 | /* |
10913 | * btrfs doesn't support the bmap operation because swapfiles | |
10914 | * use bmap to make a mapping of extents in the file. They assume | |
10915 | * these extents won't change over the life of the file and they | |
10916 | * use the bmap result to do IO directly to the drive. | |
10917 | * | |
10918 | * the btrfs bmap call would return logical addresses that aren't | |
10919 | * suitable for IO and they also will change frequently as COW | |
10920 | * operations happen. So, swapfile + btrfs == corruption. | |
10921 | * | |
10922 | * For now we're avoiding this by dropping bmap. | |
10923 | */ | |
7f09410b | 10924 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10925 | .readpage = btrfs_readpage, |
10926 | .writepage = btrfs_writepage, | |
b293f02e | 10927 | .writepages = btrfs_writepages, |
3ab2fb5a | 10928 | .readpages = btrfs_readpages, |
16432985 | 10929 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10930 | .invalidatepage = btrfs_invalidatepage, |
10931 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10932 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10933 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10934 | }; |
10935 | ||
7f09410b | 10936 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10937 | .readpage = btrfs_readpage, |
10938 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10939 | .invalidatepage = btrfs_invalidatepage, |
10940 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10941 | }; |
10942 | ||
6e1d5dcc | 10943 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10944 | .getattr = btrfs_getattr, |
10945 | .setattr = btrfs_setattr, | |
5103e947 | 10946 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10947 | .permission = btrfs_permission, |
1506fcc8 | 10948 | .fiemap = btrfs_fiemap, |
4e34e719 | 10949 | .get_acl = btrfs_get_acl, |
996a710d | 10950 | .set_acl = btrfs_set_acl, |
e41f941a | 10951 | .update_time = btrfs_update_time, |
39279cc3 | 10952 | }; |
6e1d5dcc | 10953 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10954 | .getattr = btrfs_getattr, |
10955 | .setattr = btrfs_setattr, | |
fdebe2bd | 10956 | .permission = btrfs_permission, |
33268eaf | 10957 | .listxattr = btrfs_listxattr, |
4e34e719 | 10958 | .get_acl = btrfs_get_acl, |
996a710d | 10959 | .set_acl = btrfs_set_acl, |
e41f941a | 10960 | .update_time = btrfs_update_time, |
618e21d5 | 10961 | }; |
6e1d5dcc | 10962 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10963 | .get_link = page_get_link, |
f209561a | 10964 | .getattr = btrfs_getattr, |
22c44fe6 | 10965 | .setattr = btrfs_setattr, |
fdebe2bd | 10966 | .permission = btrfs_permission, |
0279b4cd | 10967 | .listxattr = btrfs_listxattr, |
e41f941a | 10968 | .update_time = btrfs_update_time, |
39279cc3 | 10969 | }; |
76dda93c | 10970 | |
82d339d9 | 10971 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10972 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10973 | .d_release = btrfs_dentry_release, |
76dda93c | 10974 | }; |