projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fs/btrfs: drop if around WARN_ON
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / btrfs / inode.c
CommitLineData
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>
33#include <linux/statfs.h>
34#include <linux/compat.h>
9ebefb18 35#include <linux/bit_spinlock.h>
5103e947 36#include <linux/xattr.h>
33268eaf 37#include <linux/posix_acl.h>
d899e052 38#include <linux/falloc.h>
5a0e3ad6 39#include <linux/slab.h>
7a36ddec 40#include <linux/ratelimit.h>
22c44fe6 41#include <linux/mount.h>
4b4e25f2 42#include "compat.h"
39279cc3
CM		 43#include "ctree.h"
44#include "disk-io.h"
45#include "transaction.h"
46#include "btrfs_inode.h"
47#include "ioctl.h"
48#include "print-tree.h"
e6dcd2dc 49#include "ordered-data.h"
95819c05 50#include "xattr.h"
e02119d5 51#include "tree-log.h"
4a54c8c1 52#include "volumes.h"
c8b97818 53#include "compression.h"
b4ce94de 54#include "locking.h"
dc89e982 55#include "free-space-cache.h"
581bb050 56#include "inode-map.h"
39279cc3
CM		 57
58struct btrfs_iget_args {
59 u64 ino;
60 struct btrfs_root *root;
61};
62
6e1d5dcc
AD		 63static const struct inode_operations btrfs_dir_inode_operations;
64static const struct inode_operations btrfs_symlink_inode_operations;
65static const struct inode_operations btrfs_dir_ro_inode_operations;
66static const struct inode_operations btrfs_special_inode_operations;
67static const struct inode_operations btrfs_file_inode_operations;
7f09410b
AD		 68static const struct address_space_operations btrfs_aops;
69static const struct address_space_operations btrfs_symlink_aops;
828c0950 70static const struct file_operations btrfs_dir_file_operations;
d1310b2e 71static struct extent_io_ops btrfs_extent_io_ops;
39279cc3
CM		 72
73static struct kmem_cache *btrfs_inode_cachep;
8ccf6f19 74static struct kmem_cache *btrfs_delalloc_work_cachep;
39279cc3
CM		 75struct kmem_cache *btrfs_trans_handle_cachep;
76struct kmem_cache *btrfs_transaction_cachep;
39279cc3 77struct kmem_cache *btrfs_path_cachep;
dc89e982 78struct kmem_cache *btrfs_free_space_cachep;
39279cc3
CM		 79
80#define S_SHIFT 12
81static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
89};
90
a41ad394
JB		 91static int btrfs_setsize(struct inode *inode, loff_t newsize);
92static int btrfs_truncate(struct inode *inode);
5fd02043 93static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
771ed689
CM		 94static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, int *page_started,
97 unsigned long *nr_written, int unlock);
7b128766 98
f34f57a3 99static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
2a7dba39
EP		 100 struct inode *inode, struct inode *dir,
101 const struct qstr *qstr)
0279b4cd
JO		 102{
103 int err;
104
f34f57a3 105 err = btrfs_init_acl(trans, inode, dir);
0279b4cd 106 if (!err)
2a7dba39 107 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
0279b4cd
JO		 108 return err;
109}
110
c8b97818
CM		 111/*
112 * this does all the hard work for inserting an inline extent into
113 * the btree. The caller should have done a btrfs_drop_extents so that
114 * no overlapping inline items exist in the btree
115 */
d397712b 116static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
c8b97818
CM		 117 struct btrfs_root *root, struct inode *inode,
118 u64 start, size_t size, size_t compressed_size,
fe3f566c 119 int compress_type,
c8b97818
CM		 120 struct page **compressed_pages)
121{
122 struct btrfs_key key;
123 struct btrfs_path *path;
124 struct extent_buffer *leaf;
125 struct page *page = NULL;
126 char *kaddr;
127 unsigned long ptr;
128 struct btrfs_file_extent_item *ei;
129 int err = 0;
130 int ret;
131 size_t cur_size = size;
132 size_t datasize;
133 unsigned long offset;
c8b97818 134
fe3f566c 135 if (compressed_size && compressed_pages)
c8b97818 136 cur_size = compressed_size;
c8b97818 137
d397712b
CM		 138 path = btrfs_alloc_path();
139 if (!path)
c8b97818
CM		 140 return -ENOMEM;
141
b9473439 142 path->leave_spinning = 1;
c8b97818 143
33345d01 144 key.objectid = btrfs_ino(inode);
c8b97818
CM		 145 key.offset = start;
146 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
c8b97818
CM		 147 datasize = btrfs_file_extent_calc_inline_size(cur_size);
148
149 inode_add_bytes(inode, size);
150 ret = btrfs_insert_empty_item(trans, root, path, &key,
151 datasize);
c8b97818
CM		 152 if (ret) {
153 err = ret;
c8b97818
CM		 154 goto fail;
155 }
156 leaf = path->nodes[0];
157 ei = btrfs_item_ptr(leaf, path->slots[0],
158 struct btrfs_file_extent_item);
159 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
160 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
161 btrfs_set_file_extent_encryption(leaf, ei, 0);
162 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
163 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
164 ptr = btrfs_file_extent_inline_start(ei);
165
261507a0 166 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818
CM		 167 struct page *cpage;
168 int i = 0;
d397712b 169 while (compressed_size > 0) {
c8b97818 170 cpage = compressed_pages[i];
5b050f04 171 cur_size = min_t(unsigned long, compressed_size,
c8b97818
CM		 172 PAGE_CACHE_SIZE);
173
7ac687d9 174 kaddr = kmap_atomic(cpage);
c8b97818 175 write_extent_buffer(leaf, kaddr, ptr, cur_size);
7ac687d9 176 kunmap_atomic(kaddr);
c8b97818
CM		 177
178 i++;
179 ptr += cur_size;
180 compressed_size -= cur_size;
181 }
182 btrfs_set_file_extent_compression(leaf, ei,
261507a0 183 compress_type);
c8b97818
CM		 184 } else {
185 page = find_get_page(inode->i_mapping,
186 start >> PAGE_CACHE_SHIFT);
187 btrfs_set_file_extent_compression(leaf, ei, 0);
7ac687d9 188 kaddr = kmap_atomic(page);
c8b97818
CM		 189 offset = start & (PAGE_CACHE_SIZE - 1);
190 write_extent_buffer(leaf, kaddr + offset, ptr, size);
7ac687d9 191 kunmap_atomic(kaddr);
c8b97818
CM		 192 page_cache_release(page);
193 }
194 btrfs_mark_buffer_dirty(leaf);
195 btrfs_free_path(path);
196
c2167754
YZ		 197 /*
198 * we're an inline extent, so nobody can
199 * extend the file past i_size without locking
200 * a page we already have locked.
201 *
202 * We must do any isize and inode updates
203 * before we unlock the pages. Otherwise we
204 * could end up racing with unlink.
205 */
c8b97818 206 BTRFS_I(inode)->disk_i_size = inode->i_size;
79787eaa 207 ret = btrfs_update_inode(trans, root, inode);
c2167754 208
79787eaa 209 return ret;
c8b97818
CM		 210fail:
211 btrfs_free_path(path);
212 return err;
213}
214
215
216/*
217 * conditionally insert an inline extent into the file. This
218 * does the checks required to make sure the data is small enough
219 * to fit as an inline extent.
220 */
7f366cfe 221static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
c8b97818
CM		 222 struct btrfs_root *root,
223 struct inode *inode, u64 start, u64 end,
fe3f566c 224 size_t compressed_size, int compress_type,
c8b97818
CM		 225 struct page **compressed_pages)
226{
227 u64 isize = i_size_read(inode);
228 u64 actual_end = min(end + 1, isize);
229 u64 inline_len = actual_end - start;
230 u64 aligned_end = (end + root->sectorsize - 1) &
231 ~((u64)root->sectorsize - 1);
c8b97818
CM		 232 u64 data_len = inline_len;
233 int ret;
234
235 if (compressed_size)
236 data_len = compressed_size;
237
238 if (start > 0 ||
70b99e69 239 actual_end >= PAGE_CACHE_SIZE ||
c8b97818
CM		 240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (!compressed_size &&
242 (actual_end & (root->sectorsize - 1)) == 0) ||
243 end + 1 < isize ||
244 data_len > root->fs_info->max_inline) {
245 return 1;
246 }
247
2671485d 248 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
79787eaa
JM		 249 if (ret)
250 return ret;
c8b97818
CM		 251
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
fe3f566c 256 compress_type, compressed_pages);
2adcac1a 257 if (ret && ret != -ENOSPC) {
79787eaa
JM		 258 btrfs_abort_transaction(trans, root, ret);
259 return ret;
2adcac1a
JB		 260 } else if (ret == -ENOSPC) {
261 return 1;
79787eaa 262 }
2adcac1a 263
0ca1f7ce 264 btrfs_delalloc_release_metadata(inode, end + 1 - start);
a1ed835e 265 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
c8b97818
CM		 266 return 0;
267}
268
771ed689
CM		 269struct async_extent {
270 u64 start;
271 u64 ram_size;
272 u64 compressed_size;
273 struct page **pages;
274 unsigned long nr_pages;
261507a0 275 int compress_type;
771ed689
CM		 276 struct list_head list;
277};
278
279struct async_cow {
280 struct inode *inode;
281 struct btrfs_root *root;
282 struct page *locked_page;
283 u64 start;
284 u64 end;
285 struct list_head extents;
286 struct btrfs_work work;
287};
288
289static noinline int add_async_extent(struct async_cow *cow,
290 u64 start, u64 ram_size,
291 u64 compressed_size,
292 struct page **pages,
261507a0
LZ		 293 unsigned long nr_pages,
294 int compress_type)
771ed689
CM		 295{
296 struct async_extent *async_extent;
297
298 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
79787eaa 299 BUG_ON(!async_extent); /* -ENOMEM */
771ed689
CM		 300 async_extent->start = start;
301 async_extent->ram_size = ram_size;
302 async_extent->compressed_size = compressed_size;
303 async_extent->pages = pages;
304 async_extent->nr_pages = nr_pages;
261507a0 305 async_extent->compress_type = compress_type;
771ed689
CM		 306 list_add_tail(&async_extent->list, &cow->extents);
307 return 0;
308}
309
d352ac68 310/*
771ed689
CM		 311 * we create compressed extents in two phases. The first
312 * phase compresses a range of pages that have already been
313 * locked (both pages and state bits are locked).
c8b97818 314 *
771ed689
CM		 315 * This is done inside an ordered work queue, and the compression
316 * is spread across many cpus. The actual IO submission is step
317 * two, and the ordered work queue takes care of making sure that
318 * happens in the same order things were put onto the queue by
319 * writepages and friends.
c8b97818 320 *
771ed689
CM		 321 * If this code finds it can't get good compression, it puts an
322 * entry onto the work queue to write the uncompressed bytes. This
323 * makes sure that both compressed inodes and uncompressed inodes
b2570314
AB		 324 * are written in the same order that the flusher thread sent them
325 * down.
d352ac68 326 */
771ed689
CM		 327static noinline int compress_file_range(struct inode *inode,
328 struct page *locked_page,
329 u64 start, u64 end,
330 struct async_cow *async_cow,
331 int *num_added)
b888db2b
CM		 332{
333 struct btrfs_root *root = BTRFS_I(inode)->root;
334 struct btrfs_trans_handle *trans;
db94535d 335 u64 num_bytes;
db94535d 336 u64 blocksize = root->sectorsize;
c8b97818 337 u64 actual_end;
42dc7bab 338 u64 isize = i_size_read(inode);
e6dcd2dc 339 int ret = 0;
c8b97818
CM		 340 struct page **pages = NULL;
341 unsigned long nr_pages;
342 unsigned long nr_pages_ret = 0;
343 unsigned long total_compressed = 0;
344 unsigned long total_in = 0;
345 unsigned long max_compressed = 128 * 1024;
771ed689 346 unsigned long max_uncompressed = 128 * 1024;
c8b97818
CM		 347 int i;
348 int will_compress;
261507a0 349 int compress_type = root->fs_info->compress_type;
b888db2b 350
4cb13e5d
LB		 351 /* if this is a small write inside eof, kick off a defrag */
352 if ((end - start + 1) < 16 * 1024 &&
353 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 354 btrfs_add_inode_defrag(NULL, inode);
355
42dc7bab 356 actual_end = min_t(u64, isize, end + 1);
c8b97818
CM		 357again:
358 will_compress = 0;
359 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
360 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
be20aa9d 361
f03d9301
CM		 362 /*
363 * we don't want to send crud past the end of i_size through
364 * compression, that's just a waste of CPU time. So, if the
365 * end of the file is before the start of our current
366 * requested range of bytes, we bail out to the uncompressed
367 * cleanup code that can deal with all of this.
368 *
369 * It isn't really the fastest way to fix things, but this is a
370 * very uncommon corner.
371 */
372 if (actual_end <= start)
373 goto cleanup_and_bail_uncompressed;
374
c8b97818
CM		 375 total_compressed = actual_end - start;
376
377 /* we want to make sure that amount of ram required to uncompress
378 * an extent is reasonable, so we limit the total size in ram
771ed689
CM		 379 * of a compressed extent to 128k. This is a crucial number
380 * because it also controls how easily we can spread reads across
381 * cpus for decompression.
382 *
383 * We also want to make sure the amount of IO required to do
384 * a random read is reasonably small, so we limit the size of
385 * a compressed extent to 128k.
c8b97818
CM		 386 */
387 total_compressed = min(total_compressed, max_uncompressed);
db94535d 388 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
be20aa9d 389 num_bytes = max(blocksize, num_bytes);
c8b97818
CM		 390 total_in = 0;
391 ret = 0;
db94535d 392
771ed689
CM		 393 /*
394 * we do compression for mount -o compress and when the
395 * inode has not been flagged as nocompress. This flag can
396 * change at any time if we discover bad compression ratios.
c8b97818 397 */
6cbff00f 398 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
1e701a32 399 (btrfs_test_opt(root, COMPRESS) ||
75e7cb7f
LB		 400 (BTRFS_I(inode)->force_compress) ||
401 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
c8b97818 402 WARN_ON(pages);
cfbc246e 403 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
560f7d75
LZ		 404 if (!pages) {
405 /* just bail out to the uncompressed code */
406 goto cont;
407 }
c8b97818 408
261507a0
LZ		 409 if (BTRFS_I(inode)->force_compress)
410 compress_type = BTRFS_I(inode)->force_compress;
411
412 ret = btrfs_compress_pages(compress_type,
413 inode->i_mapping, start,
414 total_compressed, pages,
415 nr_pages, &nr_pages_ret,
416 &total_in,
417 &total_compressed,
418 max_compressed);
c8b97818
CM		 419
420 if (!ret) {
421 unsigned long offset = total_compressed &
422 (PAGE_CACHE_SIZE - 1);
423 struct page *page = pages[nr_pages_ret - 1];
424 char *kaddr;
425
426 /* zero the tail end of the last page, we might be
427 * sending it down to disk
428 */
429 if (offset) {
7ac687d9 430 kaddr = kmap_atomic(page);
c8b97818
CM		 431 memset(kaddr + offset, 0,
432 PAGE_CACHE_SIZE - offset);
7ac687d9 433 kunmap_atomic(kaddr);
c8b97818
CM		 434 }
435 will_compress = 1;
436 }
437 }
560f7d75 438cont:
c8b97818 439 if (start == 0) {
7a7eaa40 440 trans = btrfs_join_transaction(root);
79787eaa
JM		 441 if (IS_ERR(trans)) {
442 ret = PTR_ERR(trans);
443 trans = NULL;
444 goto cleanup_and_out;
445 }
0ca1f7ce 446 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
771ed689 447
c8b97818 448 /* lets try to make an inline extent */
771ed689 449 if (ret || total_in < (actual_end - start)) {
c8b97818 450 /* we didn't compress the entire range, try
771ed689 451 * to make an uncompressed inline extent.
c8b97818
CM		 452 */
453 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 454 start, end, 0, 0, NULL);
c8b97818 455 } else {
771ed689 456 /* try making a compressed inline extent */
c8b97818
CM		 457 ret = cow_file_range_inline(trans, root, inode,
458 start, end,
fe3f566c
LZ		 459 total_compressed,
460 compress_type, pages);
c8b97818 461 }
79787eaa 462 if (ret <= 0) {
771ed689 463 /*
79787eaa
JM		 464 * inline extent creation worked or returned error,
465 * we don't need to create any more async work items.
466 * Unlock and free up our temp pages.
771ed689 467 */
c8b97818 468 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 469 &BTRFS_I(inode)->io_tree,
470 start, end, NULL,
471 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
a3429ab7 472 EXTENT_CLEAR_DELALLOC |
a791e35e 473 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
c2167754
YZ		 474
475 btrfs_end_transaction(trans, root);
c8b97818
CM		 476 goto free_pages_out;
477 }
c2167754 478 btrfs_end_transaction(trans, root);
c8b97818
CM		 479 }
480
481 if (will_compress) {
482 /*
483 * we aren't doing an inline extent round the compressed size
484 * up to a block size boundary so the allocator does sane
485 * things
486 */
487 total_compressed = (total_compressed + blocksize - 1) &
488 ~(blocksize - 1);
489
490 /*
491 * one last check to make sure the compression is really a
492 * win, compare the page count read with the blocks on disk
493 */
494 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
495 ~(PAGE_CACHE_SIZE - 1);
496 if (total_compressed >= total_in) {
497 will_compress = 0;
498 } else {
c8b97818
CM		 499 num_bytes = total_in;
500 }
501 }
502 if (!will_compress && pages) {
503 /*
504 * the compression code ran but failed to make things smaller,
505 * free any pages it allocated and our page pointer array
506 */
507 for (i = 0; i < nr_pages_ret; i++) {
70b99e69 508 WARN_ON(pages[i]->mapping);
c8b97818
CM		 509 page_cache_release(pages[i]);
510 }
511 kfree(pages);
512 pages = NULL;
513 total_compressed = 0;
514 nr_pages_ret = 0;
515
516 /* flag the file so we don't compress in the future */
1e701a32
CM		 517 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
518 !(BTRFS_I(inode)->force_compress)) {
a555f810 519 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
1e701a32 520 }
c8b97818 521 }
771ed689
CM		 522 if (will_compress) {
523 *num_added += 1;
c8b97818 524
771ed689
CM		 525 /* the async work queues will take care of doing actual
526 * allocation on disk for these compressed pages,
527 * and will submit them to the elevator.
528 */
529 add_async_extent(async_cow, start, num_bytes,
261507a0
LZ		 530 total_compressed, pages, nr_pages_ret,
531 compress_type);
179e29e4 532
24ae6365 533 if (start + num_bytes < end) {
771ed689
CM		 534 start += num_bytes;
535 pages = NULL;
536 cond_resched();
537 goto again;
538 }
539 } else {
f03d9301 540cleanup_and_bail_uncompressed:
771ed689
CM		 541 /*
542 * No compression, but we still need to write the pages in
543 * the file we've been given so far. redirty the locked
544 * page if it corresponds to our extent and set things up
545 * for the async work queue to run cow_file_range to do
546 * the normal delalloc dance
547 */
548 if (page_offset(locked_page) >= start &&
549 page_offset(locked_page) <= end) {
550 __set_page_dirty_nobuffers(locked_page);
551 /* unlocked later on in the async handlers */
552 }
261507a0
LZ		 553 add_async_extent(async_cow, start, end - start + 1,
554 0, NULL, 0, BTRFS_COMPRESS_NONE);
771ed689
CM		 555 *num_added += 1;
556 }
3b951516 557
771ed689 558out:
79787eaa 559 return ret;
771ed689
CM		 560
561free_pages_out:
562 for (i = 0; i < nr_pages_ret; i++) {
563 WARN_ON(pages[i]->mapping);
564 page_cache_release(pages[i]);
565 }
d397712b 566 kfree(pages);
771ed689
CM		 567
568 goto out;
79787eaa
JM		 569
570cleanup_and_out:
571 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
572 start, end, NULL,
573 EXTENT_CLEAR_UNLOCK_PAGE |
574 EXTENT_CLEAR_DIRTY |
575 EXTENT_CLEAR_DELALLOC |
576 EXTENT_SET_WRITEBACK |
577 EXTENT_END_WRITEBACK);
578 if (!trans || IS_ERR(trans))
579 btrfs_error(root->fs_info, ret, "Failed to join transaction");
580 else
581 btrfs_abort_transaction(trans, root, ret);
582 goto free_pages_out;
771ed689
CM		 583}
584
585/*
586 * phase two of compressed writeback. This is the ordered portion
587 * of the code, which only gets called in the order the work was
588 * queued. We walk all the async extents created by compress_file_range
589 * and send them down to the disk.
590 */
591static noinline int submit_compressed_extents(struct inode *inode,
592 struct async_cow *async_cow)
593{
594 struct async_extent *async_extent;
595 u64 alloc_hint = 0;
596 struct btrfs_trans_handle *trans;
597 struct btrfs_key ins;
598 struct extent_map *em;
599 struct btrfs_root *root = BTRFS_I(inode)->root;
600 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
601 struct extent_io_tree *io_tree;
f5a84ee3 602 int ret = 0;
771ed689
CM		 603
604 if (list_empty(&async_cow->extents))
605 return 0;
606
771ed689 607
d397712b 608 while (!list_empty(&async_cow->extents)) {
771ed689
CM		 609 async_extent = list_entry(async_cow->extents.next,
610 struct async_extent, list);
611 list_del(&async_extent->list);
c8b97818 612
771ed689
CM		 613 io_tree = &BTRFS_I(inode)->io_tree;
614
f5a84ee3 615retry:
771ed689
CM		 616 /* did the compression code fall back to uncompressed IO? */
617 if (!async_extent->pages) {
618 int page_started = 0;
619 unsigned long nr_written = 0;
620
621 lock_extent(io_tree, async_extent->start,
2ac55d41 622 async_extent->start +
d0082371 623 async_extent->ram_size - 1);
771ed689
CM		 624
625 /* allocate blocks */
f5a84ee3
JB		 626 ret = cow_file_range(inode, async_cow->locked_page,
627 async_extent->start,
628 async_extent->start +
629 async_extent->ram_size - 1,
630 &page_started, &nr_written, 0);
771ed689 631
79787eaa
JM		 632 /* JDM XXX */
633
771ed689
CM		 634 /*
635 * if page_started, cow_file_range inserted an
636 * inline extent and took care of all the unlocking
637 * and IO for us. Otherwise, we need to submit
638 * all those pages down to the drive.
639 */
f5a84ee3 640 if (!page_started && !ret)
771ed689
CM		 641 extent_write_locked_range(io_tree,
642 inode, async_extent->start,
d397712b 643 async_extent->start +
771ed689
CM		 644 async_extent->ram_size - 1,
645 btrfs_get_extent,
646 WB_SYNC_ALL);
647 kfree(async_extent);
648 cond_resched();
649 continue;
650 }
651
652 lock_extent(io_tree, async_extent->start,
d0082371 653 async_extent->start + async_extent->ram_size - 1);
771ed689 654
7a7eaa40 655 trans = btrfs_join_transaction(root);
79787eaa
JM		 656 if (IS_ERR(trans)) {
657 ret = PTR_ERR(trans);
658 } else {
659 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
660 ret = btrfs_reserve_extent(trans, root,
771ed689
CM		 661 async_extent->compressed_size,
662 async_extent->compressed_size,
81c9ad23 663 0, alloc_hint, &ins, 1);
962197ba 664 if (ret && ret != -ENOSPC)
79787eaa
JM		 665 btrfs_abort_transaction(trans, root, ret);
666 btrfs_end_transaction(trans, root);
667 }
c2167754 668
f5a84ee3
JB		 669 if (ret) {
670 int i;
671 for (i = 0; i < async_extent->nr_pages; i++) {
672 WARN_ON(async_extent->pages[i]->mapping);
673 page_cache_release(async_extent->pages[i]);
674 }
675 kfree(async_extent->pages);
676 async_extent->nr_pages = 0;
677 async_extent->pages = NULL;
678 unlock_extent(io_tree, async_extent->start,
679 async_extent->start +
d0082371 680 async_extent->ram_size - 1);
79787eaa
JM		 681 if (ret == -ENOSPC)
682 goto retry;
683 goto out_free; /* JDM: Requeue? */
f5a84ee3
JB		 684 }
685
c2167754
YZ		 686 /*
687 * here we're doing allocation and writeback of the
688 * compressed pages
689 */
690 btrfs_drop_extent_cache(inode, async_extent->start,
691 async_extent->start +
692 async_extent->ram_size - 1, 0);
693
172ddd60 694 em = alloc_extent_map();
79787eaa 695 BUG_ON(!em); /* -ENOMEM */
771ed689
CM		 696 em->start = async_extent->start;
697 em->len = async_extent->ram_size;
445a6944 698 em->orig_start = em->start;
c8b97818 699
771ed689
CM		 700 em->block_start = ins.objectid;
701 em->block_len = ins.offset;
702 em->bdev = root->fs_info->fs_devices->latest_bdev;
261507a0 703 em->compress_type = async_extent->compress_type;
771ed689
CM		 704 set_bit(EXTENT_FLAG_PINNED, &em->flags);
705 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
706
d397712b 707 while (1) {
890871be 708 write_lock(&em_tree->lock);
771ed689 709 ret = add_extent_mapping(em_tree, em);
890871be 710 write_unlock(&em_tree->lock);
771ed689
CM		 711 if (ret != -EEXIST) {
712 free_extent_map(em);
713 break;
714 }
715 btrfs_drop_extent_cache(inode, async_extent->start,
716 async_extent->start +
717 async_extent->ram_size - 1, 0);
718 }
719
261507a0
LZ		 720 ret = btrfs_add_ordered_extent_compress(inode,
721 async_extent->start,
722 ins.objectid,
723 async_extent->ram_size,
724 ins.offset,
725 BTRFS_ORDERED_COMPRESSED,
726 async_extent->compress_type);
79787eaa 727 BUG_ON(ret); /* -ENOMEM */
771ed689 728
771ed689
CM		 729 /*
730 * clear dirty, set writeback and unlock the pages.
731 */
732 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 733 &BTRFS_I(inode)->io_tree,
734 async_extent->start,
735 async_extent->start +
736 async_extent->ram_size - 1,
737 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
738 EXTENT_CLEAR_UNLOCK |
a3429ab7 739 EXTENT_CLEAR_DELALLOC |
a791e35e 740 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
771ed689
CM		 741
742 ret = btrfs_submit_compressed_write(inode,
d397712b
CM		 743 async_extent->start,
744 async_extent->ram_size,
745 ins.objectid,
746 ins.offset, async_extent->pages,
747 async_extent->nr_pages);
771ed689 748
79787eaa 749 BUG_ON(ret); /* -ENOMEM */
771ed689
CM		 750 alloc_hint = ins.objectid + ins.offset;
751 kfree(async_extent);
752 cond_resched();
753 }
79787eaa
JM		 754 ret = 0;
755out:
756 return ret;
757out_free:
758 kfree(async_extent);
759 goto out;
771ed689
CM		 760}
761
4b46fce2
JB		 762static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
763 u64 num_bytes)
764{
765 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
766 struct extent_map *em;
767 u64 alloc_hint = 0;
768
769 read_lock(&em_tree->lock);
770 em = search_extent_mapping(em_tree, start, num_bytes);
771 if (em) {
772 /*
773 * if block start isn't an actual block number then find the
774 * first block in this inode and use that as a hint. If that
775 * block is also bogus then just don't worry about it.
776 */
777 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
778 free_extent_map(em);
779 em = search_extent_mapping(em_tree, 0, 0);
780 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
781 alloc_hint = em->block_start;
782 if (em)
783 free_extent_map(em);
784 } else {
785 alloc_hint = em->block_start;
786 free_extent_map(em);
787 }
788 }
789 read_unlock(&em_tree->lock);
790
791 return alloc_hint;
792}
793
771ed689
CM		 794/*
795 * when extent_io.c finds a delayed allocation range in the file,
796 * the call backs end up in this code. The basic idea is to
797 * allocate extents on disk for the range, and create ordered data structs
798 * in ram to track those extents.
799 *
800 * locked_page is the page that writepage had locked already. We use
801 * it to make sure we don't do extra locks or unlocks.
802 *
803 * *page_started is set to one if we unlock locked_page and do everything
804 * required to start IO on it. It may be clean and already done with
805 * IO when we return.
806 */
b7d5b0a8
MX		 807static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
808 struct inode *inode,
809 struct btrfs_root *root,
810 struct page *locked_page,
811 u64 start, u64 end, int *page_started,
812 unsigned long *nr_written,
813 int unlock)
771ed689 814{
771ed689
CM		 815 u64 alloc_hint = 0;
816 u64 num_bytes;
817 unsigned long ram_size;
818 u64 disk_num_bytes;
819 u64 cur_alloc_size;
820 u64 blocksize = root->sectorsize;
771ed689
CM		 821 struct btrfs_key ins;
822 struct extent_map *em;
823 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
824 int ret = 0;
825
83eea1f1 826 BUG_ON(btrfs_is_free_space_inode(inode));
771ed689 827
771ed689
CM		 828 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
829 num_bytes = max(blocksize, num_bytes);
830 disk_num_bytes = num_bytes;
771ed689 831
4cb5300b 832 /* if this is a small write inside eof, kick off defrag */
4cb13e5d
LB		 833 if (num_bytes < 64 * 1024 &&
834 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
4cb5300b
CM		 835 btrfs_add_inode_defrag(trans, inode);
836
771ed689
CM		 837 if (start == 0) {
838 /* lets try to make an inline extent */
839 ret = cow_file_range_inline(trans, root, inode,
fe3f566c 840 start, end, 0, 0, NULL);
771ed689
CM		 841 if (ret == 0) {
842 extent_clear_unlock_delalloc(inode,
a791e35e
CM		 843 &BTRFS_I(inode)->io_tree,
844 start, end, NULL,
845 EXTENT_CLEAR_UNLOCK_PAGE |
846 EXTENT_CLEAR_UNLOCK |
847 EXTENT_CLEAR_DELALLOC |
848 EXTENT_CLEAR_DIRTY |
849 EXTENT_SET_WRITEBACK |
850 EXTENT_END_WRITEBACK);
c2167754 851
771ed689
CM		 852 *nr_written = *nr_written +
853 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
854 *page_started = 1;
771ed689 855 goto out;
79787eaa
JM		 856 } else if (ret < 0) {
857 btrfs_abort_transaction(trans, root, ret);
858 goto out_unlock;
771ed689
CM		 859 }
860 }
861
862 BUG_ON(disk_num_bytes >
6c41761f 863 btrfs_super_total_bytes(root->fs_info->super_copy));
771ed689 864
4b46fce2 865 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
771ed689
CM		 866 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
867
d397712b 868 while (disk_num_bytes > 0) {
a791e35e
CM		 869 unsigned long op;
870
287a0ab9 871 cur_alloc_size = disk_num_bytes;
e6dcd2dc 872 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
771ed689 873 root->sectorsize, 0, alloc_hint,
81c9ad23 874 &ins, 1);
79787eaa
JM		 875 if (ret < 0) {
876 btrfs_abort_transaction(trans, root, ret);
877 goto out_unlock;
878 }
d397712b 879
172ddd60 880 em = alloc_extent_map();
79787eaa 881 BUG_ON(!em); /* -ENOMEM */
e6dcd2dc 882 em->start = start;
445a6944 883 em->orig_start = em->start;
771ed689
CM		 884 ram_size = ins.offset;
885 em->len = ins.offset;
c8b97818 886
e6dcd2dc 887 em->block_start = ins.objectid;
c8b97818 888 em->block_len = ins.offset;
e6dcd2dc 889 em->bdev = root->fs_info->fs_devices->latest_bdev;
7f3c74fb 890 set_bit(EXTENT_FLAG_PINNED, &em->flags);
c8b97818 891
d397712b 892 while (1) {
890871be 893 write_lock(&em_tree->lock);
e6dcd2dc 894 ret = add_extent_mapping(em_tree, em);
890871be 895 write_unlock(&em_tree->lock);
e6dcd2dc
CM		 896 if (ret != -EEXIST) {
897 free_extent_map(em);
898 break;
899 }
900 btrfs_drop_extent_cache(inode, start,
c8b97818 901 start + ram_size - 1, 0);
e6dcd2dc
CM		 902 }
903
98d20f67 904 cur_alloc_size = ins.offset;
e6dcd2dc 905 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771ed689 906 ram_size, cur_alloc_size, 0);
79787eaa 907 BUG_ON(ret); /* -ENOMEM */
c8b97818 908
17d217fe
YZ		 909 if (root->root_key.objectid ==
910 BTRFS_DATA_RELOC_TREE_OBJECTID) {
911 ret = btrfs_reloc_clone_csums(inode, start,
912 cur_alloc_size);
79787eaa
JM		 913 if (ret) {
914 btrfs_abort_transaction(trans, root, ret);
915 goto out_unlock;
916 }
17d217fe
YZ		 917 }
918
d397712b 919 if (disk_num_bytes < cur_alloc_size)
3b951516 920 break;
d397712b 921
c8b97818
CM		 922 /* we're not doing compressed IO, don't unlock the first
923 * page (which the caller expects to stay locked), don't
924 * clear any dirty bits and don't set any writeback bits
8b62b72b
CM		 925 *
926 * Do set the Private2 bit so we know this page was properly
927 * setup for writepage
c8b97818 928 */
a791e35e
CM		 929 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
930 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
931 EXTENT_SET_PRIVATE2;
932
c8b97818
CM		 933 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
934 start, start + ram_size - 1,
a791e35e 935 locked_page, op);
c8b97818 936 disk_num_bytes -= cur_alloc_size;
c59f8951
CM		 937 num_bytes -= cur_alloc_size;
938 alloc_hint = ins.objectid + ins.offset;
939 start += cur_alloc_size;
b888db2b 940 }
79787eaa 941out:
be20aa9d 942 return ret;
b7d5b0a8 943
79787eaa
JM		 944out_unlock:
945 extent_clear_unlock_delalloc(inode,
946 &BTRFS_I(inode)->io_tree,
beb42dd7 947 start, end, locked_page,
79787eaa
JM		 948 EXTENT_CLEAR_UNLOCK_PAGE |
949 EXTENT_CLEAR_UNLOCK |
950 EXTENT_CLEAR_DELALLOC |
951 EXTENT_CLEAR_DIRTY |
952 EXTENT_SET_WRITEBACK |
953 EXTENT_END_WRITEBACK);
954
955 goto out;
771ed689 956}
c8b97818 957
b7d5b0a8
MX		 958static noinline int cow_file_range(struct inode *inode,
959 struct page *locked_page,
960 u64 start, u64 end, int *page_started,
961 unsigned long *nr_written,
962 int unlock)
963{
964 struct btrfs_trans_handle *trans;
965 struct btrfs_root *root = BTRFS_I(inode)->root;
966 int ret;
967
968 trans = btrfs_join_transaction(root);
969 if (IS_ERR(trans)) {
970 extent_clear_unlock_delalloc(inode,
971 &BTRFS_I(inode)->io_tree,
972 start, end, locked_page,
973 EXTENT_CLEAR_UNLOCK_PAGE |
974 EXTENT_CLEAR_UNLOCK |
975 EXTENT_CLEAR_DELALLOC |
976 EXTENT_CLEAR_DIRTY |
977 EXTENT_SET_WRITEBACK |
978 EXTENT_END_WRITEBACK);
979 return PTR_ERR(trans);
980 }
981 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
982
983 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
984 page_started, nr_written, unlock);
985
986 btrfs_end_transaction(trans, root);
987
988 return ret;
989}
990
771ed689
CM		 991/*
992 * work queue call back to started compression on a file and pages
993 */
994static noinline void async_cow_start(struct btrfs_work *work)
995{
996 struct async_cow *async_cow;
997 int num_added = 0;
998 async_cow = container_of(work, struct async_cow, work);
999
1000 compress_file_range(async_cow->inode, async_cow->locked_page,
1001 async_cow->start, async_cow->end, async_cow,
1002 &num_added);
8180ef88 1003 if (num_added == 0) {
cb77fcd8 1004 btrfs_add_delayed_iput(async_cow->inode);
771ed689 1005 async_cow->inode = NULL;
8180ef88 1006 }
771ed689
CM		 1007}
1008
1009/*
1010 * work queue call back to submit previously compressed pages
1011 */
1012static noinline void async_cow_submit(struct btrfs_work *work)
1013{
1014 struct async_cow *async_cow;
1015 struct btrfs_root *root;
1016 unsigned long nr_pages;
1017
1018 async_cow = container_of(work, struct async_cow, work);
1019
1020 root = async_cow->root;
1021 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1022 PAGE_CACHE_SHIFT;
1023
66657b31 1024 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
287082b0 1025 5 * 1024 * 1024 &&
771ed689
CM		 1026 waitqueue_active(&root->fs_info->async_submit_wait))
1027 wake_up(&root->fs_info->async_submit_wait);
1028
d397712b 1029 if (async_cow->inode)
771ed689 1030 submit_compressed_extents(async_cow->inode, async_cow);
771ed689 1031}
c8b97818 1032
771ed689
CM		 1033static noinline void async_cow_free(struct btrfs_work *work)
1034{
1035 struct async_cow *async_cow;
1036 async_cow = container_of(work, struct async_cow, work);
8180ef88 1037 if (async_cow->inode)
cb77fcd8 1038 btrfs_add_delayed_iput(async_cow->inode);
771ed689
CM		 1039 kfree(async_cow);
1040}
1041
1042static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043 u64 start, u64 end, int *page_started,
1044 unsigned long *nr_written)
1045{
1046 struct async_cow *async_cow;
1047 struct btrfs_root *root = BTRFS_I(inode)->root;
1048 unsigned long nr_pages;
1049 u64 cur_end;
287082b0 1050 int limit = 10 * 1024 * 1024;
771ed689 1051
a3429ab7
CM		 1052 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053 1, 0, NULL, GFP_NOFS);
d397712b 1054 while (start < end) {
771ed689 1055 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
79787eaa 1056 BUG_ON(!async_cow); /* -ENOMEM */
8180ef88 1057 async_cow->inode = igrab(inode);
771ed689
CM		 1058 async_cow->root = root;
1059 async_cow->locked_page = locked_page;
1060 async_cow->start = start;
1061
6cbff00f 1062 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
771ed689
CM		 1063 cur_end = end;
1064 else
1065 cur_end = min(end, start + 512 * 1024 - 1);
1066
1067 async_cow->end = cur_end;
1068 INIT_LIST_HEAD(&async_cow->extents);
1069
1070 async_cow->work.func = async_cow_start;
1071 async_cow->work.ordered_func = async_cow_submit;
1072 async_cow->work.ordered_free = async_cow_free;
1073 async_cow->work.flags = 0;
1074
771ed689
CM		 1075 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1076 PAGE_CACHE_SHIFT;
1077 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1078
1079 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1080 &async_cow->work);
1081
1082 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083 wait_event(root->fs_info->async_submit_wait,
1084 (atomic_read(&root->fs_info->async_delalloc_pages) <
1085 limit));
1086 }
1087
d397712b 1088 while (atomic_read(&root->fs_info->async_submit_draining) &&
771ed689
CM		 1089 atomic_read(&root->fs_info->async_delalloc_pages)) {
1090 wait_event(root->fs_info->async_submit_wait,
1091 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1092 0));
1093 }
1094
1095 *nr_written += nr_pages;
1096 start = cur_end + 1;
1097 }
1098 *page_started = 1;
1099 return 0;
be20aa9d
CM		 1100}
1101
d397712b 1102static noinline int csum_exist_in_range(struct btrfs_root *root,
17d217fe
YZ		 1103 u64 bytenr, u64 num_bytes)
1104{
1105 int ret;
1106 struct btrfs_ordered_sum *sums;
1107 LIST_HEAD(list);
1108
07d400a6 1109 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
a2de733c 1110 bytenr + num_bytes - 1, &list, 0);
17d217fe
YZ		 1111 if (ret == 0 && list_empty(&list))
1112 return 0;
1113
1114 while (!list_empty(&list)) {
1115 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116 list_del(&sums->list);
1117 kfree(sums);
1118 }
1119 return 1;
1120}
1121
d352ac68
CM		 1122/*
1123 * when nowcow writeback call back. This checks for snapshots or COW copies
1124 * of the extents that exist in the file, and COWs the file as required.
1125 *
1126 * If no cow copies or snapshots exist, we write directly to the existing
1127 * blocks on disk
1128 */
7f366cfe
CM		 1129static noinline int run_delalloc_nocow(struct inode *inode,
1130 struct page *locked_page,
771ed689
CM		 1131 u64 start, u64 end, int *page_started, int force,
1132 unsigned long *nr_written)
be20aa9d 1133{
be20aa9d 1134 struct btrfs_root *root = BTRFS_I(inode)->root;
7ea394f1 1135 struct btrfs_trans_handle *trans;
be20aa9d 1136 struct extent_buffer *leaf;
be20aa9d 1137 struct btrfs_path *path;
80ff3856 1138 struct btrfs_file_extent_item *fi;
be20aa9d 1139 struct btrfs_key found_key;
80ff3856
YZ		 1140 u64 cow_start;
1141 u64 cur_offset;
1142 u64 extent_end;
5d4f98a2 1143 u64 extent_offset;
80ff3856
YZ		 1144 u64 disk_bytenr;
1145 u64 num_bytes;
1146 int extent_type;
79787eaa 1147 int ret, err;
d899e052 1148 int type;
80ff3856
YZ		 1149 int nocow;
1150 int check_prev = 1;
82d5902d 1151 bool nolock;
33345d01 1152 u64 ino = btrfs_ino(inode);
be20aa9d
CM		 1153
1154 path = btrfs_alloc_path();
17ca04af
JB		 1155 if (!path) {
1156 extent_clear_unlock_delalloc(inode,
1157 &BTRFS_I(inode)->io_tree,
1158 start, end, locked_page,
1159 EXTENT_CLEAR_UNLOCK_PAGE |
1160 EXTENT_CLEAR_UNLOCK |
1161 EXTENT_CLEAR_DELALLOC |
1162 EXTENT_CLEAR_DIRTY |
1163 EXTENT_SET_WRITEBACK |
1164 EXTENT_END_WRITEBACK);
d8926bb3 1165 return -ENOMEM;
17ca04af 1166 }
82d5902d 1167
83eea1f1 1168 nolock = btrfs_is_free_space_inode(inode);
82d5902d
LZ		 1169
1170 if (nolock)
7a7eaa40 1171 trans = btrfs_join_transaction_nolock(root);
82d5902d 1172 else
7a7eaa40 1173 trans = btrfs_join_transaction(root);
ff5714cc 1174
79787eaa 1175 if (IS_ERR(trans)) {
17ca04af
JB		 1176 extent_clear_unlock_delalloc(inode,
1177 &BTRFS_I(inode)->io_tree,
1178 start, end, locked_page,
1179 EXTENT_CLEAR_UNLOCK_PAGE |
1180 EXTENT_CLEAR_UNLOCK |
1181 EXTENT_CLEAR_DELALLOC |
1182 EXTENT_CLEAR_DIRTY |
1183 EXTENT_SET_WRITEBACK |
1184 EXTENT_END_WRITEBACK);
79787eaa
JM		 1185 btrfs_free_path(path);
1186 return PTR_ERR(trans);
1187 }
1188
74b21075 1189 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
be20aa9d 1190
80ff3856
YZ		 1191 cow_start = (u64)-1;
1192 cur_offset = start;
1193 while (1) {
33345d01 1194 ret = btrfs_lookup_file_extent(trans, root, path, ino,
80ff3856 1195 cur_offset, 0);
79787eaa
JM		 1196 if (ret < 0) {
1197 btrfs_abort_transaction(trans, root, ret);
1198 goto error;
1199 }
80ff3856
YZ		 1200 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201 leaf = path->nodes[0];
1202 btrfs_item_key_to_cpu(leaf, &found_key,
1203 path->slots[0] - 1);
33345d01 1204 if (found_key.objectid == ino &&
80ff3856
YZ		 1205 found_key.type == BTRFS_EXTENT_DATA_KEY)
1206 path->slots[0]--;
1207 }
1208 check_prev = 0;
1209next_slot:
1210 leaf = path->nodes[0];
1211 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212 ret = btrfs_next_leaf(root, path);
79787eaa
JM		 1213 if (ret < 0) {
1214 btrfs_abort_transaction(trans, root, ret);
1215 goto error;
1216 }
80ff3856
YZ		 1217 if (ret > 0)
1218 break;
1219 leaf = path->nodes[0];
1220 }
be20aa9d 1221
80ff3856
YZ		 1222 nocow = 0;
1223 disk_bytenr = 0;
17d217fe 1224 num_bytes = 0;
80ff3856
YZ		 1225 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1226
33345d01 1227 if (found_key.objectid > ino ||
80ff3856
YZ		 1228 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229 found_key.offset > end)
1230 break;
1231
1232 if (found_key.offset > cur_offset) {
1233 extent_end = found_key.offset;
e9061e21 1234 extent_type = 0;
80ff3856
YZ		 1235 goto out_check;
1236 }
1237
1238 fi = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_file_extent_item);
1240 extent_type = btrfs_file_extent_type(leaf, fi);
1241
d899e052
YZ		 1242 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
80ff3856 1244 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5d4f98a2 1245 extent_offset = btrfs_file_extent_offset(leaf, fi);
80ff3856
YZ		 1246 extent_end = found_key.offset +
1247 btrfs_file_extent_num_bytes(leaf, fi);
1248 if (extent_end <= start) {
1249 path->slots[0]++;
1250 goto next_slot;
1251 }
17d217fe
YZ		 1252 if (disk_bytenr == 0)
1253 goto out_check;
80ff3856
YZ		 1254 if (btrfs_file_extent_compression(leaf, fi) ||
1255 btrfs_file_extent_encryption(leaf, fi) ||
1256 btrfs_file_extent_other_encoding(leaf, fi))
1257 goto out_check;
d899e052
YZ		 1258 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1259 goto out_check;
d2fb3437 1260 if (btrfs_extent_readonly(root, disk_bytenr))
80ff3856 1261 goto out_check;
33345d01 1262 if (btrfs_cross_ref_exist(trans, root, ino,
5d4f98a2
YZ		 1263 found_key.offset -
1264 extent_offset, disk_bytenr))
17d217fe 1265 goto out_check;
5d4f98a2 1266 disk_bytenr += extent_offset;
17d217fe
YZ		 1267 disk_bytenr += cur_offset - found_key.offset;
1268 num_bytes = min(end + 1, extent_end) - cur_offset;
1269 /*
1270 * force cow if csum exists in the range.
1271 * this ensure that csum for a given extent are
1272 * either valid or do not exist.
1273 */
1274 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1275 goto out_check;
80ff3856
YZ		 1276 nocow = 1;
1277 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278 extent_end = found_key.offset +
1279 btrfs_file_extent_inline_len(leaf, fi);
1280 extent_end = ALIGN(extent_end, root->sectorsize);
1281 } else {
1282 BUG_ON(1);
1283 }
1284out_check:
1285 if (extent_end <= start) {
1286 path->slots[0]++;
1287 goto next_slot;
1288 }
1289 if (!nocow) {
1290 if (cow_start == (u64)-1)
1291 cow_start = cur_offset;
1292 cur_offset = extent_end;
1293 if (cur_offset > end)
1294 break;
1295 path->slots[0]++;
1296 goto next_slot;
7ea394f1
YZ		 1297 }
1298
b3b4aa74 1299 btrfs_release_path(path);
80ff3856 1300 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1301 ret = __cow_file_range(trans, inode, root, locked_page,
1302 cow_start, found_key.offset - 1,
1303 page_started, nr_written, 1);
79787eaa
JM		 1304 if (ret) {
1305 btrfs_abort_transaction(trans, root, ret);
1306 goto error;
1307 }
80ff3856 1308 cow_start = (u64)-1;
7ea394f1 1309 }
80ff3856 1310
d899e052
YZ		 1311 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312 struct extent_map *em;
1313 struct extent_map_tree *em_tree;
1314 em_tree = &BTRFS_I(inode)->extent_tree;
172ddd60 1315 em = alloc_extent_map();
79787eaa 1316 BUG_ON(!em); /* -ENOMEM */
d899e052 1317 em->start = cur_offset;
445a6944 1318 em->orig_start = em->start;
d899e052
YZ		 1319 em->len = num_bytes;
1320 em->block_len = num_bytes;
1321 em->block_start = disk_bytenr;
1322 em->bdev = root->fs_info->fs_devices->latest_bdev;
1323 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4e2f84e6 1324 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
d899e052 1325 while (1) {
890871be 1326 write_lock(&em_tree->lock);
d899e052 1327 ret = add_extent_mapping(em_tree, em);
890871be 1328 write_unlock(&em_tree->lock);
d899e052
YZ		 1329 if (ret != -EEXIST) {
1330 free_extent_map(em);
1331 break;
1332 }
1333 btrfs_drop_extent_cache(inode, em->start,
1334 em->start + em->len - 1, 0);
1335 }
1336 type = BTRFS_ORDERED_PREALLOC;
1337 } else {
1338 type = BTRFS_ORDERED_NOCOW;
1339 }
80ff3856
YZ		 1340
1341 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
d899e052 1342 num_bytes, num_bytes, type);
79787eaa 1343 BUG_ON(ret); /* -ENOMEM */
771ed689 1344
efa56464
YZ		 1345 if (root->root_key.objectid ==
1346 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1348 num_bytes);
79787eaa
JM		 1349 if (ret) {
1350 btrfs_abort_transaction(trans, root, ret);
1351 goto error;
1352 }
efa56464
YZ		 1353 }
1354
d899e052 1355 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
a791e35e
CM		 1356 cur_offset, cur_offset + num_bytes - 1,
1357 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359 EXTENT_SET_PRIVATE2);
80ff3856
YZ		 1360 cur_offset = extent_end;
1361 if (cur_offset > end)
1362 break;
be20aa9d 1363 }
b3b4aa74 1364 btrfs_release_path(path);
80ff3856 1365
17ca04af 1366 if (cur_offset <= end && cow_start == (u64)-1) {
80ff3856 1367 cow_start = cur_offset;
17ca04af
JB		 1368 cur_offset = end;
1369 }
1370
80ff3856 1371 if (cow_start != (u64)-1) {
b7d5b0a8
MX		 1372 ret = __cow_file_range(trans, inode, root, locked_page,
1373 cow_start, end,
1374 page_started, nr_written, 1);
79787eaa
JM		 1375 if (ret) {
1376 btrfs_abort_transaction(trans, root, ret);
1377 goto error;
1378 }
80ff3856
YZ		 1379 }
1380
79787eaa 1381error:
a698d075 1382 err = btrfs_end_transaction(trans, root);
79787eaa
JM		 1383 if (!ret)
1384 ret = err;
1385
17ca04af
JB		 1386 if (ret && cur_offset < end)
1387 extent_clear_unlock_delalloc(inode,
1388 &BTRFS_I(inode)->io_tree,
1389 cur_offset, end, locked_page,
1390 EXTENT_CLEAR_UNLOCK_PAGE |
1391 EXTENT_CLEAR_UNLOCK |
1392 EXTENT_CLEAR_DELALLOC |
1393 EXTENT_CLEAR_DIRTY |
1394 EXTENT_SET_WRITEBACK |
1395 EXTENT_END_WRITEBACK);
1396
7ea394f1 1397 btrfs_free_path(path);
79787eaa 1398 return ret;
be20aa9d
CM		 1399}
1400
d352ac68
CM		 1401/*
1402 * extent_io.c call back to do delayed allocation processing
1403 */
c8b97818 1404static int run_delalloc_range(struct inode *inode, struct page *locked_page,
771ed689
CM		 1405 u64 start, u64 end, int *page_started,
1406 unsigned long *nr_written)
be20aa9d 1407{
be20aa9d 1408 int ret;
7f366cfe 1409 struct btrfs_root *root = BTRFS_I(inode)->root;
a2135011 1410
7ddf5a42 1411 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
c8b97818 1412 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1413 page_started, 1, nr_written);
7ddf5a42 1414 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
d899e052 1415 ret = run_delalloc_nocow(inode, locked_page, start, end,
d397712b 1416 page_started, 0, nr_written);
7ddf5a42
JB		 1417 } else if (!btrfs_test_opt(root, COMPRESS) &&
1418 !(BTRFS_I(inode)->force_compress) &&
1419 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
7f366cfe
CM		 1420 ret = cow_file_range(inode, locked_page, start, end,
1421 page_started, nr_written, 1);
7ddf5a42
JB		 1422 } else {
1423 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424 &BTRFS_I(inode)->runtime_flags);
771ed689 1425 ret = cow_file_range_async(inode, locked_page, start, end,
d397712b 1426 page_started, nr_written);
7ddf5a42 1427 }
b888db2b
CM		 1428 return ret;
1429}
1430
1bf85046
JM		 1431static void btrfs_split_extent_hook(struct inode *inode,
1432 struct extent_state *orig, u64 split)
9ed74f2d 1433{
0ca1f7ce 1434 /* not delalloc, ignore it */
9ed74f2d 1435 if (!(orig->state & EXTENT_DELALLOC))
1bf85046 1436 return;
9ed74f2d 1437
9e0baf60
JB		 1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents++;
1440 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1441}
1442
1443/*
1444 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445 * extents so we can keep track of new extents that are just merged onto old
1446 * extents, such as when we are doing sequential writes, so we can properly
1447 * account for the metadata space we'll need.
1448 */
1bf85046
JM		 1449static void btrfs_merge_extent_hook(struct inode *inode,
1450 struct extent_state *new,
1451 struct extent_state *other)
9ed74f2d 1452{
9ed74f2d
JB		 1453 /* not delalloc, ignore it */
1454 if (!(other->state & EXTENT_DELALLOC))
1bf85046 1455 return;
9ed74f2d 1456
9e0baf60
JB		 1457 spin_lock(&BTRFS_I(inode)->lock);
1458 BTRFS_I(inode)->outstanding_extents--;
1459 spin_unlock(&BTRFS_I(inode)->lock);
9ed74f2d
JB		 1460}
1461
d352ac68
CM		 1462/*
1463 * extent_io.c set_bit_hook, used to track delayed allocation
1464 * bytes in this file, and to maintain the list of inodes that
1465 * have pending delalloc work to be done.
1466 */
1bf85046
JM		 1467static void btrfs_set_bit_hook(struct inode *inode,
1468 struct extent_state *state, int *bits)
291d673e 1469{
9ed74f2d 1470
75eff68e
CM		 1471 /*
1472 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1473 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1474 * bit, which is only set or cleared with irqs on
1475 */
0ca1f7ce 1476 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1477 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1478 u64 len = state->end + 1 - state->start;
83eea1f1 1479 bool do_list = !btrfs_is_free_space_inode(inode);
9ed74f2d 1480
9e0baf60 1481 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1482 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1483 } else {
1484 spin_lock(&BTRFS_I(inode)->lock);
1485 BTRFS_I(inode)->outstanding_extents++;
1486 spin_unlock(&BTRFS_I(inode)->lock);
1487 }
287a0ab9 1488
75eff68e 1489 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1490 BTRFS_I(inode)->delalloc_bytes += len;
1491 root->fs_info->delalloc_bytes += len;
0cb59c99 1492 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
ea8c2819
CM		 1493 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494 &root->fs_info->delalloc_inodes);
1495 }
75eff68e 1496 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1497 }
291d673e
CM		 1498}
1499
d352ac68
CM		 1500/*
1501 * extent_io.c clear_bit_hook, see set_bit_hook for why
1502 */
1bf85046
JM		 1503static void btrfs_clear_bit_hook(struct inode *inode,
1504 struct extent_state *state, int *bits)
291d673e 1505{
75eff68e
CM		 1506 /*
1507 * set_bit and clear bit hooks normally require _irqsave/restore
27160b6b 1508 * but in this case, we are only testing for the DELALLOC
75eff68e
CM		 1509 * bit, which is only set or cleared with irqs on
1510 */
0ca1f7ce 1511 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
291d673e 1512 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1513 u64 len = state->end + 1 - state->start;
83eea1f1 1514 bool do_list = !btrfs_is_free_space_inode(inode);
bcbfce8a 1515
9e0baf60 1516 if (*bits & EXTENT_FIRST_DELALLOC) {
0ca1f7ce 1517 *bits &= ~EXTENT_FIRST_DELALLOC;
9e0baf60
JB		 1518 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519 spin_lock(&BTRFS_I(inode)->lock);
1520 BTRFS_I(inode)->outstanding_extents--;
1521 spin_unlock(&BTRFS_I(inode)->lock);
1522 }
0ca1f7ce
YZ		 1523
1524 if (*bits & EXTENT_DO_ACCOUNTING)
1525 btrfs_delalloc_release_metadata(inode, len);
1526
0cb59c99
JB		 1527 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1528 && do_list)
0ca1f7ce 1529 btrfs_free_reserved_data_space(inode, len);
9ed74f2d 1530
75eff68e 1531 spin_lock(&root->fs_info->delalloc_lock);
0ca1f7ce
YZ		 1532 root->fs_info->delalloc_bytes -= len;
1533 BTRFS_I(inode)->delalloc_bytes -= len;
1534
0cb59c99 1535 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
ea8c2819
CM		 1536 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1538 }
75eff68e 1539 spin_unlock(&root->fs_info->delalloc_lock);
291d673e 1540 }
291d673e
CM		 1541}
1542
d352ac68
CM		 1543/*
1544 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545 * we don't create bios that span stripes or chunks
1546 */
239b14b3 1547int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
c8b97818
CM		 1548 size_t size, struct bio *bio,
1549 unsigned long bio_flags)
239b14b3
CM		 1550{
1551 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552 struct btrfs_mapping_tree *map_tree;
a62b9401 1553 u64 logical = (u64)bio->bi_sector << 9;
239b14b3
CM		 1554 u64 length = 0;
1555 u64 map_length;
239b14b3
CM		 1556 int ret;
1557
771ed689
CM		 1558 if (bio_flags & EXTENT_BIO_COMPRESSED)
1559 return 0;
1560
f2d8d74d 1561 length = bio->bi_size;
239b14b3
CM		 1562 map_tree = &root->fs_info->mapping_tree;
1563 map_length = length;
cea9e445 1564 ret = btrfs_map_block(map_tree, READ, logical,
f188591e 1565 &map_length, NULL, 0);
3444a972
JM		 1566 /* Will always return 0 or 1 with map_multi == NULL */
1567 BUG_ON(ret < 0);
d397712b 1568 if (map_length < length + size)
239b14b3 1569 return 1;
3444a972 1570 return 0;
239b14b3
CM		 1571}
1572
d352ac68
CM		 1573/*
1574 * in order to insert checksums into the metadata in large chunks,
1575 * we wait until bio submission time. All the pages in the bio are
1576 * checksummed and sums are attached onto the ordered extent record.
1577 *
1578 * At IO completion time the cums attached on the ordered extent record
1579 * are inserted into the btree
1580 */
d397712b
CM		 1581static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1582 struct bio *bio, int mirror_num,
eaf25d93
CM		 1583 unsigned long bio_flags,
1584 u64 bio_offset)
065631f6 1585{
065631f6 1586 struct btrfs_root *root = BTRFS_I(inode)->root;
065631f6 1587 int ret = 0;
e015640f 1588
d20f7043 1589 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
79787eaa 1590 BUG_ON(ret); /* -ENOMEM */
4a69a410
CM		 1591 return 0;
1592}
e015640f 1593
4a69a410
CM		 1594/*
1595 * in order to insert checksums into the metadata in large chunks,
1596 * we wait until bio submission time. All the pages in the bio are
1597 * checksummed and sums are attached onto the ordered extent record.
1598 *
1599 * At IO completion time the cums attached on the ordered extent record
1600 * are inserted into the btree
1601 */
b2950863 1602static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1603 int mirror_num, unsigned long bio_flags,
1604 u64 bio_offset)
4a69a410
CM		 1605{
1606 struct btrfs_root *root = BTRFS_I(inode)->root;
8b712842 1607 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
44b8bd7e
CM		 1608}
1609
d352ac68 1610/*
cad321ad
CM		 1611 * extent_io.c submission hook. This does the right thing for csum calculation
1612 * on write, or reading the csums from the tree before a read
d352ac68 1613 */
b2950863 1614static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 1615 int mirror_num, unsigned long bio_flags,
1616 u64 bio_offset)
44b8bd7e
CM		 1617{
1618 struct btrfs_root *root = BTRFS_I(inode)->root;
1619 int ret = 0;
19b9bdb0 1620 int skip_sum;
0417341e 1621 int metadata = 0;
44b8bd7e 1622
6cbff00f 1623 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
cad321ad 1624
83eea1f1 1625 if (btrfs_is_free_space_inode(inode))
0417341e
JM		 1626 metadata = 2;
1627
7b6d91da 1628 if (!(rw & REQ_WRITE)) {
5fd02043
JB		 1629 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1630 if (ret)
1631 return ret;
1632
d20f7043 1633 if (bio_flags & EXTENT_BIO_COMPRESSED) {
c8b97818
CM		 1634 return btrfs_submit_compressed_read(inode, bio,
1635 mirror_num, bio_flags);
c2db1073
TI		 1636 } else if (!skip_sum) {
1637 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1638 if (ret)
1639 return ret;
1640 }
4d1b5fb4 1641 goto mapit;
19b9bdb0 1642 } else if (!skip_sum) {
17d217fe
YZ		 1643 /* csum items have already been cloned */
1644 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1645 goto mapit;
19b9bdb0
CM		 1646 /* we're doing a write, do the async checksumming */
1647 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
44b8bd7e 1648 inode, rw, bio, mirror_num,
eaf25d93
CM		 1649 bio_flags, bio_offset,
1650 __btrfs_submit_bio_start,
4a69a410 1651 __btrfs_submit_bio_done);
19b9bdb0
CM		 1652 }
1653
0b86a832 1654mapit:
8b712842 1655 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
065631f6 1656}
6885f308 1657
d352ac68
CM		 1658/*
1659 * given a list of ordered sums record them in the inode. This happens
1660 * at IO completion time based on sums calculated at bio submission time.
1661 */
ba1da2f4 1662static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
e6dcd2dc
CM		 1663 struct inode *inode, u64 file_offset,
1664 struct list_head *list)
1665{
e6dcd2dc
CM		 1666 struct btrfs_ordered_sum *sum;
1667
c6e30871 1668 list_for_each_entry(sum, list, list) {
d20f7043
CM		 1669 btrfs_csum_file_blocks(trans,
1670 BTRFS_I(inode)->root->fs_info->csum_root, sum);
e6dcd2dc
CM		 1671 }
1672 return 0;
1673}
1674
2ac55d41
JB		 1675int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1676 struct extent_state **cached_state)
ea8c2819 1677{
6c1500f2 1678 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
ea8c2819 1679 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2ac55d41 1680 cached_state, GFP_NOFS);
ea8c2819
CM		 1681}
1682
d352ac68 1683/* see btrfs_writepage_start_hook for details on why this is required */
247e743c
CM		 1684struct btrfs_writepage_fixup {
1685 struct page *page;
1686 struct btrfs_work work;
1687};
1688
b2950863 1689static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
247e743c
CM		 1690{
1691 struct btrfs_writepage_fixup *fixup;
1692 struct btrfs_ordered_extent *ordered;
2ac55d41 1693 struct extent_state *cached_state = NULL;
247e743c
CM		 1694 struct page *page;
1695 struct inode *inode;
1696 u64 page_start;
1697 u64 page_end;
87826df0 1698 int ret;
247e743c
CM		 1699
1700 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1701 page = fixup->page;
4a096752 1702again:
247e743c
CM		 1703 lock_page(page);
1704 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1705 ClearPageChecked(page);
1706 goto out_page;
1707 }
1708
1709 inode = page->mapping->host;
1710 page_start = page_offset(page);
1711 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1712
2ac55d41 1713 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
d0082371 1714 &cached_state);
4a096752
CM		 1715
1716 /* already ordered? We're done */
8b62b72b 1717 if (PagePrivate2(page))
247e743c 1718 goto out;
4a096752
CM		 1719
1720 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1721 if (ordered) {
2ac55d41
JB		 1722 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1723 page_end, &cached_state, GFP_NOFS);
4a096752
CM		 1724 unlock_page(page);
1725 btrfs_start_ordered_extent(inode, ordered, 1);
87826df0 1726 btrfs_put_ordered_extent(ordered);
4a096752
CM		 1727 goto again;
1728 }
247e743c 1729
87826df0
JM		 1730 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1731 if (ret) {
1732 mapping_set_error(page->mapping, ret);
1733 end_extent_writepage(page, ret, page_start, page_end);
1734 ClearPageChecked(page);
1735 goto out;
1736 }
1737
2ac55d41 1738 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
247e743c 1739 ClearPageChecked(page);
87826df0 1740 set_page_dirty(page);
247e743c 1741out:
2ac55d41
JB		 1742 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1743 &cached_state, GFP_NOFS);
247e743c
CM		 1744out_page:
1745 unlock_page(page);
1746 page_cache_release(page);
b897abec 1747 kfree(fixup);
247e743c
CM		 1748}
1749
1750/*
1751 * There are a few paths in the higher layers of the kernel that directly
1752 * set the page dirty bit without asking the filesystem if it is a
1753 * good idea. This causes problems because we want to make sure COW
1754 * properly happens and the data=ordered rules are followed.
1755 *
c8b97818 1756 * In our case any range that doesn't have the ORDERED bit set
247e743c
CM		 1757 * hasn't been properly setup for IO. We kick off an async process
1758 * to fix it up. The async helper will wait for ordered extents, set
1759 * the delalloc bit and make it safe to write the page.
1760 */
b2950863 1761static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
247e743c
CM		 1762{
1763 struct inode *inode = page->mapping->host;
1764 struct btrfs_writepage_fixup *fixup;
1765 struct btrfs_root *root = BTRFS_I(inode)->root;
247e743c 1766
8b62b72b
CM		 1767 /* this page is properly in the ordered list */
1768 if (TestClearPagePrivate2(page))
247e743c
CM		 1769 return 0;
1770
1771 if (PageChecked(page))
1772 return -EAGAIN;
1773
1774 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1775 if (!fixup)
1776 return -EAGAIN;
f421950f 1777
247e743c
CM		 1778 SetPageChecked(page);
1779 page_cache_get(page);
1780 fixup->work.func = btrfs_writepage_fixup_worker;
1781 fixup->page = page;
1782 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
87826df0 1783 return -EBUSY;
247e743c
CM		 1784}
1785
d899e052
YZ		 1786static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1787 struct inode *inode, u64 file_pos,
1788 u64 disk_bytenr, u64 disk_num_bytes,
1789 u64 num_bytes, u64 ram_bytes,
1790 u8 compression, u8 encryption,
1791 u16 other_encoding, int extent_type)
1792{
1793 struct btrfs_root *root = BTRFS_I(inode)->root;
1794 struct btrfs_file_extent_item *fi;
1795 struct btrfs_path *path;
1796 struct extent_buffer *leaf;
1797 struct btrfs_key ins;
d899e052
YZ		 1798 int ret;
1799
1800 path = btrfs_alloc_path();
d8926bb3
MF		 1801 if (!path)
1802 return -ENOMEM;
d899e052 1803
b9473439 1804 path->leave_spinning = 1;
a1ed835e
CM		 1805
1806 /*
1807 * we may be replacing one extent in the tree with another.
1808 * The new extent is pinned in the extent map, and we don't want
1809 * to drop it from the cache until it is completely in the btree.
1810 *
1811 * So, tell btrfs_drop_extents to leave this extent in the cache.
1812 * the caller is expected to unpin it and allow it to be merged
1813 * with the others.
1814 */
5dc562c5 1815 ret = btrfs_drop_extents(trans, root, inode, file_pos,
2671485d 1816 file_pos + num_bytes, 0);
79787eaa
JM		 1817 if (ret)
1818 goto out;
d899e052 1819
33345d01 1820 ins.objectid = btrfs_ino(inode);
d899e052
YZ		 1821 ins.offset = file_pos;
1822 ins.type = BTRFS_EXTENT_DATA_KEY;
1823 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
79787eaa
JM		 1824 if (ret)
1825 goto out;
d899e052
YZ		 1826 leaf = path->nodes[0];
1827 fi = btrfs_item_ptr(leaf, path->slots[0],
1828 struct btrfs_file_extent_item);
1829 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1830 btrfs_set_file_extent_type(leaf, fi, extent_type);
1831 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1832 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1833 btrfs_set_file_extent_offset(leaf, fi, 0);
1834 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1835 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1836 btrfs_set_file_extent_compression(leaf, fi, compression);
1837 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1838 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
b9473439 1839
d899e052 1840 btrfs_mark_buffer_dirty(leaf);
ce195332 1841 btrfs_release_path(path);
d899e052
YZ		 1842
1843 inode_add_bytes(inode, num_bytes);
d899e052
YZ		 1844
1845 ins.objectid = disk_bytenr;
1846 ins.offset = disk_num_bytes;
1847 ins.type = BTRFS_EXTENT_ITEM_KEY;
5d4f98a2
YZ		 1848 ret = btrfs_alloc_reserved_file_extent(trans, root,
1849 root->root_key.objectid,
33345d01 1850 btrfs_ino(inode), file_pos, &ins);
79787eaa 1851out:
d899e052 1852 btrfs_free_path(path);
b9473439 1853
79787eaa 1854 return ret;
d899e052
YZ		 1855}
1856
5d13a98f
CM		 1857/*
1858 * helper function for btrfs_finish_ordered_io, this
1859 * just reads in some of the csum leaves to prime them into ram
1860 * before we start the transaction. It limits the amount of btree
1861 * reads required while inside the transaction.
1862 */
d352ac68
CM		 1863/* as ordered data IO finishes, this gets called so we can finish
1864 * an ordered extent if the range of bytes in the file it covers are
1865 * fully written.
1866 */
5fd02043 1867static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
e6dcd2dc 1868{
5fd02043 1869 struct inode *inode = ordered_extent->inode;
e6dcd2dc 1870 struct btrfs_root *root = BTRFS_I(inode)->root;
0ca1f7ce 1871 struct btrfs_trans_handle *trans = NULL;
e6dcd2dc 1872 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2ac55d41 1873 struct extent_state *cached_state = NULL;
261507a0 1874 int compress_type = 0;
e6dcd2dc 1875 int ret;
82d5902d 1876 bool nolock;
e6dcd2dc 1877
83eea1f1 1878 nolock = btrfs_is_free_space_inode(inode);
0cb59c99 1879
5fd02043
JB		 1880 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1881 ret = -EIO;
1882 goto out;
1883 }
1884
c2167754 1885 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
79787eaa 1886 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
c2167754
YZ		 1887 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1888 if (!ret) {
0cb59c99 1889 if (nolock)
7a7eaa40 1890 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1891 else
7a7eaa40 1892 trans = btrfs_join_transaction(root);
d280e5be
LB		 1893 if (IS_ERR(trans)) {
1894 ret = PTR_ERR(trans);
1895 trans = NULL;
1896 goto out;
1897 }
0ca1f7ce 1898 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2115133f 1899 ret = btrfs_update_inode_fallback(trans, root, inode);
79787eaa
JM		 1900 if (ret) /* -ENOMEM or corruption */
1901 btrfs_abort_transaction(trans, root, ret);
c2167754
YZ		 1902 }
1903 goto out;
1904 }
e6dcd2dc 1905
2ac55d41
JB		 1906 lock_extent_bits(io_tree, ordered_extent->file_offset,
1907 ordered_extent->file_offset + ordered_extent->len - 1,
d0082371 1908 0, &cached_state);
e6dcd2dc 1909
0cb59c99 1910 if (nolock)
7a7eaa40 1911 trans = btrfs_join_transaction_nolock(root);
0cb59c99 1912 else
7a7eaa40 1913 trans = btrfs_join_transaction(root);
79787eaa
JM		 1914 if (IS_ERR(trans)) {
1915 ret = PTR_ERR(trans);
1916 trans = NULL;
1917 goto out_unlock;
1918 }
0ca1f7ce 1919 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
c2167754 1920
c8b97818 1921 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
261507a0 1922 compress_type = ordered_extent->compress_type;
d899e052 1923 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
261507a0 1924 BUG_ON(compress_type);
920bbbfb 1925 ret = btrfs_mark_extent_written(trans, inode,
d899e052
YZ		 1926 ordered_extent->file_offset,
1927 ordered_extent->file_offset +
1928 ordered_extent->len);
d899e052 1929 } else {
0af3d00b 1930 BUG_ON(root == root->fs_info->tree_root);
d899e052
YZ		 1931 ret = insert_reserved_file_extent(trans, inode,
1932 ordered_extent->file_offset,
1933 ordered_extent->start,
1934 ordered_extent->disk_len,
1935 ordered_extent->len,
1936 ordered_extent->len,
261507a0 1937 compress_type, 0, 0,
d899e052 1938 BTRFS_FILE_EXTENT_REG);
d899e052 1939 }
5dc562c5
JB		 1940 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1941 ordered_extent->file_offset, ordered_extent->len,
1942 trans->transid);
79787eaa
JM		 1943 if (ret < 0) {
1944 btrfs_abort_transaction(trans, root, ret);
5fd02043 1945 goto out_unlock;
79787eaa 1946 }
2ac55d41 1947
e6dcd2dc
CM		 1948 add_pending_csums(trans, inode, ordered_extent->file_offset,
1949 &ordered_extent->list);
1950
1ef30be1 1951 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
a39f7521 1952 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2115133f 1953 ret = btrfs_update_inode_fallback(trans, root, inode);
79787eaa
JM		 1954 if (ret) { /* -ENOMEM or corruption */
1955 btrfs_abort_transaction(trans, root, ret);
5fd02043 1956 goto out_unlock;
79787eaa 1957 }
7c735313
JB		 1958 } else {
1959 btrfs_set_inode_last_trans(trans, inode);
1ef30be1
JB		 1960 }
1961 ret = 0;
5fd02043
JB		 1962out_unlock:
1963 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1964 ordered_extent->file_offset +
1965 ordered_extent->len - 1, &cached_state, GFP_NOFS);
c2167754 1966out:
5b0e95bf 1967 if (root != root->fs_info->tree_root)
0cb59c99 1968 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
a698d075
MX		 1969 if (trans)
1970 btrfs_end_transaction(trans, root);
0cb59c99 1971
5fd02043
JB		 1972 if (ret)
1973 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1974 ordered_extent->file_offset +
1975 ordered_extent->len - 1, NULL, GFP_NOFS);
1976
1977 /*
8bad3c02
LB		 1978 * This needs to be done to make sure anybody waiting knows we are done
1979 * updating everything for this ordered extent.
5fd02043
JB		 1980 */
1981 btrfs_remove_ordered_extent(inode, ordered_extent);
1982
e6dcd2dc
CM		 1983 /* once for us */
1984 btrfs_put_ordered_extent(ordered_extent);
1985 /* once for the tree */
1986 btrfs_put_ordered_extent(ordered_extent);
1987
5fd02043
JB		 1988 return ret;
1989}
1990
1991static void finish_ordered_fn(struct btrfs_work *work)
1992{
1993 struct btrfs_ordered_extent *ordered_extent;
1994 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1995 btrfs_finish_ordered_io(ordered_extent);
e6dcd2dc
CM		 1996}
1997
b2950863 1998static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
211f90e6
CM		 1999 struct extent_state *state, int uptodate)
2000{
5fd02043
JB		 2001 struct inode *inode = page->mapping->host;
2002 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 struct btrfs_ordered_extent *ordered_extent = NULL;
2004 struct btrfs_workers *workers;
2005
1abe9b8a 2006 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2007
8b62b72b 2008 ClearPagePrivate2(page);
5fd02043
JB		 2009 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2010 end - start + 1, uptodate))
2011 return 0;
2012
2013 ordered_extent->work.func = finish_ordered_fn;
2014 ordered_extent->work.flags = 0;
2015
83eea1f1 2016 if (btrfs_is_free_space_inode(inode))
5fd02043
JB		 2017 workers = &root->fs_info->endio_freespace_worker;
2018 else
2019 workers = &root->fs_info->endio_write_workers;
2020 btrfs_queue_worker(workers, &ordered_extent->work);
2021
2022 return 0;
211f90e6
CM		 2023}
2024
d352ac68
CM		 2025/*
2026 * when reads are done, we need to check csums to verify the data is correct
4a54c8c1
JS		 2027 * if there's a match, we allow the bio to finish. If not, the code in
2028 * extent_io.c will try to find good copies for us.
d352ac68 2029 */
b2950863 2030static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 2031 struct extent_state *state, int mirror)
07157aac 2032{
35ebb934 2033 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
07157aac 2034 struct inode *inode = page->mapping->host;
d1310b2e 2035 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
07157aac 2036 char *kaddr;
aadfeb6e 2037 u64 private = ~(u32)0;
07157aac 2038 int ret;
ff79f819
CM		 2039 struct btrfs_root *root = BTRFS_I(inode)->root;
2040 u32 csum = ~(u32)0;
d1310b2e 2041
d20f7043
CM		 2042 if (PageChecked(page)) {
2043 ClearPageChecked(page);
2044 goto good;
2045 }
6cbff00f
CH		 2046
2047 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
08d2f347 2048 goto good;
17d217fe
YZ		 2049
2050 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
9655d298 2051 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
17d217fe
YZ		 2052 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2053 GFP_NOFS);
b6cda9bc 2054 return 0;
17d217fe 2055 }
d20f7043 2056
c2e639f0 2057 if (state && state->start == start) {
70dec807
CM		 2058 private = state->private;
2059 ret = 0;
2060 } else {
2061 ret = get_state_private(io_tree, start, &private);
2062 }
7ac687d9 2063 kaddr = kmap_atomic(page);
d397712b 2064 if (ret)
07157aac 2065 goto zeroit;
d397712b 2066
ff79f819
CM		 2067 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2068 btrfs_csum_final(csum, (char *)&csum);
d397712b 2069 if (csum != private)
07157aac 2070 goto zeroit;
d397712b 2071
7ac687d9 2072 kunmap_atomic(kaddr);
d20f7043 2073good:
07157aac
CM		 2074 return 0;
2075
2076zeroit:
945d8962 2077 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
33345d01
LZ		 2078 "private %llu\n",
2079 (unsigned long long)btrfs_ino(page->mapping->host),
193f284d
CM		 2080 (unsigned long long)start, csum,
2081 (unsigned long long)private);
db94535d
CM		 2082 memset(kaddr + offset, 1, end - start + 1);
2083 flush_dcache_page(page);
7ac687d9 2084 kunmap_atomic(kaddr);
3b951516
CM		 2085 if (private == 0)
2086 return 0;
7e38326f 2087 return -EIO;
07157aac 2088}
b888db2b 2089
24bbcf04
YZ		 2090struct delayed_iput {
2091 struct list_head list;
2092 struct inode *inode;
2093};
2094
79787eaa
JM		 2095/* JDM: If this is fs-wide, why can't we add a pointer to
2096 * btrfs_inode instead and avoid the allocation? */
24bbcf04
YZ		 2097void btrfs_add_delayed_iput(struct inode *inode)
2098{
2099 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2100 struct delayed_iput *delayed;
2101
2102 if (atomic_add_unless(&inode->i_count, -1, 1))
2103 return;
2104
2105 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2106 delayed->inode = inode;
2107
2108 spin_lock(&fs_info->delayed_iput_lock);
2109 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2110 spin_unlock(&fs_info->delayed_iput_lock);
2111}
2112
2113void btrfs_run_delayed_iputs(struct btrfs_root *root)
2114{
2115 LIST_HEAD(list);
2116 struct btrfs_fs_info *fs_info = root->fs_info;
2117 struct delayed_iput *delayed;
2118 int empty;
2119
2120 spin_lock(&fs_info->delayed_iput_lock);
2121 empty = list_empty(&fs_info->delayed_iputs);
2122 spin_unlock(&fs_info->delayed_iput_lock);
2123 if (empty)
2124 return;
2125
24bbcf04
YZ		 2126 spin_lock(&fs_info->delayed_iput_lock);
2127 list_splice_init(&fs_info->delayed_iputs, &list);
2128 spin_unlock(&fs_info->delayed_iput_lock);
2129
2130 while (!list_empty(&list)) {
2131 delayed = list_entry(list.next, struct delayed_iput, list);
2132 list_del(&delayed->list);
2133 iput(delayed->inode);
2134 kfree(delayed);
2135 }
24bbcf04
YZ		 2136}
2137
d68fc57b
YZ		 2138enum btrfs_orphan_cleanup_state {
2139 ORPHAN_CLEANUP_STARTED = 1,
2140 ORPHAN_CLEANUP_DONE = 2,
2141};
2142
2143/*
42b2aa86 2144 * This is called in transaction commit time. If there are no orphan
d68fc57b
YZ		 2145 * files in the subvolume, it removes orphan item and frees block_rsv
2146 * structure.
2147 */
2148void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *root)
2150{
90290e19 2151 struct btrfs_block_rsv *block_rsv;
d68fc57b
YZ		 2152 int ret;
2153
8a35d95f 2154 if (atomic_read(&root->orphan_inodes) ||
d68fc57b
YZ		 2155 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2156 return;
2157
90290e19 2158 spin_lock(&root->orphan_lock);
8a35d95f 2159 if (atomic_read(&root->orphan_inodes)) {
90290e19
JB		 2160 spin_unlock(&root->orphan_lock);
2161 return;
2162 }
2163
2164 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2165 spin_unlock(&root->orphan_lock);
2166 return;
2167 }
2168
2169 block_rsv = root->orphan_block_rsv;
2170 root->orphan_block_rsv = NULL;
2171 spin_unlock(&root->orphan_lock);
2172
d68fc57b
YZ		 2173 if (root->orphan_item_inserted &&
2174 btrfs_root_refs(&root->root_item) > 0) {
2175 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2176 root->root_key.objectid);
2177 BUG_ON(ret);
2178 root->orphan_item_inserted = 0;
2179 }
2180
90290e19
JB		 2181 if (block_rsv) {
2182 WARN_ON(block_rsv->size > 0);
2183 btrfs_free_block_rsv(root, block_rsv);
d68fc57b
YZ		 2184 }
2185}
2186
7b128766
JB		 2187/*
2188 * This creates an orphan entry for the given inode in case something goes
2189 * wrong in the middle of an unlink/truncate.
d68fc57b
YZ		 2190 *
2191 * NOTE: caller of this function should reserve 5 units of metadata for
2192 * this function.
7b128766
JB		 2193 */
2194int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2195{
2196 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2197 struct btrfs_block_rsv *block_rsv = NULL;
2198 int reserve = 0;
2199 int insert = 0;
2200 int ret;
7b128766 2201
d68fc57b 2202 if (!root->orphan_block_rsv) {
66d8f3dd 2203 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
b532402e
TI		 2204 if (!block_rsv)
2205 return -ENOMEM;
d68fc57b 2206 }
7b128766 2207
d68fc57b
YZ		 2208 spin_lock(&root->orphan_lock);
2209 if (!root->orphan_block_rsv) {
2210 root->orphan_block_rsv = block_rsv;
2211 } else if (block_rsv) {
2212 btrfs_free_block_rsv(root, block_rsv);
2213 block_rsv = NULL;
7b128766 2214 }
7b128766 2215
8a35d95f
JB		 2216 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2217 &BTRFS_I(inode)->runtime_flags)) {
d68fc57b
YZ		 2218#if 0
2219 /*
2220 * For proper ENOSPC handling, we should do orphan
2221 * cleanup when mounting. But this introduces backward
2222 * compatibility issue.
2223 */
2224 if (!xchg(&root->orphan_item_inserted, 1))
2225 insert = 2;
2226 else
2227 insert = 1;
2228#endif
2229 insert = 1;
321f0e70 2230 atomic_inc(&root->orphan_inodes);
7b128766
JB		 2231 }
2232
72ac3c0d
JB		 2233 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2234 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2235 reserve = 1;
d68fc57b 2236 spin_unlock(&root->orphan_lock);
7b128766 2237
d68fc57b
YZ		 2238 /* grab metadata reservation from transaction handle */
2239 if (reserve) {
2240 ret = btrfs_orphan_reserve_metadata(trans, inode);
79787eaa 2241 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
d68fc57b 2242 }
7b128766 2243
d68fc57b
YZ		 2244 /* insert an orphan item to track this unlinked/truncated file */
2245 if (insert >= 1) {
33345d01 2246 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 2247 if (ret && ret != -EEXIST) {
8a35d95f
JB		 2248 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249 &BTRFS_I(inode)->runtime_flags);
79787eaa
JM		 2250 btrfs_abort_transaction(trans, root, ret);
2251 return ret;
2252 }
2253 ret = 0;
d68fc57b
YZ		 2254 }
2255
2256 /* insert an orphan item to track subvolume contains orphan files */
2257 if (insert >= 2) {
2258 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2259 root->root_key.objectid);
79787eaa
JM		 2260 if (ret && ret != -EEXIST) {
2261 btrfs_abort_transaction(trans, root, ret);
2262 return ret;
2263 }
d68fc57b
YZ		 2264 }
2265 return 0;
7b128766
JB		 2266}
2267
2268/*
2269 * We have done the truncate/delete so we can go ahead and remove the orphan
2270 * item for this particular inode.
2271 */
2272int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2273{
2274 struct btrfs_root *root = BTRFS_I(inode)->root;
d68fc57b
YZ		 2275 int delete_item = 0;
2276 int release_rsv = 0;
7b128766
JB		 2277 int ret = 0;
2278
d68fc57b 2279 spin_lock(&root->orphan_lock);
8a35d95f
JB		 2280 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2282 delete_item = 1;
7b128766 2283
72ac3c0d
JB		 2284 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2285 &BTRFS_I(inode)->runtime_flags))
d68fc57b 2286 release_rsv = 1;
d68fc57b 2287 spin_unlock(&root->orphan_lock);
7b128766 2288
d68fc57b 2289 if (trans && delete_item) {
33345d01 2290 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
79787eaa 2291 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
d68fc57b 2292 }
7b128766 2293
8a35d95f 2294 if (release_rsv) {
d68fc57b 2295 btrfs_orphan_release_metadata(inode);
8a35d95f
JB		 2296 atomic_dec(&root->orphan_inodes);
2297 }
7b128766 2298
d68fc57b 2299 return 0;
7b128766
JB		 2300}
2301
2302/*
2303 * this cleans up any orphans that may be left on the list from the last use
2304 * of this root.
2305 */
66b4ffd1 2306int btrfs_orphan_cleanup(struct btrfs_root *root)
7b128766
JB		 2307{
2308 struct btrfs_path *path;
2309 struct extent_buffer *leaf;
7b128766
JB		 2310 struct btrfs_key key, found_key;
2311 struct btrfs_trans_handle *trans;
2312 struct inode *inode;
8f6d7f4f 2313 u64 last_objectid = 0;
7b128766
JB		 2314 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2315
d68fc57b 2316 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
66b4ffd1 2317 return 0;
c71bf099
YZ		 2318
2319 path = btrfs_alloc_path();
66b4ffd1
JB		 2320 if (!path) {
2321 ret = -ENOMEM;
2322 goto out;
2323 }
7b128766
JB		 2324 path->reada = -1;
2325
2326 key.objectid = BTRFS_ORPHAN_OBJECTID;
2327 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2328 key.offset = (u64)-1;
2329
7b128766
JB		 2330 while (1) {
2331 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
66b4ffd1
JB		 2332 if (ret < 0)
2333 goto out;
7b128766
JB		 2334
2335 /*
2336 * if ret == 0 means we found what we were searching for, which
25985edc 2337 * is weird, but possible, so only screw with path if we didn't
7b128766
JB		 2338 * find the key and see if we have stuff that matches
2339 */
2340 if (ret > 0) {
66b4ffd1 2341 ret = 0;
7b128766
JB		 2342 if (path->slots[0] == 0)
2343 break;
2344 path->slots[0]--;
2345 }
2346
2347 /* pull out the item */
2348 leaf = path->nodes[0];
7b128766
JB		 2349 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2350
2351 /* make sure the item matches what we want */
2352 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2353 break;
2354 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2355 break;
2356
2357 /* release the path since we're done with it */
b3b4aa74 2358 btrfs_release_path(path);
7b128766
JB		 2359
2360 /*
2361 * this is where we are basically btrfs_lookup, without the
2362 * crossing root thing. we store the inode number in the
2363 * offset of the orphan item.
2364 */
8f6d7f4f
JB		 2365
2366 if (found_key.offset == last_objectid) {
2367 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2368 "stopping orphan cleanup\n");
2369 ret = -EINVAL;
2370 goto out;
2371 }
2372
2373 last_objectid = found_key.offset;
2374
5d4f98a2
YZ		 2375 found_key.objectid = found_key.offset;
2376 found_key.type = BTRFS_INODE_ITEM_KEY;
2377 found_key.offset = 0;
73f73415 2378 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
a8c9e576
JB		 2379 ret = PTR_RET(inode);
2380 if (ret && ret != -ESTALE)
66b4ffd1 2381 goto out;
7b128766 2382
f8e9e0b0
AJ		 2383 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2384 struct btrfs_root *dead_root;
2385 struct btrfs_fs_info *fs_info = root->fs_info;
2386 int is_dead_root = 0;
2387
2388 /*
2389 * this is an orphan in the tree root. Currently these
2390 * could come from 2 sources:
2391 * a) a snapshot deletion in progress
2392 * b) a free space cache inode
2393 * We need to distinguish those two, as the snapshot
2394 * orphan must not get deleted.
2395 * find_dead_roots already ran before us, so if this
2396 * is a snapshot deletion, we should find the root
2397 * in the dead_roots list
2398 */
2399 spin_lock(&fs_info->trans_lock);
2400 list_for_each_entry(dead_root, &fs_info->dead_roots,
2401 root_list) {
2402 if (dead_root->root_key.objectid ==
2403 found_key.objectid) {
2404 is_dead_root = 1;
2405 break;
2406 }
2407 }
2408 spin_unlock(&fs_info->trans_lock);
2409 if (is_dead_root) {
2410 /* prevent this orphan from being found again */
2411 key.offset = found_key.objectid - 1;
2412 continue;
2413 }
2414 }
7b128766 2415 /*
a8c9e576
JB		 2416 * Inode is already gone but the orphan item is still there,
2417 * kill the orphan item.
7b128766 2418 */
a8c9e576
JB		 2419 if (ret == -ESTALE) {
2420 trans = btrfs_start_transaction(root, 1);
66b4ffd1
JB		 2421 if (IS_ERR(trans)) {
2422 ret = PTR_ERR(trans);
2423 goto out;
2424 }
8a35d95f
JB		 2425 printk(KERN_ERR "auto deleting %Lu\n",
2426 found_key.objectid);
a8c9e576
JB		 2427 ret = btrfs_del_orphan_item(trans, root,
2428 found_key.objectid);
79787eaa 2429 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
5b21f2ed 2430 btrfs_end_transaction(trans, root);
7b128766
JB		 2431 continue;
2432 }
2433
a8c9e576
JB		 2434 /*
2435 * add this inode to the orphan list so btrfs_orphan_del does
2436 * the proper thing when we hit it
2437 */
8a35d95f
JB		 2438 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2439 &BTRFS_I(inode)->runtime_flags);
a8c9e576 2440
7b128766
JB		 2441 /* if we have links, this was a truncate, lets do that */
2442 if (inode->i_nlink) {
a41ad394
JB		 2443 if (!S_ISREG(inode->i_mode)) {
2444 WARN_ON(1);
2445 iput(inode);
2446 continue;
2447 }
7b128766 2448 nr_truncate++;
66b4ffd1 2449 ret = btrfs_truncate(inode);
7b128766
JB		 2450 } else {
2451 nr_unlink++;
2452 }
2453
2454 /* this will do delete_inode and everything for us */
2455 iput(inode);
66b4ffd1
JB		 2456 if (ret)
2457 goto out;
7b128766 2458 }
3254c876
MX		 2459 /* release the path since we're done with it */
2460 btrfs_release_path(path);
2461
d68fc57b
YZ		 2462 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2463
2464 if (root->orphan_block_rsv)
2465 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2466 (u64)-1);
2467
2468 if (root->orphan_block_rsv || root->orphan_item_inserted) {
7a7eaa40 2469 trans = btrfs_join_transaction(root);
66b4ffd1
JB		 2470 if (!IS_ERR(trans))
2471 btrfs_end_transaction(trans, root);
d68fc57b 2472 }
7b128766
JB		 2473
2474 if (nr_unlink)
2475 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2476 if (nr_truncate)
2477 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
66b4ffd1
JB		 2478
2479out:
2480 if (ret)
2481 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2482 btrfs_free_path(path);
2483 return ret;
7b128766
JB		 2484}
2485
46a53cca
CM		 2486/*
2487 * very simple check to peek ahead in the leaf looking for xattrs. If we
2488 * don't find any xattrs, we know there can't be any acls.
2489 *
2490 * slot is the slot the inode is in, objectid is the objectid of the inode
2491 */
2492static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2493 int slot, u64 objectid)
2494{
2495 u32 nritems = btrfs_header_nritems(leaf);
2496 struct btrfs_key found_key;
2497 int scanned = 0;
2498
2499 slot++;
2500 while (slot < nritems) {
2501 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2502
2503 /* we found a different objectid, there must not be acls */
2504 if (found_key.objectid != objectid)
2505 return 0;
2506
2507 /* we found an xattr, assume we've got an acl */
2508 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2509 return 1;
2510
2511 /*
2512 * we found a key greater than an xattr key, there can't
2513 * be any acls later on
2514 */
2515 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2516 return 0;
2517
2518 slot++;
2519 scanned++;
2520
2521 /*
2522 * it goes inode, inode backrefs, xattrs, extents,
2523 * so if there are a ton of hard links to an inode there can
2524 * be a lot of backrefs. Don't waste time searching too hard,
2525 * this is just an optimization
2526 */
2527 if (scanned >= 8)
2528 break;
2529 }
2530 /* we hit the end of the leaf before we found an xattr or
2531 * something larger than an xattr. We have to assume the inode
2532 * has acls
2533 */
2534 return 1;
2535}
2536
d352ac68
CM		 2537/*
2538 * read an inode from the btree into the in-memory inode
2539 */
5d4f98a2 2540static void btrfs_read_locked_inode(struct inode *inode)
39279cc3
CM		 2541{
2542 struct btrfs_path *path;
5f39d397 2543 struct extent_buffer *leaf;
39279cc3 2544 struct btrfs_inode_item *inode_item;
0b86a832 2545 struct btrfs_timespec *tspec;
39279cc3
CM		 2546 struct btrfs_root *root = BTRFS_I(inode)->root;
2547 struct btrfs_key location;
46a53cca 2548 int maybe_acls;
618e21d5 2549 u32 rdev;
39279cc3 2550 int ret;
2f7e33d4
MX		 2551 bool filled = false;
2552
2553 ret = btrfs_fill_inode(inode, &rdev);
2554 if (!ret)
2555 filled = true;
39279cc3
CM		 2556
2557 path = btrfs_alloc_path();
1748f843
MF		 2558 if (!path)
2559 goto make_bad;
2560
d90c7321 2561 path->leave_spinning = 1;
39279cc3 2562 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
dc17ff8f 2563
39279cc3 2564 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
5f39d397 2565 if (ret)
39279cc3 2566 goto make_bad;
39279cc3 2567
5f39d397 2568 leaf = path->nodes[0];
2f7e33d4
MX		 2569
2570 if (filled)
2571 goto cache_acl;
2572
5f39d397
CM		 2573 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2574 struct btrfs_inode_item);
5f39d397 2575 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
bfe86848 2576 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2f2f43d3
EB		 2577 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2578 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
dbe674a9 2579 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
5f39d397
CM		 2580
2581 tspec = btrfs_inode_atime(inode_item);
2582 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2584
2585 tspec = btrfs_inode_mtime(inode_item);
2586 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2588
2589 tspec = btrfs_inode_ctime(inode_item);
2590 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2591 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2592
a76a3cd4 2593 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
e02119d5 2594 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
5dc562c5
JB		 2595 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2596
2597 /*
2598 * If we were modified in the current generation and evicted from memory
2599 * and then re-read we need to do a full sync since we don't have any
2600 * idea about which extents were modified before we were evicted from
2601 * cache.
2602 */
2603 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2604 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2605 &BTRFS_I(inode)->runtime_flags);
2606
0c4d2d95 2607 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
e02119d5 2608 inode->i_generation = BTRFS_I(inode)->generation;
618e21d5 2609 inode->i_rdev = 0;
5f39d397
CM		 2610 rdev = btrfs_inode_rdev(leaf, inode_item);
2611
aec7477b 2612 BTRFS_I(inode)->index_cnt = (u64)-1;
d2fb3437 2613 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2f7e33d4 2614cache_acl:
46a53cca
CM		 2615 /*
2616 * try to precache a NULL acl entry for files that don't have
2617 * any xattrs or acls
2618 */
33345d01
LZ		 2619 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2620 btrfs_ino(inode));
72c04902
AV		 2621 if (!maybe_acls)
2622 cache_no_acl(inode);
46a53cca 2623
39279cc3 2624 btrfs_free_path(path);
39279cc3 2625
39279cc3 2626 switch (inode->i_mode & S_IFMT) {
39279cc3
CM		 2627 case S_IFREG:
2628 inode->i_mapping->a_ops = &btrfs_aops;
04160088 2629 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 2630 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3
CM		 2631 inode->i_fop = &btrfs_file_operations;
2632 inode->i_op = &btrfs_file_inode_operations;
2633 break;
2634 case S_IFDIR:
2635 inode->i_fop = &btrfs_dir_file_operations;
2636 if (root == root->fs_info->tree_root)
2637 inode->i_op = &btrfs_dir_ro_inode_operations;
2638 else
2639 inode->i_op = &btrfs_dir_inode_operations;
2640 break;
2641 case S_IFLNK:
2642 inode->i_op = &btrfs_symlink_inode_operations;
2643 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 2644 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
39279cc3 2645 break;
618e21d5 2646 default:
0279b4cd 2647 inode->i_op = &btrfs_special_inode_operations;
618e21d5
JB		 2648 init_special_inode(inode, inode->i_mode, rdev);
2649 break;
39279cc3 2650 }
6cbff00f
CH		 2651
2652 btrfs_update_iflags(inode);
39279cc3
CM		 2653 return;
2654
2655make_bad:
39279cc3 2656 btrfs_free_path(path);
39279cc3
CM		 2657 make_bad_inode(inode);
2658}
2659
d352ac68
CM		 2660/*
2661 * given a leaf and an inode, copy the inode fields into the leaf
2662 */
e02119d5
CM		 2663static void fill_inode_item(struct btrfs_trans_handle *trans,
2664 struct extent_buffer *leaf,
5f39d397 2665 struct btrfs_inode_item *item,
39279cc3
CM		 2666 struct inode *inode)
2667{
2f2f43d3
EB		 2668 btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2669 btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
dbe674a9 2670 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
5f39d397
CM		 2671 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2672 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2673
2674 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2675 inode->i_atime.tv_sec);
2676 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2677 inode->i_atime.tv_nsec);
2678
2679 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2680 inode->i_mtime.tv_sec);
2681 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2682 inode->i_mtime.tv_nsec);
2683
2684 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2685 inode->i_ctime.tv_sec);
2686 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2687 inode->i_ctime.tv_nsec);
2688
a76a3cd4 2689 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
e02119d5 2690 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
0c4d2d95 2691 btrfs_set_inode_sequence(leaf, item, inode->i_version);
e02119d5 2692 btrfs_set_inode_transid(leaf, item, trans->transid);
5f39d397 2693 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
b98b6767 2694 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
d82a6f1d 2695 btrfs_set_inode_block_group(leaf, item, 0);
39279cc3
CM		 2696}
2697
d352ac68
CM		 2698/*
2699 * copy everything in the in-memory inode into the btree.
2700 */
2115133f 2701static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
d397712b 2702 struct btrfs_root *root, struct inode *inode)
39279cc3
CM		 2703{
2704 struct btrfs_inode_item *inode_item;
2705 struct btrfs_path *path;
5f39d397 2706 struct extent_buffer *leaf;
39279cc3
CM		 2707 int ret;
2708
2709 path = btrfs_alloc_path();
16cdcec7
MX		 2710 if (!path)
2711 return -ENOMEM;
2712
b9473439 2713 path->leave_spinning = 1;
16cdcec7
MX		 2714 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2715 1);
39279cc3
CM		 2716 if (ret) {
2717 if (ret > 0)
2718 ret = -ENOENT;
2719 goto failed;
2720 }
2721
b4ce94de 2722 btrfs_unlock_up_safe(path, 1);
5f39d397
CM		 2723 leaf = path->nodes[0];
2724 inode_item = btrfs_item_ptr(leaf, path->slots[0],
16cdcec7 2725 struct btrfs_inode_item);
39279cc3 2726
e02119d5 2727 fill_inode_item(trans, leaf, inode_item, inode);
5f39d397 2728 btrfs_mark_buffer_dirty(leaf);
15ee9bc7 2729 btrfs_set_inode_last_trans(trans, inode);
39279cc3
CM		 2730 ret = 0;
2731failed:
39279cc3
CM		 2732 btrfs_free_path(path);
2733 return ret;
2734}
2735
2115133f
CM		 2736/*
2737 * copy everything in the in-memory inode into the btree.
2738 */
2739noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2740 struct btrfs_root *root, struct inode *inode)
2741{
2742 int ret;
2743
2744 /*
2745 * If the inode is a free space inode, we can deadlock during commit
2746 * if we put it into the delayed code.
2747 *
2748 * The data relocation inode should also be directly updated
2749 * without delay
2750 */
83eea1f1 2751 if (!btrfs_is_free_space_inode(inode)
2115133f 2752 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
8ea05e3a
AB		 2753 btrfs_update_root_times(trans, root);
2754
2115133f
CM		 2755 ret = btrfs_delayed_update_inode(trans, root, inode);
2756 if (!ret)
2757 btrfs_set_inode_last_trans(trans, inode);
2758 return ret;
2759 }
2760
2761 return btrfs_update_inode_item(trans, root, inode);
2762}
2763
be6aef60
JB		 2764noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 struct inode *inode)
2115133f
CM		 2767{
2768 int ret;
2769
2770 ret = btrfs_update_inode(trans, root, inode);
2771 if (ret == -ENOSPC)
2772 return btrfs_update_inode_item(trans, root, inode);
2773 return ret;
2774}
2775
d352ac68
CM		 2776/*
2777 * unlink helper that gets used here in inode.c and in the tree logging
2778 * recovery code. It remove a link in a directory with a given name, and
2779 * also drops the back refs in the inode to the directory
2780 */
92986796
AV		 2781static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2782 struct btrfs_root *root,
2783 struct inode *dir, struct inode *inode,
2784 const char *name, int name_len)
39279cc3
CM		 2785{
2786 struct btrfs_path *path;
39279cc3 2787 int ret = 0;
5f39d397 2788 struct extent_buffer *leaf;
39279cc3 2789 struct btrfs_dir_item *di;
5f39d397 2790 struct btrfs_key key;
aec7477b 2791 u64 index;
33345d01
LZ		 2792 u64 ino = btrfs_ino(inode);
2793 u64 dir_ino = btrfs_ino(dir);
39279cc3
CM		 2794
2795 path = btrfs_alloc_path();
54aa1f4d
CM		 2796 if (!path) {
2797 ret = -ENOMEM;
554233a6 2798 goto out;
54aa1f4d
CM		 2799 }
2800
b9473439 2801 path->leave_spinning = 1;
33345d01 2802 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
39279cc3
CM		 2803 name, name_len, -1);
2804 if (IS_ERR(di)) {
2805 ret = PTR_ERR(di);
2806 goto err;
2807 }
2808 if (!di) {
2809 ret = -ENOENT;
2810 goto err;
2811 }
5f39d397
CM		 2812 leaf = path->nodes[0];
2813 btrfs_dir_item_key_to_cpu(leaf, di, &key);
39279cc3 2814 ret = btrfs_delete_one_dir_name(trans, root, path, di);
54aa1f4d
CM		 2815 if (ret)
2816 goto err;
b3b4aa74 2817 btrfs_release_path(path);
39279cc3 2818
33345d01
LZ		 2819 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2820 dir_ino, &index);
aec7477b 2821 if (ret) {
d397712b 2822 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
33345d01
LZ		 2823 "inode %llu parent %llu\n", name_len, name,
2824 (unsigned long long)ino, (unsigned long long)dir_ino);
79787eaa 2825 btrfs_abort_transaction(trans, root, ret);
aec7477b
JB		 2826 goto err;
2827 }
2828
16cdcec7 2829 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 2830 if (ret) {
2831 btrfs_abort_transaction(trans, root, ret);
39279cc3 2832 goto err;
79787eaa 2833 }
39279cc3 2834
e02119d5 2835 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
33345d01 2836 inode, dir_ino);
79787eaa
JM		 2837 if (ret != 0 && ret != -ENOENT) {
2838 btrfs_abort_transaction(trans, root, ret);
2839 goto err;
2840 }
e02119d5
CM		 2841
2842 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2843 dir, index);
6418c961
CM		 2844 if (ret == -ENOENT)
2845 ret = 0;
39279cc3
CM		 2846err:
2847 btrfs_free_path(path);
e02119d5
CM		 2848 if (ret)
2849 goto out;
2850
2851 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95
JB		 2852 inode_inc_iversion(inode);
2853 inode_inc_iversion(dir);
e02119d5 2854 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
b9959295 2855 ret = btrfs_update_inode(trans, root, dir);
e02119d5 2856out:
39279cc3
CM		 2857 return ret;
2858}
2859
92986796
AV		 2860int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2861 struct btrfs_root *root,
2862 struct inode *dir, struct inode *inode,
2863 const char *name, int name_len)
2864{
2865 int ret;
2866 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2867 if (!ret) {
2868 btrfs_drop_nlink(inode);
2869 ret = btrfs_update_inode(trans, root, inode);
2870 }
2871 return ret;
2872}
2873
2874
a22285a6
YZ		 2875/* helper to check if there is any shared block in the path */
2876static int check_path_shared(struct btrfs_root *root,
2877 struct btrfs_path *path)
39279cc3 2878{
a22285a6
YZ		 2879 struct extent_buffer *eb;
2880 int level;
0e4dcbef 2881 u64 refs = 1;
5df6a9f6 2882
a22285a6 2883 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
dedefd72
JB		 2884 int ret;
2885
a22285a6
YZ		 2886 if (!path->nodes[level])
2887 break;
2888 eb = path->nodes[level];
2889 if (!btrfs_block_can_be_shared(root, eb))
2890 continue;
2891 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2892 &refs, NULL);
2893 if (refs > 1)
2894 return 1;
5df6a9f6 2895 }
dedefd72 2896 return 0;
39279cc3
CM		 2897}
2898
a22285a6
YZ		 2899/*
2900 * helper to start transaction for unlink and rmdir.
2901 *
2902 * unlink and rmdir are special in btrfs, they do not always free space.
2903 * so in enospc case, we should make sure they will free space before
2904 * allowing them to use the global metadata reservation.
2905 */
2906static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2907 struct dentry *dentry)
4df27c4d 2908{
39279cc3 2909 struct btrfs_trans_handle *trans;
a22285a6 2910 struct btrfs_root *root = BTRFS_I(dir)->root;
4df27c4d 2911 struct btrfs_path *path;
4df27c4d 2912 struct btrfs_dir_item *di;
7b128766 2913 struct inode *inode = dentry->d_inode;
4df27c4d 2914 u64 index;
a22285a6
YZ		 2915 int check_link = 1;
2916 int err = -ENOSPC;
4df27c4d 2917 int ret;
33345d01
LZ		 2918 u64 ino = btrfs_ino(inode);
2919 u64 dir_ino = btrfs_ino(dir);
4df27c4d 2920
e70bea5f
JB		 2921 /*
2922 * 1 for the possible orphan item
2923 * 1 for the dir item
2924 * 1 for the dir index
2925 * 1 for the inode ref
2926 * 1 for the inode ref in the tree log
2927 * 2 for the dir entries in the log
2928 * 1 for the inode
2929 */
2930 trans = btrfs_start_transaction(root, 8);
a22285a6
YZ		 2931 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2932 return trans;
4df27c4d 2933
33345d01 2934 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
a22285a6 2935 return ERR_PTR(-ENOSPC);
4df27c4d 2936
a22285a6
YZ		 2937 /* check if there is someone else holds reference */
2938 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2939 return ERR_PTR(-ENOSPC);
4df27c4d 2940
a22285a6
YZ		 2941 if (atomic_read(&inode->i_count) > 2)
2942 return ERR_PTR(-ENOSPC);
4df27c4d 2943
a22285a6
YZ		 2944 if (xchg(&root->fs_info->enospc_unlink, 1))
2945 return ERR_PTR(-ENOSPC);
2946
2947 path = btrfs_alloc_path();
2948 if (!path) {
2949 root->fs_info->enospc_unlink = 0;
2950 return ERR_PTR(-ENOMEM);
4df27c4d
YZ		 2951 }
2952
3880a1b4
JB		 2953 /* 1 for the orphan item */
2954 trans = btrfs_start_transaction(root, 1);
5df6a9f6 2955 if (IS_ERR(trans)) {
a22285a6
YZ		 2956 btrfs_free_path(path);
2957 root->fs_info->enospc_unlink = 0;
2958 return trans;
2959 }
4df27c4d 2960
a22285a6
YZ		 2961 path->skip_locking = 1;
2962 path->search_commit_root = 1;
4df27c4d 2963
a22285a6
YZ		 2964 ret = btrfs_lookup_inode(trans, root, path,
2965 &BTRFS_I(dir)->location, 0);
2966 if (ret < 0) {
2967 err = ret;
2968 goto out;
2969 }
2970 if (ret == 0) {
2971 if (check_path_shared(root, path))
2972 goto out;
2973 } else {
2974 check_link = 0;
5df6a9f6 2975 }
b3b4aa74 2976 btrfs_release_path(path);
a22285a6
YZ		 2977
2978 ret = btrfs_lookup_inode(trans, root, path,
2979 &BTRFS_I(inode)->location, 0);
2980 if (ret < 0) {
2981 err = ret;
2982 goto out;
2983 }
2984 if (ret == 0) {
2985 if (check_path_shared(root, path))
2986 goto out;
2987 } else {
2988 check_link = 0;
2989 }
b3b4aa74 2990 btrfs_release_path(path);
a22285a6
YZ		 2991
2992 if (ret == 0 && S_ISREG(inode->i_mode)) {
2993 ret = btrfs_lookup_file_extent(trans, root, path,
33345d01 2994 ino, (u64)-1, 0);
a22285a6
YZ		 2995 if (ret < 0) {
2996 err = ret;
2997 goto out;
2998 }
79787eaa 2999 BUG_ON(ret == 0); /* Corruption */
a22285a6
YZ		 3000 if (check_path_shared(root, path))
3001 goto out;
b3b4aa74 3002 btrfs_release_path(path);
a22285a6
YZ		 3003 }
3004
3005 if (!check_link) {
3006 err = 0;
3007 goto out;
3008 }
3009
33345d01 3010 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
a22285a6
YZ		 3011 dentry->d_name.name, dentry->d_name.len, 0);
3012 if (IS_ERR(di)) {
3013 err = PTR_ERR(di);
3014 goto out;
3015 }
3016 if (di) {
3017 if (check_path_shared(root, path))
3018 goto out;
3019 } else {
3020 err = 0;
3021 goto out;
3022 }
b3b4aa74 3023 btrfs_release_path(path);
a22285a6 3024
f186373f
MF		 3025 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3026 dentry->d_name.len, ino, dir_ino, 0,
3027 &index);
3028 if (ret) {
3029 err = ret;
a22285a6
YZ		 3030 goto out;
3031 }
f186373f 3032
a22285a6
YZ		 3033 if (check_path_shared(root, path))
3034 goto out;
f186373f 3035
b3b4aa74 3036 btrfs_release_path(path);
a22285a6 3037
16cdcec7
MX		 3038 /*
3039 * This is a commit root search, if we can lookup inode item and other
3040 * relative items in the commit root, it means the transaction of
3041 * dir/file creation has been committed, and the dir index item that we
3042 * delay to insert has also been inserted into the commit root. So
3043 * we needn't worry about the delayed insertion of the dir index item
3044 * here.
3045 */
33345d01 3046 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
a22285a6
YZ		 3047 dentry->d_name.name, dentry->d_name.len, 0);
3048 if (IS_ERR(di)) {
3049 err = PTR_ERR(di);
3050 goto out;
3051 }
3052 BUG_ON(ret == -ENOENT);
3053 if (check_path_shared(root, path))
3054 goto out;
3055
3056 err = 0;
3057out:
3058 btrfs_free_path(path);
3880a1b4
JB		 3059 /* Migrate the orphan reservation over */
3060 if (!err)
3061 err = btrfs_block_rsv_migrate(trans->block_rsv,
3062 &root->fs_info->global_block_rsv,
5a77d76c 3063 trans->bytes_reserved);
3880a1b4 3064
a22285a6
YZ		 3065 if (err) {
3066 btrfs_end_transaction(trans, root);
3067 root->fs_info->enospc_unlink = 0;
3068 return ERR_PTR(err);
3069 }
3070
3071 trans->block_rsv = &root->fs_info->global_block_rsv;
3072 return trans;
3073}
3074
3075static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3076 struct btrfs_root *root)
3077{
66d8f3dd 3078 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
5a77d76c
JB		 3079 btrfs_block_rsv_release(root, trans->block_rsv,
3080 trans->bytes_reserved);
3081 trans->block_rsv = &root->fs_info->trans_block_rsv;
a22285a6
YZ		 3082 BUG_ON(!root->fs_info->enospc_unlink);
3083 root->fs_info->enospc_unlink = 0;
3084 }
7ad85bb7 3085 btrfs_end_transaction(trans, root);
a22285a6
YZ		 3086}
3087
3088static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3089{
3090 struct btrfs_root *root = BTRFS_I(dir)->root;
3091 struct btrfs_trans_handle *trans;
3092 struct inode *inode = dentry->d_inode;
3093 int ret;
3094 unsigned long nr = 0;
3095
3096 trans = __unlink_start_trans(dir, dentry);
3097 if (IS_ERR(trans))
3098 return PTR_ERR(trans);
5f39d397 3099
12fcfd22
CM		 3100 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3101
e02119d5
CM		 3102 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3103 dentry->d_name.name, dentry->d_name.len);
b532402e
TI		 3104 if (ret)
3105 goto out;
7b128766 3106
a22285a6 3107 if (inode->i_nlink == 0) {
7b128766 3108 ret = btrfs_orphan_add(trans, inode);
b532402e
TI		 3109 if (ret)
3110 goto out;
a22285a6 3111 }
7b128766 3112
b532402e 3113out:
d3c2fdcf 3114 nr = trans->blocks_used;
a22285a6 3115 __unlink_end_trans(trans, root);
d3c2fdcf 3116 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 3117 return ret;
3118}
3119
4df27c4d
YZ		 3120int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3121 struct btrfs_root *root,
3122 struct inode *dir, u64 objectid,
3123 const char *name, int name_len)
3124{
3125 struct btrfs_path *path;
3126 struct extent_buffer *leaf;
3127 struct btrfs_dir_item *di;
3128 struct btrfs_key key;
3129 u64 index;
3130 int ret;
33345d01 3131 u64 dir_ino = btrfs_ino(dir);
4df27c4d
YZ		 3132
3133 path = btrfs_alloc_path();
3134 if (!path)
3135 return -ENOMEM;
3136
33345d01 3137 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4df27c4d 3138 name, name_len, -1);
79787eaa
JM		 3139 if (IS_ERR_OR_NULL(di)) {
3140 if (!di)
3141 ret = -ENOENT;
3142 else
3143 ret = PTR_ERR(di);
3144 goto out;
3145 }
4df27c4d
YZ		 3146
3147 leaf = path->nodes[0];
3148 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3149 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3150 ret = btrfs_delete_one_dir_name(trans, root, path, di);
79787eaa
JM		 3151 if (ret) {
3152 btrfs_abort_transaction(trans, root, ret);
3153 goto out;
3154 }
b3b4aa74 3155 btrfs_release_path(path);
4df27c4d
YZ		 3156
3157 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3158 objectid, root->root_key.objectid,
33345d01 3159 dir_ino, &index, name, name_len);
4df27c4d 3160 if (ret < 0) {
79787eaa
JM		 3161 if (ret != -ENOENT) {
3162 btrfs_abort_transaction(trans, root, ret);
3163 goto out;
3164 }
33345d01 3165 di = btrfs_search_dir_index_item(root, path, dir_ino,
4df27c4d 3166 name, name_len);
79787eaa
JM		 3167 if (IS_ERR_OR_NULL(di)) {
3168 if (!di)
3169 ret = -ENOENT;
3170 else
3171 ret = PTR_ERR(di);
3172 btrfs_abort_transaction(trans, root, ret);
3173 goto out;
3174 }
4df27c4d
YZ		 3175
3176 leaf = path->nodes[0];
3177 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 3178 btrfs_release_path(path);
4df27c4d
YZ		 3179 index = key.offset;
3180 }
945d8962 3181 btrfs_release_path(path);
4df27c4d 3182
16cdcec7 3183 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
79787eaa
JM		 3184 if (ret) {
3185 btrfs_abort_transaction(trans, root, ret);
3186 goto out;
3187 }
4df27c4d
YZ		 3188
3189 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
0c4d2d95 3190 inode_inc_iversion(dir);
4df27c4d 3191 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
5a24e84c 3192 ret = btrfs_update_inode_fallback(trans, root, dir);
79787eaa
JM		 3193 if (ret)
3194 btrfs_abort_transaction(trans, root, ret);
3195out:
71d7aed0 3196 btrfs_free_path(path);
79787eaa 3197 return ret;
4df27c4d
YZ		 3198}
3199
39279cc3
CM		 3200static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3201{
3202 struct inode *inode = dentry->d_inode;
1832a6d5 3203 int err = 0;
39279cc3 3204 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3 3205 struct btrfs_trans_handle *trans;
1832a6d5 3206 unsigned long nr = 0;
39279cc3 3207
b3ae244e 3208 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
134d4512 3209 return -ENOTEMPTY;
b3ae244e
DS		 3210 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3211 return -EPERM;
134d4512 3212
a22285a6
YZ		 3213 trans = __unlink_start_trans(dir, dentry);
3214 if (IS_ERR(trans))
5df6a9f6 3215 return PTR_ERR(trans);
5df6a9f6 3216
33345d01 3217 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4df27c4d
YZ		 3218 err = btrfs_unlink_subvol(trans, root, dir,
3219 BTRFS_I(inode)->location.objectid,
3220 dentry->d_name.name,
3221 dentry->d_name.len);
3222 goto out;
3223 }
3224
7b128766
JB		 3225 err = btrfs_orphan_add(trans, inode);
3226 if (err)
4df27c4d 3227 goto out;
7b128766 3228
39279cc3 3229 /* now the directory is empty */
e02119d5
CM		 3230 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3231 dentry->d_name.name, dentry->d_name.len);
d397712b 3232 if (!err)
dbe674a9 3233 btrfs_i_size_write(inode, 0);
4df27c4d 3234out:
d3c2fdcf 3235 nr = trans->blocks_used;
a22285a6 3236 __unlink_end_trans(trans, root);
d3c2fdcf 3237 btrfs_btree_balance_dirty(root, nr);
3954401f 3238
39279cc3
CM		 3239 return err;
3240}
3241
39279cc3
CM		 3242/*
3243 * this can truncate away extent items, csum items and directory items.
3244 * It starts at a high offset and removes keys until it can't find
d352ac68 3245 * any higher than new_size
39279cc3
CM		 3246 *
3247 * csum items that cross the new i_size are truncated to the new size
3248 * as well.
7b128766
JB		 3249 *
3250 * min_type is the minimum key type to truncate down to. If set to 0, this
3251 * will kill all the items on this inode, including the INODE_ITEM_KEY.
39279cc3 3252 */
8082510e
YZ		 3253int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3254 struct btrfs_root *root,
3255 struct inode *inode,
3256 u64 new_size, u32 min_type)
39279cc3 3257{
39279cc3 3258 struct btrfs_path *path;
5f39d397 3259 struct extent_buffer *leaf;
39279cc3 3260 struct btrfs_file_extent_item *fi;
8082510e
YZ		 3261 struct btrfs_key key;
3262 struct btrfs_key found_key;
39279cc3 3263 u64 extent_start = 0;
db94535d 3264 u64 extent_num_bytes = 0;
5d4f98a2 3265 u64 extent_offset = 0;
39279cc3 3266 u64 item_end = 0;
8082510e
YZ		 3267 u64 mask = root->sectorsize - 1;
3268 u32 found_type = (u8)-1;
39279cc3
CM		 3269 int found_extent;
3270 int del_item;
85e21bac
CM		 3271 int pending_del_nr = 0;
3272 int pending_del_slot = 0;
179e29e4 3273 int extent_type = -1;
8082510e
YZ		 3274 int ret;
3275 int err = 0;
33345d01 3276 u64 ino = btrfs_ino(inode);
8082510e
YZ		 3277
3278 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
39279cc3 3279
0eb0e19c
MF		 3280 path = btrfs_alloc_path();
3281 if (!path)
3282 return -ENOMEM;
3283 path->reada = -1;
3284
5dc562c5
JB		 3285 /*
3286 * We want to drop from the next block forward in case this new size is
3287 * not block aligned since we will be keeping the last block of the
3288 * extent just the way it is.
3289 */
0af3d00b 3290 if (root->ref_cows || root == root->fs_info->tree_root)
5dc562c5 3291 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
8082510e 3292
16cdcec7
MX		 3293 /*
3294 * This function is also used to drop the items in the log tree before
3295 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3296 * it is used to drop the loged items. So we shouldn't kill the delayed
3297 * items.
3298 */
3299 if (min_type == 0 && root == BTRFS_I(inode)->root)
3300 btrfs_kill_delayed_inode_items(inode);
3301
33345d01 3302 key.objectid = ino;
39279cc3 3303 key.offset = (u64)-1;
5f39d397
CM		 3304 key.type = (u8)-1;
3305
85e21bac 3306search_again:
b9473439 3307 path->leave_spinning = 1;
85e21bac 3308 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8082510e
YZ		 3309 if (ret < 0) {
3310 err = ret;
3311 goto out;
3312 }
d397712b 3313
85e21bac 3314 if (ret > 0) {
e02119d5
CM		 3315 /* there are no items in the tree for us to truncate, we're
3316 * done
3317 */
8082510e
YZ		 3318 if (path->slots[0] == 0)
3319 goto out;
85e21bac
CM		 3320 path->slots[0]--;
3321 }
3322
d397712b 3323 while (1) {
39279cc3 3324 fi = NULL;
5f39d397
CM		 3325 leaf = path->nodes[0];
3326 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3327 found_type = btrfs_key_type(&found_key);
39279cc3 3328
33345d01 3329 if (found_key.objectid != ino)
39279cc3 3330 break;
5f39d397 3331
85e21bac 3332 if (found_type < min_type)
39279cc3
CM		 3333 break;
3334
5f39d397 3335 item_end = found_key.offset;
39279cc3 3336 if (found_type == BTRFS_EXTENT_DATA_KEY) {
5f39d397 3337 fi = btrfs_item_ptr(leaf, path->slots[0],
39279cc3 3338 struct btrfs_file_extent_item);
179e29e4
CM		 3339 extent_type = btrfs_file_extent_type(leaf, fi);
3340 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
5f39d397 3341 item_end +=
db94535d 3342 btrfs_file_extent_num_bytes(leaf, fi);
179e29e4 3343 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
179e29e4 3344 item_end += btrfs_file_extent_inline_len(leaf,
c8b97818 3345 fi);
39279cc3 3346 }
008630c1 3347 item_end--;
39279cc3 3348 }
8082510e
YZ		 3349 if (found_type > min_type) {
3350 del_item = 1;
3351 } else {
3352 if (item_end < new_size)
b888db2b 3353 break;
8082510e
YZ		 3354 if (found_key.offset >= new_size)
3355 del_item = 1;
3356 else
3357 del_item = 0;
39279cc3 3358 }
39279cc3 3359 found_extent = 0;
39279cc3 3360 /* FIXME, shrink the extent if the ref count is only 1 */
179e29e4
CM		 3361 if (found_type != BTRFS_EXTENT_DATA_KEY)
3362 goto delete;
3363
3364 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
39279cc3 3365 u64 num_dec;
db94535d 3366 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
f70a9a6b 3367 if (!del_item) {
db94535d
CM		 3368 u64 orig_num_bytes =
3369 btrfs_file_extent_num_bytes(leaf, fi);
e02119d5 3370 extent_num_bytes = new_size -
5f39d397 3371 found_key.offset + root->sectorsize - 1;
b1632b10
Y		 3372 extent_num_bytes = extent_num_bytes &
3373 ~((u64)root->sectorsize - 1);
db94535d
CM		 3374 btrfs_set_file_extent_num_bytes(leaf, fi,
3375 extent_num_bytes);
3376 num_dec = (orig_num_bytes -
9069218d 3377 extent_num_bytes);
e02119d5 3378 if (root->ref_cows && extent_start != 0)
a76a3cd4 3379 inode_sub_bytes(inode, num_dec);
5f39d397 3380 btrfs_mark_buffer_dirty(leaf);
39279cc3 3381 } else {
db94535d
CM		 3382 extent_num_bytes =
3383 btrfs_file_extent_disk_num_bytes(leaf,
3384 fi);
5d4f98a2
YZ		 3385 extent_offset = found_key.offset -
3386 btrfs_file_extent_offset(leaf, fi);
3387
39279cc3 3388 /* FIXME blocksize != 4096 */
9069218d 3389 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
39279cc3
CM		 3390 if (extent_start != 0) {
3391 found_extent = 1;
e02119d5 3392 if (root->ref_cows)
a76a3cd4 3393 inode_sub_bytes(inode, num_dec);
e02119d5 3394 }
39279cc3 3395 }
9069218d 3396 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
c8b97818
CM		 3397 /*
3398 * we can't truncate inline items that have had
3399 * special encodings
3400 */
3401 if (!del_item &&
3402 btrfs_file_extent_compression(leaf, fi) == 0 &&
3403 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3404 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
e02119d5
CM		 3405 u32 size = new_size - found_key.offset;
3406
3407 if (root->ref_cows) {
a76a3cd4
YZ		 3408 inode_sub_bytes(inode, item_end + 1 -
3409 new_size);
e02119d5
CM		 3410 }
3411 size =
3412 btrfs_file_extent_calc_inline_size(size);
143bede5
JM		 3413 btrfs_truncate_item(trans, root, path,
3414 size, 1);
e02119d5 3415 } else if (root->ref_cows) {
a76a3cd4
YZ		 3416 inode_sub_bytes(inode, item_end + 1 -
3417 found_key.offset);
9069218d 3418 }
39279cc3 3419 }
179e29e4 3420delete:
39279cc3 3421 if (del_item) {
85e21bac
CM		 3422 if (!pending_del_nr) {
3423 /* no pending yet, add ourselves */
3424 pending_del_slot = path->slots[0];
3425 pending_del_nr = 1;
3426 } else if (pending_del_nr &&
3427 path->slots[0] + 1 == pending_del_slot) {
3428 /* hop on the pending chunk */
3429 pending_del_nr++;
3430 pending_del_slot = path->slots[0];
3431 } else {
d397712b 3432 BUG();
85e21bac 3433 }
39279cc3
CM		 3434 } else {
3435 break;
3436 }
0af3d00b
JB		 3437 if (found_extent && (root->ref_cows ||
3438 root == root->fs_info->tree_root)) {
b9473439 3439 btrfs_set_path_blocking(path);
39279cc3 3440 ret = btrfs_free_extent(trans, root, extent_start,
5d4f98a2
YZ		 3441 extent_num_bytes, 0,
3442 btrfs_header_owner(leaf),
66d7e7f0 3443 ino, extent_offset, 0);
39279cc3
CM		 3444 BUG_ON(ret);
3445 }
85e21bac 3446
8082510e
YZ		 3447 if (found_type == BTRFS_INODE_ITEM_KEY)
3448 break;
3449
3450 if (path->slots[0] == 0 ||
3451 path->slots[0] != pending_del_slot) {
8082510e
YZ		 3452 if (pending_del_nr) {
3453 ret = btrfs_del_items(trans, root, path,
3454 pending_del_slot,
3455 pending_del_nr);
79787eaa
JM		 3456 if (ret) {
3457 btrfs_abort_transaction(trans,
3458 root, ret);
3459 goto error;
3460 }
8082510e
YZ		 3461 pending_del_nr = 0;
3462 }
b3b4aa74 3463 btrfs_release_path(path);
85e21bac 3464 goto search_again;
8082510e
YZ		 3465 } else {
3466 path->slots[0]--;
85e21bac 3467 }
39279cc3 3468 }
8082510e 3469out:
85e21bac
CM		 3470 if (pending_del_nr) {
3471 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3472 pending_del_nr);
79787eaa
JM		 3473 if (ret)
3474 btrfs_abort_transaction(trans, root, ret);
85e21bac 3475 }
79787eaa 3476error:
39279cc3 3477 btrfs_free_path(path);
8082510e 3478 return err;
39279cc3
CM		 3479}
3480
3481/*
2aaa6655
JB		 3482 * btrfs_truncate_page - read, zero a chunk and write a page
3483 * @inode - inode that we're zeroing
3484 * @from - the offset to start zeroing
3485 * @len - the length to zero, 0 to zero the entire range respective to the
3486 * offset
3487 * @front - zero up to the offset instead of from the offset on
3488 *
3489 * This will find the page for the "from" offset and cow the page and zero the
3490 * part we want to zero. This is used with truncate and hole punching.
39279cc3 3491 */
2aaa6655
JB		 3492int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3493 int front)
39279cc3 3494{
2aaa6655 3495 struct address_space *mapping = inode->i_mapping;
db94535d 3496 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 3497 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3498 struct btrfs_ordered_extent *ordered;
2ac55d41 3499 struct extent_state *cached_state = NULL;
e6dcd2dc 3500 char *kaddr;
db94535d 3501 u32 blocksize = root->sectorsize;
39279cc3
CM		 3502 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3503 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3504 struct page *page;
3b16a4e3 3505 gfp_t mask = btrfs_alloc_write_mask(mapping);
39279cc3 3506 int ret = 0;
a52d9a80 3507 u64 page_start;
e6dcd2dc 3508 u64 page_end;
39279cc3 3509
2aaa6655
JB		 3510 if ((offset & (blocksize - 1)) == 0 &&
3511 (!len || ((len & (blocksize - 1)) == 0)))
39279cc3 3512 goto out;
0ca1f7ce 3513 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5d5e103a
JB		 3514 if (ret)
3515 goto out;
39279cc3
CM		 3516
3517 ret = -ENOMEM;
211c17f5 3518again:
3b16a4e3 3519 page = find_or_create_page(mapping, index, mask);
5d5e103a 3520 if (!page) {
0ca1f7ce 3521 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3 3522 goto out;
5d5e103a 3523 }
e6dcd2dc
CM		 3524
3525 page_start = page_offset(page);
3526 page_end = page_start + PAGE_CACHE_SIZE - 1;
3527
39279cc3 3528 if (!PageUptodate(page)) {
9ebefb18 3529 ret = btrfs_readpage(NULL, page);
39279cc3 3530 lock_page(page);
211c17f5
CM		 3531 if (page->mapping != mapping) {
3532 unlock_page(page);
3533 page_cache_release(page);
3534 goto again;
3535 }
39279cc3
CM		 3536 if (!PageUptodate(page)) {
3537 ret = -EIO;
89642229 3538 goto out_unlock;
39279cc3
CM		 3539 }
3540 }
211c17f5 3541 wait_on_page_writeback(page);
e6dcd2dc 3542
d0082371 3543 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 3544 set_page_extent_mapped(page);
3545
3546 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3547 if (ordered) {
2ac55d41
JB		 3548 unlock_extent_cached(io_tree, page_start, page_end,
3549 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 3550 unlock_page(page);
3551 page_cache_release(page);
eb84ae03 3552 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 3553 btrfs_put_ordered_extent(ordered);
3554 goto again;
3555 }
3556
2ac55d41 3557 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 3558 EXTENT_DIRTY | EXTENT_DELALLOC |
3559 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 3560 0, 0, &cached_state, GFP_NOFS);
5d5e103a 3561
2ac55d41
JB		 3562 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3563 &cached_state);
9ed74f2d 3564 if (ret) {
2ac55d41
JB		 3565 unlock_extent_cached(io_tree, page_start, page_end,
3566 &cached_state, GFP_NOFS);
9ed74f2d
JB		 3567 goto out_unlock;
3568 }
3569
e6dcd2dc
CM		 3570 ret = 0;
3571 if (offset != PAGE_CACHE_SIZE) {
2aaa6655
JB		 3572 if (!len)
3573 len = PAGE_CACHE_SIZE - offset;
e6dcd2dc 3574 kaddr = kmap(page);
2aaa6655
JB		 3575 if (front)
3576 memset(kaddr, 0, offset);
3577 else
3578 memset(kaddr + offset, 0, len);
e6dcd2dc
CM		 3579 flush_dcache_page(page);
3580 kunmap(page);
3581 }
247e743c 3582 ClearPageChecked(page);
e6dcd2dc 3583 set_page_dirty(page);
2ac55d41
JB		 3584 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3585 GFP_NOFS);
39279cc3 3586
89642229 3587out_unlock:
5d5e103a 3588 if (ret)
0ca1f7ce 3589 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
39279cc3
CM		 3590 unlock_page(page);
3591 page_cache_release(page);
3592out:
3593 return ret;
3594}
3595
695a0d0d
JB		 3596/*
3597 * This function puts in dummy file extents for the area we're creating a hole
3598 * for. So if we are truncating this file to a larger size we need to insert
3599 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3600 * the range between oldsize and size
3601 */
a41ad394 3602int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
39279cc3 3603{
9036c102
YZ		 3604 struct btrfs_trans_handle *trans;
3605 struct btrfs_root *root = BTRFS_I(inode)->root;
3606 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a22285a6 3607 struct extent_map *em = NULL;
2ac55d41 3608 struct extent_state *cached_state = NULL;
5dc562c5 3609 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9036c102 3610 u64 mask = root->sectorsize - 1;
a41ad394 3611 u64 hole_start = (oldsize + mask) & ~mask;
9036c102
YZ		 3612 u64 block_end = (size + mask) & ~mask;
3613 u64 last_byte;
3614 u64 cur_offset;
3615 u64 hole_size;
9ed74f2d 3616 int err = 0;
39279cc3 3617
9036c102
YZ		 3618 if (size <= hole_start)
3619 return 0;
3620
9036c102
YZ		 3621 while (1) {
3622 struct btrfs_ordered_extent *ordered;
3623 btrfs_wait_ordered_range(inode, hole_start,
3624 block_end - hole_start);
2ac55d41 3625 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
d0082371 3626 &cached_state);
9036c102
YZ		 3627 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3628 if (!ordered)
3629 break;
2ac55d41
JB		 3630 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3631 &cached_state, GFP_NOFS);
9036c102
YZ		 3632 btrfs_put_ordered_extent(ordered);
3633 }
39279cc3 3634
9036c102
YZ		 3635 cur_offset = hole_start;
3636 while (1) {
3637 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3638 block_end - cur_offset, 0);
79787eaa
JM		 3639 if (IS_ERR(em)) {
3640 err = PTR_ERR(em);
3641 break;
3642 }
9036c102
YZ		 3643 last_byte = min(extent_map_end(em), block_end);
3644 last_byte = (last_byte + mask) & ~mask;
8082510e 3645 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5dc562c5 3646 struct extent_map *hole_em;
9036c102 3647 hole_size = last_byte - cur_offset;
9ed74f2d 3648
3642320e 3649 trans = btrfs_start_transaction(root, 3);
a22285a6
YZ		 3650 if (IS_ERR(trans)) {
3651 err = PTR_ERR(trans);
9ed74f2d 3652 break;
a22285a6 3653 }
8082510e 3654
5dc562c5
JB		 3655 err = btrfs_drop_extents(trans, root, inode,
3656 cur_offset,
2671485d 3657 cur_offset + hole_size, 1);
5b397377 3658 if (err) {
79787eaa 3659 btrfs_abort_transaction(trans, root, err);
5b397377 3660 btrfs_end_transaction(trans, root);
3893e33b 3661 break;
5b397377 3662 }
8082510e 3663
9036c102 3664 err = btrfs_insert_file_extent(trans, root,
33345d01 3665 btrfs_ino(inode), cur_offset, 0,
9036c102
YZ		 3666 0, hole_size, 0, hole_size,
3667 0, 0, 0);
5b397377 3668 if (err) {
79787eaa 3669 btrfs_abort_transaction(trans, root, err);
5b397377 3670 btrfs_end_transaction(trans, root);
3893e33b 3671 break;
5b397377 3672 }
8082510e 3673
5dc562c5
JB		 3674 btrfs_drop_extent_cache(inode, cur_offset,
3675 cur_offset + hole_size - 1, 0);
3676 hole_em = alloc_extent_map();
3677 if (!hole_em) {
3678 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3679 &BTRFS_I(inode)->runtime_flags);
3680 goto next;
3681 }
3682 hole_em->start = cur_offset;
3683 hole_em->len = hole_size;
3684 hole_em->orig_start = cur_offset;
8082510e 3685
5dc562c5
JB		 3686 hole_em->block_start = EXTENT_MAP_HOLE;
3687 hole_em->block_len = 0;
3688 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3689 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3690 hole_em->generation = trans->transid;
8082510e 3691
5dc562c5
JB		 3692 while (1) {
3693 write_lock(&em_tree->lock);
3694 err = add_extent_mapping(em_tree, hole_em);
3695 if (!err)
3696 list_move(&hole_em->list,
3697 &em_tree->modified_extents);
3698 write_unlock(&em_tree->lock);
3699 if (err != -EEXIST)
3700 break;
3701 btrfs_drop_extent_cache(inode, cur_offset,
3702 cur_offset +
3703 hole_size - 1, 0);
3704 }
3705 free_extent_map(hole_em);
3706next:
3642320e 3707 btrfs_update_inode(trans, root, inode);
8082510e 3708 btrfs_end_transaction(trans, root);
9036c102
YZ		 3709 }
3710 free_extent_map(em);
a22285a6 3711 em = NULL;
9036c102 3712 cur_offset = last_byte;
8082510e 3713 if (cur_offset >= block_end)
9036c102
YZ		 3714 break;
3715 }
1832a6d5 3716
a22285a6 3717 free_extent_map(em);
2ac55d41
JB		 3718 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3719 GFP_NOFS);
9036c102
YZ		 3720 return err;
3721}
39279cc3 3722
a41ad394 3723static int btrfs_setsize(struct inode *inode, loff_t newsize)
8082510e 3724{
f4a2f4c5
MX		 3725 struct btrfs_root *root = BTRFS_I(inode)->root;
3726 struct btrfs_trans_handle *trans;
a41ad394 3727 loff_t oldsize = i_size_read(inode);
8082510e
YZ		 3728 int ret;
3729
a41ad394 3730 if (newsize == oldsize)
8082510e
YZ		 3731 return 0;
3732
a41ad394 3733 if (newsize > oldsize) {
a41ad394
JB		 3734 truncate_pagecache(inode, oldsize, newsize);
3735 ret = btrfs_cont_expand(inode, oldsize, newsize);
f4a2f4c5 3736 if (ret)
8082510e 3737 return ret;
8082510e 3738
f4a2f4c5
MX		 3739 trans = btrfs_start_transaction(root, 1);
3740 if (IS_ERR(trans))
3741 return PTR_ERR(trans);
3742
3743 i_size_write(inode, newsize);
3744 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3745 ret = btrfs_update_inode(trans, root, inode);
7ad85bb7 3746 btrfs_end_transaction(trans, root);
a41ad394 3747 } else {
8082510e 3748
a41ad394
JB		 3749 /*
3750 * We're truncating a file that used to have good data down to
3751 * zero. Make sure it gets into the ordered flush list so that
3752 * any new writes get down to disk quickly.
3753 */
3754 if (newsize == 0)
72ac3c0d
JB		 3755 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3756 &BTRFS_I(inode)->runtime_flags);
8082510e 3757
a41ad394
JB		 3758 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3759 truncate_setsize(inode, newsize);
3760 ret = btrfs_truncate(inode);
8082510e
YZ		 3761 }
3762
a41ad394 3763 return ret;
8082510e
YZ		 3764}
3765
9036c102
YZ		 3766static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3767{
3768 struct inode *inode = dentry->d_inode;
b83cc969 3769 struct btrfs_root *root = BTRFS_I(inode)->root;
9036c102 3770 int err;
39279cc3 3771
b83cc969
LZ		 3772 if (btrfs_root_readonly(root))
3773 return -EROFS;
3774
9036c102
YZ		 3775 err = inode_change_ok(inode, attr);
3776 if (err)
3777 return err;
2bf5a725 3778
5a3f23d5 3779 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
a41ad394 3780 err = btrfs_setsize(inode, attr->ia_size);
8082510e
YZ		 3781 if (err)
3782 return err;
39279cc3 3783 }
9036c102 3784
1025774c
CH		 3785 if (attr->ia_valid) {
3786 setattr_copy(inode, attr);
0c4d2d95 3787 inode_inc_iversion(inode);
22c44fe6 3788 err = btrfs_dirty_inode(inode);
1025774c 3789
22c44fe6 3790 if (!err && attr->ia_valid & ATTR_MODE)
1025774c
CH		 3791 err = btrfs_acl_chmod(inode);
3792 }
33268eaf 3793
39279cc3
CM		 3794 return err;
3795}
61295eb8 3796
bd555975 3797void btrfs_evict_inode(struct inode *inode)
39279cc3
CM		 3798{
3799 struct btrfs_trans_handle *trans;
3800 struct btrfs_root *root = BTRFS_I(inode)->root;
726c35fa 3801 struct btrfs_block_rsv *rsv, *global_rsv;
07127184 3802 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
d3c2fdcf 3803 unsigned long nr;
39279cc3
CM		 3804 int ret;
3805
1abe9b8a 3806 trace_btrfs_inode_evict(inode);
3807
39279cc3 3808 truncate_inode_pages(&inode->i_data, 0);
0af3d00b 3809 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
83eea1f1 3810 btrfs_is_free_space_inode(inode)))
bd555975
AV		 3811 goto no_delete;
3812
39279cc3 3813 if (is_bad_inode(inode)) {
7b128766 3814 btrfs_orphan_del(NULL, inode);
39279cc3
CM		 3815 goto no_delete;
3816 }
bd555975 3817 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4a096752 3818 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5f39d397 3819
c71bf099 3820 if (root->fs_info->log_root_recovering) {
6bf02314 3821 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8a35d95f 3822 &BTRFS_I(inode)->runtime_flags));
c71bf099
YZ		 3823 goto no_delete;
3824 }
3825
76dda93c
YZ		 3826 if (inode->i_nlink > 0) {
3827 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3828 goto no_delete;
3829 }
3830
66d8f3dd 3831 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4289a667
JB		 3832 if (!rsv) {
3833 btrfs_orphan_del(NULL, inode);
3834 goto no_delete;
3835 }
4a338542 3836 rsv->size = min_size;
ca7e70f5 3837 rsv->failfast = 1;
726c35fa 3838 global_rsv = &root->fs_info->global_block_rsv;
4289a667 3839
dbe674a9 3840 btrfs_i_size_write(inode, 0);
5f39d397 3841
4289a667 3842 /*
8407aa46
MX		 3843 * This is a bit simpler than btrfs_truncate since we've already
3844 * reserved our space for our orphan item in the unlink, so we just
3845 * need to reserve some slack space in case we add bytes and update
3846 * inode item when doing the truncate.
4289a667 3847 */
8082510e 3848 while (1) {
08e007d2
MX		 3849 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3850 BTRFS_RESERVE_FLUSH_LIMIT);
726c35fa
JB		 3851
3852 /*
3853 * Try and steal from the global reserve since we will
3854 * likely not use this space anyway, we want to try as
3855 * hard as possible to get this to work.
3856 */
3857 if (ret)
3858 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
d68fc57b 3859
d68fc57b 3860 if (ret) {
4289a667 3861 printk(KERN_WARNING "Could not get space for a "
482e6dc5 3862 "delete, will truncate on mount %d\n", ret);
4289a667
JB		 3863 btrfs_orphan_del(NULL, inode);
3864 btrfs_free_block_rsv(root, rsv);
3865 goto no_delete;
d68fc57b 3866 }
7b128766 3867
08e007d2 3868 trans = btrfs_start_transaction_lflush(root, 1);
4289a667
JB		 3869 if (IS_ERR(trans)) {
3870 btrfs_orphan_del(NULL, inode);
3871 btrfs_free_block_rsv(root, rsv);
3872 goto no_delete;
d68fc57b 3873 }
7b128766 3874
4289a667
JB		 3875 trans->block_rsv = rsv;
3876
d68fc57b 3877 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
ca7e70f5 3878 if (ret != -ENOSPC)
8082510e 3879 break;
85e21bac 3880
8407aa46
MX		 3881 trans->block_rsv = &root->fs_info->trans_block_rsv;
3882 ret = btrfs_update_inode(trans, root, inode);
3883 BUG_ON(ret);
3884
8082510e
YZ		 3885 nr = trans->blocks_used;
3886 btrfs_end_transaction(trans, root);
3887 trans = NULL;
3888 btrfs_btree_balance_dirty(root, nr);
3889 }
5f39d397 3890
4289a667
JB		 3891 btrfs_free_block_rsv(root, rsv);
3892
8082510e 3893 if (ret == 0) {
4289a667 3894 trans->block_rsv = root->orphan_block_rsv;
8082510e
YZ		 3895 ret = btrfs_orphan_del(trans, inode);
3896 BUG_ON(ret);
3897 }
54aa1f4d 3898
4289a667 3899 trans->block_rsv = &root->fs_info->trans_block_rsv;
581bb050
LZ		 3900 if (!(root == root->fs_info->tree_root ||
3901 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
33345d01 3902 btrfs_return_ino(root, btrfs_ino(inode));
581bb050 3903
d3c2fdcf 3904 nr = trans->blocks_used;
54aa1f4d 3905 btrfs_end_transaction(trans, root);
d3c2fdcf 3906 btrfs_btree_balance_dirty(root, nr);
39279cc3 3907no_delete:
dbd5768f 3908 clear_inode(inode);
8082510e 3909 return;
39279cc3
CM		 3910}
3911
3912/*
3913 * this returns the key found in the dir entry in the location pointer.
3914 * If no dir entries were found, location->objectid is 0.
3915 */
3916static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3917 struct btrfs_key *location)
3918{
3919 const char *name = dentry->d_name.name;
3920 int namelen = dentry->d_name.len;
3921 struct btrfs_dir_item *di;
3922 struct btrfs_path *path;
3923 struct btrfs_root *root = BTRFS_I(dir)->root;
0d9f7f3e 3924 int ret = 0;
39279cc3
CM		 3925
3926 path = btrfs_alloc_path();
d8926bb3
MF		 3927 if (!path)
3928 return -ENOMEM;
3954401f 3929
33345d01 3930 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
39279cc3 3931 namelen, 0);
0d9f7f3e
Y		 3932 if (IS_ERR(di))
3933 ret = PTR_ERR(di);
d397712b 3934
c704005d 3935 if (IS_ERR_OR_NULL(di))
3954401f 3936 goto out_err;
d397712b 3937
5f39d397 3938 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
39279cc3 3939out:
39279cc3
CM		 3940 btrfs_free_path(path);
3941 return ret;
3954401f
CM		 3942out_err:
3943 location->objectid = 0;
3944 goto out;
39279cc3
CM		 3945}
3946
3947/*
3948 * when we hit a tree root in a directory, the btrfs part of the inode
3949 * needs to be changed to reflect the root directory of the tree root. This
3950 * is kind of like crossing a mount point.
3951 */
3952static int fixup_tree_root_location(struct btrfs_root *root,
4df27c4d
YZ		 3953 struct inode *dir,
3954 struct dentry *dentry,
3955 struct btrfs_key *location,
3956 struct btrfs_root **sub_root)
39279cc3 3957{
4df27c4d
YZ		 3958 struct btrfs_path *path;
3959 struct btrfs_root *new_root;
3960 struct btrfs_root_ref *ref;
3961 struct extent_buffer *leaf;
3962 int ret;
3963 int err = 0;
39279cc3 3964
4df27c4d
YZ		 3965 path = btrfs_alloc_path();
3966 if (!path) {
3967 err = -ENOMEM;
3968 goto out;
3969 }
39279cc3 3970
4df27c4d
YZ		 3971 err = -ENOENT;
3972 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3973 BTRFS_I(dir)->root->root_key.objectid,
3974 location->objectid);
3975 if (ret) {
3976 if (ret < 0)
3977 err = ret;
3978 goto out;
3979 }
39279cc3 3980
4df27c4d
YZ		 3981 leaf = path->nodes[0];
3982 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
33345d01 3983 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4df27c4d
YZ		 3984 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3985 goto out;
39279cc3 3986
4df27c4d
YZ		 3987 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3988 (unsigned long)(ref + 1),
3989 dentry->d_name.len);
3990 if (ret)
3991 goto out;
3992
b3b4aa74 3993 btrfs_release_path(path);
4df27c4d
YZ		 3994
3995 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3996 if (IS_ERR(new_root)) {
3997 err = PTR_ERR(new_root);
3998 goto out;
3999 }
4000
4001 if (btrfs_root_refs(&new_root->root_item) == 0) {
4002 err = -ENOENT;
4003 goto out;
4004 }
4005
4006 *sub_root = new_root;
4007 location->objectid = btrfs_root_dirid(&new_root->root_item);
4008 location->type = BTRFS_INODE_ITEM_KEY;
4009 location->offset = 0;
4010 err = 0;
4011out:
4012 btrfs_free_path(path);
4013 return err;
39279cc3
CM		 4014}
4015
5d4f98a2
YZ		 4016static void inode_tree_add(struct inode *inode)
4017{
4018 struct btrfs_root *root = BTRFS_I(inode)->root;
4019 struct btrfs_inode *entry;
03e860bd
FNP		 4020 struct rb_node **p;
4021 struct rb_node *parent;
33345d01 4022 u64 ino = btrfs_ino(inode);
03e860bd
FNP		 4023again:
4024 p = &root->inode_tree.rb_node;
4025 parent = NULL;
5d4f98a2 4026
1d3382cb 4027 if (inode_unhashed(inode))
76dda93c
YZ		 4028 return;
4029
5d4f98a2
YZ		 4030 spin_lock(&root->inode_lock);
4031 while (*p) {
4032 parent = *p;
4033 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4034
33345d01 4035 if (ino < btrfs_ino(&entry->vfs_inode))
03e860bd 4036 p = &parent->rb_left;
33345d01 4037 else if (ino > btrfs_ino(&entry->vfs_inode))
03e860bd 4038 p = &parent->rb_right;
5d4f98a2
YZ		 4039 else {
4040 WARN_ON(!(entry->vfs_inode.i_state &
a4ffdde6 4041 (I_WILL_FREE | I_FREEING)));
03e860bd
FNP		 4042 rb_erase(parent, &root->inode_tree);
4043 RB_CLEAR_NODE(parent);
4044 spin_unlock(&root->inode_lock);
4045 goto again;
5d4f98a2
YZ		 4046 }
4047 }
4048 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4049 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4050 spin_unlock(&root->inode_lock);
4051}
4052
4053static void inode_tree_del(struct inode *inode)
4054{
4055 struct btrfs_root *root = BTRFS_I(inode)->root;
76dda93c 4056 int empty = 0;
5d4f98a2 4057
03e860bd 4058 spin_lock(&root->inode_lock);
5d4f98a2 4059 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5d4f98a2 4060 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5d4f98a2 4061 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
76dda93c 4062 empty = RB_EMPTY_ROOT(&root->inode_tree);
5d4f98a2 4063 }
03e860bd 4064 spin_unlock(&root->inode_lock);
76dda93c 4065
0af3d00b
JB		 4066 /*
4067 * Free space cache has inodes in the tree root, but the tree root has a
4068 * root_refs of 0, so this could end up dropping the tree root as a
4069 * snapshot, so we need the extra !root->fs_info->tree_root check to
4070 * make sure we don't drop it.
4071 */
4072 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4073 root != root->fs_info->tree_root) {
76dda93c
YZ		 4074 synchronize_srcu(&root->fs_info->subvol_srcu);
4075 spin_lock(&root->inode_lock);
4076 empty = RB_EMPTY_ROOT(&root->inode_tree);
4077 spin_unlock(&root->inode_lock);
4078 if (empty)
4079 btrfs_add_dead_root(root);
4080 }
4081}
4082
143bede5 4083void btrfs_invalidate_inodes(struct btrfs_root *root)
76dda93c
YZ		 4084{
4085 struct rb_node *node;
4086 struct rb_node *prev;
4087 struct btrfs_inode *entry;
4088 struct inode *inode;
4089 u64 objectid = 0;
4090
4091 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4092
4093 spin_lock(&root->inode_lock);
4094again:
4095 node = root->inode_tree.rb_node;
4096 prev = NULL;
4097 while (node) {
4098 prev = node;
4099 entry = rb_entry(node, struct btrfs_inode, rb_node);
4100
33345d01 4101 if (objectid < btrfs_ino(&entry->vfs_inode))
76dda93c 4102 node = node->rb_left;
33345d01 4103 else if (objectid > btrfs_ino(&entry->vfs_inode))
76dda93c
YZ		 4104 node = node->rb_right;
4105 else
4106 break;
4107 }
4108 if (!node) {
4109 while (prev) {
4110 entry = rb_entry(prev, struct btrfs_inode, rb_node);
33345d01 4111 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
76dda93c
YZ		 4112 node = prev;
4113 break;
4114 }
4115 prev = rb_next(prev);
4116 }
4117 }
4118 while (node) {
4119 entry = rb_entry(node, struct btrfs_inode, rb_node);
33345d01 4120 objectid = btrfs_ino(&entry->vfs_inode) + 1;
76dda93c
YZ		 4121 inode = igrab(&entry->vfs_inode);
4122 if (inode) {
4123 spin_unlock(&root->inode_lock);
4124 if (atomic_read(&inode->i_count) > 1)
4125 d_prune_aliases(inode);
4126 /*
45321ac5 4127 * btrfs_drop_inode will have it removed from
76dda93c
YZ		 4128 * the inode cache when its usage count
4129 * hits zero.
4130 */
4131 iput(inode);
4132 cond_resched();
4133 spin_lock(&root->inode_lock);
4134 goto again;
4135 }
4136
4137 if (cond_resched_lock(&root->inode_lock))
4138 goto again;
4139
4140 node = rb_next(node);
4141 }
4142 spin_unlock(&root->inode_lock);
5d4f98a2
YZ		 4143}
4144
e02119d5
CM		 4145static int btrfs_init_locked_inode(struct inode *inode, void *p)
4146{
4147 struct btrfs_iget_args *args = p;
4148 inode->i_ino = args->ino;
e02119d5 4149 BTRFS_I(inode)->root = args->root;
39279cc3
CM		 4150 return 0;
4151}
4152
4153static int btrfs_find_actor(struct inode *inode, void *opaque)
4154{
4155 struct btrfs_iget_args *args = opaque;
33345d01 4156 return args->ino == btrfs_ino(inode) &&
d397712b 4157 args->root == BTRFS_I(inode)->root;
39279cc3
CM		 4158}
4159
5d4f98a2
YZ		 4160static struct inode *btrfs_iget_locked(struct super_block *s,
4161 u64 objectid,
4162 struct btrfs_root *root)
39279cc3
CM		 4163{
4164 struct inode *inode;
4165 struct btrfs_iget_args args;
4166 args.ino = objectid;
4167 args.root = root;
4168
4169 inode = iget5_locked(s, objectid, btrfs_find_actor,
4170 btrfs_init_locked_inode,
4171 (void *)&args);
4172 return inode;
4173}
4174
1a54ef8c
BR		 4175/* Get an inode object given its location and corresponding root.
4176 * Returns in *is_new if the inode was read from disk
4177 */
4178struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
73f73415 4179 struct btrfs_root *root, int *new)
1a54ef8c
BR		 4180{
4181 struct inode *inode;
4182
4183 inode = btrfs_iget_locked(s, location->objectid, root);
4184 if (!inode)
5d4f98a2 4185 return ERR_PTR(-ENOMEM);
1a54ef8c
BR		 4186
4187 if (inode->i_state & I_NEW) {
4188 BTRFS_I(inode)->root = root;
4189 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4190 btrfs_read_locked_inode(inode);
1748f843
MF		 4191 if (!is_bad_inode(inode)) {
4192 inode_tree_add(inode);
4193 unlock_new_inode(inode);
4194 if (new)
4195 *new = 1;
4196 } else {
e0b6d65b
ST		 4197 unlock_new_inode(inode);
4198 iput(inode);
4199 inode = ERR_PTR(-ESTALE);
1748f843
MF		 4200 }
4201 }
4202
1a54ef8c
BR		 4203 return inode;
4204}
4205
4df27c4d
YZ		 4206static struct inode *new_simple_dir(struct super_block *s,
4207 struct btrfs_key *key,
4208 struct btrfs_root *root)
4209{
4210 struct inode *inode = new_inode(s);
4211
4212 if (!inode)
4213 return ERR_PTR(-ENOMEM);
4214
4df27c4d
YZ		 4215 BTRFS_I(inode)->root = root;
4216 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
72ac3c0d 4217 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4df27c4d
YZ		 4218
4219 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
848cce0d 4220 inode->i_op = &btrfs_dir_ro_inode_operations;
4df27c4d
YZ		 4221 inode->i_fop = &simple_dir_operations;
4222 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4223 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4224
4225 return inode;
4226}
4227
3de4586c 4228struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
39279cc3 4229{
d397712b 4230 struct inode *inode;
4df27c4d 4231 struct btrfs_root *root = BTRFS_I(dir)->root;
39279cc3
CM		 4232 struct btrfs_root *sub_root = root;
4233 struct btrfs_key location;
76dda93c 4234 int index;
b4aff1f8 4235 int ret = 0;
39279cc3
CM		 4236
4237 if (dentry->d_name.len > BTRFS_NAME_LEN)
4238 return ERR_PTR(-ENAMETOOLONG);
5f39d397 4239
b4aff1f8
JB		 4240 if (unlikely(d_need_lookup(dentry))) {
4241 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4242 kfree(dentry->d_fsdata);
4243 dentry->d_fsdata = NULL;
a66e7cc6
JB		 4244 /* This thing is hashed, drop it for now */
4245 d_drop(dentry);
b4aff1f8
JB		 4246 } else {
4247 ret = btrfs_inode_by_name(dir, dentry, &location);
4248 }
5f39d397 4249
39279cc3
CM		 4250 if (ret < 0)
4251 return ERR_PTR(ret);
5f39d397 4252
4df27c4d
YZ		 4253 if (location.objectid == 0)
4254 return NULL;
4255
4256 if (location.type == BTRFS_INODE_ITEM_KEY) {
73f73415 4257 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4df27c4d
YZ		 4258 return inode;
4259 }
4260
4261 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4262
76dda93c 4263 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4df27c4d
YZ		 4264 ret = fixup_tree_root_location(root, dir, dentry,
4265 &location, &sub_root);
4266 if (ret < 0) {
4267 if (ret != -ENOENT)
4268 inode = ERR_PTR(ret);
4269 else
4270 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4271 } else {
73f73415 4272 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
39279cc3 4273 }
76dda93c
YZ		 4274 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4275
34d19bad 4276 if (!IS_ERR(inode) && root != sub_root) {
c71bf099
YZ		 4277 down_read(&root->fs_info->cleanup_work_sem);
4278 if (!(inode->i_sb->s_flags & MS_RDONLY))
66b4ffd1 4279 ret = btrfs_orphan_cleanup(sub_root);
c71bf099 4280 up_read(&root->fs_info->cleanup_work_sem);
66b4ffd1
JB		 4281 if (ret)
4282 inode = ERR_PTR(ret);
c71bf099
YZ		 4283 }
4284
3de4586c
CM		 4285 return inode;
4286}
4287
fe15ce44 4288static int btrfs_dentry_delete(const struct dentry *dentry)
76dda93c
YZ		 4289{
4290 struct btrfs_root *root;
848cce0d 4291 struct inode *inode = dentry->d_inode;
76dda93c 4292
848cce0d
LZ		 4293 if (!inode && !IS_ROOT(dentry))
4294 inode = dentry->d_parent->d_inode;
76dda93c 4295
848cce0d
LZ		 4296 if (inode) {
4297 root = BTRFS_I(inode)->root;
efefb143
YZ		 4298 if (btrfs_root_refs(&root->root_item) == 0)
4299 return 1;
848cce0d
LZ		 4300
4301 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4302 return 1;
efefb143 4303 }
76dda93c
YZ		 4304 return 0;
4305}
4306
b4aff1f8
JB		 4307static void btrfs_dentry_release(struct dentry *dentry)
4308{
4309 if (dentry->d_fsdata)
4310 kfree(dentry->d_fsdata);
4311}
4312
3de4586c 4313static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
00cd8dd3 4314 unsigned int flags)
3de4586c 4315{
a66e7cc6
JB		 4316 struct dentry *ret;
4317
4318 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4319 if (unlikely(d_need_lookup(dentry))) {
4320 spin_lock(&dentry->d_lock);
4321 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4322 spin_unlock(&dentry->d_lock);
4323 }
4324 return ret;
39279cc3
CM		 4325}
4326
16cdcec7 4327unsigned char btrfs_filetype_table[] = {
39279cc3
CM		 4328 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4329};
4330
cbdf5a24
DW		 4331static int btrfs_real_readdir(struct file *filp, void *dirent,
4332 filldir_t filldir)
39279cc3 4333{
6da6abae 4334 struct inode *inode = filp->f_dentry->d_inode;
39279cc3
CM		 4335 struct btrfs_root *root = BTRFS_I(inode)->root;
4336 struct btrfs_item *item;
4337 struct btrfs_dir_item *di;
4338 struct btrfs_key key;
5f39d397 4339 struct btrfs_key found_key;
39279cc3 4340 struct btrfs_path *path;
16cdcec7
MX		 4341 struct list_head ins_list;
4342 struct list_head del_list;
39279cc3 4343 int ret;
5f39d397 4344 struct extent_buffer *leaf;
39279cc3 4345 int slot;
39279cc3
CM		 4346 unsigned char d_type;
4347 int over = 0;
4348 u32 di_cur;
4349 u32 di_total;
4350 u32 di_len;
4351 int key_type = BTRFS_DIR_INDEX_KEY;
5f39d397
CM		 4352 char tmp_name[32];
4353 char *name_ptr;
4354 int name_len;
16cdcec7 4355 int is_curr = 0; /* filp->f_pos points to the current index? */
39279cc3
CM		 4356
4357 /* FIXME, use a real flag for deciding about the key type */
4358 if (root->fs_info->tree_root == root)
4359 key_type = BTRFS_DIR_ITEM_KEY;
5f39d397 4360
3954401f
CM		 4361 /* special case for "." */
4362 if (filp->f_pos == 0) {
3765fefa
HS		 4363 over = filldir(dirent, ".", 1,
4364 filp->f_pos, btrfs_ino(inode), DT_DIR);
3954401f
CM		 4365 if (over)
4366 return 0;
4367 filp->f_pos = 1;
4368 }
3954401f
CM		 4369 /* special case for .., just use the back ref */
4370 if (filp->f_pos == 1) {
5ecc7e5d 4371 u64 pino = parent_ino(filp->f_path.dentry);
3954401f 4372 over = filldir(dirent, "..", 2,
3765fefa 4373 filp->f_pos, pino, DT_DIR);
3954401f 4374 if (over)
49593bfa 4375 return 0;
3954401f
CM		 4376 filp->f_pos = 2;
4377 }
49593bfa 4378 path = btrfs_alloc_path();
16cdcec7
MX		 4379 if (!path)
4380 return -ENOMEM;
ff5714cc 4381
026fd317 4382 path->reada = 1;
49593bfa 4383
16cdcec7
MX		 4384 if (key_type == BTRFS_DIR_INDEX_KEY) {
4385 INIT_LIST_HEAD(&ins_list);
4386 INIT_LIST_HEAD(&del_list);
4387 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4388 }
4389
39279cc3
CM		 4390 btrfs_set_key_type(&key, key_type);
4391 key.offset = filp->f_pos;
33345d01 4392 key.objectid = btrfs_ino(inode);
5f39d397 4393
39279cc3
CM		 4394 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4395 if (ret < 0)
4396 goto err;
49593bfa
DW		 4397
4398 while (1) {
5f39d397 4399 leaf = path->nodes[0];
39279cc3 4400 slot = path->slots[0];
b9e03af0
LZ		 4401 if (slot >= btrfs_header_nritems(leaf)) {
4402 ret = btrfs_next_leaf(root, path);
4403 if (ret < 0)
4404 goto err;
4405 else if (ret > 0)
4406 break;
4407 continue;
39279cc3 4408 }
3de4586c 4409
5f39d397
CM		 4410 item = btrfs_item_nr(leaf, slot);
4411 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4412
4413 if (found_key.objectid != key.objectid)
39279cc3 4414 break;
5f39d397 4415 if (btrfs_key_type(&found_key) != key_type)
39279cc3 4416 break;
5f39d397 4417 if (found_key.offset < filp->f_pos)
b9e03af0 4418 goto next;
16cdcec7
MX		 4419 if (key_type == BTRFS_DIR_INDEX_KEY &&
4420 btrfs_should_delete_dir_index(&del_list,
4421 found_key.offset))
4422 goto next;
5f39d397
CM		 4423
4424 filp->f_pos = found_key.offset;
16cdcec7 4425 is_curr = 1;
49593bfa 4426
39279cc3
CM		 4427 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4428 di_cur = 0;
5f39d397 4429 di_total = btrfs_item_size(leaf, item);
49593bfa
DW		 4430
4431 while (di_cur < di_total) {
5f39d397
CM		 4432 struct btrfs_key location;
4433
22a94d44
JB		 4434 if (verify_dir_item(root, leaf, di))
4435 break;
4436
5f39d397 4437 name_len = btrfs_dir_name_len(leaf, di);
49593bfa 4438 if (name_len <= sizeof(tmp_name)) {
5f39d397
CM		 4439 name_ptr = tmp_name;
4440 } else {
4441 name_ptr = kmalloc(name_len, GFP_NOFS);
49593bfa
DW		 4442 if (!name_ptr) {
4443 ret = -ENOMEM;
4444 goto err;
4445 }
5f39d397
CM		 4446 }
4447 read_extent_buffer(leaf, name_ptr,
4448 (unsigned long)(di + 1), name_len);
4449
4450 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4451 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3de4586c 4452
fede766f 4453
3de4586c 4454 /* is this a reference to our own snapshot? If so
8c9c2bf7
AJ		 4455 * skip it.
4456 *
4457 * In contrast to old kernels, we insert the snapshot's
4458 * dir item and dir index after it has been created, so
4459 * we won't find a reference to our own snapshot. We
4460 * still keep the following code for backward
4461 * compatibility.
3de4586c
CM		 4462 */
4463 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4464 location.objectid == root->root_key.objectid) {
4465 over = 0;
4466 goto skip;
4467 }
5f39d397 4468 over = filldir(dirent, name_ptr, name_len,
49593bfa 4469 found_key.offset, location.objectid,
39279cc3 4470 d_type);
5f39d397 4471
3de4586c 4472skip:
5f39d397
CM		 4473 if (name_ptr != tmp_name)
4474 kfree(name_ptr);
4475
39279cc3
CM		 4476 if (over)
4477 goto nopos;
5103e947 4478 di_len = btrfs_dir_name_len(leaf, di) +
49593bfa 4479 btrfs_dir_data_len(leaf, di) + sizeof(*di);
39279cc3
CM		 4480 di_cur += di_len;
4481 di = (struct btrfs_dir_item *)((char *)di + di_len);
4482 }
b9e03af0
LZ		 4483next:
4484 path->slots[0]++;
39279cc3 4485 }
49593bfa 4486
16cdcec7
MX		 4487 if (key_type == BTRFS_DIR_INDEX_KEY) {
4488 if (is_curr)
4489 filp->f_pos++;
4490 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4491 &ins_list);
4492 if (ret)
4493 goto nopos;
4494 }
4495
49593bfa 4496 /* Reached end of directory/root. Bump pos past the last item. */
5e591a07 4497 if (key_type == BTRFS_DIR_INDEX_KEY)
406266ab
JE		 4498 /*
4499 * 32-bit glibc will use getdents64, but then strtol -
4500 * so the last number we can serve is this.
4501 */
4502 filp->f_pos = 0x7fffffff;
5e591a07
YZ		 4503 else
4504 filp->f_pos++;
39279cc3
CM		 4505nopos:
4506 ret = 0;
4507err:
16cdcec7
MX		 4508 if (key_type == BTRFS_DIR_INDEX_KEY)
4509 btrfs_put_delayed_items(&ins_list, &del_list);
39279cc3 4510 btrfs_free_path(path);
39279cc3
CM		 4511 return ret;
4512}
4513
a9185b41 4514int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
39279cc3
CM		 4515{
4516 struct btrfs_root *root = BTRFS_I(inode)->root;
4517 struct btrfs_trans_handle *trans;
4518 int ret = 0;
0af3d00b 4519 bool nolock = false;
39279cc3 4520
72ac3c0d 4521 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4ca8b41e
CM		 4522 return 0;
4523
83eea1f1 4524 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
82d5902d 4525 nolock = true;
0af3d00b 4526
a9185b41 4527 if (wbc->sync_mode == WB_SYNC_ALL) {
0af3d00b 4528 if (nolock)
7a7eaa40 4529 trans = btrfs_join_transaction_nolock(root);
0af3d00b 4530 else
7a7eaa40 4531 trans = btrfs_join_transaction(root);
3612b495
TI		 4532 if (IS_ERR(trans))
4533 return PTR_ERR(trans);
a698d075 4534 ret = btrfs_commit_transaction(trans, root);
39279cc3
CM		 4535 }
4536 return ret;
4537}
4538
4539/*
54aa1f4d 4540 * This is somewhat expensive, updating the tree every time the
39279cc3
CM		 4541 * inode changes. But, it is most likely to find the inode in cache.
4542 * FIXME, needs more benchmarking...there are no reasons other than performance
4543 * to keep or drop this code.
4544 */
22c44fe6 4545int btrfs_dirty_inode(struct inode *inode)
39279cc3
CM		 4546{
4547 struct btrfs_root *root = BTRFS_I(inode)->root;
4548 struct btrfs_trans_handle *trans;
8929ecfa
YZ		 4549 int ret;
4550
72ac3c0d 4551 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
22c44fe6 4552 return 0;
39279cc3 4553
7a7eaa40 4554 trans = btrfs_join_transaction(root);
22c44fe6
JB		 4555 if (IS_ERR(trans))
4556 return PTR_ERR(trans);
8929ecfa
YZ		 4557
4558 ret = btrfs_update_inode(trans, root, inode);
94b60442
CM		 4559 if (ret && ret == -ENOSPC) {
4560 /* whoops, lets try again with the full transaction */
4561 btrfs_end_transaction(trans, root);
4562 trans = btrfs_start_transaction(root, 1);
22c44fe6
JB		 4563 if (IS_ERR(trans))
4564 return PTR_ERR(trans);
8929ecfa 4565
94b60442 4566 ret = btrfs_update_inode(trans, root, inode);
94b60442 4567 }
39279cc3 4568 btrfs_end_transaction(trans, root);
16cdcec7
MX		 4569 if (BTRFS_I(inode)->delayed_node)
4570 btrfs_balance_delayed_items(root);
22c44fe6
JB		 4571
4572 return ret;
4573}
4574
4575/*
4576 * This is a copy of file_update_time. We need this so we can return error on
4577 * ENOSPC for updating the inode in the case of file write and mmap writes.
4578 */
e41f941a
JB		 4579static int btrfs_update_time(struct inode *inode, struct timespec *now,
4580 int flags)
22c44fe6 4581{
2bc55652
AB		 4582 struct btrfs_root *root = BTRFS_I(inode)->root;
4583
4584 if (btrfs_root_readonly(root))
4585 return -EROFS;
4586
e41f941a 4587 if (flags & S_VERSION)
22c44fe6 4588 inode_inc_iversion(inode);
e41f941a
JB		 4589 if (flags & S_CTIME)
4590 inode->i_ctime = *now;
4591 if (flags & S_MTIME)
4592 inode->i_mtime = *now;
4593 if (flags & S_ATIME)
4594 inode->i_atime = *now;
4595 return btrfs_dirty_inode(inode);
39279cc3
CM		 4596}
4597
d352ac68
CM		 4598/*
4599 * find the highest existing sequence number in a directory
4600 * and then set the in-memory index_cnt variable to reflect
4601 * free sequence numbers
4602 */
aec7477b
JB		 4603static int btrfs_set_inode_index_count(struct inode *inode)
4604{
4605 struct btrfs_root *root = BTRFS_I(inode)->root;
4606 struct btrfs_key key, found_key;
4607 struct btrfs_path *path;
4608 struct extent_buffer *leaf;
4609 int ret;
4610
33345d01 4611 key.objectid = btrfs_ino(inode);
aec7477b
JB		 4612 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4613 key.offset = (u64)-1;
4614
4615 path = btrfs_alloc_path();
4616 if (!path)
4617 return -ENOMEM;
4618
4619 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4620 if (ret < 0)
4621 goto out;
4622 /* FIXME: we should be able to handle this */
4623 if (ret == 0)
4624 goto out;
4625 ret = 0;
4626
4627 /*
4628 * MAGIC NUMBER EXPLANATION:
4629 * since we search a directory based on f_pos we have to start at 2
4630 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4631 * else has to start at 2
4632 */
4633 if (path->slots[0] == 0) {
4634 BTRFS_I(inode)->index_cnt = 2;
4635 goto out;
4636 }
4637
4638 path->slots[0]--;
4639
4640 leaf = path->nodes[0];
4641 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4642
33345d01 4643 if (found_key.objectid != btrfs_ino(inode) ||
aec7477b
JB		 4644 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4645 BTRFS_I(inode)->index_cnt = 2;
4646 goto out;
4647 }
4648
4649 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4650out:
4651 btrfs_free_path(path);
4652 return ret;
4653}
4654
d352ac68
CM		 4655/*
4656 * helper to find a free sequence number in a given directory. This current
4657 * code is very simple, later versions will do smarter things in the btree
4658 */
3de4586c 4659int btrfs_set_inode_index(struct inode *dir, u64 *index)
aec7477b
JB		 4660{
4661 int ret = 0;
4662
4663 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
16cdcec7
MX		 4664 ret = btrfs_inode_delayed_dir_index_count(dir);
4665 if (ret) {
4666 ret = btrfs_set_inode_index_count(dir);
4667 if (ret)
4668 return ret;
4669 }
aec7477b
JB		 4670 }
4671
00e4e6b3 4672 *index = BTRFS_I(dir)->index_cnt;
aec7477b
JB		 4673 BTRFS_I(dir)->index_cnt++;
4674
4675 return ret;
4676}
4677
39279cc3
CM		 4678static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4679 struct btrfs_root *root,
aec7477b 4680 struct inode *dir,
9c58309d 4681 const char *name, int name_len,
175a4eb7
AV		 4682 u64 ref_objectid, u64 objectid,
4683 umode_t mode, u64 *index)
39279cc3
CM		 4684{
4685 struct inode *inode;
5f39d397 4686 struct btrfs_inode_item *inode_item;
39279cc3 4687 struct btrfs_key *location;
5f39d397 4688 struct btrfs_path *path;
9c58309d
CM		 4689 struct btrfs_inode_ref *ref;
4690 struct btrfs_key key[2];
4691 u32 sizes[2];
4692 unsigned long ptr;
39279cc3
CM		 4693 int ret;
4694 int owner;
4695
5f39d397 4696 path = btrfs_alloc_path();
d8926bb3
MF		 4697 if (!path)
4698 return ERR_PTR(-ENOMEM);
5f39d397 4699
39279cc3 4700 inode = new_inode(root->fs_info->sb);
8fb27640
YS		 4701 if (!inode) {
4702 btrfs_free_path(path);
39279cc3 4703 return ERR_PTR(-ENOMEM);
8fb27640 4704 }
39279cc3 4705
581bb050
LZ		 4706 /*
4707 * we have to initialize this early, so we can reclaim the inode
4708 * number if we fail afterwards in this function.
4709 */
4710 inode->i_ino = objectid;
4711
aec7477b 4712 if (dir) {
1abe9b8a 4713 trace_btrfs_inode_request(dir);
4714
3de4586c 4715 ret = btrfs_set_inode_index(dir, index);
09771430 4716 if (ret) {
8fb27640 4717 btrfs_free_path(path);
09771430 4718 iput(inode);
aec7477b 4719 return ERR_PTR(ret);
09771430 4720 }
aec7477b
JB		 4721 }
4722 /*
4723 * index_cnt is ignored for everything but a dir,
4724 * btrfs_get_inode_index_count has an explanation for the magic
4725 * number
4726 */
4727 BTRFS_I(inode)->index_cnt = 2;
39279cc3 4728 BTRFS_I(inode)->root = root;
e02119d5 4729 BTRFS_I(inode)->generation = trans->transid;
76195853 4730 inode->i_generation = BTRFS_I(inode)->generation;
b888db2b 4731
5dc562c5
JB		 4732 /*
4733 * We could have gotten an inode number from somebody who was fsynced
4734 * and then removed in this same transaction, so let's just set full
4735 * sync since it will be a full sync anyway and this will blow away the
4736 * old info in the log.
4737 */
4738 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4739
569254b0 4740 if (S_ISDIR(mode))
39279cc3
CM		 4741 owner = 0;
4742 else
4743 owner = 1;
9c58309d
CM		 4744
4745 key[0].objectid = objectid;
4746 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4747 key[0].offset = 0;
4748
f186373f
MF		 4749 /*
4750 * Start new inodes with an inode_ref. This is slightly more
4751 * efficient for small numbers of hard links since they will
4752 * be packed into one item. Extended refs will kick in if we
4753 * add more hard links than can fit in the ref item.
4754 */
9c58309d
CM		 4755 key[1].objectid = objectid;
4756 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4757 key[1].offset = ref_objectid;
4758
4759 sizes[0] = sizeof(struct btrfs_inode_item);
4760 sizes[1] = name_len + sizeof(*ref);
4761
b9473439 4762 path->leave_spinning = 1;
9c58309d
CM		 4763 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4764 if (ret != 0)
5f39d397
CM		 4765 goto fail;
4766
ecc11fab 4767 inode_init_owner(inode, dir, mode);
a76a3cd4 4768 inode_set_bytes(inode, 0);
39279cc3 4769 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5f39d397
CM		 4770 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4771 struct btrfs_inode_item);
293f7e07
LZ		 4772 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4773 sizeof(*inode_item));
e02119d5 4774 fill_inode_item(trans, path->nodes[0], inode_item, inode);
9c58309d
CM		 4775
4776 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4777 struct btrfs_inode_ref);
4778 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
00e4e6b3 4779 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
9c58309d
CM		 4780 ptr = (unsigned long)(ref + 1);
4781 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4782
5f39d397
CM		 4783 btrfs_mark_buffer_dirty(path->nodes[0]);
4784 btrfs_free_path(path);
4785
39279cc3
CM		 4786 location = &BTRFS_I(inode)->location;
4787 location->objectid = objectid;
39279cc3
CM		 4788 location->offset = 0;
4789 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4790
6cbff00f
CH		 4791 btrfs_inherit_iflags(inode, dir);
4792
569254b0 4793 if (S_ISREG(mode)) {
94272164
CM		 4794 if (btrfs_test_opt(root, NODATASUM))
4795 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
75e7cb7f
LB		 4796 if (btrfs_test_opt(root, NODATACOW) ||
4797 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
94272164
CM		 4798 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4799 }
4800
39279cc3 4801 insert_inode_hash(inode);
5d4f98a2 4802 inode_tree_add(inode);
1abe9b8a 4803
4804 trace_btrfs_inode_new(inode);
1973f0fa 4805 btrfs_set_inode_last_trans(trans, inode);
1abe9b8a 4806
8ea05e3a
AB		 4807 btrfs_update_root_times(trans, root);
4808
39279cc3 4809 return inode;
5f39d397 4810fail:
aec7477b
JB		 4811 if (dir)
4812 BTRFS_I(dir)->index_cnt--;
5f39d397 4813 btrfs_free_path(path);
09771430 4814 iput(inode);
5f39d397 4815 return ERR_PTR(ret);
39279cc3
CM		 4816}
4817
4818static inline u8 btrfs_inode_type(struct inode *inode)
4819{
4820 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4821}
4822
d352ac68
CM		 4823/*
4824 * utility function to add 'inode' into 'parent_inode' with
4825 * a give name and a given sequence number.
4826 * if 'add_backref' is true, also insert a backref from the
4827 * inode to the parent directory.
4828 */
e02119d5
CM		 4829int btrfs_add_link(struct btrfs_trans_handle *trans,
4830 struct inode *parent_inode, struct inode *inode,
4831 const char *name, int name_len, int add_backref, u64 index)
39279cc3 4832{
4df27c4d 4833 int ret = 0;
39279cc3 4834 struct btrfs_key key;
e02119d5 4835 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
33345d01
LZ		 4836 u64 ino = btrfs_ino(inode);
4837 u64 parent_ino = btrfs_ino(parent_inode);
5f39d397 4838
33345d01 4839 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4840 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4841 } else {
33345d01 4842 key.objectid = ino;
4df27c4d
YZ		 4843 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4844 key.offset = 0;
4845 }
4846
33345d01 4847 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 4848 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4849 key.objectid, root->root_key.objectid,
33345d01 4850 parent_ino, index, name, name_len);
4df27c4d 4851 } else if (add_backref) {
33345d01
LZ		 4852 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4853 parent_ino, index);
4df27c4d 4854 }
39279cc3 4855
79787eaa
JM		 4856 /* Nothing to clean up yet */
4857 if (ret)
4858 return ret;
4df27c4d 4859
79787eaa
JM		 4860 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4861 parent_inode, &key,
4862 btrfs_inode_type(inode), index);
4863 if (ret == -EEXIST)
4864 goto fail_dir_item;
4865 else if (ret) {
4866 btrfs_abort_transaction(trans, root, ret);
4867 return ret;
39279cc3 4868 }
79787eaa
JM		 4869
4870 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4871 name_len * 2);
0c4d2d95 4872 inode_inc_iversion(parent_inode);
79787eaa
JM		 4873 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4874 ret = btrfs_update_inode(trans, root, parent_inode);
4875 if (ret)
4876 btrfs_abort_transaction(trans, root, ret);
39279cc3 4877 return ret;
fe66a05a
CM		 4878
4879fail_dir_item:
4880 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4881 u64 local_index;
4882 int err;
4883 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4884 key.objectid, root->root_key.objectid,
4885 parent_ino, &local_index, name, name_len);
4886
4887 } else if (add_backref) {
4888 u64 local_index;
4889 int err;
4890
4891 err = btrfs_del_inode_ref(trans, root, name, name_len,
4892 ino, parent_ino, &local_index);
4893 }
4894 return ret;
39279cc3
CM		 4895}
4896
4897static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
a1b075d2
JB		 4898 struct inode *dir, struct dentry *dentry,
4899 struct inode *inode, int backref, u64 index)
39279cc3 4900{
a1b075d2
JB		 4901 int err = btrfs_add_link(trans, dir, inode,
4902 dentry->d_name.name, dentry->d_name.len,
4903 backref, index);
39279cc3
CM		 4904 if (err > 0)
4905 err = -EEXIST;
4906 return err;
4907}
4908
618e21d5 4909static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 4910 umode_t mode, dev_t rdev)
618e21d5
JB		 4911{
4912 struct btrfs_trans_handle *trans;
4913 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4914 struct inode *inode = NULL;
618e21d5
JB		 4915 int err;
4916 int drop_inode = 0;
4917 u64 objectid;
1832a6d5 4918 unsigned long nr = 0;
00e4e6b3 4919 u64 index = 0;
618e21d5
JB		 4920
4921 if (!new_valid_dev(rdev))
4922 return -EINVAL;
4923
9ed74f2d
JB		 4924 /*
4925 * 2 for inode item and ref
4926 * 2 for dir items
4927 * 1 for xattr if selinux is on
4928 */
a22285a6
YZ		 4929 trans = btrfs_start_transaction(root, 5);
4930 if (IS_ERR(trans))
4931 return PTR_ERR(trans);
1832a6d5 4932
581bb050
LZ		 4933 err = btrfs_find_free_ino(root, &objectid);
4934 if (err)
4935 goto out_unlock;
4936
aec7477b 4937 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 4938 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 4939 mode, &index);
7cf96da3
TI		 4940 if (IS_ERR(inode)) {
4941 err = PTR_ERR(inode);
618e21d5 4942 goto out_unlock;
7cf96da3 4943 }
618e21d5 4944
2a7dba39 4945 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 4946 if (err) {
4947 drop_inode = 1;
4948 goto out_unlock;
4949 }
4950
ad19db71
CS		 4951 /*
4952 * If the active LSM wants to access the inode during
4953 * d_instantiate it needs these. Smack checks to see
4954 * if the filesystem supports xattrs by looking at the
4955 * ops vector.
4956 */
4957
4958 inode->i_op = &btrfs_special_inode_operations;
a1b075d2 4959 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
618e21d5
JB		 4960 if (err)
4961 drop_inode = 1;
4962 else {
618e21d5 4963 init_special_inode(inode, inode->i_mode, rdev);
1b4ab1bb 4964 btrfs_update_inode(trans, root, inode);
08c422c2 4965 d_instantiate(dentry, inode);
618e21d5 4966 }
618e21d5 4967out_unlock:
d3c2fdcf 4968 nr = trans->blocks_used;
7ad85bb7 4969 btrfs_end_transaction(trans, root);
a22285a6 4970 btrfs_btree_balance_dirty(root, nr);
618e21d5
JB		 4971 if (drop_inode) {
4972 inode_dec_link_count(inode);
4973 iput(inode);
4974 }
618e21d5
JB		 4975 return err;
4976}
4977
39279cc3 4978static int btrfs_create(struct inode *dir, struct dentry *dentry,
ebfc3b49 4979 umode_t mode, bool excl)
39279cc3
CM		 4980{
4981 struct btrfs_trans_handle *trans;
4982 struct btrfs_root *root = BTRFS_I(dir)->root;
1832a6d5 4983 struct inode *inode = NULL;
39279cc3 4984 int drop_inode = 0;
a22285a6 4985 int err;
1832a6d5 4986 unsigned long nr = 0;
39279cc3 4987 u64 objectid;
00e4e6b3 4988 u64 index = 0;
39279cc3 4989
9ed74f2d
JB		 4990 /*
4991 * 2 for inode item and ref
4992 * 2 for dir items
4993 * 1 for xattr if selinux is on
4994 */
a22285a6
YZ		 4995 trans = btrfs_start_transaction(root, 5);
4996 if (IS_ERR(trans))
4997 return PTR_ERR(trans);
9ed74f2d 4998
581bb050
LZ		 4999 err = btrfs_find_free_ino(root, &objectid);
5000 if (err)
5001 goto out_unlock;
5002
aec7477b 5003 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5004 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5005 mode, &index);
7cf96da3
TI		 5006 if (IS_ERR(inode)) {
5007 err = PTR_ERR(inode);
39279cc3 5008 goto out_unlock;
7cf96da3 5009 }
39279cc3 5010
2a7dba39 5011 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5012 if (err) {
5013 drop_inode = 1;
5014 goto out_unlock;
5015 }
5016
ad19db71
CS		 5017 /*
5018 * If the active LSM wants to access the inode during
5019 * d_instantiate it needs these. Smack checks to see
5020 * if the filesystem supports xattrs by looking at the
5021 * ops vector.
5022 */
5023 inode->i_fop = &btrfs_file_operations;
5024 inode->i_op = &btrfs_file_inode_operations;
5025
a1b075d2 5026 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 5027 if (err)
5028 drop_inode = 1;
5029 else {
5030 inode->i_mapping->a_ops = &btrfs_aops;
04160088 5031 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 5032 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
08c422c2 5033 d_instantiate(dentry, inode);
39279cc3 5034 }
39279cc3 5035out_unlock:
d3c2fdcf 5036 nr = trans->blocks_used;
7ad85bb7 5037 btrfs_end_transaction(trans, root);
39279cc3
CM		 5038 if (drop_inode) {
5039 inode_dec_link_count(inode);
5040 iput(inode);
5041 }
d3c2fdcf 5042 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 5043 return err;
5044}
5045
5046static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5047 struct dentry *dentry)
5048{
5049 struct btrfs_trans_handle *trans;
5050 struct btrfs_root *root = BTRFS_I(dir)->root;
5051 struct inode *inode = old_dentry->d_inode;
00e4e6b3 5052 u64 index;
1832a6d5 5053 unsigned long nr = 0;
39279cc3
CM		 5054 int err;
5055 int drop_inode = 0;
5056
4a8be425
TH		 5057 /* do not allow sys_link's with other subvols of the same device */
5058 if (root->objectid != BTRFS_I(inode)->root->objectid)
3ab3564f 5059 return -EXDEV;
4a8be425 5060
f186373f 5061 if (inode->i_nlink >= BTRFS_LINK_MAX)
c055e99e 5062 return -EMLINK;
4a8be425 5063
3de4586c 5064 err = btrfs_set_inode_index(dir, &index);
aec7477b
JB		 5065 if (err)
5066 goto fail;
5067
a22285a6 5068 /*
7e6b6465 5069 * 2 items for inode and inode ref
a22285a6 5070 * 2 items for dir items
7e6b6465 5071 * 1 item for parent inode
a22285a6 5072 */
7e6b6465 5073 trans = btrfs_start_transaction(root, 5);
a22285a6
YZ		 5074 if (IS_ERR(trans)) {
5075 err = PTR_ERR(trans);
5076 goto fail;
5077 }
5f39d397 5078
3153495d 5079 btrfs_inc_nlink(inode);
0c4d2d95 5080 inode_inc_iversion(inode);
3153495d 5081 inode->i_ctime = CURRENT_TIME;
7de9c6ee 5082 ihold(inode);
aec7477b 5083
a1b075d2 5084 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5f39d397 5085
a5719521 5086 if (err) {
54aa1f4d 5087 drop_inode = 1;
a5719521 5088 } else {
10d9f309 5089 struct dentry *parent = dentry->d_parent;
a5719521 5090 err = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 5091 if (err)
5092 goto fail;
08c422c2 5093 d_instantiate(dentry, inode);
6a912213 5094 btrfs_log_new_name(trans, inode, NULL, parent);
a5719521 5095 }
39279cc3 5096
d3c2fdcf 5097 nr = trans->blocks_used;
7ad85bb7 5098 btrfs_end_transaction(trans, root);
1832a6d5 5099fail:
39279cc3
CM		 5100 if (drop_inode) {
5101 inode_dec_link_count(inode);
5102 iput(inode);
5103 }
d3c2fdcf 5104 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 5105 return err;
5106}
5107
18bb1db3 5108static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
39279cc3 5109{
b9d86667 5110 struct inode *inode = NULL;
39279cc3
CM		 5111 struct btrfs_trans_handle *trans;
5112 struct btrfs_root *root = BTRFS_I(dir)->root;
5113 int err = 0;
5114 int drop_on_err = 0;
b9d86667 5115 u64 objectid = 0;
00e4e6b3 5116 u64 index = 0;
d3c2fdcf 5117 unsigned long nr = 1;
39279cc3 5118
9ed74f2d
JB		 5119 /*
5120 * 2 items for inode and ref
5121 * 2 items for dir items
5122 * 1 for xattr if selinux is on
5123 */
a22285a6
YZ		 5124 trans = btrfs_start_transaction(root, 5);
5125 if (IS_ERR(trans))
5126 return PTR_ERR(trans);
39279cc3 5127
581bb050
LZ		 5128 err = btrfs_find_free_ino(root, &objectid);
5129 if (err)
5130 goto out_fail;
5131
aec7477b 5132 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 5133 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 5134 S_IFDIR | mode, &index);
39279cc3
CM		 5135 if (IS_ERR(inode)) {
5136 err = PTR_ERR(inode);
5137 goto out_fail;
5138 }
5f39d397 5139
39279cc3 5140 drop_on_err = 1;
33268eaf 5141
2a7dba39 5142 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 5143 if (err)
5144 goto out_fail;
5145
39279cc3
CM		 5146 inode->i_op = &btrfs_dir_inode_operations;
5147 inode->i_fop = &btrfs_dir_file_operations;
39279cc3 5148
dbe674a9 5149 btrfs_i_size_write(inode, 0);
39279cc3
CM		 5150 err = btrfs_update_inode(trans, root, inode);
5151 if (err)
5152 goto out_fail;
5f39d397 5153
a1b075d2
JB		 5154 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5155 dentry->d_name.len, 0, index);
39279cc3
CM		 5156 if (err)
5157 goto out_fail;
5f39d397 5158
39279cc3
CM		 5159 d_instantiate(dentry, inode);
5160 drop_on_err = 0;
39279cc3
CM		 5161
5162out_fail:
d3c2fdcf 5163 nr = trans->blocks_used;
7ad85bb7 5164 btrfs_end_transaction(trans, root);
39279cc3
CM		 5165 if (drop_on_err)
5166 iput(inode);
d3c2fdcf 5167 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 5168 return err;
5169}
5170
d352ac68
CM		 5171/* helper for btfs_get_extent. Given an existing extent in the tree,
5172 * and an extent that you want to insert, deal with overlap and insert
5173 * the new extent into the tree.
5174 */
3b951516
CM		 5175static int merge_extent_mapping(struct extent_map_tree *em_tree,
5176 struct extent_map *existing,
e6dcd2dc
CM		 5177 struct extent_map *em,
5178 u64 map_start, u64 map_len)
3b951516
CM		 5179{
5180 u64 start_diff;
3b951516 5181
e6dcd2dc
CM		 5182 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5183 start_diff = map_start - em->start;
5184 em->start = map_start;
5185 em->len = map_len;
c8b97818
CM		 5186 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5187 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
e6dcd2dc 5188 em->block_start += start_diff;
c8b97818
CM		 5189 em->block_len -= start_diff;
5190 }
e6dcd2dc 5191 return add_extent_mapping(em_tree, em);
3b951516
CM		 5192}
5193
c8b97818
CM		 5194static noinline int uncompress_inline(struct btrfs_path *path,
5195 struct inode *inode, struct page *page,
5196 size_t pg_offset, u64 extent_offset,
5197 struct btrfs_file_extent_item *item)
5198{
5199 int ret;
5200 struct extent_buffer *leaf = path->nodes[0];
5201 char *tmp;
5202 size_t max_size;
5203 unsigned long inline_size;
5204 unsigned long ptr;
261507a0 5205 int compress_type;
c8b97818
CM		 5206
5207 WARN_ON(pg_offset != 0);
261507a0 5208 compress_type = btrfs_file_extent_compression(leaf, item);
c8b97818
CM		 5209 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5210 inline_size = btrfs_file_extent_inline_item_len(leaf,
5211 btrfs_item_nr(leaf, path->slots[0]));
5212 tmp = kmalloc(inline_size, GFP_NOFS);
8d413713
TI		 5213 if (!tmp)
5214 return -ENOMEM;
c8b97818
CM		 5215 ptr = btrfs_file_extent_inline_start(item);
5216
5217 read_extent_buffer(leaf, tmp, ptr, inline_size);
5218
5b050f04 5219 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
261507a0
LZ		 5220 ret = btrfs_decompress(compress_type, tmp, page,
5221 extent_offset, inline_size, max_size);
c8b97818 5222 if (ret) {
7ac687d9 5223 char *kaddr = kmap_atomic(page);
c8b97818
CM		 5224 unsigned long copy_size = min_t(u64,
5225 PAGE_CACHE_SIZE - pg_offset,
5226 max_size - extent_offset);
5227 memset(kaddr + pg_offset, 0, copy_size);
7ac687d9 5228 kunmap_atomic(kaddr);
c8b97818
CM		 5229 }
5230 kfree(tmp);
5231 return 0;
5232}
5233
d352ac68
CM		 5234/*
5235 * a bit scary, this does extent mapping from logical file offset to the disk.
d397712b
CM		 5236 * the ugly parts come from merging extents from the disk with the in-ram
5237 * representation. This gets more complex because of the data=ordered code,
d352ac68
CM		 5238 * where the in-ram extents might be locked pending data=ordered completion.
5239 *
5240 * This also copies inline extents directly into the page.
5241 */
d397712b 5242
a52d9a80 5243struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
70dec807 5244 size_t pg_offset, u64 start, u64 len,
a52d9a80
CM		 5245 int create)
5246{
5247 int ret;
5248 int err = 0;
db94535d 5249 u64 bytenr;
a52d9a80
CM		 5250 u64 extent_start = 0;
5251 u64 extent_end = 0;
33345d01 5252 u64 objectid = btrfs_ino(inode);
a52d9a80 5253 u32 found_type;
f421950f 5254 struct btrfs_path *path = NULL;
a52d9a80
CM		 5255 struct btrfs_root *root = BTRFS_I(inode)->root;
5256 struct btrfs_file_extent_item *item;
5f39d397
CM		 5257 struct extent_buffer *leaf;
5258 struct btrfs_key found_key;
a52d9a80
CM		 5259 struct extent_map *em = NULL;
5260 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
d1310b2e 5261 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
a52d9a80 5262 struct btrfs_trans_handle *trans = NULL;
261507a0 5263 int compress_type;
a52d9a80 5264
a52d9a80 5265again:
890871be 5266 read_lock(&em_tree->lock);
d1310b2e 5267 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 5268 if (em)
5269 em->bdev = root->fs_info->fs_devices->latest_bdev;
890871be 5270 read_unlock(&em_tree->lock);
d1310b2e 5271
a52d9a80 5272 if (em) {
e1c4b745
CM		 5273 if (em->start > start || em->start + em->len <= start)
5274 free_extent_map(em);
5275 else if (em->block_start == EXTENT_MAP_INLINE && page)
70dec807
CM		 5276 free_extent_map(em);
5277 else
5278 goto out;
a52d9a80 5279 }
172ddd60 5280 em = alloc_extent_map();
a52d9a80 5281 if (!em) {
d1310b2e
CM		 5282 err = -ENOMEM;
5283 goto out;
a52d9a80 5284 }
e6dcd2dc 5285 em->bdev = root->fs_info->fs_devices->latest_bdev;
d1310b2e 5286 em->start = EXTENT_MAP_HOLE;
445a6944 5287 em->orig_start = EXTENT_MAP_HOLE;
d1310b2e 5288 em->len = (u64)-1;
c8b97818 5289 em->block_len = (u64)-1;
f421950f
CM		 5290
5291 if (!path) {
5292 path = btrfs_alloc_path();
026fd317
JB		 5293 if (!path) {
5294 err = -ENOMEM;
5295 goto out;
5296 }
5297 /*
5298 * Chances are we'll be called again, so go ahead and do
5299 * readahead
5300 */
5301 path->reada = 1;
f421950f
CM		 5302 }
5303
179e29e4
CM		 5304 ret = btrfs_lookup_file_extent(trans, root, path,
5305 objectid, start, trans != NULL);
a52d9a80
CM		 5306 if (ret < 0) {
5307 err = ret;
5308 goto out;
5309 }
5310
5311 if (ret != 0) {
5312 if (path->slots[0] == 0)
5313 goto not_found;
5314 path->slots[0]--;
5315 }
5316
5f39d397
CM		 5317 leaf = path->nodes[0];
5318 item = btrfs_item_ptr(leaf, path->slots[0],
a52d9a80 5319 struct btrfs_file_extent_item);
a52d9a80 5320 /* are we inside the extent that was found? */
5f39d397
CM		 5321 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5322 found_type = btrfs_key_type(&found_key);
5323 if (found_key.objectid != objectid ||
a52d9a80
CM		 5324 found_type != BTRFS_EXTENT_DATA_KEY) {
5325 goto not_found;
5326 }
5327
5f39d397
CM		 5328 found_type = btrfs_file_extent_type(leaf, item);
5329 extent_start = found_key.offset;
261507a0 5330 compress_type = btrfs_file_extent_compression(leaf, item);
d899e052
YZ		 5331 if (found_type == BTRFS_FILE_EXTENT_REG ||
5332 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
a52d9a80 5333 extent_end = extent_start +
db94535d 5334 btrfs_file_extent_num_bytes(leaf, item);
9036c102
YZ		 5335 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5336 size_t size;
5337 size = btrfs_file_extent_inline_len(leaf, item);
5338 extent_end = (extent_start + size + root->sectorsize - 1) &
5339 ~((u64)root->sectorsize - 1);
5340 }
5341
5342 if (start >= extent_end) {
5343 path->slots[0]++;
5344 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5345 ret = btrfs_next_leaf(root, path);
5346 if (ret < 0) {
5347 err = ret;
5348 goto out;
a52d9a80 5349 }
9036c102
YZ		 5350 if (ret > 0)
5351 goto not_found;
5352 leaf = path->nodes[0];
a52d9a80 5353 }
9036c102
YZ		 5354 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5355 if (found_key.objectid != objectid ||
5356 found_key.type != BTRFS_EXTENT_DATA_KEY)
5357 goto not_found;
5358 if (start + len <= found_key.offset)
5359 goto not_found;
5360 em->start = start;
5361 em->len = found_key.offset - start;
5362 goto not_found_em;
5363 }
5364
d899e052
YZ		 5365 if (found_type == BTRFS_FILE_EXTENT_REG ||
5366 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
9036c102
YZ		 5367 em->start = extent_start;
5368 em->len = extent_end - extent_start;
ff5b7ee3
YZ		 5369 em->orig_start = extent_start -
5370 btrfs_file_extent_offset(leaf, item);
db94535d
CM		 5371 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5372 if (bytenr == 0) {
5f39d397 5373 em->block_start = EXTENT_MAP_HOLE;
a52d9a80
CM		 5374 goto insert;
5375 }
261507a0 5376 if (compress_type != BTRFS_COMPRESS_NONE) {
c8b97818 5377 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0 5378 em->compress_type = compress_type;
c8b97818
CM		 5379 em->block_start = bytenr;
5380 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5381 item);
5382 } else {
5383 bytenr += btrfs_file_extent_offset(leaf, item);
5384 em->block_start = bytenr;
5385 em->block_len = em->len;
d899e052
YZ		 5386 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5387 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
c8b97818 5388 }
a52d9a80
CM		 5389 goto insert;
5390 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5f39d397 5391 unsigned long ptr;
a52d9a80 5392 char *map;
3326d1b0
CM		 5393 size_t size;
5394 size_t extent_offset;
5395 size_t copy_size;
a52d9a80 5396
689f9346 5397 em->block_start = EXTENT_MAP_INLINE;
c8b97818 5398 if (!page || create) {
689f9346 5399 em->start = extent_start;
9036c102 5400 em->len = extent_end - extent_start;
689f9346
Y		 5401 goto out;
5402 }
5f39d397 5403
9036c102
YZ		 5404 size = btrfs_file_extent_inline_len(leaf, item);
5405 extent_offset = page_offset(page) + pg_offset - extent_start;
70dec807 5406 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3326d1b0 5407 size - extent_offset);
3326d1b0 5408 em->start = extent_start + extent_offset;
70dec807
CM		 5409 em->len = (copy_size + root->sectorsize - 1) &
5410 ~((u64)root->sectorsize - 1);
ff5b7ee3 5411 em->orig_start = EXTENT_MAP_INLINE;
261507a0 5412 if (compress_type) {
c8b97818 5413 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
261507a0
LZ		 5414 em->compress_type = compress_type;
5415 }
689f9346 5416 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
179e29e4 5417 if (create == 0 && !PageUptodate(page)) {
261507a0
LZ		 5418 if (btrfs_file_extent_compression(leaf, item) !=
5419 BTRFS_COMPRESS_NONE) {
c8b97818
CM		 5420 ret = uncompress_inline(path, inode, page,
5421 pg_offset,
5422 extent_offset, item);
79787eaa 5423 BUG_ON(ret); /* -ENOMEM */
c8b97818
CM		 5424 } else {
5425 map = kmap(page);
5426 read_extent_buffer(leaf, map + pg_offset, ptr,
5427 copy_size);
93c82d57
CM		 5428 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5429 memset(map + pg_offset + copy_size, 0,
5430 PAGE_CACHE_SIZE - pg_offset -
5431 copy_size);
5432 }
c8b97818
CM		 5433 kunmap(page);
5434 }
179e29e4
CM		 5435 flush_dcache_page(page);
5436 } else if (create && PageUptodate(page)) {
6bf7e080 5437 BUG();
179e29e4
CM		 5438 if (!trans) {
5439 kunmap(page);
5440 free_extent_map(em);
5441 em = NULL;
ff5714cc 5442
b3b4aa74 5443 btrfs_release_path(path);
7a7eaa40 5444 trans = btrfs_join_transaction(root);
ff5714cc 5445
3612b495
TI		 5446 if (IS_ERR(trans))
5447 return ERR_CAST(trans);
179e29e4
CM		 5448 goto again;
5449 }
c8b97818 5450 map = kmap(page);
70dec807 5451 write_extent_buffer(leaf, map + pg_offset, ptr,
179e29e4 5452 copy_size);
c8b97818 5453 kunmap(page);
179e29e4 5454 btrfs_mark_buffer_dirty(leaf);
a52d9a80 5455 }
d1310b2e 5456 set_extent_uptodate(io_tree, em->start,
507903b8 5457 extent_map_end(em) - 1, NULL, GFP_NOFS);
a52d9a80
CM		 5458 goto insert;
5459 } else {
31b1a2bd 5460 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
a52d9a80
CM		 5461 }
5462not_found:
5463 em->start = start;
d1310b2e 5464 em->len = len;
a52d9a80 5465not_found_em:
5f39d397 5466 em->block_start = EXTENT_MAP_HOLE;
9036c102 5467 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
a52d9a80 5468insert:
b3b4aa74 5469 btrfs_release_path(path);
d1310b2e 5470 if (em->start > start || extent_map_end(em) <= start) {
d397712b
CM		 5471 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5472 "[%llu %llu]\n", (unsigned long long)em->start,
5473 (unsigned long long)em->len,
5474 (unsigned long long)start,
5475 (unsigned long long)len);
a52d9a80
CM		 5476 err = -EIO;
5477 goto out;
5478 }
d1310b2e
CM		 5479
5480 err = 0;
890871be 5481 write_lock(&em_tree->lock);
a52d9a80 5482 ret = add_extent_mapping(em_tree, em);
3b951516
CM		 5483 /* it is possible that someone inserted the extent into the tree
5484 * while we had the lock dropped. It is also possible that
5485 * an overlapping map exists in the tree
5486 */
a52d9a80 5487 if (ret == -EEXIST) {
3b951516 5488 struct extent_map *existing;
e6dcd2dc
CM		 5489
5490 ret = 0;
5491
3b951516 5492 existing = lookup_extent_mapping(em_tree, start, len);
e1c4b745
CM		 5493 if (existing && (existing->start > start ||
5494 existing->start + existing->len <= start)) {
5495 free_extent_map(existing);
5496 existing = NULL;
5497 }
3b951516
CM		 5498 if (!existing) {
5499 existing = lookup_extent_mapping(em_tree, em->start,
5500 em->len);
5501 if (existing) {
5502 err = merge_extent_mapping(em_tree, existing,
e6dcd2dc
CM		 5503 em, start,
5504 root->sectorsize);
3b951516
CM		 5505 free_extent_map(existing);
5506 if (err) {
5507 free_extent_map(em);
5508 em = NULL;
5509 }
5510 } else {
5511 err = -EIO;
3b951516
CM		 5512 free_extent_map(em);
5513 em = NULL;
5514 }
5515 } else {
5516 free_extent_map(em);
5517 em = existing;
e6dcd2dc 5518 err = 0;
a52d9a80 5519 }
a52d9a80 5520 }
890871be 5521 write_unlock(&em_tree->lock);
a52d9a80 5522out:
1abe9b8a 5523
f0bd95ea
TI		 5524 if (em)
5525 trace_btrfs_get_extent(root, em);
1abe9b8a 5526
f421950f
CM		 5527 if (path)
5528 btrfs_free_path(path);
a52d9a80
CM		 5529 if (trans) {
5530 ret = btrfs_end_transaction(trans, root);
d397712b 5531 if (!err)
a52d9a80
CM		 5532 err = ret;
5533 }
a52d9a80
CM		 5534 if (err) {
5535 free_extent_map(em);
a52d9a80
CM		 5536 return ERR_PTR(err);
5537 }
79787eaa 5538 BUG_ON(!em); /* Error is always set */
a52d9a80
CM		 5539 return em;
5540}
5541
ec29ed5b
CM		 5542struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5543 size_t pg_offset, u64 start, u64 len,
5544 int create)
5545{
5546 struct extent_map *em;
5547 struct extent_map *hole_em = NULL;
5548 u64 range_start = start;
5549 u64 end;
5550 u64 found;
5551 u64 found_end;
5552 int err = 0;
5553
5554 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5555 if (IS_ERR(em))
5556 return em;
5557 if (em) {
5558 /*
5559 * if our em maps to a hole, there might
5560 * actually be delalloc bytes behind it
5561 */
5562 if (em->block_start != EXTENT_MAP_HOLE)
5563 return em;
5564 else
5565 hole_em = em;
5566 }
5567
5568 /* check to see if we've wrapped (len == -1 or similar) */
5569 end = start + len;
5570 if (end < start)
5571 end = (u64)-1;
5572 else
5573 end -= 1;
5574
5575 em = NULL;
5576
5577 /* ok, we didn't find anything, lets look for delalloc */
5578 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5579 end, len, EXTENT_DELALLOC, 1);
5580 found_end = range_start + found;
5581 if (found_end < range_start)
5582 found_end = (u64)-1;
5583
5584 /*
5585 * we didn't find anything useful, return
5586 * the original results from get_extent()
5587 */
5588 if (range_start > end || found_end <= start) {
5589 em = hole_em;
5590 hole_em = NULL;
5591 goto out;
5592 }
5593
5594 /* adjust the range_start to make sure it doesn't
5595 * go backwards from the start they passed in
5596 */
5597 range_start = max(start,range_start);
5598 found = found_end - range_start;
5599
5600 if (found > 0) {
5601 u64 hole_start = start;
5602 u64 hole_len = len;
5603
172ddd60 5604 em = alloc_extent_map();
ec29ed5b
CM		 5605 if (!em) {
5606 err = -ENOMEM;
5607 goto out;
5608 }
5609 /*
5610 * when btrfs_get_extent can't find anything it
5611 * returns one huge hole
5612 *
5613 * make sure what it found really fits our range, and
5614 * adjust to make sure it is based on the start from
5615 * the caller
5616 */
5617 if (hole_em) {
5618 u64 calc_end = extent_map_end(hole_em);
5619
5620 if (calc_end <= start || (hole_em->start > end)) {
5621 free_extent_map(hole_em);
5622 hole_em = NULL;
5623 } else {
5624 hole_start = max(hole_em->start, start);
5625 hole_len = calc_end - hole_start;
5626 }
5627 }
5628 em->bdev = NULL;
5629 if (hole_em && range_start > hole_start) {
5630 /* our hole starts before our delalloc, so we
5631 * have to return just the parts of the hole
5632 * that go until the delalloc starts
5633 */
5634 em->len = min(hole_len,
5635 range_start - hole_start);
5636 em->start = hole_start;
5637 em->orig_start = hole_start;
5638 /*
5639 * don't adjust block start at all,
5640 * it is fixed at EXTENT_MAP_HOLE
5641 */
5642 em->block_start = hole_em->block_start;
5643 em->block_len = hole_len;
5644 } else {
5645 em->start = range_start;
5646 em->len = found;
5647 em->orig_start = range_start;
5648 em->block_start = EXTENT_MAP_DELALLOC;
5649 em->block_len = found;
5650 }
5651 } else if (hole_em) {
5652 return hole_em;
5653 }
5654out:
5655
5656 free_extent_map(hole_em);
5657 if (err) {
5658 free_extent_map(em);
5659 return ERR_PTR(err);
5660 }
5661 return em;
5662}
5663
4b46fce2 5664static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
16d299ac 5665 struct extent_map *em,
4b46fce2
JB		 5666 u64 start, u64 len)
5667{
5668 struct btrfs_root *root = BTRFS_I(inode)->root;
5669 struct btrfs_trans_handle *trans;
4b46fce2
JB		 5670 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5671 struct btrfs_key ins;
5672 u64 alloc_hint;
5673 int ret;
16d299ac 5674 bool insert = false;
4b46fce2 5675
16d299ac
JB		 5676 /*
5677 * Ok if the extent map we looked up is a hole and is for the exact
5678 * range we want, there is no reason to allocate a new one, however if
5679 * it is not right then we need to free this one and drop the cache for
5680 * our range.
5681 */
5682 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5683 em->len != len) {
5684 free_extent_map(em);
5685 em = NULL;
5686 insert = true;
5687 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5688 }
4b46fce2 5689
7a7eaa40 5690 trans = btrfs_join_transaction(root);
3612b495
TI		 5691 if (IS_ERR(trans))
5692 return ERR_CAST(trans);
4b46fce2 5693
4cb5300b
CM		 5694 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5695 btrfs_add_inode_defrag(trans, inode);
5696
4b46fce2
JB		 5697 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5698
5699 alloc_hint = get_extent_allocation_hint(inode, start, len);
5700 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
81c9ad23 5701 alloc_hint, &ins, 1);
4b46fce2
JB		 5702 if (ret) {
5703 em = ERR_PTR(ret);
5704 goto out;
5705 }
5706
4b46fce2 5707 if (!em) {
172ddd60 5708 em = alloc_extent_map();
16d299ac
JB		 5709 if (!em) {
5710 em = ERR_PTR(-ENOMEM);
5711 goto out;
5712 }
4b46fce2
JB		 5713 }
5714
5715 em->start = start;
5716 em->orig_start = em->start;
5717 em->len = ins.offset;
5718
5719 em->block_start = ins.objectid;
5720 em->block_len = ins.offset;
5721 em->bdev = root->fs_info->fs_devices->latest_bdev;
16d299ac
JB		 5722
5723 /*
5724 * We need to do this because if we're using the original em we searched
5725 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5726 */
5727 em->flags = 0;
4b46fce2
JB		 5728 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5729
16d299ac 5730 while (insert) {
4b46fce2
JB		 5731 write_lock(&em_tree->lock);
5732 ret = add_extent_mapping(em_tree, em);
5733 write_unlock(&em_tree->lock);
5734 if (ret != -EEXIST)
5735 break;
5736 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5737 }
5738
5739 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5740 ins.offset, ins.offset, 0);
5741 if (ret) {
5742 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5743 em = ERR_PTR(ret);
5744 }
5745out:
5746 btrfs_end_transaction(trans, root);
5747 return em;
5748}
5749
46bfbb5c
CM		 5750/*
5751 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5752 * block must be cow'd
5753 */
5754static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5755 struct inode *inode, u64 offset, u64 len)
5756{
5757 struct btrfs_path *path;
5758 int ret;
5759 struct extent_buffer *leaf;
5760 struct btrfs_root *root = BTRFS_I(inode)->root;
5761 struct btrfs_file_extent_item *fi;
5762 struct btrfs_key key;
5763 u64 disk_bytenr;
5764 u64 backref_offset;
5765 u64 extent_end;
5766 u64 num_bytes;
5767 int slot;
5768 int found_type;
5769
5770 path = btrfs_alloc_path();
5771 if (!path)
5772 return -ENOMEM;
5773
33345d01 5774 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
46bfbb5c
CM		 5775 offset, 0);
5776 if (ret < 0)
5777 goto out;
5778
5779 slot = path->slots[0];
5780 if (ret == 1) {
5781 if (slot == 0) {
5782 /* can't find the item, must cow */
5783 ret = 0;
5784 goto out;
5785 }
5786 slot--;
5787 }
5788 ret = 0;
5789 leaf = path->nodes[0];
5790 btrfs_item_key_to_cpu(leaf, &key, slot);
33345d01 5791 if (key.objectid != btrfs_ino(inode) ||
46bfbb5c
CM		 5792 key.type != BTRFS_EXTENT_DATA_KEY) {
5793 /* not our file or wrong item type, must cow */
5794 goto out;
5795 }
5796
5797 if (key.offset > offset) {
5798 /* Wrong offset, must cow */
5799 goto out;
5800 }
5801
5802 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5803 found_type = btrfs_file_extent_type(leaf, fi);
5804 if (found_type != BTRFS_FILE_EXTENT_REG &&
5805 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5806 /* not a regular extent, must cow */
5807 goto out;
5808 }
5809 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5810 backref_offset = btrfs_file_extent_offset(leaf, fi);
5811
5812 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5813 if (extent_end < offset + len) {
5814 /* extent doesn't include our full range, must cow */
5815 goto out;
5816 }
5817
5818 if (btrfs_extent_readonly(root, disk_bytenr))
5819 goto out;
5820
5821 /*
5822 * look for other files referencing this extent, if we
5823 * find any we must cow
5824 */
33345d01 5825 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
46bfbb5c
CM		 5826 key.offset - backref_offset, disk_bytenr))
5827 goto out;
5828
5829 /*
5830 * adjust disk_bytenr and num_bytes to cover just the bytes
5831 * in this extent we are about to write. If there
5832 * are any csums in that range we have to cow in order
5833 * to keep the csums correct
5834 */
5835 disk_bytenr += backref_offset;
5836 disk_bytenr += offset - key.offset;
5837 num_bytes = min(offset + len, extent_end) - offset;
5838 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5839 goto out;
5840 /*
5841 * all of the above have passed, it is safe to overwrite this extent
5842 * without cow
5843 */
5844 ret = 1;
5845out:
5846 btrfs_free_path(path);
5847 return ret;
5848}
5849
eb838e73
JB		 5850static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5851 struct extent_state **cached_state, int writing)
5852{
5853 struct btrfs_ordered_extent *ordered;
5854 int ret = 0;
5855
5856 while (1) {
5857 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5858 0, cached_state);
5859 /*
5860 * We're concerned with the entire range that we're going to be
5861 * doing DIO to, so we need to make sure theres no ordered
5862 * extents in this range.
5863 */
5864 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5865 lockend - lockstart + 1);
5866
5867 /*
5868 * We need to make sure there are no buffered pages in this
5869 * range either, we could have raced between the invalidate in
5870 * generic_file_direct_write and locking the extent. The
5871 * invalidate needs to happen so that reads after a write do not
5872 * get stale data.
5873 */
5874 if (!ordered && (!writing ||
5875 !test_range_bit(&BTRFS_I(inode)->io_tree,
5876 lockstart, lockend, EXTENT_UPTODATE, 0,
5877 *cached_state)))
5878 break;
5879
5880 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5881 cached_state, GFP_NOFS);
5882
5883 if (ordered) {
5884 btrfs_start_ordered_extent(inode, ordered, 1);
5885 btrfs_put_ordered_extent(ordered);
5886 } else {
5887 /* Screw you mmap */
5888 ret = filemap_write_and_wait_range(inode->i_mapping,
5889 lockstart,
5890 lockend);
5891 if (ret)
5892 break;
5893
5894 /*
5895 * If we found a page that couldn't be invalidated just
5896 * fall back to buffered.
5897 */
5898 ret = invalidate_inode_pages2_range(inode->i_mapping,
5899 lockstart >> PAGE_CACHE_SHIFT,
5900 lockend >> PAGE_CACHE_SHIFT);
5901 if (ret)
5902 break;
5903 }
5904
5905 cond_resched();
5906 }
5907
5908 return ret;
5909}
5910
69ffb543
JB		 5911static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5912 u64 len, u64 orig_start,
5913 u64 block_start, u64 block_len,
5914 int type)
5915{
5916 struct extent_map_tree *em_tree;
5917 struct extent_map *em;
5918 struct btrfs_root *root = BTRFS_I(inode)->root;
5919 int ret;
5920
5921 em_tree = &BTRFS_I(inode)->extent_tree;
5922 em = alloc_extent_map();
5923 if (!em)
5924 return ERR_PTR(-ENOMEM);
5925
5926 em->start = start;
5927 em->orig_start = orig_start;
5928 em->len = len;
5929 em->block_len = block_len;
5930 em->block_start = block_start;
5931 em->bdev = root->fs_info->fs_devices->latest_bdev;
5932 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5933 if (type == BTRFS_ORDERED_PREALLOC)
5934 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5935
5936 do {
5937 btrfs_drop_extent_cache(inode, em->start,
5938 em->start + em->len - 1, 0);
5939 write_lock(&em_tree->lock);
5940 ret = add_extent_mapping(em_tree, em);
5941 write_unlock(&em_tree->lock);
5942 } while (ret == -EEXIST);
5943
5944 if (ret) {
5945 free_extent_map(em);
5946 return ERR_PTR(ret);
5947 }
5948
5949 return em;
5950}
5951
5952
4b46fce2
JB		 5953static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5954 struct buffer_head *bh_result, int create)
5955{
5956 struct extent_map *em;
5957 struct btrfs_root *root = BTRFS_I(inode)->root;
eb838e73 5958 struct extent_state *cached_state = NULL;
4b46fce2 5959 u64 start = iblock << inode->i_blkbits;
eb838e73 5960 u64 lockstart, lockend;
4b46fce2 5961 u64 len = bh_result->b_size;
46bfbb5c 5962 struct btrfs_trans_handle *trans;
eb838e73
JB		 5963 int unlock_bits = EXTENT_LOCKED;
5964 int ret;
5965
eb838e73
JB		 5966 if (create) {
5967 ret = btrfs_delalloc_reserve_space(inode, len);
5968 if (ret)
5969 return ret;
5970 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
c329861d
JB		 5971 } else {
5972 len = min_t(u64, len, root->sectorsize);
eb838e73
JB		 5973 }
5974
c329861d
JB		 5975 lockstart = start;
5976 lockend = start + len - 1;
5977
eb838e73
JB		 5978 /*
5979 * If this errors out it's because we couldn't invalidate pagecache for
5980 * this range and we need to fallback to buffered.
5981 */
5982 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5983 return -ENOTBLK;
5984
5985 if (create) {
5986 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5987 lockend, EXTENT_DELALLOC, NULL,
5988 &cached_state, GFP_NOFS);
5989 if (ret)
5990 goto unlock_err;
5991 }
4b46fce2
JB		 5992
5993 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
eb838e73
JB		 5994 if (IS_ERR(em)) {
5995 ret = PTR_ERR(em);
5996 goto unlock_err;
5997 }
4b46fce2
JB		 5998
5999 /*
6000 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6001 * io. INLINE is special, and we could probably kludge it in here, but
6002 * it's still buffered so for safety lets just fall back to the generic
6003 * buffered path.
6004 *
6005 * For COMPRESSED we _have_ to read the entire extent in so we can
6006 * decompress it, so there will be buffering required no matter what we
6007 * do, so go ahead and fallback to buffered.
6008 *
6009 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6010 * to buffered IO. Don't blame me, this is the price we pay for using
6011 * the generic code.
6012 */
6013 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6014 em->block_start == EXTENT_MAP_INLINE) {
6015 free_extent_map(em);
eb838e73
JB		 6016 ret = -ENOTBLK;
6017 goto unlock_err;
4b46fce2
JB		 6018 }
6019
6020 /* Just a good old fashioned hole, return */
6021 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6022 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6023 free_extent_map(em);
eb838e73
JB		 6024 ret = 0;
6025 goto unlock_err;
4b46fce2
JB		 6026 }
6027
6028 /*
6029 * We don't allocate a new extent in the following cases
6030 *
6031 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6032 * existing extent.
6033 * 2) The extent is marked as PREALLOC. We're good to go here and can
6034 * just use the extent.
6035 *
6036 */
46bfbb5c 6037 if (!create) {
eb838e73
JB		 6038 len = min(len, em->len - (start - em->start));
6039 lockstart = start + len;
6040 goto unlock;
46bfbb5c 6041 }
4b46fce2
JB		 6042
6043 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6044 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6045 em->block_start != EXTENT_MAP_HOLE)) {
4b46fce2
JB		 6046 int type;
6047 int ret;
46bfbb5c 6048 u64 block_start;
4b46fce2
JB		 6049
6050 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6051 type = BTRFS_ORDERED_PREALLOC;
6052 else
6053 type = BTRFS_ORDERED_NOCOW;
46bfbb5c 6054 len = min(len, em->len - (start - em->start));
4b46fce2 6055 block_start = em->block_start + (start - em->start);
46bfbb5c
CM		 6056
6057 /*
6058 * we're not going to log anything, but we do need
6059 * to make sure the current transaction stays open
6060 * while we look for nocow cross refs
6061 */
7a7eaa40 6062 trans = btrfs_join_transaction(root);
3612b495 6063 if (IS_ERR(trans))
46bfbb5c
CM		 6064 goto must_cow;
6065
6066 if (can_nocow_odirect(trans, inode, start, len) == 1) {
69ffb543
JB		 6067 u64 orig_start = em->start;
6068
6069 if (type == BTRFS_ORDERED_PREALLOC) {
6070 free_extent_map(em);
6071 em = create_pinned_em(inode, start, len,
6072 orig_start,
6073 block_start, len, type);
6074 if (IS_ERR(em)) {
6075 btrfs_end_transaction(trans, root);
6076 goto unlock_err;
6077 }
6078 }
6079
46bfbb5c
CM		 6080 ret = btrfs_add_ordered_extent_dio(inode, start,
6081 block_start, len, len, type);
6082 btrfs_end_transaction(trans, root);
6083 if (ret) {
6084 free_extent_map(em);
eb838e73 6085 goto unlock_err;
46bfbb5c
CM		 6086 }
6087 goto unlock;
4b46fce2 6088 }
46bfbb5c 6089 btrfs_end_transaction(trans, root);
4b46fce2 6090 }
46bfbb5c
CM		 6091must_cow:
6092 /*
6093 * this will cow the extent, reset the len in case we changed
6094 * it above
6095 */
6096 len = bh_result->b_size;
16d299ac 6097 em = btrfs_new_extent_direct(inode, em, start, len);
eb838e73
JB		 6098 if (IS_ERR(em)) {
6099 ret = PTR_ERR(em);
6100 goto unlock_err;
6101 }
46bfbb5c
CM		 6102 len = min(len, em->len - (start - em->start));
6103unlock:
4b46fce2
JB		 6104 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6105 inode->i_blkbits;
46bfbb5c 6106 bh_result->b_size = len;
4b46fce2
JB		 6107 bh_result->b_bdev = em->bdev;
6108 set_buffer_mapped(bh_result);
c3473e83
JB		 6109 if (create) {
6110 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6111 set_buffer_new(bh_result);
6112
6113 /*
6114 * Need to update the i_size under the extent lock so buffered
6115 * readers will get the updated i_size when we unlock.
6116 */
6117 if (start + len > i_size_read(inode))
6118 i_size_write(inode, start + len);
6119 }
4b46fce2 6120
eb838e73
JB		 6121 /*
6122 * In the case of write we need to clear and unlock the entire range,
6123 * in the case of read we need to unlock only the end area that we
6124 * aren't using if there is any left over space.
6125 */
24c03fa5
LB		 6126 if (lockstart < lockend) {
6127 if (create && len < lockend - lockstart) {
6128 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
9e8a4a8b
LB		 6129 lockstart + len - 1,
6130 unlock_bits | EXTENT_DEFRAG, 1, 0,
24c03fa5
LB		 6131 &cached_state, GFP_NOFS);
6132 /*
6133 * Beside unlock, we also need to cleanup reserved space
6134 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6135 */
6136 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6137 lockstart + len, lockend,
9e8a4a8b
LB		 6138 unlock_bits | EXTENT_DO_ACCOUNTING |
6139 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
24c03fa5
LB		 6140 } else {
6141 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6142 lockend, unlock_bits, 1, 0,
6143 &cached_state, GFP_NOFS);
6144 }
6145 } else {
eb838e73 6146 free_extent_state(cached_state);
24c03fa5 6147 }
eb838e73 6148
4b46fce2
JB		 6149 free_extent_map(em);
6150
6151 return 0;
eb838e73
JB		 6152
6153unlock_err:
6154 if (create)
6155 unlock_bits |= EXTENT_DO_ACCOUNTING;
6156
6157 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6158 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6159 return ret;
4b46fce2
JB		 6160}
6161
6162struct btrfs_dio_private {
6163 struct inode *inode;
6164 u64 logical_offset;
6165 u64 disk_bytenr;
6166 u64 bytes;
4b46fce2 6167 void *private;
e65e1535
MX		 6168
6169 /* number of bios pending for this dio */
6170 atomic_t pending_bios;
6171
6172 /* IO errors */
6173 int errors;
6174
6175 struct bio *orig_bio;
4b46fce2
JB		 6176};
6177
6178static void btrfs_endio_direct_read(struct bio *bio, int err)
6179{
e65e1535 6180 struct btrfs_dio_private *dip = bio->bi_private;
4b46fce2
JB		 6181 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6182 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2
JB		 6183 struct inode *inode = dip->inode;
6184 struct btrfs_root *root = BTRFS_I(inode)->root;
6185 u64 start;
4b46fce2
JB		 6186
6187 start = dip->logical_offset;
6188 do {
6189 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6190 struct page *page = bvec->bv_page;
6191 char *kaddr;
6192 u32 csum = ~(u32)0;
c329861d 6193 u64 private = ~(u32)0;
4b46fce2
JB		 6194 unsigned long flags;
6195
c329861d
JB		 6196 if (get_state_private(&BTRFS_I(inode)->io_tree,
6197 start, &private))
6198 goto failed;
4b46fce2 6199 local_irq_save(flags);
7ac687d9 6200 kaddr = kmap_atomic(page);
4b46fce2
JB		 6201 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6202 csum, bvec->bv_len);
6203 btrfs_csum_final(csum, (char *)&csum);
7ac687d9 6204 kunmap_atomic(kaddr);
4b46fce2
JB		 6205 local_irq_restore(flags);
6206
6207 flush_dcache_page(bvec->bv_page);
c329861d
JB		 6208 if (csum != private) {
6209failed:
33345d01 6210 printk(KERN_ERR "btrfs csum failed ino %llu off"
4b46fce2 6211 " %llu csum %u private %u\n",
33345d01
LZ		 6212 (unsigned long long)btrfs_ino(inode),
6213 (unsigned long long)start,
c329861d 6214 csum, (unsigned)private);
4b46fce2
JB		 6215 err = -EIO;
6216 }
6217 }
6218
6219 start += bvec->bv_len;
4b46fce2
JB		 6220 bvec++;
6221 } while (bvec <= bvec_end);
6222
6223 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
d0082371 6224 dip->logical_offset + dip->bytes - 1);
4b46fce2
JB		 6225 bio->bi_private = dip->private;
6226
4b46fce2 6227 kfree(dip);
c0da7aa1
JB		 6228
6229 /* If we had a csum failure make sure to clear the uptodate flag */
6230 if (err)
6231 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 6232 dio_end_io(bio, err);
6233}
6234
6235static void btrfs_endio_direct_write(struct bio *bio, int err)
6236{
6237 struct btrfs_dio_private *dip = bio->bi_private;
6238 struct inode *inode = dip->inode;
6239 struct btrfs_root *root = BTRFS_I(inode)->root;
4b46fce2 6240 struct btrfs_ordered_extent *ordered = NULL;
163cf09c
CM		 6241 u64 ordered_offset = dip->logical_offset;
6242 u64 ordered_bytes = dip->bytes;
4b46fce2
JB		 6243 int ret;
6244
6245 if (err)
6246 goto out_done;
163cf09c
CM		 6247again:
6248 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6249 &ordered_offset,
5fd02043 6250 ordered_bytes, !err);
4b46fce2 6251 if (!ret)
163cf09c 6252 goto out_test;
4b46fce2 6253
5fd02043
JB		 6254 ordered->work.func = finish_ordered_fn;
6255 ordered->work.flags = 0;
6256 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6257 &ordered->work);
163cf09c
CM		 6258out_test:
6259 /*
6260 * our bio might span multiple ordered extents. If we haven't
6261 * completed the accounting for the whole dio, go back and try again
6262 */
6263 if (ordered_offset < dip->logical_offset + dip->bytes) {
6264 ordered_bytes = dip->logical_offset + dip->bytes -
6265 ordered_offset;
5fd02043 6266 ordered = NULL;
163cf09c
CM		 6267 goto again;
6268 }
4b46fce2
JB		 6269out_done:
6270 bio->bi_private = dip->private;
6271
4b46fce2 6272 kfree(dip);
c0da7aa1
JB		 6273
6274 /* If we had an error make sure to clear the uptodate flag */
6275 if (err)
6276 clear_bit(BIO_UPTODATE, &bio->bi_flags);
4b46fce2
JB		 6277 dio_end_io(bio, err);
6278}
6279
eaf25d93
CM		 6280static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6281 struct bio *bio, int mirror_num,
6282 unsigned long bio_flags, u64 offset)
6283{
6284 int ret;
6285 struct btrfs_root *root = BTRFS_I(inode)->root;
6286 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
79787eaa 6287 BUG_ON(ret); /* -ENOMEM */
eaf25d93
CM		 6288 return 0;
6289}
6290
e65e1535
MX		 6291static void btrfs_end_dio_bio(struct bio *bio, int err)
6292{
6293 struct btrfs_dio_private *dip = bio->bi_private;
6294
6295 if (err) {
33345d01 6296 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
3dd1462e 6297 "sector %#Lx len %u err no %d\n",
33345d01 6298 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
3dd1462e 6299 (unsigned long long)bio->bi_sector, bio->bi_size, err);
e65e1535
MX		 6300 dip->errors = 1;
6301
6302 /*
6303 * before atomic variable goto zero, we must make sure
6304 * dip->errors is perceived to be set.
6305 */
6306 smp_mb__before_atomic_dec();
6307 }
6308
6309 /* if there are more bios still pending for this dio, just exit */
6310 if (!atomic_dec_and_test(&dip->pending_bios))
6311 goto out;
6312
6313 if (dip->errors)
6314 bio_io_error(dip->orig_bio);
6315 else {
6316 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6317 bio_endio(dip->orig_bio, 0);
6318 }
6319out:
6320 bio_put(bio);
6321}
6322
6323static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6324 u64 first_sector, gfp_t gfp_flags)
6325{
6326 int nr_vecs = bio_get_nr_vecs(bdev);
6327 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6328}
6329
6330static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6331 int rw, u64 file_offset, int skip_sum,
c329861d 6332 int async_submit)
e65e1535
MX		 6333{
6334 int write = rw & REQ_WRITE;
6335 struct btrfs_root *root = BTRFS_I(inode)->root;
6336 int ret;
6337
6338 bio_get(bio);
5fd02043
JB		 6339
6340 if (!write) {
6341 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6342 if (ret)
6343 goto err;
6344 }
e65e1535 6345
1ae39938
JB		 6346 if (skip_sum)
6347 goto map;
6348
6349 if (write && async_submit) {
e65e1535
MX		 6350 ret = btrfs_wq_submit_bio(root->fs_info,
6351 inode, rw, bio, 0, 0,
6352 file_offset,
6353 __btrfs_submit_bio_start_direct_io,
6354 __btrfs_submit_bio_done);
6355 goto err;
1ae39938
JB		 6356 } else if (write) {
6357 /*
6358 * If we aren't doing async submit, calculate the csum of the
6359 * bio now.
6360 */
6361 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6362 if (ret)
6363 goto err;
c2db1073 6364 } else if (!skip_sum) {
c329861d 6365 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
c2db1073
TI		 6366 if (ret)
6367 goto err;
6368 }
e65e1535 6369
1ae39938
JB		 6370map:
6371 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
e65e1535
MX		 6372err:
6373 bio_put(bio);
6374 return ret;
6375}
6376
6377static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6378 int skip_sum)
6379{
6380 struct inode *inode = dip->inode;
6381 struct btrfs_root *root = BTRFS_I(inode)->root;
6382 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6383 struct bio *bio;
6384 struct bio *orig_bio = dip->orig_bio;
6385 struct bio_vec *bvec = orig_bio->bi_io_vec;
6386 u64 start_sector = orig_bio->bi_sector;
6387 u64 file_offset = dip->logical_offset;
6388 u64 submit_len = 0;
6389 u64 map_length;
6390 int nr_pages = 0;
e65e1535 6391 int ret = 0;
1ae39938 6392 int async_submit = 0;
e65e1535 6393
e65e1535
MX		 6394 map_length = orig_bio->bi_size;
6395 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6396 &map_length, NULL, 0);
6397 if (ret) {
64728bbb 6398 bio_put(orig_bio);
e65e1535
MX		 6399 return -EIO;
6400 }
6401
02f57c7a
JB		 6402 if (map_length >= orig_bio->bi_size) {
6403 bio = orig_bio;
6404 goto submit;
6405 }
6406
1ae39938 6407 async_submit = 1;
02f57c7a
JB		 6408 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6409 if (!bio)
6410 return -ENOMEM;
6411 bio->bi_private = dip;
6412 bio->bi_end_io = btrfs_end_dio_bio;
6413 atomic_inc(&dip->pending_bios);
6414
e65e1535
MX		 6415 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6416 if (unlikely(map_length < submit_len + bvec->bv_len ||
6417 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6418 bvec->bv_offset) < bvec->bv_len)) {
6419 /*
6420 * inc the count before we submit the bio so
6421 * we know the end IO handler won't happen before
6422 * we inc the count. Otherwise, the dip might get freed
6423 * before we're done setting it up
6424 */
6425 atomic_inc(&dip->pending_bios);
6426 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6427 file_offset, skip_sum,
c329861d 6428 async_submit);
e65e1535
MX		 6429 if (ret) {
6430 bio_put(bio);
6431 atomic_dec(&dip->pending_bios);
6432 goto out_err;
6433 }
6434
e65e1535
MX		 6435 start_sector += submit_len >> 9;
6436 file_offset += submit_len;
6437
6438 submit_len = 0;
6439 nr_pages = 0;
6440
6441 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6442 start_sector, GFP_NOFS);
6443 if (!bio)
6444 goto out_err;
6445 bio->bi_private = dip;
6446 bio->bi_end_io = btrfs_end_dio_bio;
6447
6448 map_length = orig_bio->bi_size;
6449 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6450 &map_length, NULL, 0);
6451 if (ret) {
6452 bio_put(bio);
6453 goto out_err;
6454 }
6455 } else {
6456 submit_len += bvec->bv_len;
6457 nr_pages ++;
6458 bvec++;
6459 }
6460 }
6461
02f57c7a 6462submit:
e65e1535 6463 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
c329861d 6464 async_submit);
e65e1535
MX		 6465 if (!ret)
6466 return 0;
6467
6468 bio_put(bio);
6469out_err:
6470 dip->errors = 1;
6471 /*
6472 * before atomic variable goto zero, we must
6473 * make sure dip->errors is perceived to be set.
6474 */
6475 smp_mb__before_atomic_dec();
6476 if (atomic_dec_and_test(&dip->pending_bios))
6477 bio_io_error(dip->orig_bio);
6478
6479 /* bio_end_io() will handle error, so we needn't return it */
6480 return 0;
6481}
6482
4b46fce2
JB		 6483static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6484 loff_t file_offset)
6485{
6486 struct btrfs_root *root = BTRFS_I(inode)->root;
6487 struct btrfs_dio_private *dip;
6488 struct bio_vec *bvec = bio->bi_io_vec;
4b46fce2 6489 int skip_sum;
7b6d91da 6490 int write = rw & REQ_WRITE;
4b46fce2
JB		 6491 int ret = 0;
6492
6493 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6494
6495 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6496 if (!dip) {
6497 ret = -ENOMEM;
6498 goto free_ordered;
6499 }
4b46fce2
JB		 6500
6501 dip->private = bio->bi_private;
6502 dip->inode = inode;
6503 dip->logical_offset = file_offset;
6504
4b46fce2
JB		 6505 dip->bytes = 0;
6506 do {
6507 dip->bytes += bvec->bv_len;
6508 bvec++;
6509 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6510
46bfbb5c 6511 dip->disk_bytenr = (u64)bio->bi_sector << 9;
4b46fce2 6512 bio->bi_private = dip;
e65e1535
MX		 6513 dip->errors = 0;
6514 dip->orig_bio = bio;
6515 atomic_set(&dip->pending_bios, 0);
4b46fce2
JB		 6516
6517 if (write)
6518 bio->bi_end_io = btrfs_endio_direct_write;
6519 else
6520 bio->bi_end_io = btrfs_endio_direct_read;
6521
e65e1535
MX		 6522 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6523 if (!ret)
eaf25d93 6524 return;
4b46fce2
JB		 6525free_ordered:
6526 /*
6527 * If this is a write, we need to clean up the reserved space and kill
6528 * the ordered extent.
6529 */
6530 if (write) {
6531 struct btrfs_ordered_extent *ordered;
955256f2 6532 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
4b46fce2
JB		 6533 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6534 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6535 btrfs_free_reserved_extent(root, ordered->start,
6536 ordered->disk_len);
6537 btrfs_put_ordered_extent(ordered);
6538 btrfs_put_ordered_extent(ordered);
6539 }
6540 bio_endio(bio, ret);
6541}
6542
5a5f79b5
CM		 6543static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6544 const struct iovec *iov, loff_t offset,
6545 unsigned long nr_segs)
6546{
6547 int seg;
a1b75f7d 6548 int i;
5a5f79b5
CM		 6549 size_t size;
6550 unsigned long addr;
6551 unsigned blocksize_mask = root->sectorsize - 1;
6552 ssize_t retval = -EINVAL;
6553 loff_t end = offset;
6554
6555 if (offset & blocksize_mask)
6556 goto out;
6557
6558 /* Check the memory alignment. Blocks cannot straddle pages */
6559 for (seg = 0; seg < nr_segs; seg++) {
6560 addr = (unsigned long)iov[seg].iov_base;
6561 size = iov[seg].iov_len;
6562 end += size;
a1b75f7d 6563 if ((addr & blocksize_mask) || (size & blocksize_mask))
5a5f79b5 6564 goto out;
a1b75f7d
JB		 6565
6566 /* If this is a write we don't need to check anymore */
6567 if (rw & WRITE)
6568 continue;
6569
6570 /*
6571 * Check to make sure we don't have duplicate iov_base's in this
6572 * iovec, if so return EINVAL, otherwise we'll get csum errors
6573 * when reading back.
6574 */
6575 for (i = seg + 1; i < nr_segs; i++) {
6576 if (iov[seg].iov_base == iov[i].iov_base)
6577 goto out;
6578 }
5a5f79b5
CM		 6579 }
6580 retval = 0;
6581out:
6582 return retval;
6583}
eb838e73 6584
16432985
CM		 6585static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6586 const struct iovec *iov, loff_t offset,
6587 unsigned long nr_segs)
6588{
4b46fce2
JB		 6589 struct file *file = iocb->ki_filp;
6590 struct inode *inode = file->f_mapping->host;
4b46fce2 6591
5a5f79b5 6592 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
eb838e73 6593 offset, nr_segs))
5a5f79b5 6594 return 0;
3f7c579c 6595
eb838e73 6596 return __blockdev_direct_IO(rw, iocb, inode,
5a5f79b5
CM		 6597 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6598 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6599 btrfs_submit_direct, 0);
16432985
CM		 6600}
6601
1506fcc8
YS		 6602static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6603 __u64 start, __u64 len)
6604{
ec29ed5b 6605 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
1506fcc8
YS		 6606}
6607
a52d9a80 6608int btrfs_readpage(struct file *file, struct page *page)
9ebefb18 6609{
d1310b2e
CM		 6610 struct extent_io_tree *tree;
6611 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 6612 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
9ebefb18 6613}
1832a6d5 6614
a52d9a80 6615static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
39279cc3 6616{
d1310b2e 6617 struct extent_io_tree *tree;
b888db2b
CM		 6618
6619
6620 if (current->flags & PF_MEMALLOC) {
6621 redirty_page_for_writepage(wbc, page);
6622 unlock_page(page);
6623 return 0;
6624 }
d1310b2e 6625 tree = &BTRFS_I(page->mapping->host)->io_tree;
a52d9a80 6626 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
9ebefb18
CM		 6627}
6628
f421950f
CM		 6629int btrfs_writepages(struct address_space *mapping,
6630 struct writeback_control *wbc)
b293f02e 6631{
d1310b2e 6632 struct extent_io_tree *tree;
771ed689 6633
d1310b2e 6634 tree = &BTRFS_I(mapping->host)->io_tree;
b293f02e
CM		 6635 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6636}
6637
3ab2fb5a
CM		 6638static int
6639btrfs_readpages(struct file *file, struct address_space *mapping,
6640 struct list_head *pages, unsigned nr_pages)
6641{
d1310b2e
CM		 6642 struct extent_io_tree *tree;
6643 tree = &BTRFS_I(mapping->host)->io_tree;
3ab2fb5a
CM		 6644 return extent_readpages(tree, mapping, pages, nr_pages,
6645 btrfs_get_extent);
6646}
e6dcd2dc 6647static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
9ebefb18 6648{
d1310b2e
CM		 6649 struct extent_io_tree *tree;
6650 struct extent_map_tree *map;
a52d9a80 6651 int ret;
8c2383c3 6652
d1310b2e
CM		 6653 tree = &BTRFS_I(page->mapping->host)->io_tree;
6654 map = &BTRFS_I(page->mapping->host)->extent_tree;
70dec807 6655 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
a52d9a80
CM		 6656 if (ret == 1) {
6657 ClearPagePrivate(page);
6658 set_page_private(page, 0);
6659 page_cache_release(page);
39279cc3 6660 }
a52d9a80 6661 return ret;
39279cc3
CM		 6662}
6663
e6dcd2dc
CM		 6664static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6665{
98509cfc
CM		 6666 if (PageWriteback(page) || PageDirty(page))
6667 return 0;
b335b003 6668 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
e6dcd2dc
CM		 6669}
6670
a52d9a80 6671static void btrfs_invalidatepage(struct page *page, unsigned long offset)
39279cc3 6672{
5fd02043 6673 struct inode *inode = page->mapping->host;
d1310b2e 6674 struct extent_io_tree *tree;
e6dcd2dc 6675 struct btrfs_ordered_extent *ordered;
2ac55d41 6676 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6677 u64 page_start = page_offset(page);
6678 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
39279cc3 6679
8b62b72b
CM		 6680 /*
6681 * we have the page locked, so new writeback can't start,
6682 * and the dirty bit won't be cleared while we are here.
6683 *
6684 * Wait for IO on this page so that we can safely clear
6685 * the PagePrivate2 bit and do ordered accounting
6686 */
e6dcd2dc 6687 wait_on_page_writeback(page);
8b62b72b 6688
5fd02043 6689 tree = &BTRFS_I(inode)->io_tree;
e6dcd2dc
CM		 6690 if (offset) {
6691 btrfs_releasepage(page, GFP_NOFS);
6692 return;
6693 }
d0082371 6694 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
5fd02043 6695 ordered = btrfs_lookup_ordered_extent(inode,
e6dcd2dc
CM		 6696 page_offset(page));
6697 if (ordered) {
eb84ae03
CM		 6698 /*
6699 * IO on this page will never be started, so we need
6700 * to account for any ordered extents now
6701 */
e6dcd2dc
CM		 6702 clear_extent_bit(tree, page_start, page_end,
6703 EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 6704 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6705 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
8b62b72b
CM		 6706 /*
6707 * whoever cleared the private bit is responsible
6708 * for the finish_ordered_io
6709 */
5fd02043
JB		 6710 if (TestClearPagePrivate2(page) &&
6711 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6712 PAGE_CACHE_SIZE, 1)) {
6713 btrfs_finish_ordered_io(ordered);
8b62b72b 6714 }
e6dcd2dc 6715 btrfs_put_ordered_extent(ordered);
2ac55d41 6716 cached_state = NULL;
d0082371 6717 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 6718 }
6719 clear_extent_bit(tree, page_start, page_end,
32c00aff 6720 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
9e8a4a8b
LB		 6721 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6722 &cached_state, GFP_NOFS);
e6dcd2dc
CM		 6723 __btrfs_releasepage(page, GFP_NOFS);
6724
4a096752 6725 ClearPageChecked(page);
9ad6b7bc 6726 if (PagePrivate(page)) {
9ad6b7bc
CM		 6727 ClearPagePrivate(page);
6728 set_page_private(page, 0);
6729 page_cache_release(page);
6730 }
39279cc3
CM		 6731}
6732
9ebefb18
CM		 6733/*
6734 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6735 * called from a page fault handler when a page is first dirtied. Hence we must
6736 * be careful to check for EOF conditions here. We set the page up correctly
6737 * for a written page which means we get ENOSPC checking when writing into
6738 * holes and correct delalloc and unwritten extent mapping on filesystems that
6739 * support these features.
6740 *
6741 * We are not allowed to take the i_mutex here so we have to play games to
6742 * protect against truncate races as the page could now be beyond EOF. Because
6743 * vmtruncate() writes the inode size before removing pages, once we have the
6744 * page lock we can determine safely if the page is beyond EOF. If it is not
6745 * beyond EOF, then the page is guaranteed safe against truncation until we
6746 * unlock the page.
6747 */
c2ec175c 6748int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
9ebefb18 6749{
c2ec175c 6750 struct page *page = vmf->page;
6da6abae 6751 struct inode *inode = fdentry(vma->vm_file)->d_inode;
1832a6d5 6752 struct btrfs_root *root = BTRFS_I(inode)->root;
e6dcd2dc
CM		 6753 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6754 struct btrfs_ordered_extent *ordered;
2ac55d41 6755 struct extent_state *cached_state = NULL;
e6dcd2dc
CM		 6756 char *kaddr;
6757 unsigned long zero_start;
9ebefb18 6758 loff_t size;
1832a6d5 6759 int ret;
9998eb70 6760 int reserved = 0;
a52d9a80 6761 u64 page_start;
e6dcd2dc 6762 u64 page_end;
9ebefb18 6763
b2b5ef5c 6764 sb_start_pagefault(inode->i_sb);
0ca1f7ce 6765 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
9998eb70 6766 if (!ret) {
e41f941a 6767 ret = file_update_time(vma->vm_file);
9998eb70
CM		 6768 reserved = 1;
6769 }
56a76f82
NP		 6770 if (ret) {
6771 if (ret == -ENOMEM)
6772 ret = VM_FAULT_OOM;
6773 else /* -ENOSPC, -EIO, etc */
6774 ret = VM_FAULT_SIGBUS;
9998eb70
CM		 6775 if (reserved)
6776 goto out;
6777 goto out_noreserve;
56a76f82 6778 }
1832a6d5 6779
56a76f82 6780 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
e6dcd2dc 6781again:
9ebefb18 6782 lock_page(page);
9ebefb18 6783 size = i_size_read(inode);
e6dcd2dc
CM		 6784 page_start = page_offset(page);
6785 page_end = page_start + PAGE_CACHE_SIZE - 1;
a52d9a80 6786
9ebefb18 6787 if ((page->mapping != inode->i_mapping) ||
e6dcd2dc 6788 (page_start >= size)) {
9ebefb18
CM		 6789 /* page got truncated out from underneath us */
6790 goto out_unlock;
6791 }
e6dcd2dc
CM		 6792 wait_on_page_writeback(page);
6793
d0082371 6794 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
e6dcd2dc
CM		 6795 set_page_extent_mapped(page);
6796
eb84ae03
CM		 6797 /*
6798 * we can't set the delalloc bits if there are pending ordered
6799 * extents. Drop our locks and wait for them to finish
6800 */
e6dcd2dc
CM		 6801 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6802 if (ordered) {
2ac55d41
JB		 6803 unlock_extent_cached(io_tree, page_start, page_end,
6804 &cached_state, GFP_NOFS);
e6dcd2dc 6805 unlock_page(page);
eb84ae03 6806 btrfs_start_ordered_extent(inode, ordered, 1);
e6dcd2dc
CM		 6807 btrfs_put_ordered_extent(ordered);
6808 goto again;
6809 }
6810
fbf19087
JB		 6811 /*
6812 * XXX - page_mkwrite gets called every time the page is dirtied, even
6813 * if it was already dirty, so for space accounting reasons we need to
6814 * clear any delalloc bits for the range we are fixing to save. There
6815 * is probably a better way to do this, but for now keep consistent with
6816 * prepare_pages in the normal write path.
6817 */
2ac55d41 6818 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
9e8a4a8b
LB		 6819 EXTENT_DIRTY | EXTENT_DELALLOC |
6820 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
2ac55d41 6821 0, 0, &cached_state, GFP_NOFS);
fbf19087 6822
2ac55d41
JB		 6823 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6824 &cached_state);
9ed74f2d 6825 if (ret) {
2ac55d41
JB		 6826 unlock_extent_cached(io_tree, page_start, page_end,
6827 &cached_state, GFP_NOFS);
9ed74f2d
JB		 6828 ret = VM_FAULT_SIGBUS;
6829 goto out_unlock;
6830 }
e6dcd2dc 6831 ret = 0;
9ebefb18
CM		 6832
6833 /* page is wholly or partially inside EOF */
a52d9a80 6834 if (page_start + PAGE_CACHE_SIZE > size)
e6dcd2dc 6835 zero_start = size & ~PAGE_CACHE_MASK;
9ebefb18 6836 else
e6dcd2dc 6837 zero_start = PAGE_CACHE_SIZE;
9ebefb18 6838
e6dcd2dc
CM		 6839 if (zero_start != PAGE_CACHE_SIZE) {
6840 kaddr = kmap(page);
6841 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6842 flush_dcache_page(page);
6843 kunmap(page);
6844 }
247e743c 6845 ClearPageChecked(page);
e6dcd2dc 6846 set_page_dirty(page);
50a9b214 6847 SetPageUptodate(page);
5a3f23d5 6848
257c62e1
CM		 6849 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6850 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
46d8bc34 6851 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
257c62e1 6852
2ac55d41 6853 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9ebefb18
CM		 6854
6855out_unlock:
b2b5ef5c
JK		 6856 if (!ret) {
6857 sb_end_pagefault(inode->i_sb);
50a9b214 6858 return VM_FAULT_LOCKED;
b2b5ef5c 6859 }
9ebefb18 6860 unlock_page(page);
1832a6d5 6861out:
ec39e180 6862 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
9998eb70 6863out_noreserve:
b2b5ef5c 6864 sb_end_pagefault(inode->i_sb);
9ebefb18
CM		 6865 return ret;
6866}
6867
a41ad394 6868static int btrfs_truncate(struct inode *inode)
39279cc3
CM		 6869{
6870 struct btrfs_root *root = BTRFS_I(inode)->root;
fcb80c2a 6871 struct btrfs_block_rsv *rsv;
39279cc3 6872 int ret;
3893e33b 6873 int err = 0;
39279cc3 6874 struct btrfs_trans_handle *trans;
d3c2fdcf 6875 unsigned long nr;
dbe674a9 6876 u64 mask = root->sectorsize - 1;
07127184 6877 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
39279cc3 6878
2aaa6655 6879 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
5d5e103a 6880 if (ret)
a41ad394 6881 return ret;
8082510e 6882
4a096752 6883 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
8082510e 6884 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
39279cc3 6885
fcb80c2a
JB		 6886 /*
6887 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6888 * 3 things going on here
6889 *
6890 * 1) We need to reserve space for our orphan item and the space to
6891 * delete our orphan item. Lord knows we don't want to have a dangling
6892 * orphan item because we didn't reserve space to remove it.
6893 *
6894 * 2) We need to reserve space to update our inode.
6895 *
6896 * 3) We need to have something to cache all the space that is going to
6897 * be free'd up by the truncate operation, but also have some slack
6898 * space reserved in case it uses space during the truncate (thank you
6899 * very much snapshotting).
6900 *
6901 * And we need these to all be seperate. The fact is we can use alot of
6902 * space doing the truncate, and we have no earthly idea how much space
6903 * we will use, so we need the truncate reservation to be seperate so it
6904 * doesn't end up using space reserved for updating the inode or
6905 * removing the orphan item. We also need to be able to stop the
6906 * transaction and start a new one, which means we need to be able to
6907 * update the inode several times, and we have no idea of knowing how
6908 * many times that will be, so we can't just reserve 1 item for the
6909 * entirety of the opration, so that has to be done seperately as well.
6910 * Then there is the orphan item, which does indeed need to be held on
6911 * to for the whole operation, and we need nobody to touch this reserved
6912 * space except the orphan code.
6913 *
6914 * So that leaves us with
6915 *
6916 * 1) root->orphan_block_rsv - for the orphan deletion.
6917 * 2) rsv - for the truncate reservation, which we will steal from the
6918 * transaction reservation.
6919 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6920 * updating the inode.
6921 */
66d8f3dd 6922 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
fcb80c2a
JB		 6923 if (!rsv)
6924 return -ENOMEM;
4a338542 6925 rsv->size = min_size;
ca7e70f5 6926 rsv->failfast = 1;
f0cd846e 6927
907cbceb 6928 /*
07127184 6929 * 1 for the truncate slack space
907cbceb
JB		 6930 * 1 for the orphan item we're going to add
6931 * 1 for the orphan item deletion
6932 * 1 for updating the inode.
6933 */
fcb80c2a
JB		 6934 trans = btrfs_start_transaction(root, 4);
6935 if (IS_ERR(trans)) {
6936 err = PTR_ERR(trans);
6937 goto out;
6938 }
f0cd846e 6939
907cbceb
JB		 6940 /* Migrate the slack space for the truncate to our reserve */
6941 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6942 min_size);
fcb80c2a 6943 BUG_ON(ret);
f0cd846e
JB		 6944
6945 ret = btrfs_orphan_add(trans, inode);
6946 if (ret) {
6947 btrfs_end_transaction(trans, root);
fcb80c2a 6948 goto out;
f0cd846e
JB		 6949 }
6950
5a3f23d5
CM		 6951 /*
6952 * setattr is responsible for setting the ordered_data_close flag,
6953 * but that is only tested during the last file release. That
6954 * could happen well after the next commit, leaving a great big
6955 * window where new writes may get lost if someone chooses to write
6956 * to this file after truncating to zero
6957 *
6958 * The inode doesn't have any dirty data here, and so if we commit
6959 * this is a noop. If someone immediately starts writing to the inode
6960 * it is very likely we'll catch some of their writes in this
6961 * transaction, and the commit will find this file on the ordered
6962 * data list with good things to send down.
6963 *
6964 * This is a best effort solution, there is still a window where
6965 * using truncate to replace the contents of the file will
6966 * end up with a zero length file after a crash.
6967 */
72ac3c0d
JB		 6968 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6969 &BTRFS_I(inode)->runtime_flags))
5a3f23d5
CM		 6970 btrfs_add_ordered_operation(trans, root, inode);
6971
5dc562c5
JB		 6972 /*
6973 * So if we truncate and then write and fsync we normally would just
6974 * write the extents that changed, which is a problem if we need to
6975 * first truncate that entire inode. So set this flag so we write out
6976 * all of the extents in the inode to the sync log so we're completely
6977 * safe.
6978 */
6979 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
ca7e70f5 6980 trans->block_rsv = rsv;
907cbceb 6981
8082510e
YZ		 6982 while (1) {
6983 ret = btrfs_truncate_inode_items(trans, root, inode,
6984 inode->i_size,
6985 BTRFS_EXTENT_DATA_KEY);
ca7e70f5 6986 if (ret != -ENOSPC) {
3893e33b 6987 err = ret;
8082510e 6988 break;
3893e33b 6989 }
39279cc3 6990
fcb80c2a 6991 trans->block_rsv = &root->fs_info->trans_block_rsv;
8082510e 6992 ret = btrfs_update_inode(trans, root, inode);
3893e33b
JB		 6993 if (ret) {
6994 err = ret;
6995 break;
6996 }
ca7e70f5 6997
8082510e
YZ		 6998 nr = trans->blocks_used;
6999 btrfs_end_transaction(trans, root);
7000 btrfs_btree_balance_dirty(root, nr);
ca7e70f5
JB		 7001
7002 trans = btrfs_start_transaction(root, 2);
7003 if (IS_ERR(trans)) {
7004 ret = err = PTR_ERR(trans);
7005 trans = NULL;
7006 break;
7007 }
7008
7009 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7010 rsv, min_size);
7011 BUG_ON(ret); /* shouldn't happen */
7012 trans->block_rsv = rsv;
8082510e
YZ		 7013 }
7014
7015 if (ret == 0 && inode->i_nlink > 0) {
fcb80c2a 7016 trans->block_rsv = root->orphan_block_rsv;
8082510e 7017 ret = btrfs_orphan_del(trans, inode);
3893e33b
JB		 7018 if (ret)
7019 err = ret;
ded5db9d
JB		 7020 } else if (ret && inode->i_nlink > 0) {
7021 /*
7022 * Failed to do the truncate, remove us from the in memory
7023 * orphan list.
7024 */
7025 ret = btrfs_orphan_del(NULL, inode);
8082510e
YZ		 7026 }
7027
917c16b2
CM		 7028 if (trans) {
7029 trans->block_rsv = &root->fs_info->trans_block_rsv;
7030 ret = btrfs_update_inode(trans, root, inode);
7031 if (ret && !err)
7032 err = ret;
7b128766 7033
917c16b2 7034 nr = trans->blocks_used;
7ad85bb7 7035 ret = btrfs_end_transaction(trans, root);
917c16b2
CM		 7036 btrfs_btree_balance_dirty(root, nr);
7037 }
fcb80c2a
JB		 7038
7039out:
7040 btrfs_free_block_rsv(root, rsv);
7041
3893e33b
JB		 7042 if (ret && !err)
7043 err = ret;
a41ad394 7044
3893e33b 7045 return err;
39279cc3
CM		 7046}
7047
d352ac68
CM		 7048/*
7049 * create a new subvolume directory/inode (helper for the ioctl).
7050 */
d2fb3437 7051int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
d82a6f1d 7052 struct btrfs_root *new_root, u64 new_dirid)
39279cc3 7053{
39279cc3 7054 struct inode *inode;
76dda93c 7055 int err;
00e4e6b3 7056 u64 index = 0;
39279cc3 7057
12fc9d09
FA		 7058 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7059 new_dirid, new_dirid,
7060 S_IFDIR | (~current_umask() & S_IRWXUGO),
7061 &index);
54aa1f4d 7062 if (IS_ERR(inode))
f46b5a66 7063 return PTR_ERR(inode);
39279cc3
CM		 7064 inode->i_op = &btrfs_dir_inode_operations;
7065 inode->i_fop = &btrfs_dir_file_operations;
7066
bfe86848 7067 set_nlink(inode, 1);
dbe674a9 7068 btrfs_i_size_write(inode, 0);
3b96362c 7069
76dda93c 7070 err = btrfs_update_inode(trans, new_root, inode);
cb8e7090 7071
76dda93c 7072 iput(inode);
ce598979 7073 return err;
39279cc3
CM		 7074}
7075
39279cc3
CM		 7076struct inode *btrfs_alloc_inode(struct super_block *sb)
7077{
7078 struct btrfs_inode *ei;
2ead6ae7 7079 struct inode *inode;
39279cc3
CM		 7080
7081 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7082 if (!ei)
7083 return NULL;
2ead6ae7
YZ		 7084
7085 ei->root = NULL;
2ead6ae7 7086 ei->generation = 0;
15ee9bc7 7087 ei->last_trans = 0;
257c62e1 7088 ei->last_sub_trans = 0;
e02119d5 7089 ei->logged_trans = 0;
2ead6ae7 7090 ei->delalloc_bytes = 0;
2ead6ae7
YZ		 7091 ei->disk_i_size = 0;
7092 ei->flags = 0;
7709cde3 7093 ei->csum_bytes = 0;
2ead6ae7
YZ		 7094 ei->index_cnt = (u64)-1;
7095 ei->last_unlink_trans = 0;
46d8bc34 7096 ei->last_log_commit = 0;
2ead6ae7 7097
9e0baf60
JB		 7098 spin_lock_init(&ei->lock);
7099 ei->outstanding_extents = 0;
7100 ei->reserved_extents = 0;
2ead6ae7 7101
72ac3c0d 7102 ei->runtime_flags = 0;
261507a0 7103 ei->force_compress = BTRFS_COMPRESS_NONE;
2ead6ae7 7104
16cdcec7
MX		 7105 ei->delayed_node = NULL;
7106
2ead6ae7 7107 inode = &ei->vfs_inode;
a8067e02 7108 extent_map_tree_init(&ei->extent_tree);
f993c883
DS		 7109 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7110 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
0b32f4bb
JB		 7111 ei->io_tree.track_uptodate = 1;
7112 ei->io_failure_tree.track_uptodate = 1;
2ead6ae7 7113 mutex_init(&ei->log_mutex);
f248679e 7114 mutex_init(&ei->delalloc_mutex);
e6dcd2dc 7115 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
2ead6ae7 7116 INIT_LIST_HEAD(&ei->delalloc_inodes);
5a3f23d5 7117 INIT_LIST_HEAD(&ei->ordered_operations);
2ead6ae7
YZ		 7118 RB_CLEAR_NODE(&ei->rb_node);
7119
7120 return inode;
39279cc3
CM		 7121}
7122
fa0d7e3d
NP		 7123static void btrfs_i_callback(struct rcu_head *head)
7124{
7125 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP		 7126 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7127}
7128
39279cc3
CM		 7129void btrfs_destroy_inode(struct inode *inode)
7130{
e6dcd2dc 7131 struct btrfs_ordered_extent *ordered;
5a3f23d5
CM		 7132 struct btrfs_root *root = BTRFS_I(inode)->root;
7133
b3d9b7a3 7134 WARN_ON(!hlist_empty(&inode->i_dentry));
39279cc3 7135 WARN_ON(inode->i_data.nrpages);
9e0baf60
JB		 7136 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7137 WARN_ON(BTRFS_I(inode)->reserved_extents);
7709cde3
JB		 7138 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7139 WARN_ON(BTRFS_I(inode)->csum_bytes);
39279cc3 7140
a6dbd429
JB		 7141 /*
7142 * This can happen where we create an inode, but somebody else also
7143 * created the same inode and we need to destroy the one we already
7144 * created.
7145 */
7146 if (!root)
7147 goto free;
7148
5a3f23d5
CM		 7149 /*
7150 * Make sure we're properly removed from the ordered operation
7151 * lists.
7152 */
7153 smp_mb();
7154 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7155 spin_lock(&root->fs_info->ordered_extent_lock);
7156 list_del_init(&BTRFS_I(inode)->ordered_operations);
7157 spin_unlock(&root->fs_info->ordered_extent_lock);
7158 }
7159
8a35d95f
JB		 7160 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7161 &BTRFS_I(inode)->runtime_flags)) {
33345d01
LZ		 7162 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7163 (unsigned long long)btrfs_ino(inode));
8a35d95f 7164 atomic_dec(&root->orphan_inodes);
7b128766 7165 }
7b128766 7166
d397712b 7167 while (1) {
e6dcd2dc
CM		 7168 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7169 if (!ordered)
7170 break;
7171 else {
d397712b
CM		 7172 printk(KERN_ERR "btrfs found ordered "
7173 "extent %llu %llu on inode cleanup\n",
7174 (unsigned long long)ordered->file_offset,
7175 (unsigned long long)ordered->len);
e6dcd2dc
CM		 7176 btrfs_remove_ordered_extent(inode, ordered);
7177 btrfs_put_ordered_extent(ordered);
7178 btrfs_put_ordered_extent(ordered);
7179 }
7180 }
5d4f98a2 7181 inode_tree_del(inode);
5b21f2ed 7182 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
a6dbd429 7183free:
16cdcec7 7184 btrfs_remove_delayed_node(inode);
fa0d7e3d 7185 call_rcu(&inode->i_rcu, btrfs_i_callback);
39279cc3
CM		 7186}
7187
45321ac5 7188int btrfs_drop_inode(struct inode *inode)
76dda93c
YZ		 7189{
7190 struct btrfs_root *root = BTRFS_I(inode)->root;
45321ac5 7191
0af3d00b 7192 if (btrfs_root_refs(&root->root_item) == 0 &&
83eea1f1 7193 !btrfs_is_free_space_inode(inode))
45321ac5 7194 return 1;
76dda93c 7195 else
45321ac5 7196 return generic_drop_inode(inode);
76dda93c
YZ		 7197}
7198
0ee0fda0 7199static void init_once(void *foo)
39279cc3
CM		 7200{
7201 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7202
7203 inode_init_once(&ei->vfs_inode);
7204}
7205
7206void btrfs_destroy_cachep(void)
7207{
8c0a8537
KS		 7208 /*
7209 * Make sure all delayed rcu free inodes are flushed before we
7210 * destroy cache.
7211 */
7212 rcu_barrier();
39279cc3
CM		 7213 if (btrfs_inode_cachep)
7214 kmem_cache_destroy(btrfs_inode_cachep);
7215 if (btrfs_trans_handle_cachep)
7216 kmem_cache_destroy(btrfs_trans_handle_cachep);
7217 if (btrfs_transaction_cachep)
7218 kmem_cache_destroy(btrfs_transaction_cachep);
39279cc3
CM		 7219 if (btrfs_path_cachep)
7220 kmem_cache_destroy(btrfs_path_cachep);
dc89e982
JB		 7221 if (btrfs_free_space_cachep)
7222 kmem_cache_destroy(btrfs_free_space_cachep);
8ccf6f19
MX		 7223 if (btrfs_delalloc_work_cachep)
7224 kmem_cache_destroy(btrfs_delalloc_work_cachep);
39279cc3
CM		 7225}
7226
7227int btrfs_init_cachep(void)
7228{
837e1972 7229 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9601e3f6
CH		 7230 sizeof(struct btrfs_inode), 0,
7231 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
39279cc3
CM		 7232 if (!btrfs_inode_cachep)
7233 goto fail;
9601e3f6 7234
837e1972 7235 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9601e3f6
CH		 7236 sizeof(struct btrfs_trans_handle), 0,
7237 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7238 if (!btrfs_trans_handle_cachep)
7239 goto fail;
9601e3f6 7240
837e1972 7241 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9601e3f6
CH		 7242 sizeof(struct btrfs_transaction), 0,
7243 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7244 if (!btrfs_transaction_cachep)
7245 goto fail;
9601e3f6 7246
837e1972 7247 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9601e3f6
CH		 7248 sizeof(struct btrfs_path), 0,
7249 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
39279cc3
CM		 7250 if (!btrfs_path_cachep)
7251 goto fail;
9601e3f6 7252
837e1972 7253 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
dc89e982
JB		 7254 sizeof(struct btrfs_free_space), 0,
7255 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7256 if (!btrfs_free_space_cachep)
7257 goto fail;
7258
8ccf6f19
MX		 7259 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7260 sizeof(struct btrfs_delalloc_work), 0,
7261 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7262 NULL);
7263 if (!btrfs_delalloc_work_cachep)
7264 goto fail;
7265
39279cc3
CM		 7266 return 0;
7267fail:
7268 btrfs_destroy_cachep();
7269 return -ENOMEM;
7270}
7271
7272static int btrfs_getattr(struct vfsmount *mnt,
7273 struct dentry *dentry, struct kstat *stat)
7274{
7275 struct inode *inode = dentry->d_inode;
fadc0d8b
DS		 7276 u32 blocksize = inode->i_sb->s_blocksize;
7277
39279cc3 7278 generic_fillattr(inode, stat);
0ee5dc67 7279 stat->dev = BTRFS_I(inode)->root->anon_dev;
d6667462 7280 stat->blksize = PAGE_CACHE_SIZE;
fadc0d8b
DS		 7281 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7282 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
39279cc3
CM		 7283 return 0;
7284}
7285
75e7cb7f
LB		 7286/*
7287 * If a file is moved, it will inherit the cow and compression flags of the new
7288 * directory.
7289 */
7290static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7291{
7292 struct btrfs_inode *b_dir = BTRFS_I(dir);
7293 struct btrfs_inode *b_inode = BTRFS_I(inode);
7294
7295 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7296 b_inode->flags |= BTRFS_INODE_NODATACOW;
7297 else
7298 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7299
bc178237 7300 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
75e7cb7f 7301 b_inode->flags |= BTRFS_INODE_COMPRESS;
bc178237
LB		 7302 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7303 } else {
7304 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7305 BTRFS_INODE_NOCOMPRESS);
7306 }
75e7cb7f
LB		 7307}
7308
d397712b
CM		 7309static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7310 struct inode *new_dir, struct dentry *new_dentry)
39279cc3
CM		 7311{
7312 struct btrfs_trans_handle *trans;
7313 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4df27c4d 7314 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
39279cc3
CM		 7315 struct inode *new_inode = new_dentry->d_inode;
7316 struct inode *old_inode = old_dentry->d_inode;
7317 struct timespec ctime = CURRENT_TIME;
00e4e6b3 7318 u64 index = 0;
4df27c4d 7319 u64 root_objectid;
39279cc3 7320 int ret;
33345d01 7321 u64 old_ino = btrfs_ino(old_inode);
39279cc3 7322
33345d01 7323 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
f679a840
YZ		 7324 return -EPERM;
7325
4df27c4d 7326 /* we only allow rename subvolume link between subvolumes */
33345d01 7327 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
3394e160
CM		 7328 return -EXDEV;
7329
33345d01
LZ		 7330 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7331 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
39279cc3 7332 return -ENOTEMPTY;
5f39d397 7333
4df27c4d
YZ		 7334 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7335 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7336 return -ENOTEMPTY;
5a3f23d5
CM		 7337 /*
7338 * we're using rename to replace one file with another.
7339 * and the replacement file is large. Start IO on it now so
7340 * we don't add too much work to the end of the transaction
7341 */
4baf8c92 7342 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5a3f23d5
CM		 7343 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7344 filemap_flush(old_inode->i_mapping);
7345
76dda93c 7346 /* close the racy window with snapshot create/destroy ioctl */
33345d01 7347 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7348 down_read(&root->fs_info->subvol_sem);
a22285a6
YZ		 7349 /*
7350 * We want to reserve the absolute worst case amount of items. So if
7351 * both inodes are subvols and we need to unlink them then that would
7352 * require 4 item modifications, but if they are both normal inodes it
7353 * would require 5 item modifications, so we'll assume their normal
7354 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7355 * should cover the worst case number of items we'll modify.
7356 */
7357 trans = btrfs_start_transaction(root, 20);
b44c59a8
JL		 7358 if (IS_ERR(trans)) {
7359 ret = PTR_ERR(trans);
7360 goto out_notrans;
7361 }
76dda93c 7362
4df27c4d
YZ		 7363 if (dest != root)
7364 btrfs_record_root_in_trans(trans, dest);
5f39d397 7365
a5719521
YZ		 7366 ret = btrfs_set_inode_index(new_dir, &index);
7367 if (ret)
7368 goto out_fail;
5a3f23d5 7369
33345d01 7370 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7371 /* force full log commit if subvolume involved. */
7372 root->fs_info->last_trans_log_full_commit = trans->transid;
7373 } else {
a5719521
YZ		 7374 ret = btrfs_insert_inode_ref(trans, dest,
7375 new_dentry->d_name.name,
7376 new_dentry->d_name.len,
33345d01
LZ		 7377 old_ino,
7378 btrfs_ino(new_dir), index);
a5719521
YZ		 7379 if (ret)
7380 goto out_fail;
4df27c4d
YZ		 7381 /*
7382 * this is an ugly little race, but the rename is required
7383 * to make sure that if we crash, the inode is either at the
7384 * old name or the new one. pinning the log transaction lets
7385 * us make sure we don't allow a log commit to come in after
7386 * we unlink the name but before we add the new name back in.
7387 */
7388 btrfs_pin_log_trans(root);
7389 }
5a3f23d5
CM		 7390 /*
7391 * make sure the inode gets flushed if it is replacing
7392 * something.
7393 */
33345d01 7394 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
5a3f23d5 7395 btrfs_add_ordered_operation(trans, root, old_inode);
5a3f23d5 7396
0c4d2d95
JB		 7397 inode_inc_iversion(old_dir);
7398 inode_inc_iversion(new_dir);
7399 inode_inc_iversion(old_inode);
39279cc3
CM		 7400 old_dir->i_ctime = old_dir->i_mtime = ctime;
7401 new_dir->i_ctime = new_dir->i_mtime = ctime;
7402 old_inode->i_ctime = ctime;
5f39d397 7403
12fcfd22
CM		 7404 if (old_dentry->d_parent != new_dentry->d_parent)
7405 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7406
33345d01 7407 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4df27c4d
YZ		 7408 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7409 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7410 old_dentry->d_name.name,
7411 old_dentry->d_name.len);
7412 } else {
92986796
AV		 7413 ret = __btrfs_unlink_inode(trans, root, old_dir,
7414 old_dentry->d_inode,
7415 old_dentry->d_name.name,
7416 old_dentry->d_name.len);
7417 if (!ret)
7418 ret = btrfs_update_inode(trans, root, old_inode);
4df27c4d 7419 }
79787eaa
JM		 7420 if (ret) {
7421 btrfs_abort_transaction(trans, root, ret);
7422 goto out_fail;
7423 }
39279cc3
CM		 7424
7425 if (new_inode) {
0c4d2d95 7426 inode_inc_iversion(new_inode);
39279cc3 7427 new_inode->i_ctime = CURRENT_TIME;
33345d01 7428 if (unlikely(btrfs_ino(new_inode) ==
4df27c4d
YZ		 7429 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7430 root_objectid = BTRFS_I(new_inode)->location.objectid;
7431 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7432 root_objectid,
7433 new_dentry->d_name.name,
7434 new_dentry->d_name.len);
7435 BUG_ON(new_inode->i_nlink == 0);
7436 } else {
7437 ret = btrfs_unlink_inode(trans, dest, new_dir,
7438 new_dentry->d_inode,
7439 new_dentry->d_name.name,
7440 new_dentry->d_name.len);
7441 }
79787eaa 7442 if (!ret && new_inode->i_nlink == 0) {
e02119d5 7443 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4df27c4d 7444 BUG_ON(ret);
7b128766 7445 }
79787eaa
JM		 7446 if (ret) {
7447 btrfs_abort_transaction(trans, root, ret);
7448 goto out_fail;
7449 }
39279cc3 7450 }
aec7477b 7451
75e7cb7f
LB		 7452 fixup_inode_flags(new_dir, old_inode);
7453
4df27c4d
YZ		 7454 ret = btrfs_add_link(trans, new_dir, old_inode,
7455 new_dentry->d_name.name,
a5719521 7456 new_dentry->d_name.len, 0, index);
79787eaa
JM		 7457 if (ret) {
7458 btrfs_abort_transaction(trans, root, ret);
7459 goto out_fail;
7460 }
39279cc3 7461
33345d01 7462 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
10d9f309 7463 struct dentry *parent = new_dentry->d_parent;
6a912213 7464 btrfs_log_new_name(trans, old_inode, old_dir, parent);
4df27c4d
YZ		 7465 btrfs_end_log_trans(root);
7466 }
39279cc3 7467out_fail:
7ad85bb7 7468 btrfs_end_transaction(trans, root);
b44c59a8 7469out_notrans:
33345d01 7470 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
76dda93c 7471 up_read(&root->fs_info->subvol_sem);
9ed74f2d 7472
39279cc3
CM		 7473 return ret;
7474}
7475
8ccf6f19
MX		 7476static void btrfs_run_delalloc_work(struct btrfs_work *work)
7477{
7478 struct btrfs_delalloc_work *delalloc_work;
7479
7480 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7481 work);
7482 if (delalloc_work->wait)
7483 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7484 else
7485 filemap_flush(delalloc_work->inode->i_mapping);
7486
7487 if (delalloc_work->delay_iput)
7488 btrfs_add_delayed_iput(delalloc_work->inode);
7489 else
7490 iput(delalloc_work->inode);
7491 complete(&delalloc_work->completion);
7492}
7493
7494struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7495 int wait, int delay_iput)
7496{
7497 struct btrfs_delalloc_work *work;
7498
7499 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7500 if (!work)
7501 return NULL;
7502
7503 init_completion(&work->completion);
7504 INIT_LIST_HEAD(&work->list);
7505 work->inode = inode;
7506 work->wait = wait;
7507 work->delay_iput = delay_iput;
7508 work->work.func = btrfs_run_delalloc_work;
7509
7510 return work;
7511}
7512
7513void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7514{
7515 wait_for_completion(&work->completion);
7516 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7517}
7518
d352ac68
CM		 7519/*
7520 * some fairly slow code that needs optimization. This walks the list
7521 * of all the inodes with pending delalloc and forces them to disk.
7522 */
24bbcf04 7523int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
ea8c2819
CM		 7524{
7525 struct list_head *head = &root->fs_info->delalloc_inodes;
7526 struct btrfs_inode *binode;
5b21f2ed 7527 struct inode *inode;
8ccf6f19
MX		 7528 struct btrfs_delalloc_work *work, *next;
7529 struct list_head works;
7530 int ret = 0;
ea8c2819 7531
c146afad
YZ		 7532 if (root->fs_info->sb->s_flags & MS_RDONLY)
7533 return -EROFS;
7534
8ccf6f19
MX		 7535 INIT_LIST_HEAD(&works);
7536
75eff68e 7537 spin_lock(&root->fs_info->delalloc_lock);
d397712b 7538 while (!list_empty(head)) {
ea8c2819
CM		 7539 binode = list_entry(head->next, struct btrfs_inode,
7540 delalloc_inodes);
5b21f2ed
ZY		 7541 inode = igrab(&binode->vfs_inode);
7542 if (!inode)
7543 list_del_init(&binode->delalloc_inodes);
75eff68e 7544 spin_unlock(&root->fs_info->delalloc_lock);
5b21f2ed 7545 if (inode) {
8ccf6f19
MX		 7546 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7547 if (!work) {
7548 ret = -ENOMEM;
7549 goto out;
7550 }
7551 list_add_tail(&work->list, &works);
7552 btrfs_queue_worker(&root->fs_info->flush_workers,
7553 &work->work);
5b21f2ed
ZY		 7554 }
7555 cond_resched();
75eff68e 7556 spin_lock(&root->fs_info->delalloc_lock);
ea8c2819 7557 }
75eff68e 7558 spin_unlock(&root->fs_info->delalloc_lock);
8c8bee1d
CM		 7559
7560 /* the filemap_flush will queue IO into the worker threads, but
7561 * we have to make sure the IO is actually started and that
7562 * ordered extents get created before we return
7563 */
7564 atomic_inc(&root->fs_info->async_submit_draining);
d397712b 7565 while (atomic_read(&root->fs_info->nr_async_submits) ||
771ed689 7566 atomic_read(&root->fs_info->async_delalloc_pages)) {
8c8bee1d 7567 wait_event(root->fs_info->async_submit_wait,
771ed689
CM		 7568 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7569 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8c8bee1d
CM		 7570 }
7571 atomic_dec(&root->fs_info->async_submit_draining);
8ccf6f19
MX		 7572out:
7573 list_for_each_entry_safe(work, next, &works, list) {
7574 list_del_init(&work->list);
7575 btrfs_wait_and_free_delalloc_work(work);
7576 }
7577 return ret;
ea8c2819
CM		 7578}
7579
39279cc3
CM		 7580static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7581 const char *symname)
7582{
7583 struct btrfs_trans_handle *trans;
7584 struct btrfs_root *root = BTRFS_I(dir)->root;
7585 struct btrfs_path *path;
7586 struct btrfs_key key;
1832a6d5 7587 struct inode *inode = NULL;
39279cc3
CM		 7588 int err;
7589 int drop_inode = 0;
7590 u64 objectid;
00e4e6b3 7591 u64 index = 0 ;
39279cc3
CM		 7592 int name_len;
7593 int datasize;
5f39d397 7594 unsigned long ptr;
39279cc3 7595 struct btrfs_file_extent_item *ei;
5f39d397 7596 struct extent_buffer *leaf;
1832a6d5 7597 unsigned long nr = 0;
39279cc3
CM		 7598
7599 name_len = strlen(symname) + 1;
7600 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7601 return -ENAMETOOLONG;
1832a6d5 7602
9ed74f2d
JB		 7603 /*
7604 * 2 items for inode item and ref
7605 * 2 items for dir items
7606 * 1 item for xattr if selinux is on
7607 */
a22285a6
YZ		 7608 trans = btrfs_start_transaction(root, 5);
7609 if (IS_ERR(trans))
7610 return PTR_ERR(trans);
1832a6d5 7611
581bb050
LZ		 7612 err = btrfs_find_free_ino(root, &objectid);
7613 if (err)
7614 goto out_unlock;
7615
aec7477b 7616 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
33345d01 7617 dentry->d_name.len, btrfs_ino(dir), objectid,
d82a6f1d 7618 S_IFLNK|S_IRWXUGO, &index);
7cf96da3
TI		 7619 if (IS_ERR(inode)) {
7620 err = PTR_ERR(inode);
39279cc3 7621 goto out_unlock;
7cf96da3 7622 }
39279cc3 7623
2a7dba39 7624 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
33268eaf
JB		 7625 if (err) {
7626 drop_inode = 1;
7627 goto out_unlock;
7628 }
7629
ad19db71
CS		 7630 /*
7631 * If the active LSM wants to access the inode during
7632 * d_instantiate it needs these. Smack checks to see
7633 * if the filesystem supports xattrs by looking at the
7634 * ops vector.
7635 */
7636 inode->i_fop = &btrfs_file_operations;
7637 inode->i_op = &btrfs_file_inode_operations;
7638
a1b075d2 7639 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
39279cc3
CM		 7640 if (err)
7641 drop_inode = 1;
7642 else {
7643 inode->i_mapping->a_ops = &btrfs_aops;
04160088 7644 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d1310b2e 7645 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
39279cc3 7646 }
39279cc3
CM		 7647 if (drop_inode)
7648 goto out_unlock;
7649
7650 path = btrfs_alloc_path();
d8926bb3
MF		 7651 if (!path) {
7652 err = -ENOMEM;
7653 drop_inode = 1;
7654 goto out_unlock;
7655 }
33345d01 7656 key.objectid = btrfs_ino(inode);
39279cc3 7657 key.offset = 0;
39279cc3
CM		 7658 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7659 datasize = btrfs_file_extent_calc_inline_size(name_len);
7660 err = btrfs_insert_empty_item(trans, root, path, &key,
7661 datasize);
54aa1f4d
CM		 7662 if (err) {
7663 drop_inode = 1;
b0839166 7664 btrfs_free_path(path);
54aa1f4d
CM		 7665 goto out_unlock;
7666 }
5f39d397
CM		 7667 leaf = path->nodes[0];
7668 ei = btrfs_item_ptr(leaf, path->slots[0],
7669 struct btrfs_file_extent_item);
7670 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7671 btrfs_set_file_extent_type(leaf, ei,
39279cc3 7672 BTRFS_FILE_EXTENT_INLINE);
c8b97818
CM		 7673 btrfs_set_file_extent_encryption(leaf, ei, 0);
7674 btrfs_set_file_extent_compression(leaf, ei, 0);
7675 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7676 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7677
39279cc3 7678 ptr = btrfs_file_extent_inline_start(ei);
5f39d397
CM		 7679 write_extent_buffer(leaf, symname, ptr, name_len);
7680 btrfs_mark_buffer_dirty(leaf);
39279cc3 7681 btrfs_free_path(path);
5f39d397 7682
39279cc3
CM		 7683 inode->i_op = &btrfs_symlink_inode_operations;
7684 inode->i_mapping->a_ops = &btrfs_symlink_aops;
04160088 7685 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
d899e052 7686 inode_set_bytes(inode, name_len);
dbe674a9 7687 btrfs_i_size_write(inode, name_len - 1);
54aa1f4d
CM		 7688 err = btrfs_update_inode(trans, root, inode);
7689 if (err)
7690 drop_inode = 1;
39279cc3
CM		 7691
7692out_unlock:
08c422c2
AV		 7693 if (!err)
7694 d_instantiate(dentry, inode);
d3c2fdcf 7695 nr = trans->blocks_used;
7ad85bb7 7696 btrfs_end_transaction(trans, root);
39279cc3
CM		 7697 if (drop_inode) {
7698 inode_dec_link_count(inode);
7699 iput(inode);
7700 }
d3c2fdcf 7701 btrfs_btree_balance_dirty(root, nr);
39279cc3
CM		 7702 return err;
7703}
16432985 7704
0af3d00b
JB		 7705static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7706 u64 start, u64 num_bytes, u64 min_size,
7707 loff_t actual_len, u64 *alloc_hint,
7708 struct btrfs_trans_handle *trans)
d899e052 7709{
5dc562c5
JB		 7710 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7711 struct extent_map *em;
d899e052
YZ		 7712 struct btrfs_root *root = BTRFS_I(inode)->root;
7713 struct btrfs_key ins;
d899e052 7714 u64 cur_offset = start;
55a61d1d 7715 u64 i_size;
d899e052 7716 int ret = 0;
0af3d00b 7717 bool own_trans = true;
d899e052 7718
0af3d00b
JB		 7719 if (trans)
7720 own_trans = false;
d899e052 7721 while (num_bytes > 0) {
0af3d00b
JB		 7722 if (own_trans) {
7723 trans = btrfs_start_transaction(root, 3);
7724 if (IS_ERR(trans)) {
7725 ret = PTR_ERR(trans);
7726 break;
7727 }
5a303d5d
YZ		 7728 }
7729
efa56464 7730 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
81c9ad23 7731 0, *alloc_hint, &ins, 1);
5a303d5d 7732 if (ret) {
0af3d00b
JB		 7733 if (own_trans)
7734 btrfs_end_transaction(trans, root);
a22285a6 7735 break;
d899e052 7736 }
5a303d5d 7737
d899e052
YZ		 7738 ret = insert_reserved_file_extent(trans, inode,
7739 cur_offset, ins.objectid,
7740 ins.offset, ins.offset,
920bbbfb 7741 ins.offset, 0, 0, 0,
d899e052 7742 BTRFS_FILE_EXTENT_PREALLOC);
79787eaa
JM		 7743 if (ret) {
7744 btrfs_abort_transaction(trans, root, ret);
7745 if (own_trans)
7746 btrfs_end_transaction(trans, root);
7747 break;
7748 }
a1ed835e
CM		 7749 btrfs_drop_extent_cache(inode, cur_offset,
7750 cur_offset + ins.offset -1, 0);
5a303d5d 7751
5dc562c5
JB		 7752 em = alloc_extent_map();
7753 if (!em) {
7754 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7755 &BTRFS_I(inode)->runtime_flags);
7756 goto next;
7757 }
7758
7759 em->start = cur_offset;
7760 em->orig_start = cur_offset;
7761 em->len = ins.offset;
7762 em->block_start = ins.objectid;
7763 em->block_len = ins.offset;
7764 em->bdev = root->fs_info->fs_devices->latest_bdev;
7765 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7766 em->generation = trans->transid;
7767
7768 while (1) {
7769 write_lock(&em_tree->lock);
7770 ret = add_extent_mapping(em_tree, em);
7771 if (!ret)
7772 list_move(&em->list,
7773 &em_tree->modified_extents);
7774 write_unlock(&em_tree->lock);
7775 if (ret != -EEXIST)
7776 break;
7777 btrfs_drop_extent_cache(inode, cur_offset,
7778 cur_offset + ins.offset - 1,
7779 0);
7780 }
7781 free_extent_map(em);
7782next:
d899e052
YZ		 7783 num_bytes -= ins.offset;
7784 cur_offset += ins.offset;
efa56464 7785 *alloc_hint = ins.objectid + ins.offset;
5a303d5d 7786
0c4d2d95 7787 inode_inc_iversion(inode);
d899e052 7788 inode->i_ctime = CURRENT_TIME;
6cbff00f 7789 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
d899e052 7790 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
efa56464
YZ		 7791 (actual_len > inode->i_size) &&
7792 (cur_offset > inode->i_size)) {
d1ea6a61 7793 if (cur_offset > actual_len)
55a61d1d 7794 i_size = actual_len;
d1ea6a61 7795 else
55a61d1d
JB		 7796 i_size = cur_offset;
7797 i_size_write(inode, i_size);
7798 btrfs_ordered_update_i_size(inode, i_size, NULL);
5a303d5d
YZ		 7799 }
7800
d899e052 7801 ret = btrfs_update_inode(trans, root, inode);
79787eaa
JM		 7802
7803 if (ret) {
7804 btrfs_abort_transaction(trans, root, ret);
7805 if (own_trans)
7806 btrfs_end_transaction(trans, root);
7807 break;
7808 }
d899e052 7809
0af3d00b
JB		 7810 if (own_trans)
7811 btrfs_end_transaction(trans, root);
5a303d5d 7812 }
d899e052
YZ		 7813 return ret;
7814}
7815
0af3d00b
JB		 7816int btrfs_prealloc_file_range(struct inode *inode, int mode,
7817 u64 start, u64 num_bytes, u64 min_size,
7818 loff_t actual_len, u64 *alloc_hint)
7819{
7820 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7821 min_size, actual_len, alloc_hint,
7822 NULL);
7823}
7824
7825int btrfs_prealloc_file_range_trans(struct inode *inode,
7826 struct btrfs_trans_handle *trans, int mode,
7827 u64 start, u64 num_bytes, u64 min_size,
7828 loff_t actual_len, u64 *alloc_hint)
7829{
7830 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7831 min_size, actual_len, alloc_hint, trans);
7832}
7833
e6dcd2dc
CM		 7834static int btrfs_set_page_dirty(struct page *page)
7835{
e6dcd2dc
CM		 7836 return __set_page_dirty_nobuffers(page);
7837}
7838
10556cb2 7839static int btrfs_permission(struct inode *inode, int mask)
fdebe2bd 7840{
b83cc969 7841 struct btrfs_root *root = BTRFS_I(inode)->root;
cb6db4e5 7842 umode_t mode = inode->i_mode;
b83cc969 7843
cb6db4e5
JM		 7844 if (mask & MAY_WRITE &&
7845 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7846 if (btrfs_root_readonly(root))
7847 return -EROFS;
7848 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7849 return -EACCES;
7850 }
2830ba7f 7851 return generic_permission(inode, mask);
fdebe2bd 7852}
39279cc3 7853
6e1d5dcc 7854static const struct inode_operations btrfs_dir_inode_operations = {
3394e160 7855 .getattr = btrfs_getattr,
39279cc3
CM		 7856 .lookup = btrfs_lookup,
7857 .create = btrfs_create,
7858 .unlink = btrfs_unlink,
7859 .link = btrfs_link,
7860 .mkdir = btrfs_mkdir,
7861 .rmdir = btrfs_rmdir,
7862 .rename = btrfs_rename,
7863 .symlink = btrfs_symlink,
7864 .setattr = btrfs_setattr,
618e21d5 7865 .mknod = btrfs_mknod,
95819c05
CH		 7866 .setxattr = btrfs_setxattr,
7867 .getxattr = btrfs_getxattr,
5103e947 7868 .listxattr = btrfs_listxattr,
95819c05 7869 .removexattr = btrfs_removexattr,
fdebe2bd 7870 .permission = btrfs_permission,
4e34e719 7871 .get_acl = btrfs_get_acl,
39279cc3 7872};
6e1d5dcc 7873static const struct inode_operations btrfs_dir_ro_inode_operations = {
39279cc3 7874 .lookup = btrfs_lookup,
fdebe2bd 7875 .permission = btrfs_permission,
4e34e719 7876 .get_acl = btrfs_get_acl,
39279cc3 7877};
76dda93c 7878
828c0950 7879static const struct file_operations btrfs_dir_file_operations = {
39279cc3
CM		 7880 .llseek = generic_file_llseek,
7881 .read = generic_read_dir,
cbdf5a24 7882 .readdir = btrfs_real_readdir,
34287aa3 7883 .unlocked_ioctl = btrfs_ioctl,
39279cc3 7884#ifdef CONFIG_COMPAT
34287aa3 7885 .compat_ioctl = btrfs_ioctl,
39279cc3 7886#endif
6bf13c0c 7887 .release = btrfs_release_file,
e02119d5 7888 .fsync = btrfs_sync_file,
39279cc3
CM		 7889};
7890
d1310b2e 7891static struct extent_io_ops btrfs_extent_io_ops = {
07157aac 7892 .fill_delalloc = run_delalloc_range,
065631f6 7893 .submit_bio_hook = btrfs_submit_bio_hook,
239b14b3 7894 .merge_bio_hook = btrfs_merge_bio_hook,
07157aac 7895 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
e6dcd2dc 7896 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
247e743c 7897 .writepage_start_hook = btrfs_writepage_start_hook,
b0c68f8b
CM		 7898 .set_bit_hook = btrfs_set_bit_hook,
7899 .clear_bit_hook = btrfs_clear_bit_hook,
9ed74f2d
JB		 7900 .merge_extent_hook = btrfs_merge_extent_hook,
7901 .split_extent_hook = btrfs_split_extent_hook,
07157aac
CM		 7902};
7903
35054394
CM		 7904/*
7905 * btrfs doesn't support the bmap operation because swapfiles
7906 * use bmap to make a mapping of extents in the file. They assume
7907 * these extents won't change over the life of the file and they
7908 * use the bmap result to do IO directly to the drive.
7909 *
7910 * the btrfs bmap call would return logical addresses that aren't
7911 * suitable for IO and they also will change frequently as COW
7912 * operations happen. So, swapfile + btrfs == corruption.
7913 *
7914 * For now we're avoiding this by dropping bmap.
7915 */
7f09410b 7916static const struct address_space_operations btrfs_aops = {
39279cc3
CM		 7917 .readpage = btrfs_readpage,
7918 .writepage = btrfs_writepage,
b293f02e 7919 .writepages = btrfs_writepages,
3ab2fb5a 7920 .readpages = btrfs_readpages,
16432985 7921 .direct_IO = btrfs_direct_IO,
a52d9a80
CM		 7922 .invalidatepage = btrfs_invalidatepage,
7923 .releasepage = btrfs_releasepage,
e6dcd2dc 7924 .set_page_dirty = btrfs_set_page_dirty,
465fdd97 7925 .error_remove_page = generic_error_remove_page,
39279cc3
CM		 7926};
7927
7f09410b 7928static const struct address_space_operations btrfs_symlink_aops = {
39279cc3
CM		 7929 .readpage = btrfs_readpage,
7930 .writepage = btrfs_writepage,
2bf5a725
CM		 7931 .invalidatepage = btrfs_invalidatepage,
7932 .releasepage = btrfs_releasepage,
39279cc3
CM		 7933};
7934
6e1d5dcc 7935static const struct inode_operations btrfs_file_inode_operations = {
39279cc3
CM		 7936 .getattr = btrfs_getattr,
7937 .setattr = btrfs_setattr,
95819c05
CH		 7938 .setxattr = btrfs_setxattr,
7939 .getxattr = btrfs_getxattr,
5103e947 7940 .listxattr = btrfs_listxattr,
95819c05 7941 .removexattr = btrfs_removexattr,
fdebe2bd 7942 .permission = btrfs_permission,
1506fcc8 7943 .fiemap = btrfs_fiemap,
4e34e719 7944 .get_acl = btrfs_get_acl,
e41f941a 7945 .update_time = btrfs_update_time,
39279cc3 7946};
6e1d5dcc 7947static const struct inode_operations btrfs_special_inode_operations = {
618e21d5
JB		 7948 .getattr = btrfs_getattr,
7949 .setattr = btrfs_setattr,
fdebe2bd 7950 .permission = btrfs_permission,
95819c05
CH		 7951 .setxattr = btrfs_setxattr,
7952 .getxattr = btrfs_getxattr,
33268eaf 7953 .listxattr = btrfs_listxattr,
95819c05 7954 .removexattr = btrfs_removexattr,
4e34e719 7955 .get_acl = btrfs_get_acl,
e41f941a 7956 .update_time = btrfs_update_time,
618e21d5 7957};
6e1d5dcc 7958static const struct inode_operations btrfs_symlink_inode_operations = {
39279cc3
CM		 7959 .readlink = generic_readlink,
7960 .follow_link = page_follow_link_light,
7961 .put_link = page_put_link,
f209561a 7962 .getattr = btrfs_getattr,
22c44fe6 7963 .setattr = btrfs_setattr,
fdebe2bd 7964 .permission = btrfs_permission,
0279b4cd
JO		 7965 .setxattr = btrfs_setxattr,
7966 .getxattr = btrfs_getxattr,
7967 .listxattr = btrfs_listxattr,
7968 .removexattr = btrfs_removexattr,
4e34e719 7969 .get_acl = btrfs_get_acl,
e41f941a 7970 .update_time = btrfs_update_time,
39279cc3 7971};
76dda93c 7972
82d339d9 7973const struct dentry_operations btrfs_dentry_operations = {
76dda93c 7974 .d_delete = btrfs_dentry_delete,
b4aff1f8 7975 .d_release = btrfs_dentry_release,
76dda93c 7976};
kernel source for telekom puls (alcatel/mediatek)