Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
ac27a0ec DK |
15 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
16 | * (jj@sunsite.ms.mff.cuni.cz) | |
17 | * | |
617ba13b | 18 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
19 | */ |
20 | ||
ac27a0ec DK |
21 | #include <linux/fs.h> |
22 | #include <linux/time.h> | |
dab291af | 23 | #include <linux/jbd2.h> |
ac27a0ec DK |
24 | #include <linux/highuid.h> |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/quotaops.h> | |
27 | #include <linux/string.h> | |
28 | #include <linux/buffer_head.h> | |
29 | #include <linux/writeback.h> | |
64769240 | 30 | #include <linux/pagevec.h> |
ac27a0ec | 31 | #include <linux/mpage.h> |
e83c1397 | 32 | #include <linux/namei.h> |
ac27a0ec DK |
33 | #include <linux/uio.h> |
34 | #include <linux/bio.h> | |
4c0425ff | 35 | #include <linux/workqueue.h> |
744692dc | 36 | #include <linux/kernel.h> |
6db26ffc | 37 | #include <linux/printk.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
a8901d34 | 39 | #include <linux/ratelimit.h> |
a27bb332 | 40 | #include <linux/aio.h> |
9bffad1e | 41 | |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
ac27a0ec DK |
43 | #include "xattr.h" |
44 | #include "acl.h" | |
9f125d64 | 45 | #include "truncate.h" |
ac27a0ec | 46 | |
9bffad1e TT |
47 | #include <trace/events/ext4.h> |
48 | ||
a1d6cc56 AK |
49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
50 | ||
814525f4 DW |
51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
52 | struct ext4_inode_info *ei) | |
53 | { | |
54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
55 | __u16 csum_lo; | |
56 | __u16 csum_hi = 0; | |
57 | __u32 csum; | |
58 | ||
171a7f21 | 59 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
60 | raw->i_checksum_lo = 0; |
61 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
62 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 63 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
64 | raw->i_checksum_hi = 0; |
65 | } | |
66 | ||
67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
68 | EXT4_INODE_SIZE(inode->i_sb)); | |
69 | ||
171a7f21 | 70 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
71 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
72 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 73 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
74 | |
75 | return csum; | |
76 | } | |
77 | ||
78 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
79 | struct ext4_inode_info *ei) | |
80 | { | |
81 | __u32 provided, calculated; | |
82 | ||
83 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
84 | cpu_to_le32(EXT4_OS_LINUX) || | |
85 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
86 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
87 | return 1; | |
88 | ||
89 | provided = le16_to_cpu(raw->i_checksum_lo); | |
90 | calculated = ext4_inode_csum(inode, raw, ei); | |
91 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
92 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
93 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
94 | else | |
95 | calculated &= 0xFFFF; | |
96 | ||
97 | return provided == calculated; | |
98 | } | |
99 | ||
100 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
101 | struct ext4_inode_info *ei) | |
102 | { | |
103 | __u32 csum; | |
104 | ||
105 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
106 | cpu_to_le32(EXT4_OS_LINUX) || | |
107 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
108 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
109 | return; | |
110 | ||
111 | csum = ext4_inode_csum(inode, raw, ei); | |
112 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
113 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
114 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
115 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
116 | } | |
117 | ||
678aaf48 JK |
118 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
119 | loff_t new_size) | |
120 | { | |
7ff9c073 | 121 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
122 | /* |
123 | * If jinode is zero, then we never opened the file for | |
124 | * writing, so there's no need to call | |
125 | * jbd2_journal_begin_ordered_truncate() since there's no | |
126 | * outstanding writes we need to flush. | |
127 | */ | |
128 | if (!EXT4_I(inode)->jinode) | |
129 | return 0; | |
130 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
131 | EXT4_I(inode)->jinode, | |
132 | new_size); | |
678aaf48 JK |
133 | } |
134 | ||
64769240 | 135 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
136 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
137 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
5f163cc7 ES |
138 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
139 | struct inode *inode, struct page *page, loff_t from, | |
140 | loff_t length, int flags); | |
64769240 | 141 | |
ac27a0ec DK |
142 | /* |
143 | * Test whether an inode is a fast symlink. | |
144 | */ | |
617ba13b | 145 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 146 | { |
617ba13b | 147 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
148 | (inode->i_sb->s_blocksize >> 9) : 0; |
149 | ||
150 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
151 | } | |
152 | ||
ac27a0ec DK |
153 | /* |
154 | * Restart the transaction associated with *handle. This does a commit, | |
155 | * so before we call here everything must be consistently dirtied against | |
156 | * this transaction. | |
157 | */ | |
fa5d1113 | 158 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 159 | int nblocks) |
ac27a0ec | 160 | { |
487caeef JK |
161 | int ret; |
162 | ||
163 | /* | |
e35fd660 | 164 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
165 | * moment, get_block can be called only for blocks inside i_size since |
166 | * page cache has been already dropped and writes are blocked by | |
167 | * i_mutex. So we can safely drop the i_data_sem here. | |
168 | */ | |
0390131b | 169 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 170 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 171 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 172 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 173 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 174 | ext4_discard_preallocations(inode); |
487caeef JK |
175 | |
176 | return ret; | |
ac27a0ec DK |
177 | } |
178 | ||
179 | /* | |
180 | * Called at the last iput() if i_nlink is zero. | |
181 | */ | |
0930fcc1 | 182 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
183 | { |
184 | handle_t *handle; | |
bc965ab3 | 185 | int err; |
ac27a0ec | 186 | |
7ff9c073 | 187 | trace_ext4_evict_inode(inode); |
2581fdc8 | 188 | |
0930fcc1 | 189 | if (inode->i_nlink) { |
2d859db3 JK |
190 | /* |
191 | * When journalling data dirty buffers are tracked only in the | |
192 | * journal. So although mm thinks everything is clean and | |
193 | * ready for reaping the inode might still have some pages to | |
194 | * write in the running transaction or waiting to be | |
195 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
196 | * (via truncate_inode_pages()) to discard these buffers can | |
197 | * cause data loss. Also even if we did not discard these | |
198 | * buffers, we would have no way to find them after the inode | |
199 | * is reaped and thus user could see stale data if he tries to | |
200 | * read them before the transaction is checkpointed. So be | |
201 | * careful and force everything to disk here... We use | |
202 | * ei->i_datasync_tid to store the newest transaction | |
203 | * containing inode's data. | |
204 | * | |
205 | * Note that directories do not have this problem because they | |
206 | * don't use page cache. | |
207 | */ | |
208 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
209 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
210 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
211 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
212 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
213 | ||
d76a3a77 | 214 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
215 | filemap_write_and_wait(&inode->i_data); |
216 | } | |
0930fcc1 | 217 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 218 | ext4_ioend_shutdown(inode); |
0930fcc1 AV |
219 | goto no_delete; |
220 | } | |
221 | ||
907f4554 | 222 | if (!is_bad_inode(inode)) |
871a2931 | 223 | dquot_initialize(inode); |
907f4554 | 224 | |
678aaf48 JK |
225 | if (ext4_should_order_data(inode)) |
226 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec | 227 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 228 | ext4_ioend_shutdown(inode); |
ac27a0ec DK |
229 | |
230 | if (is_bad_inode(inode)) | |
231 | goto no_delete; | |
232 | ||
8e8ad8a5 JK |
233 | /* |
234 | * Protect us against freezing - iput() caller didn't have to have any | |
235 | * protection against it | |
236 | */ | |
237 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
238 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
239 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 240 | if (IS_ERR(handle)) { |
bc965ab3 | 241 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
242 | /* |
243 | * If we're going to skip the normal cleanup, we still need to | |
244 | * make sure that the in-core orphan linked list is properly | |
245 | * cleaned up. | |
246 | */ | |
617ba13b | 247 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 248 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
249 | goto no_delete; |
250 | } | |
251 | ||
252 | if (IS_SYNC(inode)) | |
0390131b | 253 | ext4_handle_sync(handle); |
ac27a0ec | 254 | inode->i_size = 0; |
bc965ab3 TT |
255 | err = ext4_mark_inode_dirty(handle, inode); |
256 | if (err) { | |
12062ddd | 257 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
258 | "couldn't mark inode dirty (err %d)", err); |
259 | goto stop_handle; | |
260 | } | |
ac27a0ec | 261 | if (inode->i_blocks) |
617ba13b | 262 | ext4_truncate(inode); |
bc965ab3 TT |
263 | |
264 | /* | |
265 | * ext4_ext_truncate() doesn't reserve any slop when it | |
266 | * restarts journal transactions; therefore there may not be | |
267 | * enough credits left in the handle to remove the inode from | |
268 | * the orphan list and set the dtime field. | |
269 | */ | |
0390131b | 270 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
271 | err = ext4_journal_extend(handle, 3); |
272 | if (err > 0) | |
273 | err = ext4_journal_restart(handle, 3); | |
274 | if (err != 0) { | |
12062ddd | 275 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
276 | "couldn't extend journal (err %d)", err); |
277 | stop_handle: | |
278 | ext4_journal_stop(handle); | |
45388219 | 279 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 280 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
281 | goto no_delete; |
282 | } | |
283 | } | |
284 | ||
ac27a0ec | 285 | /* |
617ba13b | 286 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 287 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 288 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 289 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 290 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
291 | * (Well, we could do this if we need to, but heck - it works) |
292 | */ | |
617ba13b MC |
293 | ext4_orphan_del(handle, inode); |
294 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
295 | |
296 | /* | |
297 | * One subtle ordering requirement: if anything has gone wrong | |
298 | * (transaction abort, IO errors, whatever), then we can still | |
299 | * do these next steps (the fs will already have been marked as | |
300 | * having errors), but we can't free the inode if the mark_dirty | |
301 | * fails. | |
302 | */ | |
617ba13b | 303 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 304 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 305 | ext4_clear_inode(inode); |
ac27a0ec | 306 | else |
617ba13b MC |
307 | ext4_free_inode(handle, inode); |
308 | ext4_journal_stop(handle); | |
8e8ad8a5 | 309 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
310 | return; |
311 | no_delete: | |
0930fcc1 | 312 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
313 | } |
314 | ||
a9e7f447 DM |
315 | #ifdef CONFIG_QUOTA |
316 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 317 | { |
a9e7f447 | 318 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 319 | } |
a9e7f447 | 320 | #endif |
9d0be502 | 321 | |
12219aea AK |
322 | /* |
323 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 324 | * to allocate a block located at @lblock |
12219aea | 325 | */ |
01f49d0b | 326 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 327 | { |
12e9b892 | 328 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 329 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 330 | |
8bb2b247 | 331 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
332 | } |
333 | ||
0637c6f4 TT |
334 | /* |
335 | * Called with i_data_sem down, which is important since we can call | |
336 | * ext4_discard_preallocations() from here. | |
337 | */ | |
5f634d06 AK |
338 | void ext4_da_update_reserve_space(struct inode *inode, |
339 | int used, int quota_claim) | |
12219aea AK |
340 | { |
341 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 342 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
343 | |
344 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 345 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 346 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 347 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 348 | "with only %d reserved data blocks", |
0637c6f4 TT |
349 | __func__, inode->i_ino, used, |
350 | ei->i_reserved_data_blocks); | |
351 | WARN_ON(1); | |
352 | used = ei->i_reserved_data_blocks; | |
353 | } | |
12219aea | 354 | |
97795d2a | 355 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
01a523eb TT |
356 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
357 | "with only %d reserved metadata blocks " | |
358 | "(releasing %d blocks with reserved %d data blocks)", | |
359 | inode->i_ino, ei->i_allocated_meta_blocks, | |
360 | ei->i_reserved_meta_blocks, used, | |
361 | ei->i_reserved_data_blocks); | |
97795d2a BF |
362 | WARN_ON(1); |
363 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; | |
364 | } | |
365 | ||
0637c6f4 TT |
366 | /* Update per-inode reservations */ |
367 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 368 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 369 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 370 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 371 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 372 | |
0637c6f4 TT |
373 | if (ei->i_reserved_data_blocks == 0) { |
374 | /* | |
375 | * We can release all of the reserved metadata blocks | |
376 | * only when we have written all of the delayed | |
377 | * allocation blocks. | |
378 | */ | |
57042651 | 379 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 380 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 381 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 382 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 383 | } |
12219aea | 384 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 385 | |
72b8ab9d ES |
386 | /* Update quota subsystem for data blocks */ |
387 | if (quota_claim) | |
7b415bf6 | 388 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 389 | else { |
5f634d06 AK |
390 | /* |
391 | * We did fallocate with an offset that is already delayed | |
392 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 393 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 394 | */ |
7b415bf6 | 395 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 396 | } |
d6014301 AK |
397 | |
398 | /* | |
399 | * If we have done all the pending block allocations and if | |
400 | * there aren't any writers on the inode, we can discard the | |
401 | * inode's preallocations. | |
402 | */ | |
0637c6f4 TT |
403 | if ((ei->i_reserved_data_blocks == 0) && |
404 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 405 | ext4_discard_preallocations(inode); |
12219aea AK |
406 | } |
407 | ||
e29136f8 | 408 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
409 | unsigned int line, |
410 | struct ext4_map_blocks *map) | |
6fd058f7 | 411 | { |
24676da4 TT |
412 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
413 | map->m_len)) { | |
c398eda0 TT |
414 | ext4_error_inode(inode, func, line, map->m_pblk, |
415 | "lblock %lu mapped to illegal pblock " | |
416 | "(length %d)", (unsigned long) map->m_lblk, | |
417 | map->m_len); | |
6fd058f7 TT |
418 | return -EIO; |
419 | } | |
420 | return 0; | |
421 | } | |
422 | ||
e29136f8 | 423 | #define check_block_validity(inode, map) \ |
c398eda0 | 424 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 425 | |
55138e0b | 426 | /* |
1f94533d TT |
427 | * Return the number of contiguous dirty pages in a given inode |
428 | * starting at page frame idx. | |
55138e0b TT |
429 | */ |
430 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
431 | unsigned int max_pages) | |
432 | { | |
433 | struct address_space *mapping = inode->i_mapping; | |
434 | pgoff_t index; | |
435 | struct pagevec pvec; | |
436 | pgoff_t num = 0; | |
437 | int i, nr_pages, done = 0; | |
438 | ||
439 | if (max_pages == 0) | |
440 | return 0; | |
441 | pagevec_init(&pvec, 0); | |
442 | while (!done) { | |
443 | index = idx; | |
444 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
445 | PAGECACHE_TAG_DIRTY, | |
446 | (pgoff_t)PAGEVEC_SIZE); | |
447 | if (nr_pages == 0) | |
448 | break; | |
449 | for (i = 0; i < nr_pages; i++) { | |
450 | struct page *page = pvec.pages[i]; | |
451 | struct buffer_head *bh, *head; | |
452 | ||
453 | lock_page(page); | |
454 | if (unlikely(page->mapping != mapping) || | |
455 | !PageDirty(page) || | |
456 | PageWriteback(page) || | |
457 | page->index != idx) { | |
458 | done = 1; | |
459 | unlock_page(page); | |
460 | break; | |
461 | } | |
1f94533d TT |
462 | if (page_has_buffers(page)) { |
463 | bh = head = page_buffers(page); | |
464 | do { | |
465 | if (!buffer_delay(bh) && | |
466 | !buffer_unwritten(bh)) | |
467 | done = 1; | |
468 | bh = bh->b_this_page; | |
469 | } while (!done && (bh != head)); | |
470 | } | |
55138e0b TT |
471 | unlock_page(page); |
472 | if (done) | |
473 | break; | |
474 | idx++; | |
475 | num++; | |
659c6009 ES |
476 | if (num >= max_pages) { |
477 | done = 1; | |
55138e0b | 478 | break; |
659c6009 | 479 | } |
55138e0b TT |
480 | } |
481 | pagevec_release(&pvec); | |
482 | } | |
483 | return num; | |
484 | } | |
485 | ||
921f266b DM |
486 | #ifdef ES_AGGRESSIVE_TEST |
487 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
488 | struct inode *inode, | |
489 | struct ext4_map_blocks *es_map, | |
490 | struct ext4_map_blocks *map, | |
491 | int flags) | |
492 | { | |
493 | int retval; | |
494 | ||
495 | map->m_flags = 0; | |
496 | /* | |
497 | * There is a race window that the result is not the same. | |
498 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
499 | * is that we lookup a block mapping in extent status tree with | |
500 | * out taking i_data_sem. So at the time the unwritten extent | |
501 | * could be converted. | |
502 | */ | |
503 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
504 | down_read((&EXT4_I(inode)->i_data_sem)); | |
505 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
506 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
507 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
508 | } else { | |
509 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
510 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
511 | } | |
512 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
513 | up_read((&EXT4_I(inode)->i_data_sem)); | |
514 | /* | |
515 | * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag | |
516 | * because it shouldn't be marked in es_map->m_flags. | |
517 | */ | |
518 | map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); | |
519 | ||
520 | /* | |
521 | * We don't check m_len because extent will be collpased in status | |
522 | * tree. So the m_len might not equal. | |
523 | */ | |
524 | if (es_map->m_lblk != map->m_lblk || | |
525 | es_map->m_flags != map->m_flags || | |
526 | es_map->m_pblk != map->m_pblk) { | |
527 | printk("ES cache assertation failed for inode: %lu " | |
528 | "es_cached ex [%d/%d/%llu/%x] != " | |
529 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
530 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
531 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
532 | map->m_len, map->m_pblk, map->m_flags, | |
533 | retval, flags); | |
534 | } | |
535 | } | |
536 | #endif /* ES_AGGRESSIVE_TEST */ | |
537 | ||
f5ab0d1f | 538 | /* |
e35fd660 | 539 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 540 | * and returns if the blocks are already mapped. |
f5ab0d1f | 541 | * |
f5ab0d1f MC |
542 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
543 | * and store the allocated blocks in the result buffer head and mark it | |
544 | * mapped. | |
545 | * | |
e35fd660 TT |
546 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
547 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
548 | * based files |
549 | * | |
550 | * On success, it returns the number of blocks being mapped or allocate. | |
551 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
552 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
553 | * the buffer head is mapped. | |
554 | * | |
555 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 556 | * that case, buffer head is unmapped |
f5ab0d1f MC |
557 | * |
558 | * It returns the error in case of allocation failure. | |
559 | */ | |
e35fd660 TT |
560 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
561 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 562 | { |
d100eef2 | 563 | struct extent_status es; |
0e855ac8 | 564 | int retval; |
921f266b DM |
565 | #ifdef ES_AGGRESSIVE_TEST |
566 | struct ext4_map_blocks orig_map; | |
567 | ||
568 | memcpy(&orig_map, map, sizeof(*map)); | |
569 | #endif | |
f5ab0d1f | 570 | |
e35fd660 TT |
571 | map->m_flags = 0; |
572 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
573 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
574 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
575 | |
576 | /* Lookup extent status tree firstly */ | |
577 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
578 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
579 | map->m_pblk = ext4_es_pblock(&es) + | |
580 | map->m_lblk - es.es_lblk; | |
581 | map->m_flags |= ext4_es_is_written(&es) ? | |
582 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
583 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
584 | if (retval > map->m_len) | |
585 | retval = map->m_len; | |
586 | map->m_len = retval; | |
587 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
588 | retval = 0; | |
589 | } else { | |
590 | BUG_ON(1); | |
591 | } | |
921f266b DM |
592 | #ifdef ES_AGGRESSIVE_TEST |
593 | ext4_map_blocks_es_recheck(handle, inode, map, | |
594 | &orig_map, flags); | |
595 | #endif | |
d100eef2 ZL |
596 | goto found; |
597 | } | |
598 | ||
4df3d265 | 599 | /* |
b920c755 TT |
600 | * Try to see if we can get the block without requesting a new |
601 | * file system block. | |
4df3d265 | 602 | */ |
729f52c6 ZL |
603 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
604 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 605 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
606 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
607 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 608 | } else { |
a4e5d88b DM |
609 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
610 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 611 | } |
f7fec032 ZL |
612 | if (retval > 0) { |
613 | int ret; | |
614 | unsigned long long status; | |
615 | ||
921f266b DM |
616 | #ifdef ES_AGGRESSIVE_TEST |
617 | if (retval != map->m_len) { | |
618 | printk("ES len assertation failed for inode: %lu " | |
619 | "retval %d != map->m_len %d " | |
620 | "in %s (lookup)\n", inode->i_ino, retval, | |
621 | map->m_len, __func__); | |
622 | } | |
623 | #endif | |
624 | ||
f7fec032 ZL |
625 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
626 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
627 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
628 | ext4_find_delalloc_range(inode, map->m_lblk, | |
629 | map->m_lblk + map->m_len - 1)) | |
630 | status |= EXTENT_STATUS_DELAYED; | |
631 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
632 | map->m_len, map->m_pblk, status); | |
633 | if (ret < 0) | |
634 | retval = ret; | |
635 | } | |
729f52c6 ZL |
636 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
637 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 638 | |
d100eef2 | 639 | found: |
e35fd660 | 640 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
f7fec032 | 641 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
642 | if (ret != 0) |
643 | return ret; | |
644 | } | |
645 | ||
f5ab0d1f | 646 | /* If it is only a block(s) look up */ |
c2177057 | 647 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
648 | return retval; |
649 | ||
650 | /* | |
651 | * Returns if the blocks have already allocated | |
652 | * | |
653 | * Note that if blocks have been preallocated | |
df3ab170 | 654 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
655 | * with buffer head unmapped. |
656 | */ | |
e35fd660 | 657 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
658 | return retval; |
659 | ||
2a8964d6 | 660 | /* |
a25a4e1a ZL |
661 | * Here we clear m_flags because after allocating an new extent, |
662 | * it will be set again. | |
2a8964d6 | 663 | */ |
a25a4e1a | 664 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 665 | |
4df3d265 | 666 | /* |
f5ab0d1f MC |
667 | * New blocks allocate and/or writing to uninitialized extent |
668 | * will possibly result in updating i_data, so we take | |
669 | * the write lock of i_data_sem, and call get_blocks() | |
670 | * with create == 1 flag. | |
4df3d265 AK |
671 | */ |
672 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
673 | |
674 | /* | |
675 | * if the caller is from delayed allocation writeout path | |
676 | * we have already reserved fs blocks for allocation | |
677 | * let the underlying get_block() function know to | |
678 | * avoid double accounting | |
679 | */ | |
c2177057 | 680 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 681 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
682 | /* |
683 | * We need to check for EXT4 here because migrate | |
684 | * could have changed the inode type in between | |
685 | */ | |
12e9b892 | 686 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 687 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 688 | } else { |
e35fd660 | 689 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 690 | |
e35fd660 | 691 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
692 | /* |
693 | * We allocated new blocks which will result in | |
694 | * i_data's format changing. Force the migrate | |
695 | * to fail by clearing migrate flags | |
696 | */ | |
19f5fb7a | 697 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 698 | } |
d2a17637 | 699 | |
5f634d06 AK |
700 | /* |
701 | * Update reserved blocks/metadata blocks after successful | |
702 | * block allocation which had been deferred till now. We don't | |
703 | * support fallocate for non extent files. So we can update | |
704 | * reserve space here. | |
705 | */ | |
706 | if ((retval > 0) && | |
1296cc85 | 707 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
708 | ext4_da_update_reserve_space(inode, retval, 1); |
709 | } | |
f7fec032 | 710 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 711 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 712 | |
f7fec032 ZL |
713 | if (retval > 0) { |
714 | int ret; | |
715 | unsigned long long status; | |
716 | ||
921f266b DM |
717 | #ifdef ES_AGGRESSIVE_TEST |
718 | if (retval != map->m_len) { | |
719 | printk("ES len assertation failed for inode: %lu " | |
720 | "retval %d != map->m_len %d " | |
721 | "in %s (allocation)\n", inode->i_ino, retval, | |
722 | map->m_len, __func__); | |
723 | } | |
724 | #endif | |
725 | ||
adb23551 ZL |
726 | /* |
727 | * If the extent has been zeroed out, we don't need to update | |
728 | * extent status tree. | |
729 | */ | |
730 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
731 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
732 | if (ext4_es_is_written(&es)) | |
733 | goto has_zeroout; | |
734 | } | |
f7fec032 ZL |
735 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
736 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
737 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
738 | ext4_find_delalloc_range(inode, map->m_lblk, | |
739 | map->m_lblk + map->m_len - 1)) | |
740 | status |= EXTENT_STATUS_DELAYED; | |
741 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
742 | map->m_pblk, status); | |
743 | if (ret < 0) | |
744 | retval = ret; | |
5356f261 AK |
745 | } |
746 | ||
adb23551 | 747 | has_zeroout: |
4df3d265 | 748 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 749 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 750 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
751 | if (ret != 0) |
752 | return ret; | |
753 | } | |
0e855ac8 AK |
754 | return retval; |
755 | } | |
756 | ||
f3bd1f3f MC |
757 | /* Maximum number of blocks we map for direct IO at once. */ |
758 | #define DIO_MAX_BLOCKS 4096 | |
759 | ||
2ed88685 TT |
760 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
761 | struct buffer_head *bh, int flags) | |
ac27a0ec | 762 | { |
3e4fdaf8 | 763 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 764 | struct ext4_map_blocks map; |
7fb5409d | 765 | int ret = 0, started = 0; |
f3bd1f3f | 766 | int dio_credits; |
ac27a0ec | 767 | |
46c7f254 TM |
768 | if (ext4_has_inline_data(inode)) |
769 | return -ERANGE; | |
770 | ||
2ed88685 TT |
771 | map.m_lblk = iblock; |
772 | map.m_len = bh->b_size >> inode->i_blkbits; | |
773 | ||
8b0f165f | 774 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 775 | /* Direct IO write... */ |
2ed88685 TT |
776 | if (map.m_len > DIO_MAX_BLOCKS) |
777 | map.m_len = DIO_MAX_BLOCKS; | |
778 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
779 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
780 | dio_credits); | |
7fb5409d | 781 | if (IS_ERR(handle)) { |
ac27a0ec | 782 | ret = PTR_ERR(handle); |
2ed88685 | 783 | return ret; |
ac27a0ec | 784 | } |
7fb5409d | 785 | started = 1; |
ac27a0ec DK |
786 | } |
787 | ||
2ed88685 | 788 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 789 | if (ret > 0) { |
2ed88685 TT |
790 | map_bh(bh, inode->i_sb, map.m_pblk); |
791 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
792 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 793 | ret = 0; |
ac27a0ec | 794 | } |
7fb5409d JK |
795 | if (started) |
796 | ext4_journal_stop(handle); | |
ac27a0ec DK |
797 | return ret; |
798 | } | |
799 | ||
2ed88685 TT |
800 | int ext4_get_block(struct inode *inode, sector_t iblock, |
801 | struct buffer_head *bh, int create) | |
802 | { | |
803 | return _ext4_get_block(inode, iblock, bh, | |
804 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
805 | } | |
806 | ||
ac27a0ec DK |
807 | /* |
808 | * `handle' can be NULL if create is zero | |
809 | */ | |
617ba13b | 810 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 811 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 812 | { |
2ed88685 TT |
813 | struct ext4_map_blocks map; |
814 | struct buffer_head *bh; | |
ac27a0ec DK |
815 | int fatal = 0, err; |
816 | ||
817 | J_ASSERT(handle != NULL || create == 0); | |
818 | ||
2ed88685 TT |
819 | map.m_lblk = block; |
820 | map.m_len = 1; | |
821 | err = ext4_map_blocks(handle, inode, &map, | |
822 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 823 | |
90b0a973 CM |
824 | /* ensure we send some value back into *errp */ |
825 | *errp = 0; | |
826 | ||
0f70b406 TT |
827 | if (create && err == 0) |
828 | err = -ENOSPC; /* should never happen */ | |
2ed88685 TT |
829 | if (err < 0) |
830 | *errp = err; | |
831 | if (err <= 0) | |
832 | return NULL; | |
2ed88685 TT |
833 | |
834 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
aebf0243 | 835 | if (unlikely(!bh)) { |
860d21e2 | 836 | *errp = -ENOMEM; |
2ed88685 | 837 | return NULL; |
ac27a0ec | 838 | } |
2ed88685 TT |
839 | if (map.m_flags & EXT4_MAP_NEW) { |
840 | J_ASSERT(create != 0); | |
841 | J_ASSERT(handle != NULL); | |
ac27a0ec | 842 | |
2ed88685 TT |
843 | /* |
844 | * Now that we do not always journal data, we should | |
845 | * keep in mind whether this should always journal the | |
846 | * new buffer as metadata. For now, regular file | |
847 | * writes use ext4_get_block instead, so it's not a | |
848 | * problem. | |
849 | */ | |
850 | lock_buffer(bh); | |
851 | BUFFER_TRACE(bh, "call get_create_access"); | |
852 | fatal = ext4_journal_get_create_access(handle, bh); | |
853 | if (!fatal && !buffer_uptodate(bh)) { | |
854 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
855 | set_buffer_uptodate(bh); | |
ac27a0ec | 856 | } |
2ed88685 TT |
857 | unlock_buffer(bh); |
858 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
859 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
860 | if (!fatal) | |
861 | fatal = err; | |
862 | } else { | |
863 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 864 | } |
2ed88685 TT |
865 | if (fatal) { |
866 | *errp = fatal; | |
867 | brelse(bh); | |
868 | bh = NULL; | |
869 | } | |
870 | return bh; | |
ac27a0ec DK |
871 | } |
872 | ||
617ba13b | 873 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 874 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 875 | { |
af5bc92d | 876 | struct buffer_head *bh; |
ac27a0ec | 877 | |
617ba13b | 878 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
879 | if (!bh) |
880 | return bh; | |
881 | if (buffer_uptodate(bh)) | |
882 | return bh; | |
65299a3b | 883 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
884 | wait_on_buffer(bh); |
885 | if (buffer_uptodate(bh)) | |
886 | return bh; | |
887 | put_bh(bh); | |
888 | *err = -EIO; | |
889 | return NULL; | |
890 | } | |
891 | ||
f19d5870 TM |
892 | int ext4_walk_page_buffers(handle_t *handle, |
893 | struct buffer_head *head, | |
894 | unsigned from, | |
895 | unsigned to, | |
896 | int *partial, | |
897 | int (*fn)(handle_t *handle, | |
898 | struct buffer_head *bh)) | |
ac27a0ec DK |
899 | { |
900 | struct buffer_head *bh; | |
901 | unsigned block_start, block_end; | |
902 | unsigned blocksize = head->b_size; | |
903 | int err, ret = 0; | |
904 | struct buffer_head *next; | |
905 | ||
af5bc92d TT |
906 | for (bh = head, block_start = 0; |
907 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 908 | block_start = block_end, bh = next) { |
ac27a0ec DK |
909 | next = bh->b_this_page; |
910 | block_end = block_start + blocksize; | |
911 | if (block_end <= from || block_start >= to) { | |
912 | if (partial && !buffer_uptodate(bh)) | |
913 | *partial = 1; | |
914 | continue; | |
915 | } | |
916 | err = (*fn)(handle, bh); | |
917 | if (!ret) | |
918 | ret = err; | |
919 | } | |
920 | return ret; | |
921 | } | |
922 | ||
923 | /* | |
924 | * To preserve ordering, it is essential that the hole instantiation and | |
925 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 926 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 927 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
928 | * prepare_write() is the right place. |
929 | * | |
36ade451 JK |
930 | * Also, this function can nest inside ext4_writepage(). In that case, we |
931 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
932 | * whole page. So we won't block on the journal in that case, which is good, | |
933 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 934 | * |
617ba13b | 935 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
936 | * quota file writes. If we were to commit the transaction while thus |
937 | * reentered, there can be a deadlock - we would be holding a quota | |
938 | * lock, and the commit would never complete if another thread had a | |
939 | * transaction open and was blocking on the quota lock - a ranking | |
940 | * violation. | |
941 | * | |
dab291af | 942 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
943 | * will _not_ run commit under these circumstances because handle->h_ref |
944 | * is elevated. We'll still have enough credits for the tiny quotafile | |
945 | * write. | |
946 | */ | |
f19d5870 TM |
947 | int do_journal_get_write_access(handle_t *handle, |
948 | struct buffer_head *bh) | |
ac27a0ec | 949 | { |
56d35a4c JK |
950 | int dirty = buffer_dirty(bh); |
951 | int ret; | |
952 | ||
ac27a0ec DK |
953 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
954 | return 0; | |
56d35a4c | 955 | /* |
ebdec241 | 956 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
957 | * the dirty bit as jbd2_journal_get_write_access() could complain |
958 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 959 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
960 | * the bit before releasing a page lock and thus writeback cannot |
961 | * ever write the buffer. | |
962 | */ | |
963 | if (dirty) | |
964 | clear_buffer_dirty(bh); | |
965 | ret = ext4_journal_get_write_access(handle, bh); | |
966 | if (!ret && dirty) | |
967 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
968 | return ret; | |
ac27a0ec DK |
969 | } |
970 | ||
8b0f165f AP |
971 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
972 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 973 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
974 | loff_t pos, unsigned len, unsigned flags, |
975 | struct page **pagep, void **fsdata) | |
ac27a0ec | 976 | { |
af5bc92d | 977 | struct inode *inode = mapping->host; |
1938a150 | 978 | int ret, needed_blocks; |
ac27a0ec DK |
979 | handle_t *handle; |
980 | int retries = 0; | |
af5bc92d | 981 | struct page *page; |
de9a55b8 | 982 | pgoff_t index; |
af5bc92d | 983 | unsigned from, to; |
bfc1af65 | 984 | |
9bffad1e | 985 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
986 | /* |
987 | * Reserve one block more for addition to orphan list in case | |
988 | * we allocate blocks but write fails for some reason | |
989 | */ | |
990 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 991 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
992 | from = pos & (PAGE_CACHE_SIZE - 1); |
993 | to = from + len; | |
ac27a0ec | 994 | |
f19d5870 TM |
995 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
996 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
997 | flags, pagep); | |
998 | if (ret < 0) | |
47564bfb TT |
999 | return ret; |
1000 | if (ret == 1) | |
1001 | return 0; | |
f19d5870 TM |
1002 | } |
1003 | ||
47564bfb TT |
1004 | /* |
1005 | * grab_cache_page_write_begin() can take a long time if the | |
1006 | * system is thrashing due to memory pressure, or if the page | |
1007 | * is being written back. So grab it first before we start | |
1008 | * the transaction handle. This also allows us to allocate | |
1009 | * the page (if needed) without using GFP_NOFS. | |
1010 | */ | |
1011 | retry_grab: | |
1012 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1013 | if (!page) | |
1014 | return -ENOMEM; | |
1015 | unlock_page(page); | |
1016 | ||
1017 | retry_journal: | |
9924a92a | 1018 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1019 | if (IS_ERR(handle)) { |
47564bfb TT |
1020 | page_cache_release(page); |
1021 | return PTR_ERR(handle); | |
7479d2b9 | 1022 | } |
ac27a0ec | 1023 | |
47564bfb TT |
1024 | lock_page(page); |
1025 | if (page->mapping != mapping) { | |
1026 | /* The page got truncated from under us */ | |
1027 | unlock_page(page); | |
1028 | page_cache_release(page); | |
cf108bca | 1029 | ext4_journal_stop(handle); |
47564bfb | 1030 | goto retry_grab; |
cf108bca | 1031 | } |
47564bfb | 1032 | wait_on_page_writeback(page); |
cf108bca | 1033 | |
744692dc | 1034 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1035 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1036 | else |
6e1db88d | 1037 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1038 | |
1039 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 TM |
1040 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1041 | from, to, NULL, | |
1042 | do_journal_get_write_access); | |
ac27a0ec | 1043 | } |
bfc1af65 NP |
1044 | |
1045 | if (ret) { | |
af5bc92d | 1046 | unlock_page(page); |
ae4d5372 | 1047 | /* |
6e1db88d | 1048 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1049 | * outside i_size. Trim these off again. Don't need |
1050 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1051 | * |
1052 | * Add inode to orphan list in case we crash before | |
1053 | * truncate finishes | |
ae4d5372 | 1054 | */ |
ffacfa7a | 1055 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1056 | ext4_orphan_add(handle, inode); |
1057 | ||
1058 | ext4_journal_stop(handle); | |
1059 | if (pos + len > inode->i_size) { | |
b9a4207d | 1060 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1061 | /* |
ffacfa7a | 1062 | * If truncate failed early the inode might |
1938a150 AK |
1063 | * still be on the orphan list; we need to |
1064 | * make sure the inode is removed from the | |
1065 | * orphan list in that case. | |
1066 | */ | |
1067 | if (inode->i_nlink) | |
1068 | ext4_orphan_del(NULL, inode); | |
1069 | } | |
bfc1af65 | 1070 | |
47564bfb TT |
1071 | if (ret == -ENOSPC && |
1072 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1073 | goto retry_journal; | |
1074 | page_cache_release(page); | |
1075 | return ret; | |
1076 | } | |
1077 | *pagep = page; | |
ac27a0ec DK |
1078 | return ret; |
1079 | } | |
1080 | ||
bfc1af65 NP |
1081 | /* For write_end() in data=journal mode */ |
1082 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1083 | { |
13fca323 | 1084 | int ret; |
ac27a0ec DK |
1085 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1086 | return 0; | |
1087 | set_buffer_uptodate(bh); | |
13fca323 TT |
1088 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1089 | clear_buffer_meta(bh); | |
1090 | clear_buffer_prio(bh); | |
1091 | return ret; | |
ac27a0ec DK |
1092 | } |
1093 | ||
eed4333f ZL |
1094 | /* |
1095 | * We need to pick up the new inode size which generic_commit_write gave us | |
1096 | * `file' can be NULL - eg, when called from page_symlink(). | |
1097 | * | |
1098 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1099 | * buffers are managed internally. | |
1100 | */ | |
1101 | static int ext4_write_end(struct file *file, | |
1102 | struct address_space *mapping, | |
1103 | loff_t pos, unsigned len, unsigned copied, | |
1104 | struct page *page, void *fsdata) | |
f8514083 | 1105 | { |
f8514083 | 1106 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f ZL |
1107 | struct inode *inode = mapping->host; |
1108 | int ret = 0, ret2; | |
1109 | int i_size_changed = 0; | |
1110 | ||
1111 | trace_ext4_write_end(inode, pos, len, copied); | |
1112 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1113 | ret = ext4_jbd2_file_inode(handle, inode); | |
1114 | if (ret) { | |
1115 | unlock_page(page); | |
1116 | page_cache_release(page); | |
1117 | goto errout; | |
1118 | } | |
1119 | } | |
f8514083 | 1120 | |
f19d5870 TM |
1121 | if (ext4_has_inline_data(inode)) |
1122 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1123 | copied, page); | |
1124 | else | |
1125 | copied = block_write_end(file, mapping, pos, | |
1126 | len, copied, page, fsdata); | |
f8514083 AK |
1127 | |
1128 | /* | |
1129 | * No need to use i_size_read() here, the i_size | |
eed4333f | 1130 | * cannot change under us because we hole i_mutex. |
f8514083 AK |
1131 | * |
1132 | * But it's important to update i_size while still holding page lock: | |
1133 | * page writeout could otherwise come in and zero beyond i_size. | |
1134 | */ | |
1135 | if (pos + copied > inode->i_size) { | |
1136 | i_size_write(inode, pos + copied); | |
1137 | i_size_changed = 1; | |
1138 | } | |
1139 | ||
eed4333f | 1140 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
f8514083 AK |
1141 | /* We need to mark inode dirty even if |
1142 | * new_i_size is less that inode->i_size | |
eed4333f | 1143 | * but greater than i_disksize. (hint delalloc) |
f8514083 AK |
1144 | */ |
1145 | ext4_update_i_disksize(inode, (pos + copied)); | |
1146 | i_size_changed = 1; | |
1147 | } | |
1148 | unlock_page(page); | |
1149 | page_cache_release(page); | |
1150 | ||
1151 | /* | |
1152 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1153 | * makes the holding time of page lock longer. Second, it forces lock | |
1154 | * ordering of page lock and transaction start for journaling | |
1155 | * filesystems. | |
1156 | */ | |
1157 | if (i_size_changed) | |
1158 | ext4_mark_inode_dirty(handle, inode); | |
1159 | ||
74d553aa TT |
1160 | if (copied < 0) |
1161 | ret = copied; | |
ffacfa7a | 1162 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1163 | /* if we have allocated more blocks and copied |
1164 | * less. We will have blocks allocated outside | |
1165 | * inode->i_size. So truncate them | |
1166 | */ | |
1167 | ext4_orphan_add(handle, inode); | |
74d553aa | 1168 | errout: |
617ba13b | 1169 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1170 | if (!ret) |
1171 | ret = ret2; | |
bfc1af65 | 1172 | |
f8514083 | 1173 | if (pos + len > inode->i_size) { |
b9a4207d | 1174 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1175 | /* |
ffacfa7a | 1176 | * If truncate failed early the inode might still be |
f8514083 AK |
1177 | * on the orphan list; we need to make sure the inode |
1178 | * is removed from the orphan list in that case. | |
1179 | */ | |
1180 | if (inode->i_nlink) | |
1181 | ext4_orphan_del(NULL, inode); | |
1182 | } | |
1183 | ||
bfc1af65 | 1184 | return ret ? ret : copied; |
ac27a0ec DK |
1185 | } |
1186 | ||
bfc1af65 | 1187 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1188 | struct address_space *mapping, |
1189 | loff_t pos, unsigned len, unsigned copied, | |
1190 | struct page *page, void *fsdata) | |
ac27a0ec | 1191 | { |
617ba13b | 1192 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1193 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1194 | int ret = 0, ret2; |
1195 | int partial = 0; | |
bfc1af65 | 1196 | unsigned from, to; |
cf17fea6 | 1197 | loff_t new_i_size; |
ac27a0ec | 1198 | |
9bffad1e | 1199 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1200 | from = pos & (PAGE_CACHE_SIZE - 1); |
1201 | to = from + len; | |
1202 | ||
441c8508 CW |
1203 | BUG_ON(!ext4_handle_valid(handle)); |
1204 | ||
3fdcfb66 TM |
1205 | if (ext4_has_inline_data(inode)) |
1206 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1207 | copied, page); | |
1208 | else { | |
1209 | if (copied < len) { | |
1210 | if (!PageUptodate(page)) | |
1211 | copied = 0; | |
1212 | page_zero_new_buffers(page, from+copied, to); | |
1213 | } | |
ac27a0ec | 1214 | |
3fdcfb66 TM |
1215 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1216 | to, &partial, write_end_fn); | |
1217 | if (!partial) | |
1218 | SetPageUptodate(page); | |
1219 | } | |
cf17fea6 AK |
1220 | new_i_size = pos + copied; |
1221 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1222 | i_size_write(inode, pos+copied); |
19f5fb7a | 1223 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1224 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1225 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1226 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1227 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1228 | if (!ret) |
1229 | ret = ret2; | |
1230 | } | |
bfc1af65 | 1231 | |
cf108bca | 1232 | unlock_page(page); |
f8514083 | 1233 | page_cache_release(page); |
ffacfa7a | 1234 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1235 | /* if we have allocated more blocks and copied |
1236 | * less. We will have blocks allocated outside | |
1237 | * inode->i_size. So truncate them | |
1238 | */ | |
1239 | ext4_orphan_add(handle, inode); | |
1240 | ||
617ba13b | 1241 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1242 | if (!ret) |
1243 | ret = ret2; | |
f8514083 | 1244 | if (pos + len > inode->i_size) { |
b9a4207d | 1245 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1246 | /* |
ffacfa7a | 1247 | * If truncate failed early the inode might still be |
f8514083 AK |
1248 | * on the orphan list; we need to make sure the inode |
1249 | * is removed from the orphan list in that case. | |
1250 | */ | |
1251 | if (inode->i_nlink) | |
1252 | ext4_orphan_del(NULL, inode); | |
1253 | } | |
bfc1af65 NP |
1254 | |
1255 | return ret ? ret : copied; | |
ac27a0ec | 1256 | } |
d2a17637 | 1257 | |
386ad67c LC |
1258 | /* |
1259 | * Reserve a metadata for a single block located at lblock | |
1260 | */ | |
1261 | static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) | |
1262 | { | |
1263 | int retries = 0; | |
1264 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1265 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1266 | unsigned int md_needed; | |
1267 | ext4_lblk_t save_last_lblock; | |
1268 | int save_len; | |
1269 | ||
1270 | /* | |
1271 | * recalculate the amount of metadata blocks to reserve | |
1272 | * in order to allocate nrblocks | |
1273 | * worse case is one extent per block | |
1274 | */ | |
1275 | repeat: | |
1276 | spin_lock(&ei->i_block_reservation_lock); | |
1277 | /* | |
1278 | * ext4_calc_metadata_amount() has side effects, which we have | |
1279 | * to be prepared undo if we fail to claim space. | |
1280 | */ | |
1281 | save_len = ei->i_da_metadata_calc_len; | |
1282 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
1283 | md_needed = EXT4_NUM_B2C(sbi, | |
1284 | ext4_calc_metadata_amount(inode, lblock)); | |
1285 | trace_ext4_da_reserve_space(inode, md_needed); | |
1286 | ||
1287 | /* | |
1288 | * We do still charge estimated metadata to the sb though; | |
1289 | * we cannot afford to run out of free blocks. | |
1290 | */ | |
1291 | if (ext4_claim_free_clusters(sbi, md_needed, 0)) { | |
1292 | ei->i_da_metadata_calc_len = save_len; | |
1293 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1294 | spin_unlock(&ei->i_block_reservation_lock); | |
1295 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
1296 | cond_resched(); | |
1297 | goto repeat; | |
1298 | } | |
1299 | return -ENOSPC; | |
1300 | } | |
1301 | ei->i_reserved_meta_blocks += md_needed; | |
1302 | spin_unlock(&ei->i_block_reservation_lock); | |
1303 | ||
1304 | return 0; /* success */ | |
1305 | } | |
1306 | ||
9d0be502 | 1307 | /* |
7b415bf6 | 1308 | * Reserve a single cluster located at lblock |
9d0be502 | 1309 | */ |
01f49d0b | 1310 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1311 | { |
030ba6bc | 1312 | int retries = 0; |
60e58e0f | 1313 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1314 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1315 | unsigned int md_needed; |
5dd4056d | 1316 | int ret; |
03179fe9 TT |
1317 | ext4_lblk_t save_last_lblock; |
1318 | int save_len; | |
1319 | ||
1320 | /* | |
1321 | * We will charge metadata quota at writeout time; this saves | |
1322 | * us from metadata over-estimation, though we may go over by | |
1323 | * a small amount in the end. Here we just reserve for data. | |
1324 | */ | |
1325 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1326 | if (ret) | |
1327 | return ret; | |
d2a17637 MC |
1328 | |
1329 | /* | |
1330 | * recalculate the amount of metadata blocks to reserve | |
1331 | * in order to allocate nrblocks | |
1332 | * worse case is one extent per block | |
1333 | */ | |
030ba6bc | 1334 | repeat: |
0637c6f4 | 1335 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1336 | /* |
1337 | * ext4_calc_metadata_amount() has side effects, which we have | |
1338 | * to be prepared undo if we fail to claim space. | |
1339 | */ | |
1340 | save_len = ei->i_da_metadata_calc_len; | |
1341 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
7b415bf6 AK |
1342 | md_needed = EXT4_NUM_B2C(sbi, |
1343 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1344 | trace_ext4_da_reserve_space(inode, md_needed); |
d2a17637 | 1345 | |
72b8ab9d ES |
1346 | /* |
1347 | * We do still charge estimated metadata to the sb though; | |
1348 | * we cannot afford to run out of free blocks. | |
1349 | */ | |
e7d5f315 | 1350 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
03179fe9 TT |
1351 | ei->i_da_metadata_calc_len = save_len; |
1352 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1353 | spin_unlock(&ei->i_block_reservation_lock); | |
030ba6bc | 1354 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
bb8b20ed | 1355 | cond_resched(); |
030ba6bc AK |
1356 | goto repeat; |
1357 | } | |
03179fe9 | 1358 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1359 | return -ENOSPC; |
1360 | } | |
9d0be502 | 1361 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1362 | ei->i_reserved_meta_blocks += md_needed; |
1363 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1364 | |
d2a17637 MC |
1365 | return 0; /* success */ |
1366 | } | |
1367 | ||
12219aea | 1368 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1369 | { |
1370 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1371 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1372 | |
cd213226 MC |
1373 | if (!to_free) |
1374 | return; /* Nothing to release, exit */ | |
1375 | ||
d2a17637 | 1376 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1377 | |
5a58ec87 | 1378 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1379 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1380 | /* |
0637c6f4 TT |
1381 | * if there aren't enough reserved blocks, then the |
1382 | * counter is messed up somewhere. Since this | |
1383 | * function is called from invalidate page, it's | |
1384 | * harmless to return without any action. | |
cd213226 | 1385 | */ |
8de5c325 | 1386 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1387 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1388 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1389 | ei->i_reserved_data_blocks); |
1390 | WARN_ON(1); | |
1391 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1392 | } |
0637c6f4 | 1393 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1394 | |
0637c6f4 TT |
1395 | if (ei->i_reserved_data_blocks == 0) { |
1396 | /* | |
1397 | * We can release all of the reserved metadata blocks | |
1398 | * only when we have written all of the delayed | |
1399 | * allocation blocks. | |
7b415bf6 AK |
1400 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1401 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1402 | */ |
57042651 | 1403 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1404 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1405 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1406 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1407 | } |
d2a17637 | 1408 | |
72b8ab9d | 1409 | /* update fs dirty data blocks counter */ |
57042651 | 1410 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1411 | |
d2a17637 | 1412 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1413 | |
7b415bf6 | 1414 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1415 | } |
1416 | ||
1417 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1418 | unsigned long offset) |
d2a17637 MC |
1419 | { |
1420 | int to_release = 0; | |
1421 | struct buffer_head *head, *bh; | |
1422 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1423 | struct inode *inode = page->mapping->host; |
1424 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1425 | int num_clusters; | |
51865fda | 1426 | ext4_fsblk_t lblk; |
d2a17637 MC |
1427 | |
1428 | head = page_buffers(page); | |
1429 | bh = head; | |
1430 | do { | |
1431 | unsigned int next_off = curr_off + bh->b_size; | |
1432 | ||
1433 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1434 | to_release++; | |
1435 | clear_buffer_delay(bh); | |
1436 | } | |
1437 | curr_off = next_off; | |
1438 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1439 | |
51865fda ZL |
1440 | if (to_release) { |
1441 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1442 | ext4_es_remove_extent(inode, lblk, to_release); | |
1443 | } | |
1444 | ||
7b415bf6 AK |
1445 | /* If we have released all the blocks belonging to a cluster, then we |
1446 | * need to release the reserved space for that cluster. */ | |
1447 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1448 | while (num_clusters > 0) { | |
7b415bf6 AK |
1449 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1450 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1451 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1452 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1453 | ext4_da_release_space(inode, 1); |
1454 | ||
1455 | num_clusters--; | |
1456 | } | |
d2a17637 | 1457 | } |
ac27a0ec | 1458 | |
64769240 AT |
1459 | /* |
1460 | * Delayed allocation stuff | |
1461 | */ | |
1462 | ||
64769240 AT |
1463 | /* |
1464 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1465 | * them with writepage() call back |
64769240 AT |
1466 | * |
1467 | * @mpd->inode: inode | |
1468 | * @mpd->first_page: first page of the extent | |
1469 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1470 | * |
1471 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1472 | * to be allocated. this may be wrong if allocation failed. | |
1473 | * | |
1474 | * As pages are already locked by write_cache_pages(), we can't use it | |
1475 | */ | |
1de3e3df TT |
1476 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1477 | struct ext4_map_blocks *map) | |
64769240 | 1478 | { |
791b7f08 AK |
1479 | struct pagevec pvec; |
1480 | unsigned long index, end; | |
1481 | int ret = 0, err, nr_pages, i; | |
1482 | struct inode *inode = mpd->inode; | |
1483 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1484 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1485 | unsigned int len, block_start; |
1486 | struct buffer_head *bh, *page_bufs = NULL; | |
1de3e3df | 1487 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1488 | struct ext4_io_submit io_submit; |
64769240 AT |
1489 | |
1490 | BUG_ON(mpd->next_page <= mpd->first_page); | |
a549984b | 1491 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1492 | /* |
1493 | * We need to start from the first_page to the next_page - 1 | |
1494 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1495 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1496 | * at the currently mapped buffer_heads. |
1497 | */ | |
64769240 AT |
1498 | index = mpd->first_page; |
1499 | end = mpd->next_page - 1; | |
1500 | ||
791b7f08 | 1501 | pagevec_init(&pvec, 0); |
64769240 | 1502 | while (index <= end) { |
791b7f08 | 1503 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1504 | if (nr_pages == 0) |
1505 | break; | |
1506 | for (i = 0; i < nr_pages; i++) { | |
f8bec370 | 1507 | int skip_page = 0; |
64769240 AT |
1508 | struct page *page = pvec.pages[i]; |
1509 | ||
791b7f08 AK |
1510 | index = page->index; |
1511 | if (index > end) | |
1512 | break; | |
cb20d518 TT |
1513 | |
1514 | if (index == size >> PAGE_CACHE_SHIFT) | |
1515 | len = size & ~PAGE_CACHE_MASK; | |
1516 | else | |
1517 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1518 | if (map) { |
1519 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1520 | inode->i_blkbits); | |
1521 | pblock = map->m_pblk + (cur_logical - | |
1522 | map->m_lblk); | |
1523 | } | |
791b7f08 AK |
1524 | index++; |
1525 | ||
1526 | BUG_ON(!PageLocked(page)); | |
1527 | BUG_ON(PageWriteback(page)); | |
1528 | ||
3ecdb3a1 TT |
1529 | bh = page_bufs = page_buffers(page); |
1530 | block_start = 0; | |
64769240 | 1531 | do { |
1de3e3df TT |
1532 | if (map && (cur_logical >= map->m_lblk) && |
1533 | (cur_logical <= (map->m_lblk + | |
1534 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1535 | if (buffer_delay(bh)) { |
1536 | clear_buffer_delay(bh); | |
1537 | bh->b_blocknr = pblock; | |
29fa89d0 | 1538 | } |
1de3e3df TT |
1539 | if (buffer_unwritten(bh) || |
1540 | buffer_mapped(bh)) | |
1541 | BUG_ON(bh->b_blocknr != pblock); | |
1542 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1543 | set_buffer_uninit(bh); | |
1544 | clear_buffer_unwritten(bh); | |
1545 | } | |
29fa89d0 | 1546 | |
13a79a47 YY |
1547 | /* |
1548 | * skip page if block allocation undone and | |
1549 | * block is dirty | |
1550 | */ | |
1551 | if (ext4_bh_delay_or_unwritten(NULL, bh)) | |
97498956 | 1552 | skip_page = 1; |
3ecdb3a1 TT |
1553 | bh = bh->b_this_page; |
1554 | block_start += bh->b_size; | |
64769240 AT |
1555 | cur_logical++; |
1556 | pblock++; | |
1de3e3df TT |
1557 | } while (bh != page_bufs); |
1558 | ||
f8bec370 JK |
1559 | if (skip_page) { |
1560 | unlock_page(page); | |
1561 | continue; | |
1562 | } | |
cb20d518 | 1563 | |
97498956 | 1564 | clear_page_dirty_for_io(page); |
fe089c77 JK |
1565 | err = ext4_bio_write_page(&io_submit, page, len, |
1566 | mpd->wbc); | |
cb20d518 | 1567 | if (!err) |
a1d6cc56 | 1568 | mpd->pages_written++; |
64769240 AT |
1569 | /* |
1570 | * In error case, we have to continue because | |
1571 | * remaining pages are still locked | |
64769240 AT |
1572 | */ |
1573 | if (ret == 0) | |
1574 | ret = err; | |
64769240 AT |
1575 | } |
1576 | pagevec_release(&pvec); | |
1577 | } | |
bd2d0210 | 1578 | ext4_io_submit(&io_submit); |
64769240 | 1579 | return ret; |
64769240 AT |
1580 | } |
1581 | ||
c7f5938a | 1582 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1583 | { |
1584 | int nr_pages, i; | |
1585 | pgoff_t index, end; | |
1586 | struct pagevec pvec; | |
1587 | struct inode *inode = mpd->inode; | |
1588 | struct address_space *mapping = inode->i_mapping; | |
51865fda | 1589 | ext4_lblk_t start, last; |
c4a0c46e | 1590 | |
c7f5938a CW |
1591 | index = mpd->first_page; |
1592 | end = mpd->next_page - 1; | |
51865fda ZL |
1593 | |
1594 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1595 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1596 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1597 | ||
66bea92c | 1598 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1599 | while (index <= end) { |
1600 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1601 | if (nr_pages == 0) | |
1602 | break; | |
1603 | for (i = 0; i < nr_pages; i++) { | |
1604 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1605 | if (page->index > end) |
c4a0c46e | 1606 | break; |
c4a0c46e AK |
1607 | BUG_ON(!PageLocked(page)); |
1608 | BUG_ON(PageWriteback(page)); | |
1609 | block_invalidatepage(page, 0); | |
1610 | ClearPageUptodate(page); | |
1611 | unlock_page(page); | |
1612 | } | |
9b1d0998 JK |
1613 | index = pvec.pages[nr_pages - 1]->index + 1; |
1614 | pagevec_release(&pvec); | |
c4a0c46e AK |
1615 | } |
1616 | return; | |
1617 | } | |
1618 | ||
df22291f AK |
1619 | static void ext4_print_free_blocks(struct inode *inode) |
1620 | { | |
1621 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1622 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1623 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1624 | |
1625 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1626 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1627 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1628 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1629 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1630 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1631 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1632 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1633 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1634 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1635 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1636 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1637 | ei->i_reserved_data_blocks); |
92b97816 | 1638 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", |
f78ee70d LC |
1639 | ei->i_reserved_meta_blocks); |
1640 | ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u", | |
1641 | ei->i_allocated_meta_blocks); | |
df22291f AK |
1642 | return; |
1643 | } | |
1644 | ||
64769240 | 1645 | /* |
5a87b7a5 TT |
1646 | * mpage_da_map_and_submit - go through given space, map them |
1647 | * if necessary, and then submit them for I/O | |
64769240 | 1648 | * |
8dc207c0 | 1649 | * @mpd - bh describing space |
64769240 AT |
1650 | * |
1651 | * The function skips space we know is already mapped to disk blocks. | |
1652 | * | |
64769240 | 1653 | */ |
5a87b7a5 | 1654 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1655 | { |
2ac3b6e0 | 1656 | int err, blks, get_blocks_flags; |
1de3e3df | 1657 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1658 | sector_t next = mpd->b_blocknr; |
1659 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1660 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1661 | handle_t *handle = NULL; | |
64769240 AT |
1662 | |
1663 | /* | |
5a87b7a5 TT |
1664 | * If the blocks are mapped already, or we couldn't accumulate |
1665 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1666 | */ |
5a87b7a5 TT |
1667 | if ((mpd->b_size == 0) || |
1668 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1669 | !(mpd->b_state & (1 << BH_Delay)) && | |
1670 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1671 | goto submit_io; | |
2fa3cdfb TT |
1672 | |
1673 | handle = ext4_journal_current_handle(); | |
1674 | BUG_ON(!handle); | |
1675 | ||
79ffab34 | 1676 | /* |
79e83036 | 1677 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1678 | * blocks, or to convert an uninitialized extent to be |
1679 | * initialized (in the case where we have written into | |
1680 | * one or more preallocated blocks). | |
1681 | * | |
1682 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1683 | * indicate that we are on the delayed allocation path. This | |
1684 | * affects functions in many different parts of the allocation | |
1685 | * call path. This flag exists primarily because we don't | |
79e83036 | 1686 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1687 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1688 | * inode's allocation semaphore is taken. |
1689 | * | |
1690 | * If the blocks in questions were delalloc blocks, set | |
1691 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1692 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1693 | */ |
2ed88685 TT |
1694 | map.m_lblk = next; |
1695 | map.m_len = max_blocks; | |
27dd4385 LC |
1696 | /* |
1697 | * We're in delalloc path and it is possible that we're going to | |
1698 | * need more metadata blocks than previously reserved. However | |
1699 | * we must not fail because we're in writeback and there is | |
1700 | * nothing we can do about it so it might result in data loss. | |
1701 | * So use reserved blocks to allocate metadata if possible. | |
1702 | */ | |
1703 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
1704 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
744692dc JZ |
1705 | if (ext4_should_dioread_nolock(mpd->inode)) |
1706 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1707 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1708 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1709 | ||
27dd4385 | 1710 | |
2ed88685 | 1711 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1712 | if (blks < 0) { |
e3570639 ES |
1713 | struct super_block *sb = mpd->inode->i_sb; |
1714 | ||
2fa3cdfb | 1715 | err = blks; |
ed5bde0b | 1716 | /* |
5a87b7a5 | 1717 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1718 | * appears to be free blocks we will just let |
1719 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1720 | */ |
1721 | if (err == -EAGAIN) | |
5a87b7a5 | 1722 | goto submit_io; |
df22291f | 1723 | |
5dee5437 | 1724 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1725 | mpd->retval = err; |
5a87b7a5 | 1726 | goto submit_io; |
df22291f AK |
1727 | } |
1728 | ||
c4a0c46e | 1729 | /* |
ed5bde0b TT |
1730 | * get block failure will cause us to loop in |
1731 | * writepages, because a_ops->writepage won't be able | |
1732 | * to make progress. The page will be redirtied by | |
1733 | * writepage and writepages will again try to write | |
1734 | * the same. | |
c4a0c46e | 1735 | */ |
e3570639 ES |
1736 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1737 | ext4_msg(sb, KERN_CRIT, | |
1738 | "delayed block allocation failed for inode %lu " | |
1739 | "at logical offset %llu with max blocks %zd " | |
1740 | "with error %d", mpd->inode->i_ino, | |
1741 | (unsigned long long) next, | |
1742 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1743 | ext4_msg(sb, KERN_CRIT, | |
01a523eb | 1744 | "This should not happen!! Data will be lost"); |
e3570639 ES |
1745 | if (err == -ENOSPC) |
1746 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1747 | } |
2fa3cdfb | 1748 | /* invalidate all the pages */ |
c7f5938a | 1749 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1750 | |
1751 | /* Mark this page range as having been completed */ | |
1752 | mpd->io_done = 1; | |
5a87b7a5 | 1753 | return; |
c4a0c46e | 1754 | } |
2fa3cdfb TT |
1755 | BUG_ON(blks == 0); |
1756 | ||
1de3e3df | 1757 | mapp = ↦ |
2ed88685 TT |
1758 | if (map.m_flags & EXT4_MAP_NEW) { |
1759 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1760 | int i; | |
64769240 | 1761 | |
2ed88685 TT |
1762 | for (i = 0; i < map.m_len; i++) |
1763 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2fa3cdfb TT |
1764 | } |
1765 | ||
1766 | /* | |
03f5d8bc | 1767 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1768 | */ |
1769 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1770 | if (disksize > i_size_read(mpd->inode)) | |
1771 | disksize = i_size_read(mpd->inode); | |
1772 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1773 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1774 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1775 | if (err) | |
1776 | ext4_error(mpd->inode->i_sb, | |
1777 | "Failed to mark inode %lu dirty", | |
1778 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1779 | } |
1780 | ||
5a87b7a5 | 1781 | submit_io: |
1de3e3df | 1782 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1783 | mpd->io_done = 1; |
64769240 AT |
1784 | } |
1785 | ||
bf068ee2 AK |
1786 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1787 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1788 | |
1789 | /* | |
1790 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1791 | * | |
1792 | * @mpd->lbh - extent of blocks | |
1793 | * @logical - logical number of the block in the file | |
b6a8e62f | 1794 | * @b_state - b_state of the buffer head added |
64769240 AT |
1795 | * |
1796 | * the function is used to collect contig. blocks in same state | |
1797 | */ | |
b6a8e62f | 1798 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical, |
8dc207c0 | 1799 | unsigned long b_state) |
64769240 | 1800 | { |
64769240 | 1801 | sector_t next; |
b6a8e62f JK |
1802 | int blkbits = mpd->inode->i_blkbits; |
1803 | int nrblocks = mpd->b_size >> blkbits; | |
64769240 | 1804 | |
c445e3e0 ES |
1805 | /* |
1806 | * XXX Don't go larger than mballoc is willing to allocate | |
1807 | * This is a stopgap solution. We eventually need to fold | |
1808 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1809 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 | 1810 | */ |
b6a8e62f | 1811 | if (nrblocks >= (8*1024*1024 >> blkbits)) |
c445e3e0 ES |
1812 | goto flush_it; |
1813 | ||
b6a8e62f JK |
1814 | /* check if the reserved journal credits might overflow */ |
1815 | if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) { | |
525f4ed8 MC |
1816 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1817 | /* | |
1818 | * With non-extent format we are limited by the journal | |
1819 | * credit available. Total credit needed to insert | |
1820 | * nrblocks contiguous blocks is dependent on the | |
1821 | * nrblocks. So limit nrblocks. | |
1822 | */ | |
1823 | goto flush_it; | |
525f4ed8 MC |
1824 | } |
1825 | } | |
64769240 AT |
1826 | /* |
1827 | * First block in the extent | |
1828 | */ | |
8dc207c0 TT |
1829 | if (mpd->b_size == 0) { |
1830 | mpd->b_blocknr = logical; | |
b6a8e62f | 1831 | mpd->b_size = 1 << blkbits; |
8dc207c0 | 1832 | mpd->b_state = b_state & BH_FLAGS; |
64769240 AT |
1833 | return; |
1834 | } | |
1835 | ||
8dc207c0 | 1836 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1837 | /* |
1838 | * Can we merge the block to our big extent? | |
1839 | */ | |
8dc207c0 | 1840 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
b6a8e62f | 1841 | mpd->b_size += 1 << blkbits; |
64769240 AT |
1842 | return; |
1843 | } | |
1844 | ||
525f4ed8 | 1845 | flush_it: |
64769240 AT |
1846 | /* |
1847 | * We couldn't merge the block to our extent, so we | |
1848 | * need to flush current extent and start new one | |
1849 | */ | |
5a87b7a5 | 1850 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1851 | return; |
64769240 AT |
1852 | } |
1853 | ||
c364b22c | 1854 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1855 | { |
c364b22c | 1856 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1857 | } |
1858 | ||
5356f261 AK |
1859 | /* |
1860 | * This function is grabs code from the very beginning of | |
1861 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1862 | * time. This function looks up the requested blocks and sets the | |
1863 | * buffer delay bit under the protection of i_data_sem. | |
1864 | */ | |
1865 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1866 | struct ext4_map_blocks *map, | |
1867 | struct buffer_head *bh) | |
1868 | { | |
d100eef2 | 1869 | struct extent_status es; |
5356f261 AK |
1870 | int retval; |
1871 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1872 | #ifdef ES_AGGRESSIVE_TEST |
1873 | struct ext4_map_blocks orig_map; | |
1874 | ||
1875 | memcpy(&orig_map, map, sizeof(*map)); | |
1876 | #endif | |
5356f261 AK |
1877 | |
1878 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1879 | invalid_block = ~0; | |
1880 | ||
1881 | map->m_flags = 0; | |
1882 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1883 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1884 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1885 | |
1886 | /* Lookup extent status tree firstly */ | |
1887 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
1888 | ||
1889 | if (ext4_es_is_hole(&es)) { | |
1890 | retval = 0; | |
1891 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1892 | goto add_delayed; | |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * Delayed extent could be allocated by fallocate. | |
1897 | * So we need to check it. | |
1898 | */ | |
1899 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1900 | map_bh(bh, inode->i_sb, invalid_block); | |
1901 | set_buffer_new(bh); | |
1902 | set_buffer_delay(bh); | |
1903 | return 0; | |
1904 | } | |
1905 | ||
1906 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1907 | retval = es.es_len - (iblock - es.es_lblk); | |
1908 | if (retval > map->m_len) | |
1909 | retval = map->m_len; | |
1910 | map->m_len = retval; | |
1911 | if (ext4_es_is_written(&es)) | |
1912 | map->m_flags |= EXT4_MAP_MAPPED; | |
1913 | else if (ext4_es_is_unwritten(&es)) | |
1914 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1915 | else | |
1916 | BUG_ON(1); | |
1917 | ||
921f266b DM |
1918 | #ifdef ES_AGGRESSIVE_TEST |
1919 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1920 | #endif | |
d100eef2 ZL |
1921 | return retval; |
1922 | } | |
1923 | ||
5356f261 AK |
1924 | /* |
1925 | * Try to see if we can get the block without requesting a new | |
1926 | * file system block. | |
1927 | */ | |
1928 | down_read((&EXT4_I(inode)->i_data_sem)); | |
9c3569b5 TM |
1929 | if (ext4_has_inline_data(inode)) { |
1930 | /* | |
1931 | * We will soon create blocks for this page, and let | |
1932 | * us pretend as if the blocks aren't allocated yet. | |
1933 | * In case of clusters, we have to handle the work | |
1934 | * of mapping from cluster so that the reserved space | |
1935 | * is calculated properly. | |
1936 | */ | |
1937 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && | |
1938 | ext4_find_delalloc_cluster(inode, map->m_lblk)) | |
1939 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
1940 | retval = 0; | |
1941 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
d100eef2 ZL |
1942 | retval = ext4_ext_map_blocks(NULL, inode, map, |
1943 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1944 | else |
d100eef2 ZL |
1945 | retval = ext4_ind_map_blocks(NULL, inode, map, |
1946 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1947 | |
d100eef2 | 1948 | add_delayed: |
5356f261 | 1949 | if (retval == 0) { |
f7fec032 | 1950 | int ret; |
5356f261 AK |
1951 | /* |
1952 | * XXX: __block_prepare_write() unmaps passed block, | |
1953 | * is it OK? | |
1954 | */ | |
386ad67c LC |
1955 | /* |
1956 | * If the block was allocated from previously allocated cluster, | |
1957 | * then we don't need to reserve it again. However we still need | |
1958 | * to reserve metadata for every block we're going to write. | |
1959 | */ | |
5356f261 | 1960 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
f7fec032 ZL |
1961 | ret = ext4_da_reserve_space(inode, iblock); |
1962 | if (ret) { | |
5356f261 | 1963 | /* not enough space to reserve */ |
f7fec032 | 1964 | retval = ret; |
5356f261 | 1965 | goto out_unlock; |
f7fec032 | 1966 | } |
386ad67c LC |
1967 | } else { |
1968 | ret = ext4_da_reserve_metadata(inode, iblock); | |
1969 | if (ret) { | |
1970 | /* not enough space to reserve */ | |
1971 | retval = ret; | |
1972 | goto out_unlock; | |
1973 | } | |
5356f261 AK |
1974 | } |
1975 | ||
f7fec032 ZL |
1976 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1977 | ~0, EXTENT_STATUS_DELAYED); | |
1978 | if (ret) { | |
1979 | retval = ret; | |
51865fda | 1980 | goto out_unlock; |
f7fec032 | 1981 | } |
51865fda | 1982 | |
5356f261 AK |
1983 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
1984 | * and it should not appear on the bh->b_state. | |
1985 | */ | |
1986 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1987 | ||
1988 | map_bh(bh, inode->i_sb, invalid_block); | |
1989 | set_buffer_new(bh); | |
1990 | set_buffer_delay(bh); | |
f7fec032 ZL |
1991 | } else if (retval > 0) { |
1992 | int ret; | |
1993 | unsigned long long status; | |
1994 | ||
921f266b DM |
1995 | #ifdef ES_AGGRESSIVE_TEST |
1996 | if (retval != map->m_len) { | |
1997 | printk("ES len assertation failed for inode: %lu " | |
1998 | "retval %d != map->m_len %d " | |
1999 | "in %s (lookup)\n", inode->i_ino, retval, | |
2000 | map->m_len, __func__); | |
2001 | } | |
2002 | #endif | |
2003 | ||
f7fec032 ZL |
2004 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
2005 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
2006 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
2007 | map->m_pblk, status); | |
2008 | if (ret != 0) | |
2009 | retval = ret; | |
5356f261 AK |
2010 | } |
2011 | ||
2012 | out_unlock: | |
2013 | up_read((&EXT4_I(inode)->i_data_sem)); | |
2014 | ||
2015 | return retval; | |
2016 | } | |
2017 | ||
64769240 | 2018 | /* |
b920c755 TT |
2019 | * This is a special get_blocks_t callback which is used by |
2020 | * ext4_da_write_begin(). It will either return mapped block or | |
2021 | * reserve space for a single block. | |
29fa89d0 AK |
2022 | * |
2023 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2024 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2025 | * | |
2026 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2027 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2028 | * initialized properly. | |
64769240 | 2029 | */ |
9c3569b5 TM |
2030 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
2031 | struct buffer_head *bh, int create) | |
64769240 | 2032 | { |
2ed88685 | 2033 | struct ext4_map_blocks map; |
64769240 AT |
2034 | int ret = 0; |
2035 | ||
2036 | BUG_ON(create == 0); | |
2ed88685 TT |
2037 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2038 | ||
2039 | map.m_lblk = iblock; | |
2040 | map.m_len = 1; | |
64769240 AT |
2041 | |
2042 | /* | |
2043 | * first, we need to know whether the block is allocated already | |
2044 | * preallocated blocks are unmapped but should treated | |
2045 | * the same as allocated blocks. | |
2046 | */ | |
5356f261 AK |
2047 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
2048 | if (ret <= 0) | |
2ed88685 | 2049 | return ret; |
64769240 | 2050 | |
2ed88685 TT |
2051 | map_bh(bh, inode->i_sb, map.m_pblk); |
2052 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2053 | ||
2054 | if (buffer_unwritten(bh)) { | |
2055 | /* A delayed write to unwritten bh should be marked | |
2056 | * new and mapped. Mapped ensures that we don't do | |
2057 | * get_block multiple times when we write to the same | |
2058 | * offset and new ensures that we do proper zero out | |
2059 | * for partial write. | |
2060 | */ | |
2061 | set_buffer_new(bh); | |
c8205636 | 2062 | set_buffer_mapped(bh); |
2ed88685 TT |
2063 | } |
2064 | return 0; | |
64769240 | 2065 | } |
61628a3f | 2066 | |
62e086be AK |
2067 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2068 | { | |
2069 | get_bh(bh); | |
2070 | return 0; | |
2071 | } | |
2072 | ||
2073 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2074 | { | |
2075 | put_bh(bh); | |
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2080 | unsigned int len) |
2081 | { | |
2082 | struct address_space *mapping = page->mapping; | |
2083 | struct inode *inode = mapping->host; | |
3fdcfb66 | 2084 | struct buffer_head *page_bufs = NULL; |
62e086be | 2085 | handle_t *handle = NULL; |
3fdcfb66 TM |
2086 | int ret = 0, err = 0; |
2087 | int inline_data = ext4_has_inline_data(inode); | |
2088 | struct buffer_head *inode_bh = NULL; | |
62e086be | 2089 | |
cb20d518 | 2090 | ClearPageChecked(page); |
3fdcfb66 TM |
2091 | |
2092 | if (inline_data) { | |
2093 | BUG_ON(page->index != 0); | |
2094 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
2095 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
2096 | if (inode_bh == NULL) | |
2097 | goto out; | |
2098 | } else { | |
2099 | page_bufs = page_buffers(page); | |
2100 | if (!page_bufs) { | |
2101 | BUG(); | |
2102 | goto out; | |
2103 | } | |
2104 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2105 | NULL, bget_one); | |
2106 | } | |
62e086be AK |
2107 | /* As soon as we unlock the page, it can go away, but we have |
2108 | * references to buffers so we are safe */ | |
2109 | unlock_page(page); | |
2110 | ||
9924a92a TT |
2111 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2112 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2113 | if (IS_ERR(handle)) { |
2114 | ret = PTR_ERR(handle); | |
2115 | goto out; | |
2116 | } | |
2117 | ||
441c8508 CW |
2118 | BUG_ON(!ext4_handle_valid(handle)); |
2119 | ||
3fdcfb66 TM |
2120 | if (inline_data) { |
2121 | ret = ext4_journal_get_write_access(handle, inode_bh); | |
62e086be | 2122 | |
3fdcfb66 TM |
2123 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
2124 | ||
2125 | } else { | |
2126 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2127 | do_journal_get_write_access); | |
2128 | ||
2129 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2130 | write_end_fn); | |
2131 | } | |
62e086be AK |
2132 | if (ret == 0) |
2133 | ret = err; | |
2d859db3 | 2134 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2135 | err = ext4_journal_stop(handle); |
2136 | if (!ret) | |
2137 | ret = err; | |
2138 | ||
3fdcfb66 TM |
2139 | if (!ext4_has_inline_data(inode)) |
2140 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2141 | NULL, bput_one); | |
19f5fb7a | 2142 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2143 | out: |
3fdcfb66 | 2144 | brelse(inode_bh); |
62e086be AK |
2145 | return ret; |
2146 | } | |
2147 | ||
61628a3f | 2148 | /* |
43ce1d23 AK |
2149 | * Note that we don't need to start a transaction unless we're journaling data |
2150 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2151 | * need to file the inode to the transaction's list in ordered mode because if | |
2152 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2153 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2154 | * transaction the data will hit the disk. In case we are journaling data, we |
2155 | * cannot start transaction directly because transaction start ranks above page | |
2156 | * lock so we have to do some magic. | |
2157 | * | |
b920c755 TT |
2158 | * This function can get called via... |
2159 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2160 | * - journal_submit_inode_data_buffers (no journal handle) | |
f6463b0d | 2161 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2162 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2163 | * |
2164 | * We don't do any block allocation in this function. If we have page with | |
2165 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2166 | * is important for mmaped based write. So if we do with blocksize 1K | |
2167 | * truncate(f, 1024); | |
2168 | * a = mmap(f, 0, 4096); | |
2169 | * a[0] = 'a'; | |
2170 | * truncate(f, 4096); | |
2171 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2172 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2173 | * do_wp_page). So writepage should write the first block. If we modify |
2174 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2175 | * page_mkwrite callback will do the block allocation and mark the | |
2176 | * buffer_heads mapped. | |
2177 | * | |
2178 | * We redirty the page if we have any buffer_heads that is either delay or | |
2179 | * unwritten in the page. | |
2180 | * | |
2181 | * We can get recursively called as show below. | |
2182 | * | |
2183 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2184 | * ext4_writepage() | |
2185 | * | |
2186 | * But since we don't do any block allocation we should not deadlock. | |
2187 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2188 | */ |
43ce1d23 | 2189 | static int ext4_writepage(struct page *page, |
62e086be | 2190 | struct writeback_control *wbc) |
64769240 | 2191 | { |
f8bec370 | 2192 | int ret = 0; |
61628a3f | 2193 | loff_t size; |
498e5f24 | 2194 | unsigned int len; |
744692dc | 2195 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2196 | struct inode *inode = page->mapping->host; |
36ade451 | 2197 | struct ext4_io_submit io_submit; |
61628a3f | 2198 | |
a9c667f8 | 2199 | trace_ext4_writepage(page); |
f0e6c985 AK |
2200 | size = i_size_read(inode); |
2201 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2202 | len = size & ~PAGE_CACHE_MASK; | |
2203 | else | |
2204 | len = PAGE_CACHE_SIZE; | |
64769240 | 2205 | |
a42afc5f | 2206 | page_bufs = page_buffers(page); |
a42afc5f | 2207 | /* |
fe386132 JK |
2208 | * We cannot do block allocation or other extent handling in this |
2209 | * function. If there are buffers needing that, we have to redirty | |
2210 | * the page. But we may reach here when we do a journal commit via | |
2211 | * journal_submit_inode_data_buffers() and in that case we must write | |
2212 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 2213 | */ |
f19d5870 TM |
2214 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2215 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2216 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
2217 | if (current->flags & PF_MEMALLOC) { |
2218 | /* | |
2219 | * For memory cleaning there's no point in writing only | |
2220 | * some buffers. So just bail out. Warn if we came here | |
2221 | * from direct reclaim. | |
2222 | */ | |
2223 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2224 | == PF_MEMALLOC); | |
f0e6c985 AK |
2225 | unlock_page(page); |
2226 | return 0; | |
2227 | } | |
a42afc5f | 2228 | } |
64769240 | 2229 | |
cb20d518 | 2230 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2231 | /* |
2232 | * It's mmapped pagecache. Add buffers and journal it. There | |
2233 | * doesn't seem much point in redirtying the page here. | |
2234 | */ | |
3f0ca309 | 2235 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2236 | |
a549984b | 2237 | memset(&io_submit, 0, sizeof(io_submit)); |
36ade451 JK |
2238 | ret = ext4_bio_write_page(&io_submit, page, len, wbc); |
2239 | ext4_io_submit(&io_submit); | |
64769240 AT |
2240 | return ret; |
2241 | } | |
2242 | ||
61628a3f | 2243 | /* |
525f4ed8 | 2244 | * This is called via ext4_da_writepages() to |
25985edc | 2245 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2246 | * a single extent allocation into a single transaction, |
2247 | * ext4_da_writpeages() will loop calling this before | |
2248 | * the block allocation. | |
61628a3f | 2249 | */ |
525f4ed8 MC |
2250 | |
2251 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2252 | { | |
2253 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2254 | ||
2255 | /* | |
2256 | * With non-extent format the journal credit needed to | |
2257 | * insert nrblocks contiguous block is dependent on | |
2258 | * number of contiguous block. So we will limit | |
2259 | * number of contiguous block to a sane value | |
2260 | */ | |
12e9b892 | 2261 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2262 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2263 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2264 | ||
2265 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2266 | } | |
61628a3f | 2267 | |
8e48dcfb TT |
2268 | /* |
2269 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2270 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2271 | * mpage_da_map_and_submit to map a single contiguous memory region |
2272 | * and then write them. | |
8e48dcfb | 2273 | */ |
9c3569b5 TM |
2274 | static int write_cache_pages_da(handle_t *handle, |
2275 | struct address_space *mapping, | |
8e48dcfb | 2276 | struct writeback_control *wbc, |
72f84e65 ES |
2277 | struct mpage_da_data *mpd, |
2278 | pgoff_t *done_index) | |
8e48dcfb | 2279 | { |
4f01b02c | 2280 | struct buffer_head *bh, *head; |
168fc022 | 2281 | struct inode *inode = mapping->host; |
4f01b02c TT |
2282 | struct pagevec pvec; |
2283 | unsigned int nr_pages; | |
2284 | sector_t logical; | |
2285 | pgoff_t index, end; | |
2286 | long nr_to_write = wbc->nr_to_write; | |
2287 | int i, tag, ret = 0; | |
8e48dcfb | 2288 | |
168fc022 TT |
2289 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2290 | mpd->wbc = wbc; | |
2291 | mpd->inode = inode; | |
8e48dcfb TT |
2292 | pagevec_init(&pvec, 0); |
2293 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2294 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2295 | ||
6e6938b6 | 2296 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2297 | tag = PAGECACHE_TAG_TOWRITE; |
2298 | else | |
2299 | tag = PAGECACHE_TAG_DIRTY; | |
2300 | ||
72f84e65 | 2301 | *done_index = index; |
4f01b02c | 2302 | while (index <= end) { |
5b41d924 | 2303 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2304 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2305 | if (nr_pages == 0) | |
4f01b02c | 2306 | return 0; |
8e48dcfb TT |
2307 | |
2308 | for (i = 0; i < nr_pages; i++) { | |
2309 | struct page *page = pvec.pages[i]; | |
2310 | ||
2311 | /* | |
2312 | * At this point, the page may be truncated or | |
2313 | * invalidated (changing page->mapping to NULL), or | |
2314 | * even swizzled back from swapper_space to tmpfs file | |
2315 | * mapping. However, page->index will not change | |
2316 | * because we have a reference on the page. | |
2317 | */ | |
4f01b02c TT |
2318 | if (page->index > end) |
2319 | goto out; | |
8e48dcfb | 2320 | |
72f84e65 ES |
2321 | *done_index = page->index + 1; |
2322 | ||
78aaced3 TT |
2323 | /* |
2324 | * If we can't merge this page, and we have | |
2325 | * accumulated an contiguous region, write it | |
2326 | */ | |
2327 | if ((mpd->next_page != page->index) && | |
2328 | (mpd->next_page != mpd->first_page)) { | |
2329 | mpage_da_map_and_submit(mpd); | |
2330 | goto ret_extent_tail; | |
2331 | } | |
2332 | ||
8e48dcfb TT |
2333 | lock_page(page); |
2334 | ||
2335 | /* | |
4f01b02c TT |
2336 | * If the page is no longer dirty, or its |
2337 | * mapping no longer corresponds to inode we | |
2338 | * are writing (which means it has been | |
2339 | * truncated or invalidated), or the page is | |
2340 | * already under writeback and we are not | |
2341 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2342 | */ |
4f01b02c TT |
2343 | if (!PageDirty(page) || |
2344 | (PageWriteback(page) && | |
2345 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2346 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2347 | unlock_page(page); |
2348 | continue; | |
2349 | } | |
2350 | ||
7cb1a535 | 2351 | wait_on_page_writeback(page); |
8e48dcfb | 2352 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2353 | |
9c3569b5 TM |
2354 | /* |
2355 | * If we have inline data and arrive here, it means that | |
2356 | * we will soon create the block for the 1st page, so | |
2357 | * we'd better clear the inline data here. | |
2358 | */ | |
2359 | if (ext4_has_inline_data(inode)) { | |
2360 | BUG_ON(ext4_test_inode_state(inode, | |
2361 | EXT4_STATE_MAY_INLINE_DATA)); | |
2362 | ext4_destroy_inline_data(handle, inode); | |
2363 | } | |
2364 | ||
168fc022 | 2365 | if (mpd->next_page != page->index) |
8eb9e5ce | 2366 | mpd->first_page = page->index; |
8eb9e5ce TT |
2367 | mpd->next_page = page->index + 1; |
2368 | logical = (sector_t) page->index << | |
2369 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2370 | ||
f8bec370 JK |
2371 | /* Add all dirty buffers to mpd */ |
2372 | head = page_buffers(page); | |
2373 | bh = head; | |
2374 | do { | |
2375 | BUG_ON(buffer_locked(bh)); | |
8eb9e5ce | 2376 | /* |
f8bec370 JK |
2377 | * We need to try to allocate unmapped blocks |
2378 | * in the same page. Otherwise we won't make | |
2379 | * progress with the page in ext4_writepage | |
8eb9e5ce | 2380 | */ |
f8bec370 JK |
2381 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
2382 | mpage_add_bh_to_extent(mpd, logical, | |
f8bec370 JK |
2383 | bh->b_state); |
2384 | if (mpd->io_done) | |
2385 | goto ret_extent_tail; | |
2386 | } else if (buffer_dirty(bh) && | |
2387 | buffer_mapped(bh)) { | |
8eb9e5ce | 2388 | /* |
f8bec370 JK |
2389 | * mapped dirty buffer. We need to |
2390 | * update the b_state because we look | |
2391 | * at b_state in mpage_da_map_blocks. | |
2392 | * We don't update b_size because if we | |
2393 | * find an unmapped buffer_head later | |
2394 | * we need to use the b_state flag of | |
2395 | * that buffer_head. | |
8eb9e5ce | 2396 | */ |
f8bec370 JK |
2397 | if (mpd->b_size == 0) |
2398 | mpd->b_state = | |
2399 | bh->b_state & BH_FLAGS; | |
2400 | } | |
2401 | logical++; | |
2402 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2403 | |
2404 | if (nr_to_write > 0) { | |
2405 | nr_to_write--; | |
2406 | if (nr_to_write == 0 && | |
4f01b02c | 2407 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2408 | /* |
2409 | * We stop writing back only if we are | |
2410 | * not doing integrity sync. In case of | |
2411 | * integrity sync we have to keep going | |
2412 | * because someone may be concurrently | |
2413 | * dirtying pages, and we might have | |
2414 | * synced a lot of newly appeared dirty | |
2415 | * pages, but have not synced all of the | |
2416 | * old dirty pages. | |
2417 | */ | |
4f01b02c | 2418 | goto out; |
8e48dcfb TT |
2419 | } |
2420 | } | |
2421 | pagevec_release(&pvec); | |
2422 | cond_resched(); | |
2423 | } | |
4f01b02c TT |
2424 | return 0; |
2425 | ret_extent_tail: | |
2426 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2427 | out: |
2428 | pagevec_release(&pvec); | |
2429 | cond_resched(); | |
8e48dcfb TT |
2430 | return ret; |
2431 | } | |
2432 | ||
2433 | ||
64769240 | 2434 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2435 | struct writeback_control *wbc) |
64769240 | 2436 | { |
22208ded AK |
2437 | pgoff_t index; |
2438 | int range_whole = 0; | |
61628a3f | 2439 | handle_t *handle = NULL; |
df22291f | 2440 | struct mpage_da_data mpd; |
5e745b04 | 2441 | struct inode *inode = mapping->host; |
498e5f24 | 2442 | int pages_written = 0; |
55138e0b | 2443 | unsigned int max_pages; |
2acf2c26 | 2444 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2445 | int needed_blocks, ret = 0; |
2446 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2447 | loff_t range_start = wbc->range_start; |
5e745b04 | 2448 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2449 | pgoff_t done_index = 0; |
5b41d924 | 2450 | pgoff_t end; |
1bce63d1 | 2451 | struct blk_plug plug; |
61628a3f | 2452 | |
9bffad1e | 2453 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2454 | |
61628a3f MC |
2455 | /* |
2456 | * No pages to write? This is mainly a kludge to avoid starting | |
2457 | * a transaction for special inodes like journal inode on last iput() | |
2458 | * because that could violate lock ordering on umount | |
2459 | */ | |
a1d6cc56 | 2460 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2461 | return 0; |
2a21e37e TT |
2462 | |
2463 | /* | |
2464 | * If the filesystem has aborted, it is read-only, so return | |
2465 | * right away instead of dumping stack traces later on that | |
2466 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2467 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2468 | * the latter could be true if the filesystem is mounted |
2469 | * read-only, and in that case, ext4_da_writepages should | |
2470 | * *never* be called, so if that ever happens, we would want | |
2471 | * the stack trace. | |
2472 | */ | |
4ab2f15b | 2473 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2474 | return -EROFS; |
2475 | ||
22208ded AK |
2476 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2477 | range_whole = 1; | |
61628a3f | 2478 | |
2acf2c26 AK |
2479 | range_cyclic = wbc->range_cyclic; |
2480 | if (wbc->range_cyclic) { | |
22208ded | 2481 | index = mapping->writeback_index; |
2acf2c26 AK |
2482 | if (index) |
2483 | cycled = 0; | |
2484 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2485 | wbc->range_end = LLONG_MAX; | |
2486 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2487 | end = -1; |
2488 | } else { | |
22208ded | 2489 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2490 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2491 | } | |
a1d6cc56 | 2492 | |
55138e0b TT |
2493 | /* |
2494 | * This works around two forms of stupidity. The first is in | |
2495 | * the writeback code, which caps the maximum number of pages | |
2496 | * written to be 1024 pages. This is wrong on multiple | |
2497 | * levels; different architectues have a different page size, | |
2498 | * which changes the maximum amount of data which gets | |
2499 | * written. Secondly, 4 megabytes is way too small. XFS | |
2500 | * forces this value to be 16 megabytes by multiplying | |
2501 | * nr_to_write parameter by four, and then relies on its | |
2502 | * allocator to allocate larger extents to make them | |
2503 | * contiguous. Unfortunately this brings us to the second | |
2504 | * stupidity, which is that ext4's mballoc code only allocates | |
2505 | * at most 2048 blocks. So we force contiguous writes up to | |
2506 | * the number of dirty blocks in the inode, or | |
2507 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2508 | */ | |
2509 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2510 | if (!range_cyclic && range_whole) { |
2511 | if (wbc->nr_to_write == LONG_MAX) | |
2512 | desired_nr_to_write = wbc->nr_to_write; | |
2513 | else | |
2514 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2515 | } else | |
55138e0b TT |
2516 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2517 | max_pages); | |
2518 | if (desired_nr_to_write > max_pages) | |
2519 | desired_nr_to_write = max_pages; | |
2520 | ||
2521 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2522 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2523 | wbc->nr_to_write = desired_nr_to_write; | |
2524 | } | |
2525 | ||
2acf2c26 | 2526 | retry: |
6e6938b6 | 2527 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2528 | tag_pages_for_writeback(mapping, index, end); |
2529 | ||
1bce63d1 | 2530 | blk_start_plug(&plug); |
22208ded | 2531 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2532 | |
2533 | /* | |
2534 | * we insert one extent at a time. So we need | |
2535 | * credit needed for single extent allocation. | |
2536 | * journalled mode is currently not supported | |
2537 | * by delalloc | |
2538 | */ | |
2539 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2540 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2541 | |
61628a3f | 2542 | /* start a new transaction*/ |
9924a92a TT |
2543 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2544 | needed_blocks); | |
61628a3f MC |
2545 | if (IS_ERR(handle)) { |
2546 | ret = PTR_ERR(handle); | |
1693918e | 2547 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2548 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2549 | wbc->nr_to_write, inode->i_ino, ret); |
3c1fcb2c | 2550 | blk_finish_plug(&plug); |
61628a3f MC |
2551 | goto out_writepages; |
2552 | } | |
f63e6005 TT |
2553 | |
2554 | /* | |
8eb9e5ce | 2555 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2556 | * contiguous region of logical blocks that need |
8eb9e5ce | 2557 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2558 | */ |
9c3569b5 TM |
2559 | ret = write_cache_pages_da(handle, mapping, |
2560 | wbc, &mpd, &done_index); | |
f63e6005 | 2561 | /* |
af901ca1 | 2562 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2563 | * haven't done the I/O yet, map the blocks and submit |
2564 | * them for I/O. | |
2565 | */ | |
2566 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2567 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2568 | ret = MPAGE_DA_EXTENT_TAIL; |
2569 | } | |
b3a3ca8c | 2570 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2571 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2572 | |
61628a3f | 2573 | ext4_journal_stop(handle); |
df22291f | 2574 | |
8f64b32e | 2575 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2576 | /* commit the transaction which would |
2577 | * free blocks released in the transaction | |
2578 | * and try again | |
2579 | */ | |
df22291f | 2580 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2581 | ret = 0; |
2582 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 | 2583 | /* |
8de49e67 KM |
2584 | * Got one extent now try with rest of the pages. |
2585 | * If mpd.retval is set -EIO, journal is aborted. | |
2586 | * So we don't need to write any more. | |
a1d6cc56 | 2587 | */ |
22208ded | 2588 | pages_written += mpd.pages_written; |
8de49e67 | 2589 | ret = mpd.retval; |
2acf2c26 | 2590 | io_done = 1; |
22208ded | 2591 | } else if (wbc->nr_to_write) |
61628a3f MC |
2592 | /* |
2593 | * There is no more writeout needed | |
2594 | * or we requested for a noblocking writeout | |
2595 | * and we found the device congested | |
2596 | */ | |
61628a3f | 2597 | break; |
a1d6cc56 | 2598 | } |
1bce63d1 | 2599 | blk_finish_plug(&plug); |
2acf2c26 AK |
2600 | if (!io_done && !cycled) { |
2601 | cycled = 1; | |
2602 | index = 0; | |
2603 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2604 | wbc->range_end = mapping->writeback_index - 1; | |
2605 | goto retry; | |
2606 | } | |
22208ded AK |
2607 | |
2608 | /* Update index */ | |
2acf2c26 | 2609 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2610 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2611 | /* | |
2612 | * set the writeback_index so that range_cyclic | |
2613 | * mode will write it back later | |
2614 | */ | |
72f84e65 | 2615 | mapping->writeback_index = done_index; |
a1d6cc56 | 2616 | |
61628a3f | 2617 | out_writepages: |
2faf2e19 | 2618 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2619 | wbc->range_start = range_start; |
9bffad1e | 2620 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2621 | return ret; |
64769240 AT |
2622 | } |
2623 | ||
79f0be8d AK |
2624 | static int ext4_nonda_switch(struct super_block *sb) |
2625 | { | |
5c1ff336 | 2626 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2627 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2628 | ||
2629 | /* | |
2630 | * switch to non delalloc mode if we are running low | |
2631 | * on free block. The free block accounting via percpu | |
179f7ebf | 2632 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2633 | * accumulated on each CPU without updating global counters |
2634 | * Delalloc need an accurate free block accounting. So switch | |
2635 | * to non delalloc when we are near to error range. | |
2636 | */ | |
5c1ff336 EW |
2637 | free_clusters = |
2638 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2639 | dirty_clusters = | |
2640 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2641 | /* |
2642 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2643 | */ | |
5c1ff336 | 2644 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2645 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2646 | |
5c1ff336 EW |
2647 | if (2 * free_clusters < 3 * dirty_clusters || |
2648 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2649 | /* |
c8afb446 ES |
2650 | * free block count is less than 150% of dirty blocks |
2651 | * or free blocks is less than watermark | |
79f0be8d AK |
2652 | */ |
2653 | return 1; | |
2654 | } | |
2655 | return 0; | |
2656 | } | |
2657 | ||
64769240 | 2658 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2659 | loff_t pos, unsigned len, unsigned flags, |
2660 | struct page **pagep, void **fsdata) | |
64769240 | 2661 | { |
72b8ab9d | 2662 | int ret, retries = 0; |
64769240 AT |
2663 | struct page *page; |
2664 | pgoff_t index; | |
64769240 AT |
2665 | struct inode *inode = mapping->host; |
2666 | handle_t *handle; | |
2667 | ||
2668 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2669 | |
2670 | if (ext4_nonda_switch(inode->i_sb)) { | |
2671 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2672 | return ext4_write_begin(file, mapping, pos, | |
2673 | len, flags, pagep, fsdata); | |
2674 | } | |
2675 | *fsdata = (void *)0; | |
9bffad1e | 2676 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2677 | |
2678 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2679 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2680 | pos, len, flags, | |
2681 | pagep, fsdata); | |
2682 | if (ret < 0) | |
47564bfb TT |
2683 | return ret; |
2684 | if (ret == 1) | |
2685 | return 0; | |
9c3569b5 TM |
2686 | } |
2687 | ||
47564bfb TT |
2688 | /* |
2689 | * grab_cache_page_write_begin() can take a long time if the | |
2690 | * system is thrashing due to memory pressure, or if the page | |
2691 | * is being written back. So grab it first before we start | |
2692 | * the transaction handle. This also allows us to allocate | |
2693 | * the page (if needed) without using GFP_NOFS. | |
2694 | */ | |
2695 | retry_grab: | |
2696 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2697 | if (!page) | |
2698 | return -ENOMEM; | |
2699 | unlock_page(page); | |
2700 | ||
64769240 AT |
2701 | /* |
2702 | * With delayed allocation, we don't log the i_disksize update | |
2703 | * if there is delayed block allocation. But we still need | |
2704 | * to journalling the i_disksize update if writes to the end | |
2705 | * of file which has an already mapped buffer. | |
2706 | */ | |
47564bfb | 2707 | retry_journal: |
9924a92a | 2708 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
64769240 | 2709 | if (IS_ERR(handle)) { |
47564bfb TT |
2710 | page_cache_release(page); |
2711 | return PTR_ERR(handle); | |
64769240 AT |
2712 | } |
2713 | ||
47564bfb TT |
2714 | lock_page(page); |
2715 | if (page->mapping != mapping) { | |
2716 | /* The page got truncated from under us */ | |
2717 | unlock_page(page); | |
2718 | page_cache_release(page); | |
d5a0d4f7 | 2719 | ext4_journal_stop(handle); |
47564bfb | 2720 | goto retry_grab; |
d5a0d4f7 | 2721 | } |
47564bfb TT |
2722 | /* In case writeback began while the page was unlocked */ |
2723 | wait_on_page_writeback(page); | |
64769240 | 2724 | |
6e1db88d | 2725 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2726 | if (ret < 0) { |
2727 | unlock_page(page); | |
2728 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2729 | /* |
2730 | * block_write_begin may have instantiated a few blocks | |
2731 | * outside i_size. Trim these off again. Don't need | |
2732 | * i_size_read because we hold i_mutex. | |
2733 | */ | |
2734 | if (pos + len > inode->i_size) | |
b9a4207d | 2735 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2736 | |
2737 | if (ret == -ENOSPC && | |
2738 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2739 | goto retry_journal; | |
2740 | ||
2741 | page_cache_release(page); | |
2742 | return ret; | |
64769240 AT |
2743 | } |
2744 | ||
47564bfb | 2745 | *pagep = page; |
64769240 AT |
2746 | return ret; |
2747 | } | |
2748 | ||
632eaeab MC |
2749 | /* |
2750 | * Check if we should update i_disksize | |
2751 | * when write to the end of file but not require block allocation | |
2752 | */ | |
2753 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2754 | unsigned long offset) |
632eaeab MC |
2755 | { |
2756 | struct buffer_head *bh; | |
2757 | struct inode *inode = page->mapping->host; | |
2758 | unsigned int idx; | |
2759 | int i; | |
2760 | ||
2761 | bh = page_buffers(page); | |
2762 | idx = offset >> inode->i_blkbits; | |
2763 | ||
af5bc92d | 2764 | for (i = 0; i < idx; i++) |
632eaeab MC |
2765 | bh = bh->b_this_page; |
2766 | ||
29fa89d0 | 2767 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2768 | return 0; |
2769 | return 1; | |
2770 | } | |
2771 | ||
64769240 | 2772 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2773 | struct address_space *mapping, |
2774 | loff_t pos, unsigned len, unsigned copied, | |
2775 | struct page *page, void *fsdata) | |
64769240 AT |
2776 | { |
2777 | struct inode *inode = mapping->host; | |
2778 | int ret = 0, ret2; | |
2779 | handle_t *handle = ext4_journal_current_handle(); | |
2780 | loff_t new_i_size; | |
632eaeab | 2781 | unsigned long start, end; |
79f0be8d AK |
2782 | int write_mode = (int)(unsigned long)fsdata; |
2783 | ||
74d553aa TT |
2784 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2785 | return ext4_write_end(file, mapping, pos, | |
2786 | len, copied, page, fsdata); | |
632eaeab | 2787 | |
9bffad1e | 2788 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2789 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2790 | end = start + copied - 1; |
64769240 AT |
2791 | |
2792 | /* | |
2793 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2794 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2795 | * into that. | |
2796 | */ | |
64769240 | 2797 | new_i_size = pos + copied; |
ea51d132 | 2798 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2799 | if (ext4_has_inline_data(inode) || |
2800 | ext4_da_should_update_i_disksize(page, end)) { | |
632eaeab | 2801 | down_write(&EXT4_I(inode)->i_data_sem); |
f3b59291 | 2802 | if (new_i_size > EXT4_I(inode)->i_disksize) |
632eaeab | 2803 | EXT4_I(inode)->i_disksize = new_i_size; |
632eaeab | 2804 | up_write(&EXT4_I(inode)->i_data_sem); |
cf17fea6 AK |
2805 | /* We need to mark inode dirty even if |
2806 | * new_i_size is less that inode->i_size | |
2807 | * bu greater than i_disksize.(hint delalloc) | |
2808 | */ | |
2809 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2810 | } |
632eaeab | 2811 | } |
9c3569b5 TM |
2812 | |
2813 | if (write_mode != CONVERT_INLINE_DATA && | |
2814 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2815 | ext4_has_inline_data(inode)) | |
2816 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2817 | page); | |
2818 | else | |
2819 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2820 | page, fsdata); |
9c3569b5 | 2821 | |
64769240 AT |
2822 | copied = ret2; |
2823 | if (ret2 < 0) | |
2824 | ret = ret2; | |
2825 | ret2 = ext4_journal_stop(handle); | |
2826 | if (!ret) | |
2827 | ret = ret2; | |
2828 | ||
2829 | return ret ? ret : copied; | |
2830 | } | |
2831 | ||
2832 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2833 | { | |
64769240 AT |
2834 | /* |
2835 | * Drop reserved blocks | |
2836 | */ | |
2837 | BUG_ON(!PageLocked(page)); | |
2838 | if (!page_has_buffers(page)) | |
2839 | goto out; | |
2840 | ||
d2a17637 | 2841 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2842 | |
2843 | out: | |
2844 | ext4_invalidatepage(page, offset); | |
2845 | ||
2846 | return; | |
2847 | } | |
2848 | ||
ccd2506b TT |
2849 | /* |
2850 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2851 | */ | |
2852 | int ext4_alloc_da_blocks(struct inode *inode) | |
2853 | { | |
fb40ba0d TT |
2854 | trace_ext4_alloc_da_blocks(inode); |
2855 | ||
ccd2506b TT |
2856 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2857 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2858 | return 0; | |
2859 | ||
2860 | /* | |
2861 | * We do something simple for now. The filemap_flush() will | |
2862 | * also start triggering a write of the data blocks, which is | |
2863 | * not strictly speaking necessary (and for users of | |
2864 | * laptop_mode, not even desirable). However, to do otherwise | |
2865 | * would require replicating code paths in: | |
de9a55b8 | 2866 | * |
ccd2506b TT |
2867 | * ext4_da_writepages() -> |
2868 | * write_cache_pages() ---> (via passed in callback function) | |
2869 | * __mpage_da_writepage() --> | |
2870 | * mpage_add_bh_to_extent() | |
2871 | * mpage_da_map_blocks() | |
2872 | * | |
2873 | * The problem is that write_cache_pages(), located in | |
2874 | * mm/page-writeback.c, marks pages clean in preparation for | |
2875 | * doing I/O, which is not desirable if we're not planning on | |
2876 | * doing I/O at all. | |
2877 | * | |
2878 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2879 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2880 | * would be ugly in the extreme. So instead we would need to |
2881 | * replicate parts of the code in the above functions, | |
25985edc | 2882 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2883 | * write out the pages, but rather only collect contiguous |
2884 | * logical block extents, call the multi-block allocator, and | |
2885 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2886 | * |
ccd2506b TT |
2887 | * For now, though, we'll cheat by calling filemap_flush(), |
2888 | * which will map the blocks, and start the I/O, but not | |
2889 | * actually wait for the I/O to complete. | |
2890 | */ | |
2891 | return filemap_flush(inode->i_mapping); | |
2892 | } | |
64769240 | 2893 | |
ac27a0ec DK |
2894 | /* |
2895 | * bmap() is special. It gets used by applications such as lilo and by | |
2896 | * the swapper to find the on-disk block of a specific piece of data. | |
2897 | * | |
2898 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2899 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2900 | * filesystem and enables swap, then they may get a nasty shock when the |
2901 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2902 | * the original zero's written out previously to the journal and | |
2903 | * awaiting writeback in the kernel's buffer cache. | |
2904 | * | |
2905 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2906 | * take extra steps to flush any blocks which might be in the cache. | |
2907 | */ | |
617ba13b | 2908 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2909 | { |
2910 | struct inode *inode = mapping->host; | |
2911 | journal_t *journal; | |
2912 | int err; | |
2913 | ||
46c7f254 TM |
2914 | /* |
2915 | * We can get here for an inline file via the FIBMAP ioctl | |
2916 | */ | |
2917 | if (ext4_has_inline_data(inode)) | |
2918 | return 0; | |
2919 | ||
64769240 AT |
2920 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2921 | test_opt(inode->i_sb, DELALLOC)) { | |
2922 | /* | |
2923 | * With delalloc we want to sync the file | |
2924 | * so that we can make sure we allocate | |
2925 | * blocks for file | |
2926 | */ | |
2927 | filemap_write_and_wait(mapping); | |
2928 | } | |
2929 | ||
19f5fb7a TT |
2930 | if (EXT4_JOURNAL(inode) && |
2931 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2932 | /* |
2933 | * This is a REALLY heavyweight approach, but the use of | |
2934 | * bmap on dirty files is expected to be extremely rare: | |
2935 | * only if we run lilo or swapon on a freshly made file | |
2936 | * do we expect this to happen. | |
2937 | * | |
2938 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2939 | * represent an unprivileged user DOS attack --- we'd be | |
2940 | * in trouble if mortal users could trigger this path at | |
2941 | * will.) | |
2942 | * | |
617ba13b | 2943 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2944 | * regular files. If somebody wants to bmap a directory |
2945 | * or symlink and gets confused because the buffer | |
2946 | * hasn't yet been flushed to disk, they deserve | |
2947 | * everything they get. | |
2948 | */ | |
2949 | ||
19f5fb7a | 2950 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2951 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2952 | jbd2_journal_lock_updates(journal); |
2953 | err = jbd2_journal_flush(journal); | |
2954 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2955 | |
2956 | if (err) | |
2957 | return 0; | |
2958 | } | |
2959 | ||
af5bc92d | 2960 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2961 | } |
2962 | ||
617ba13b | 2963 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2964 | { |
46c7f254 TM |
2965 | int ret = -EAGAIN; |
2966 | struct inode *inode = page->mapping->host; | |
2967 | ||
0562e0ba | 2968 | trace_ext4_readpage(page); |
46c7f254 TM |
2969 | |
2970 | if (ext4_has_inline_data(inode)) | |
2971 | ret = ext4_readpage_inline(inode, page); | |
2972 | ||
2973 | if (ret == -EAGAIN) | |
2974 | return mpage_readpage(page, ext4_get_block); | |
2975 | ||
2976 | return ret; | |
ac27a0ec DK |
2977 | } |
2978 | ||
2979 | static int | |
617ba13b | 2980 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2981 | struct list_head *pages, unsigned nr_pages) |
2982 | { | |
46c7f254 TM |
2983 | struct inode *inode = mapping->host; |
2984 | ||
2985 | /* If the file has inline data, no need to do readpages. */ | |
2986 | if (ext4_has_inline_data(inode)) | |
2987 | return 0; | |
2988 | ||
617ba13b | 2989 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2990 | } |
2991 | ||
617ba13b | 2992 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2993 | { |
0562e0ba JZ |
2994 | trace_ext4_invalidatepage(page, offset); |
2995 | ||
4520fb3c JK |
2996 | /* No journalling happens on data buffers when this function is used */ |
2997 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
2998 | ||
2999 | block_invalidatepage(page, offset); | |
3000 | } | |
3001 | ||
53e87268 JK |
3002 | static int __ext4_journalled_invalidatepage(struct page *page, |
3003 | unsigned long offset) | |
4520fb3c JK |
3004 | { |
3005 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3006 | ||
3007 | trace_ext4_journalled_invalidatepage(page, offset); | |
3008 | ||
ac27a0ec DK |
3009 | /* |
3010 | * If it's a full truncate we just forget about the pending dirtying | |
3011 | */ | |
3012 | if (offset == 0) | |
3013 | ClearPageChecked(page); | |
3014 | ||
53e87268 JK |
3015 | return jbd2_journal_invalidatepage(journal, page, offset); |
3016 | } | |
3017 | ||
3018 | /* Wrapper for aops... */ | |
3019 | static void ext4_journalled_invalidatepage(struct page *page, | |
3020 | unsigned long offset) | |
3021 | { | |
3022 | WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0); | |
ac27a0ec DK |
3023 | } |
3024 | ||
617ba13b | 3025 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3026 | { |
617ba13b | 3027 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3028 | |
0562e0ba JZ |
3029 | trace_ext4_releasepage(page); |
3030 | ||
e1c36595 JK |
3031 | /* Page has dirty journalled data -> cannot release */ |
3032 | if (PageChecked(page)) | |
ac27a0ec | 3033 | return 0; |
0390131b FM |
3034 | if (journal) |
3035 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3036 | else | |
3037 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3038 | } |
3039 | ||
2ed88685 TT |
3040 | /* |
3041 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3042 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3043 | * The extent will be converted to initialized after the IO is complete. | |
3044 | */ | |
f19d5870 | 3045 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3046 | struct buffer_head *bh_result, int create) |
3047 | { | |
c7064ef1 | 3048 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3049 | inode->i_ino, create); |
2ed88685 TT |
3050 | return _ext4_get_block(inode, iblock, bh_result, |
3051 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3052 | } |
3053 | ||
729f52c6 | 3054 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 3055 | struct buffer_head *bh_result, int create) |
729f52c6 | 3056 | { |
8b0f165f AP |
3057 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
3058 | inode->i_ino, create); | |
3059 | return _ext4_get_block(inode, iblock, bh_result, | |
3060 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
3061 | } |
3062 | ||
4c0425ff | 3063 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
3064 | ssize_t size, void *private, int ret, |
3065 | bool is_async) | |
4c0425ff | 3066 | { |
496ad9aa | 3067 | struct inode *inode = file_inode(iocb->ki_filp); |
4c0425ff | 3068 | ext4_io_end_t *io_end = iocb->private; |
4c0425ff | 3069 | |
a549984b TT |
3070 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3071 | if (!io_end || !size) | |
3072 | goto out; | |
4b70df18 | 3073 | |
88635ca2 | 3074 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3075 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3076 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3077 | size); | |
8d5d02e6 | 3078 | |
b5a7e970 | 3079 | iocb->private = NULL; |
a549984b TT |
3080 | |
3081 | /* if not aio dio with unwritten extents, just free io and return */ | |
3082 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { | |
3083 | ext4_free_io_end(io_end); | |
3084 | out: | |
3085 | inode_dio_done(inode); | |
3086 | if (is_async) | |
3087 | aio_complete(iocb, ret, 0); | |
3088 | return; | |
3089 | } | |
3090 | ||
4c0425ff MC |
3091 | io_end->offset = offset; |
3092 | io_end->size = size; | |
5b3ff237 JZ |
3093 | if (is_async) { |
3094 | io_end->iocb = iocb; | |
3095 | io_end->result = ret; | |
3096 | } | |
a549984b TT |
3097 | |
3098 | ext4_add_complete_io(io_end); | |
4c0425ff | 3099 | } |
c7064ef1 | 3100 | |
4c0425ff MC |
3101 | /* |
3102 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3103 | * preallocated extents, and those write extend the file, no need to | |
3104 | * fall back to buffered IO. | |
3105 | * | |
b595076a | 3106 | * For holes, we fallocate those blocks, mark them as uninitialized |
69c499d1 | 3107 | * If those blocks were preallocated, we mark sure they are split, but |
b595076a | 3108 | * still keep the range to write as uninitialized. |
4c0425ff | 3109 | * |
69c499d1 | 3110 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3111 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3112 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3113 | * when async direct IO completed. |
4c0425ff MC |
3114 | * |
3115 | * If the O_DIRECT write will extend the file then add this inode to the | |
3116 | * orphan list. So recovery will truncate it back to the original size | |
3117 | * if the machine crashes during the write. | |
3118 | * | |
3119 | */ | |
3120 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3121 | const struct iovec *iov, loff_t offset, | |
3122 | unsigned long nr_segs) | |
3123 | { | |
3124 | struct file *file = iocb->ki_filp; | |
3125 | struct inode *inode = file->f_mapping->host; | |
3126 | ssize_t ret; | |
3127 | size_t count = iov_length(iov, nr_segs); | |
69c499d1 TT |
3128 | int overwrite = 0; |
3129 | get_block_t *get_block_func = NULL; | |
3130 | int dio_flags = 0; | |
4c0425ff | 3131 | loff_t final_size = offset + count; |
729f52c6 | 3132 | |
69c499d1 TT |
3133 | /* Use the old path for reads and writes beyond i_size. */ |
3134 | if (rw != WRITE || final_size > inode->i_size) | |
3135 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
4bd809db | 3136 | |
69c499d1 | 3137 | BUG_ON(iocb->private == NULL); |
4bd809db | 3138 | |
69c499d1 TT |
3139 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3140 | overwrite = *((int *)iocb->private); | |
4bd809db | 3141 | |
69c499d1 TT |
3142 | if (overwrite) { |
3143 | atomic_inc(&inode->i_dio_count); | |
3144 | down_read(&EXT4_I(inode)->i_data_sem); | |
3145 | mutex_unlock(&inode->i_mutex); | |
3146 | } | |
8d5d02e6 | 3147 | |
69c499d1 TT |
3148 | /* |
3149 | * We could direct write to holes and fallocate. | |
3150 | * | |
3151 | * Allocated blocks to fill the hole are marked as | |
3152 | * uninitialized to prevent parallel buffered read to expose | |
3153 | * the stale data before DIO complete the data IO. | |
3154 | * | |
3155 | * As to previously fallocated extents, ext4 get_block will | |
3156 | * just simply mark the buffer mapped but still keep the | |
3157 | * extents uninitialized. | |
3158 | * | |
3159 | * For non AIO case, we will convert those unwritten extents | |
3160 | * to written after return back from blockdev_direct_IO. | |
3161 | * | |
3162 | * For async DIO, the conversion needs to be deferred when the | |
3163 | * IO is completed. The ext4 end_io callback function will be | |
3164 | * called to take care of the conversion work. Here for async | |
3165 | * case, we allocate an io_end structure to hook to the iocb. | |
3166 | */ | |
3167 | iocb->private = NULL; | |
3168 | ext4_inode_aio_set(inode, NULL); | |
3169 | if (!is_sync_kiocb(iocb)) { | |
a549984b | 3170 | ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3171 | if (!io_end) { |
3172 | ret = -ENOMEM; | |
3173 | goto retake_lock; | |
8b0f165f | 3174 | } |
69c499d1 | 3175 | io_end->flag |= EXT4_IO_END_DIRECT; |
a549984b | 3176 | iocb->private = io_end; |
8d5d02e6 | 3177 | /* |
69c499d1 TT |
3178 | * we save the io structure for current async direct |
3179 | * IO, so that later ext4_map_blocks() could flag the | |
3180 | * io structure whether there is a unwritten extents | |
3181 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3182 | */ |
69c499d1 TT |
3183 | ext4_inode_aio_set(inode, io_end); |
3184 | } | |
4bd809db | 3185 | |
69c499d1 TT |
3186 | if (overwrite) { |
3187 | get_block_func = ext4_get_block_write_nolock; | |
3188 | } else { | |
3189 | get_block_func = ext4_get_block_write; | |
3190 | dio_flags = DIO_LOCKING; | |
3191 | } | |
3192 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
3193 | inode->i_sb->s_bdev, iov, | |
3194 | offset, nr_segs, | |
3195 | get_block_func, | |
3196 | ext4_end_io_dio, | |
3197 | NULL, | |
3198 | dio_flags); | |
3199 | ||
a549984b TT |
3200 | if (iocb->private) |
3201 | ext4_inode_aio_set(inode, NULL); | |
69c499d1 | 3202 | /* |
a549984b TT |
3203 | * The io_end structure takes a reference to the inode, that |
3204 | * structure needs to be destroyed and the reference to the | |
3205 | * inode need to be dropped, when IO is complete, even with 0 | |
3206 | * byte write, or failed. | |
3207 | * | |
3208 | * In the successful AIO DIO case, the io_end structure will | |
3209 | * be destroyed and the reference to the inode will be dropped | |
3210 | * after the end_io call back function is called. | |
3211 | * | |
3212 | * In the case there is 0 byte write, or error case, since VFS | |
3213 | * direct IO won't invoke the end_io call back function, we | |
3214 | * need to free the end_io structure here. | |
69c499d1 | 3215 | */ |
a549984b TT |
3216 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { |
3217 | ext4_free_io_end(iocb->private); | |
3218 | iocb->private = NULL; | |
3219 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3220 | EXT4_STATE_DIO_UNWRITTEN)) { |
3221 | int err; | |
3222 | /* | |
3223 | * for non AIO case, since the IO is already | |
3224 | * completed, we could do the conversion right here | |
3225 | */ | |
3226 | err = ext4_convert_unwritten_extents(inode, | |
3227 | offset, ret); | |
3228 | if (err < 0) | |
3229 | ret = err; | |
3230 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3231 | } | |
4bd809db | 3232 | |
69c499d1 TT |
3233 | retake_lock: |
3234 | /* take i_mutex locking again if we do a ovewrite dio */ | |
3235 | if (overwrite) { | |
3236 | inode_dio_done(inode); | |
3237 | up_read(&EXT4_I(inode)->i_data_sem); | |
3238 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3239 | } |
8d5d02e6 | 3240 | |
69c499d1 | 3241 | return ret; |
4c0425ff MC |
3242 | } |
3243 | ||
3244 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3245 | const struct iovec *iov, loff_t offset, | |
3246 | unsigned long nr_segs) | |
3247 | { | |
3248 | struct file *file = iocb->ki_filp; | |
3249 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3250 | ssize_t ret; |
4c0425ff | 3251 | |
84ebd795 TT |
3252 | /* |
3253 | * If we are doing data journalling we don't support O_DIRECT | |
3254 | */ | |
3255 | if (ext4_should_journal_data(inode)) | |
3256 | return 0; | |
3257 | ||
46c7f254 TM |
3258 | /* Let buffer I/O handle the inline data case. */ |
3259 | if (ext4_has_inline_data(inode)) | |
3260 | return 0; | |
3261 | ||
0562e0ba | 3262 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3263 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3264 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3265 | else | |
3266 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3267 | trace_ext4_direct_IO_exit(inode, offset, | |
3268 | iov_length(iov, nr_segs), rw, ret); | |
3269 | return ret; | |
4c0425ff MC |
3270 | } |
3271 | ||
ac27a0ec | 3272 | /* |
617ba13b | 3273 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3274 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3275 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3276 | * not necessarily locked. | |
3277 | * | |
3278 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3279 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3280 | * or jbddirty because all the journalling code will explode. | |
3281 | * | |
3282 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3283 | * is called, propagate that into the buffers appropriately. | |
3284 | */ | |
617ba13b | 3285 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3286 | { |
3287 | SetPageChecked(page); | |
3288 | return __set_page_dirty_nobuffers(page); | |
3289 | } | |
3290 | ||
74d553aa | 3291 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3292 | .readpage = ext4_readpage, |
3293 | .readpages = ext4_readpages, | |
43ce1d23 | 3294 | .writepage = ext4_writepage, |
8ab22b9a | 3295 | .write_begin = ext4_write_begin, |
74d553aa | 3296 | .write_end = ext4_write_end, |
8ab22b9a HH |
3297 | .bmap = ext4_bmap, |
3298 | .invalidatepage = ext4_invalidatepage, | |
3299 | .releasepage = ext4_releasepage, | |
3300 | .direct_IO = ext4_direct_IO, | |
3301 | .migratepage = buffer_migrate_page, | |
3302 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3303 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3304 | }; |
3305 | ||
617ba13b | 3306 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3307 | .readpage = ext4_readpage, |
3308 | .readpages = ext4_readpages, | |
43ce1d23 | 3309 | .writepage = ext4_writepage, |
8ab22b9a HH |
3310 | .write_begin = ext4_write_begin, |
3311 | .write_end = ext4_journalled_write_end, | |
3312 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3313 | .bmap = ext4_bmap, | |
4520fb3c | 3314 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3315 | .releasepage = ext4_releasepage, |
84ebd795 | 3316 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3317 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3318 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3319 | }; |
3320 | ||
64769240 | 3321 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3322 | .readpage = ext4_readpage, |
3323 | .readpages = ext4_readpages, | |
43ce1d23 | 3324 | .writepage = ext4_writepage, |
8ab22b9a | 3325 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3326 | .write_begin = ext4_da_write_begin, |
3327 | .write_end = ext4_da_write_end, | |
3328 | .bmap = ext4_bmap, | |
3329 | .invalidatepage = ext4_da_invalidatepage, | |
3330 | .releasepage = ext4_releasepage, | |
3331 | .direct_IO = ext4_direct_IO, | |
3332 | .migratepage = buffer_migrate_page, | |
3333 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3334 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3335 | }; |
3336 | ||
617ba13b | 3337 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3338 | { |
3d2b1582 LC |
3339 | switch (ext4_inode_journal_mode(inode)) { |
3340 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3341 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3342 | break; |
3343 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3344 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3345 | break; |
3346 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3347 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3348 | return; |
3d2b1582 LC |
3349 | default: |
3350 | BUG(); | |
3351 | } | |
74d553aa TT |
3352 | if (test_opt(inode->i_sb, DELALLOC)) |
3353 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3354 | else | |
3355 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3356 | } |
3357 | ||
4e96b2db AH |
3358 | |
3359 | /* | |
3360 | * ext4_discard_partial_page_buffers() | |
3361 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3362 | * This function finds and locks the page containing the offset | |
3363 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3364 | * Calling functions that already have the page locked should call | |
3365 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3366 | */ | |
3367 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3368 | struct address_space *mapping, loff_t from, | |
3369 | loff_t length, int flags) | |
3370 | { | |
3371 | struct inode *inode = mapping->host; | |
3372 | struct page *page; | |
3373 | int err = 0; | |
3374 | ||
3375 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3376 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3377 | if (!page) | |
5129d05f | 3378 | return -ENOMEM; |
4e96b2db AH |
3379 | |
3380 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3381 | from, length, flags); | |
3382 | ||
3383 | unlock_page(page); | |
3384 | page_cache_release(page); | |
3385 | return err; | |
3386 | } | |
3387 | ||
3388 | /* | |
3389 | * ext4_discard_partial_page_buffers_no_lock() | |
3390 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3391 | * Buffer heads that correspond to the block aligned regions of the | |
3392 | * zeroed range will be unmapped. Unblock aligned regions | |
3393 | * will have the corresponding buffer head mapped if needed so that | |
3394 | * that region of the page can be updated with the partial zero out. | |
3395 | * | |
3396 | * This function assumes that the page has already been locked. The | |
3397 | * The range to be discarded must be contained with in the given page. | |
3398 | * If the specified range exceeds the end of the page it will be shortened | |
3399 | * to the end of the page that corresponds to 'from'. This function is | |
3400 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3401 | * zeroed for blocks that have been either released, or are going to be | |
3402 | * released. | |
3403 | * | |
3404 | * handle: The journal handle | |
3405 | * inode: The files inode | |
3406 | * page: A locked page that contains the offset "from" | |
4907cb7b | 3407 | * from: The starting byte offset (from the beginning of the file) |
4e96b2db AH |
3408 | * to begin discarding |
3409 | * len: The length of bytes to discard | |
3410 | * flags: Optional flags that may be used: | |
3411 | * | |
3412 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3413 | * Only zero the regions of the page whose buffer heads | |
3414 | * have already been unmapped. This flag is appropriate | |
4907cb7b | 3415 | * for updating the contents of a page whose blocks may |
4e96b2db AH |
3416 | * have already been released, and we only want to zero |
3417 | * out the regions that correspond to those released blocks. | |
3418 | * | |
4907cb7b | 3419 | * Returns zero on success or negative on failure. |
4e96b2db | 3420 | */ |
5f163cc7 | 3421 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
4e96b2db AH |
3422 | struct inode *inode, struct page *page, loff_t from, |
3423 | loff_t length, int flags) | |
3424 | { | |
3425 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3426 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3427 | unsigned int blocksize, max, pos; | |
4e96b2db AH |
3428 | ext4_lblk_t iblock; |
3429 | struct buffer_head *bh; | |
3430 | int err = 0; | |
3431 | ||
3432 | blocksize = inode->i_sb->s_blocksize; | |
3433 | max = PAGE_CACHE_SIZE - offset; | |
3434 | ||
3435 | if (index != page->index) | |
3436 | return -EINVAL; | |
3437 | ||
3438 | /* | |
3439 | * correct length if it does not fall between | |
3440 | * 'from' and the end of the page | |
3441 | */ | |
3442 | if (length > max || length < 0) | |
3443 | length = max; | |
3444 | ||
3445 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3446 | ||
093e6e36 YY |
3447 | if (!page_has_buffers(page)) |
3448 | create_empty_buffers(page, blocksize, 0); | |
4e96b2db AH |
3449 | |
3450 | /* Find the buffer that contains "offset" */ | |
3451 | bh = page_buffers(page); | |
3452 | pos = blocksize; | |
3453 | while (offset >= pos) { | |
3454 | bh = bh->b_this_page; | |
3455 | iblock++; | |
3456 | pos += blocksize; | |
3457 | } | |
3458 | ||
3459 | pos = offset; | |
3460 | while (pos < offset + length) { | |
e260daf2 YY |
3461 | unsigned int end_of_block, range_to_discard; |
3462 | ||
4e96b2db AH |
3463 | err = 0; |
3464 | ||
3465 | /* The length of space left to zero and unmap */ | |
3466 | range_to_discard = offset + length - pos; | |
3467 | ||
3468 | /* The length of space until the end of the block */ | |
3469 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3470 | ||
3471 | /* | |
3472 | * Do not unmap or zero past end of block | |
3473 | * for this buffer head | |
3474 | */ | |
3475 | if (range_to_discard > end_of_block) | |
3476 | range_to_discard = end_of_block; | |
3477 | ||
3478 | ||
3479 | /* | |
3480 | * Skip this buffer head if we are only zeroing unampped | |
3481 | * regions of the page | |
3482 | */ | |
3483 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3484 | buffer_mapped(bh)) | |
3485 | goto next; | |
3486 | ||
3487 | /* If the range is block aligned, unmap */ | |
3488 | if (range_to_discard == blocksize) { | |
3489 | clear_buffer_dirty(bh); | |
3490 | bh->b_bdev = NULL; | |
3491 | clear_buffer_mapped(bh); | |
3492 | clear_buffer_req(bh); | |
3493 | clear_buffer_new(bh); | |
3494 | clear_buffer_delay(bh); | |
3495 | clear_buffer_unwritten(bh); | |
3496 | clear_buffer_uptodate(bh); | |
3497 | zero_user(page, pos, range_to_discard); | |
3498 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3499 | goto next; | |
3500 | } | |
3501 | ||
3502 | /* | |
3503 | * If this block is not completely contained in the range | |
3504 | * to be discarded, then it is not going to be released. Because | |
3505 | * we need to keep this block, we need to make sure this part | |
3506 | * of the page is uptodate before we modify it by writeing | |
3507 | * partial zeros on it. | |
3508 | */ | |
3509 | if (!buffer_mapped(bh)) { | |
3510 | /* | |
3511 | * Buffer head must be mapped before we can read | |
3512 | * from the block | |
3513 | */ | |
3514 | BUFFER_TRACE(bh, "unmapped"); | |
3515 | ext4_get_block(inode, iblock, bh, 0); | |
3516 | /* unmapped? It's a hole - nothing to do */ | |
3517 | if (!buffer_mapped(bh)) { | |
3518 | BUFFER_TRACE(bh, "still unmapped"); | |
3519 | goto next; | |
3520 | } | |
3521 | } | |
3522 | ||
3523 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3524 | if (PageUptodate(page)) | |
3525 | set_buffer_uptodate(bh); | |
3526 | ||
3527 | if (!buffer_uptodate(bh)) { | |
3528 | err = -EIO; | |
3529 | ll_rw_block(READ, 1, &bh); | |
3530 | wait_on_buffer(bh); | |
3531 | /* Uhhuh. Read error. Complain and punt.*/ | |
3532 | if (!buffer_uptodate(bh)) | |
3533 | goto next; | |
3534 | } | |
3535 | ||
3536 | if (ext4_should_journal_data(inode)) { | |
3537 | BUFFER_TRACE(bh, "get write access"); | |
3538 | err = ext4_journal_get_write_access(handle, bh); | |
3539 | if (err) | |
3540 | goto next; | |
3541 | } | |
3542 | ||
3543 | zero_user(page, pos, range_to_discard); | |
3544 | ||
3545 | err = 0; | |
3546 | if (ext4_should_journal_data(inode)) { | |
3547 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3548 | } else |
4e96b2db | 3549 | mark_buffer_dirty(bh); |
4e96b2db AH |
3550 | |
3551 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3552 | next: | |
3553 | bh = bh->b_this_page; | |
3554 | iblock++; | |
3555 | pos += range_to_discard; | |
3556 | } | |
3557 | ||
3558 | return err; | |
3559 | } | |
3560 | ||
91ef4caf DG |
3561 | int ext4_can_truncate(struct inode *inode) |
3562 | { | |
91ef4caf DG |
3563 | if (S_ISREG(inode->i_mode)) |
3564 | return 1; | |
3565 | if (S_ISDIR(inode->i_mode)) | |
3566 | return 1; | |
3567 | if (S_ISLNK(inode->i_mode)) | |
3568 | return !ext4_inode_is_fast_symlink(inode); | |
3569 | return 0; | |
3570 | } | |
3571 | ||
a4bb6b64 AH |
3572 | /* |
3573 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3574 | * associated with the given offset and length | |
3575 | * | |
3576 | * @inode: File inode | |
3577 | * @offset: The offset where the hole will begin | |
3578 | * @len: The length of the hole | |
3579 | * | |
4907cb7b | 3580 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3581 | */ |
3582 | ||
3583 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3584 | { | |
496ad9aa | 3585 | struct inode *inode = file_inode(file); |
26a4c0c6 TT |
3586 | struct super_block *sb = inode->i_sb; |
3587 | ext4_lblk_t first_block, stop_block; | |
3588 | struct address_space *mapping = inode->i_mapping; | |
3589 | loff_t first_page, last_page, page_len; | |
3590 | loff_t first_page_offset, last_page_offset; | |
3591 | handle_t *handle; | |
3592 | unsigned int credits; | |
3593 | int ret = 0; | |
3594 | ||
a4bb6b64 | 3595 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3596 | return -EOPNOTSUPP; |
a4bb6b64 | 3597 | |
26a4c0c6 | 3598 | if (EXT4_SB(sb)->s_cluster_ratio > 1) { |
bab08ab9 | 3599 | /* TODO: Add support for bigalloc file systems */ |
73355192 | 3600 | return -EOPNOTSUPP; |
bab08ab9 TT |
3601 | } |
3602 | ||
aaddea81 ZL |
3603 | trace_ext4_punch_hole(inode, offset, length); |
3604 | ||
26a4c0c6 TT |
3605 | /* |
3606 | * Write out all dirty pages to avoid race conditions | |
3607 | * Then release them. | |
3608 | */ | |
3609 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3610 | ret = filemap_write_and_wait_range(mapping, offset, | |
3611 | offset + length - 1); | |
3612 | if (ret) | |
3613 | return ret; | |
3614 | } | |
3615 | ||
3616 | mutex_lock(&inode->i_mutex); | |
3617 | /* It's not possible punch hole on append only file */ | |
3618 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
3619 | ret = -EPERM; | |
3620 | goto out_mutex; | |
3621 | } | |
3622 | if (IS_SWAPFILE(inode)) { | |
3623 | ret = -ETXTBSY; | |
3624 | goto out_mutex; | |
3625 | } | |
3626 | ||
3627 | /* No need to punch hole beyond i_size */ | |
3628 | if (offset >= inode->i_size) | |
3629 | goto out_mutex; | |
3630 | ||
3631 | /* | |
3632 | * If the hole extends beyond i_size, set the hole | |
3633 | * to end after the page that contains i_size | |
3634 | */ | |
3635 | if (offset + length > inode->i_size) { | |
3636 | length = inode->i_size + | |
3637 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3638 | offset; | |
3639 | } | |
3640 | ||
3641 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
3642 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
3643 | ||
3644 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
3645 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
3646 | ||
3647 | /* Now release the pages */ | |
3648 | if (last_page_offset > first_page_offset) { | |
3649 | truncate_pagecache_range(inode, first_page_offset, | |
3650 | last_page_offset - 1); | |
3651 | } | |
3652 | ||
3653 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3654 | ext4_inode_block_unlocked_dio(inode); | |
3655 | ret = ext4_flush_unwritten_io(inode); | |
3656 | if (ret) | |
3657 | goto out_dio; | |
3658 | inode_dio_wait(inode); | |
3659 | ||
3660 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3661 | credits = ext4_writepage_trans_blocks(inode); | |
3662 | else | |
3663 | credits = ext4_blocks_for_truncate(inode); | |
3664 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3665 | if (IS_ERR(handle)) { | |
3666 | ret = PTR_ERR(handle); | |
3667 | ext4_std_error(sb, ret); | |
3668 | goto out_dio; | |
3669 | } | |
3670 | ||
3671 | /* | |
3672 | * Now we need to zero out the non-page-aligned data in the | |
3673 | * pages at the start and tail of the hole, and unmap the | |
3674 | * buffer heads for the block aligned regions of the page that | |
3675 | * were completely zeroed. | |
3676 | */ | |
3677 | if (first_page > last_page) { | |
3678 | /* | |
3679 | * If the file space being truncated is contained | |
3680 | * within a page just zero out and unmap the middle of | |
3681 | * that page | |
3682 | */ | |
3683 | ret = ext4_discard_partial_page_buffers(handle, | |
3684 | mapping, offset, length, 0); | |
3685 | ||
3686 | if (ret) | |
3687 | goto out_stop; | |
3688 | } else { | |
3689 | /* | |
3690 | * zero out and unmap the partial page that contains | |
3691 | * the start of the hole | |
3692 | */ | |
3693 | page_len = first_page_offset - offset; | |
3694 | if (page_len > 0) { | |
3695 | ret = ext4_discard_partial_page_buffers(handle, mapping, | |
3696 | offset, page_len, 0); | |
3697 | if (ret) | |
3698 | goto out_stop; | |
3699 | } | |
3700 | ||
3701 | /* | |
3702 | * zero out and unmap the partial page that contains | |
3703 | * the end of the hole | |
3704 | */ | |
3705 | page_len = offset + length - last_page_offset; | |
3706 | if (page_len > 0) { | |
3707 | ret = ext4_discard_partial_page_buffers(handle, mapping, | |
3708 | last_page_offset, page_len, 0); | |
3709 | if (ret) | |
3710 | goto out_stop; | |
3711 | } | |
3712 | } | |
3713 | ||
3714 | /* | |
3715 | * If i_size is contained in the last page, we need to | |
3716 | * unmap and zero the partial page after i_size | |
3717 | */ | |
3718 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
3719 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
3720 | page_len = PAGE_CACHE_SIZE - | |
3721 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3722 | ||
3723 | if (page_len > 0) { | |
3724 | ret = ext4_discard_partial_page_buffers(handle, | |
3725 | mapping, inode->i_size, page_len, 0); | |
3726 | ||
3727 | if (ret) | |
3728 | goto out_stop; | |
3729 | } | |
3730 | } | |
3731 | ||
3732 | first_block = (offset + sb->s_blocksize - 1) >> | |
3733 | EXT4_BLOCK_SIZE_BITS(sb); | |
3734 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3735 | ||
3736 | /* If there are no blocks to remove, return now */ | |
3737 | if (first_block >= stop_block) | |
3738 | goto out_stop; | |
3739 | ||
3740 | down_write(&EXT4_I(inode)->i_data_sem); | |
3741 | ext4_discard_preallocations(inode); | |
3742 | ||
3743 | ret = ext4_es_remove_extent(inode, first_block, | |
3744 | stop_block - first_block); | |
3745 | if (ret) { | |
3746 | up_write(&EXT4_I(inode)->i_data_sem); | |
3747 | goto out_stop; | |
3748 | } | |
3749 | ||
3750 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3751 | ret = ext4_ext_remove_space(inode, first_block, | |
3752 | stop_block - 1); | |
3753 | else | |
3754 | ret = ext4_free_hole_blocks(handle, inode, first_block, | |
3755 | stop_block); | |
3756 | ||
3757 | ext4_discard_preallocations(inode); | |
819c4920 | 3758 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3759 | if (IS_SYNC(inode)) |
3760 | ext4_handle_sync(handle); | |
26a4c0c6 TT |
3761 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3762 | ext4_mark_inode_dirty(handle, inode); | |
3763 | out_stop: | |
3764 | ext4_journal_stop(handle); | |
3765 | out_dio: | |
3766 | ext4_inode_resume_unlocked_dio(inode); | |
3767 | out_mutex: | |
3768 | mutex_unlock(&inode->i_mutex); | |
3769 | return ret; | |
a4bb6b64 AH |
3770 | } |
3771 | ||
ac27a0ec | 3772 | /* |
617ba13b | 3773 | * ext4_truncate() |
ac27a0ec | 3774 | * |
617ba13b MC |
3775 | * We block out ext4_get_block() block instantiations across the entire |
3776 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3777 | * simultaneously on behalf of the same inode. |
3778 | * | |
42b2aa86 | 3779 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3780 | * is one core, guiding principle: the file's tree must always be consistent on |
3781 | * disk. We must be able to restart the truncate after a crash. | |
3782 | * | |
3783 | * The file's tree may be transiently inconsistent in memory (although it | |
3784 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3785 | * the contents of (the filesystem + the journal) must be consistent and | |
3786 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3787 | * left-to-right works OK too). | |
3788 | * | |
3789 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3790 | * truncate against the orphan inode list. | |
3791 | * | |
3792 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3793 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3794 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3795 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3796 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3797 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3798 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3799 | */ |
617ba13b | 3800 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3801 | { |
819c4920 TT |
3802 | struct ext4_inode_info *ei = EXT4_I(inode); |
3803 | unsigned int credits; | |
3804 | handle_t *handle; | |
3805 | struct address_space *mapping = inode->i_mapping; | |
3806 | loff_t page_len; | |
3807 | ||
19b5ef61 TT |
3808 | /* |
3809 | * There is a possibility that we're either freeing the inode | |
3810 | * or it completely new indode. In those cases we might not | |
3811 | * have i_mutex locked because it's not necessary. | |
3812 | */ | |
3813 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3814 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3815 | trace_ext4_truncate_enter(inode); |
3816 | ||
91ef4caf | 3817 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3818 | return; |
3819 | ||
12e9b892 | 3820 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3821 | |
5534fb5b | 3822 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3823 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3824 | |
aef1c851 TM |
3825 | if (ext4_has_inline_data(inode)) { |
3826 | int has_inline = 1; | |
3827 | ||
3828 | ext4_inline_data_truncate(inode, &has_inline); | |
3829 | if (has_inline) | |
3830 | return; | |
3831 | } | |
3832 | ||
819c4920 TT |
3833 | /* |
3834 | * finish any pending end_io work so we won't run the risk of | |
3835 | * converting any truncated blocks to initialized later | |
3836 | */ | |
3837 | ext4_flush_unwritten_io(inode); | |
3838 | ||
3839 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3840 | credits = ext4_writepage_trans_blocks(inode); | |
3841 | else | |
3842 | credits = ext4_blocks_for_truncate(inode); | |
3843 | ||
3844 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3845 | if (IS_ERR(handle)) { | |
3846 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3847 | return; | |
3848 | } | |
3849 | ||
3850 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { | |
3851 | page_len = PAGE_CACHE_SIZE - | |
3852 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3853 | ||
3854 | if (ext4_discard_partial_page_buffers(handle, | |
3855 | mapping, inode->i_size, page_len, 0)) | |
3856 | goto out_stop; | |
3857 | } | |
3858 | ||
3859 | /* | |
3860 | * We add the inode to the orphan list, so that if this | |
3861 | * truncate spans multiple transactions, and we crash, we will | |
3862 | * resume the truncate when the filesystem recovers. It also | |
3863 | * marks the inode dirty, to catch the new size. | |
3864 | * | |
3865 | * Implication: the file must always be in a sane, consistent | |
3866 | * truncatable state while each transaction commits. | |
3867 | */ | |
3868 | if (ext4_orphan_add(handle, inode)) | |
3869 | goto out_stop; | |
3870 | ||
3871 | down_write(&EXT4_I(inode)->i_data_sem); | |
3872 | ||
3873 | ext4_discard_preallocations(inode); | |
3874 | ||
ff9893dc | 3875 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3876 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3877 | else |
819c4920 TT |
3878 | ext4_ind_truncate(handle, inode); |
3879 | ||
3880 | up_write(&ei->i_data_sem); | |
3881 | ||
3882 | if (IS_SYNC(inode)) | |
3883 | ext4_handle_sync(handle); | |
3884 | ||
3885 | out_stop: | |
3886 | /* | |
3887 | * If this was a simple ftruncate() and the file will remain alive, | |
3888 | * then we need to clear up the orphan record which we created above. | |
3889 | * However, if this was a real unlink then we were called by | |
3890 | * ext4_delete_inode(), and we allow that function to clean up the | |
3891 | * orphan info for us. | |
3892 | */ | |
3893 | if (inode->i_nlink) | |
3894 | ext4_orphan_del(handle, inode); | |
3895 | ||
3896 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3897 | ext4_mark_inode_dirty(handle, inode); | |
3898 | ext4_journal_stop(handle); | |
ac27a0ec | 3899 | |
0562e0ba | 3900 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3901 | } |
3902 | ||
ac27a0ec | 3903 | /* |
617ba13b | 3904 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3905 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3906 | * data in memory that is needed to recreate the on-disk version of this | |
3907 | * inode. | |
3908 | */ | |
617ba13b MC |
3909 | static int __ext4_get_inode_loc(struct inode *inode, |
3910 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3911 | { |
240799cd TT |
3912 | struct ext4_group_desc *gdp; |
3913 | struct buffer_head *bh; | |
3914 | struct super_block *sb = inode->i_sb; | |
3915 | ext4_fsblk_t block; | |
3916 | int inodes_per_block, inode_offset; | |
3917 | ||
3a06d778 | 3918 | iloc->bh = NULL; |
240799cd TT |
3919 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3920 | return -EIO; | |
ac27a0ec | 3921 | |
240799cd TT |
3922 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3923 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3924 | if (!gdp) | |
ac27a0ec DK |
3925 | return -EIO; |
3926 | ||
240799cd TT |
3927 | /* |
3928 | * Figure out the offset within the block group inode table | |
3929 | */ | |
00d09882 | 3930 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3931 | inode_offset = ((inode->i_ino - 1) % |
3932 | EXT4_INODES_PER_GROUP(sb)); | |
3933 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3934 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3935 | ||
3936 | bh = sb_getblk(sb, block); | |
aebf0243 | 3937 | if (unlikely(!bh)) |
860d21e2 | 3938 | return -ENOMEM; |
ac27a0ec DK |
3939 | if (!buffer_uptodate(bh)) { |
3940 | lock_buffer(bh); | |
9c83a923 HK |
3941 | |
3942 | /* | |
3943 | * If the buffer has the write error flag, we have failed | |
3944 | * to write out another inode in the same block. In this | |
3945 | * case, we don't have to read the block because we may | |
3946 | * read the old inode data successfully. | |
3947 | */ | |
3948 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3949 | set_buffer_uptodate(bh); | |
3950 | ||
ac27a0ec DK |
3951 | if (buffer_uptodate(bh)) { |
3952 | /* someone brought it uptodate while we waited */ | |
3953 | unlock_buffer(bh); | |
3954 | goto has_buffer; | |
3955 | } | |
3956 | ||
3957 | /* | |
3958 | * If we have all information of the inode in memory and this | |
3959 | * is the only valid inode in the block, we need not read the | |
3960 | * block. | |
3961 | */ | |
3962 | if (in_mem) { | |
3963 | struct buffer_head *bitmap_bh; | |
240799cd | 3964 | int i, start; |
ac27a0ec | 3965 | |
240799cd | 3966 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3967 | |
240799cd TT |
3968 | /* Is the inode bitmap in cache? */ |
3969 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3970 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3971 | goto make_io; |
3972 | ||
3973 | /* | |
3974 | * If the inode bitmap isn't in cache then the | |
3975 | * optimisation may end up performing two reads instead | |
3976 | * of one, so skip it. | |
3977 | */ | |
3978 | if (!buffer_uptodate(bitmap_bh)) { | |
3979 | brelse(bitmap_bh); | |
3980 | goto make_io; | |
3981 | } | |
240799cd | 3982 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3983 | if (i == inode_offset) |
3984 | continue; | |
617ba13b | 3985 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3986 | break; |
3987 | } | |
3988 | brelse(bitmap_bh); | |
240799cd | 3989 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3990 | /* all other inodes are free, so skip I/O */ |
3991 | memset(bh->b_data, 0, bh->b_size); | |
3992 | set_buffer_uptodate(bh); | |
3993 | unlock_buffer(bh); | |
3994 | goto has_buffer; | |
3995 | } | |
3996 | } | |
3997 | ||
3998 | make_io: | |
240799cd TT |
3999 | /* |
4000 | * If we need to do any I/O, try to pre-readahead extra | |
4001 | * blocks from the inode table. | |
4002 | */ | |
4003 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4004 | ext4_fsblk_t b, end, table; | |
4005 | unsigned num; | |
0d606e2c | 4006 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4007 | |
4008 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4009 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4010 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4011 | if (table > b) |
4012 | b = table; | |
0d606e2c | 4013 | end = b + ra_blks; |
240799cd | 4014 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4015 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4016 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4017 | table += num / inodes_per_block; |
4018 | if (end > table) | |
4019 | end = table; | |
4020 | while (b <= end) | |
4021 | sb_breadahead(sb, b++); | |
4022 | } | |
4023 | ||
ac27a0ec DK |
4024 | /* |
4025 | * There are other valid inodes in the buffer, this inode | |
4026 | * has in-inode xattrs, or we don't have this inode in memory. | |
4027 | * Read the block from disk. | |
4028 | */ | |
0562e0ba | 4029 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4030 | get_bh(bh); |
4031 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 4032 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4033 | wait_on_buffer(bh); |
4034 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4035 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4036 | "unable to read itable block"); | |
ac27a0ec DK |
4037 | brelse(bh); |
4038 | return -EIO; | |
4039 | } | |
4040 | } | |
4041 | has_buffer: | |
4042 | iloc->bh = bh; | |
4043 | return 0; | |
4044 | } | |
4045 | ||
617ba13b | 4046 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4047 | { |
4048 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4049 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4050 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4051 | } |
4052 | ||
617ba13b | 4053 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4054 | { |
617ba13b | 4055 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4056 | |
4057 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4058 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4059 | inode->i_flags |= S_SYNC; |
617ba13b | 4060 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4061 | inode->i_flags |= S_APPEND; |
617ba13b | 4062 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4063 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4064 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4065 | inode->i_flags |= S_NOATIME; |
617ba13b | 4066 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4067 | inode->i_flags |= S_DIRSYNC; |
4068 | } | |
4069 | ||
ff9ddf7e JK |
4070 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4071 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4072 | { | |
84a8dce2 DM |
4073 | unsigned int vfs_fl; |
4074 | unsigned long old_fl, new_fl; | |
4075 | ||
4076 | do { | |
4077 | vfs_fl = ei->vfs_inode.i_flags; | |
4078 | old_fl = ei->i_flags; | |
4079 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4080 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4081 | EXT4_DIRSYNC_FL); | |
4082 | if (vfs_fl & S_SYNC) | |
4083 | new_fl |= EXT4_SYNC_FL; | |
4084 | if (vfs_fl & S_APPEND) | |
4085 | new_fl |= EXT4_APPEND_FL; | |
4086 | if (vfs_fl & S_IMMUTABLE) | |
4087 | new_fl |= EXT4_IMMUTABLE_FL; | |
4088 | if (vfs_fl & S_NOATIME) | |
4089 | new_fl |= EXT4_NOATIME_FL; | |
4090 | if (vfs_fl & S_DIRSYNC) | |
4091 | new_fl |= EXT4_DIRSYNC_FL; | |
4092 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4093 | } |
de9a55b8 | 4094 | |
0fc1b451 | 4095 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4096 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4097 | { |
4098 | blkcnt_t i_blocks ; | |
8180a562 AK |
4099 | struct inode *inode = &(ei->vfs_inode); |
4100 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4101 | |
4102 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4103 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4104 | /* we are using combined 48 bit field */ | |
4105 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4106 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4107 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4108 | /* i_blocks represent file system block size */ |
4109 | return i_blocks << (inode->i_blkbits - 9); | |
4110 | } else { | |
4111 | return i_blocks; | |
4112 | } | |
0fc1b451 AK |
4113 | } else { |
4114 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4115 | } | |
4116 | } | |
ff9ddf7e | 4117 | |
152a7b0a TM |
4118 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4119 | struct ext4_inode *raw_inode, | |
4120 | struct ext4_inode_info *ei) | |
4121 | { | |
4122 | __le32 *magic = (void *)raw_inode + | |
4123 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4124 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4125 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4126 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4127 | } else |
4128 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4129 | } |
4130 | ||
1d1fe1ee | 4131 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4132 | { |
617ba13b MC |
4133 | struct ext4_iloc iloc; |
4134 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4135 | struct ext4_inode_info *ei; |
1d1fe1ee | 4136 | struct inode *inode; |
b436b9be | 4137 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4138 | long ret; |
ac27a0ec | 4139 | int block; |
08cefc7a EB |
4140 | uid_t i_uid; |
4141 | gid_t i_gid; | |
ac27a0ec | 4142 | |
1d1fe1ee DH |
4143 | inode = iget_locked(sb, ino); |
4144 | if (!inode) | |
4145 | return ERR_PTR(-ENOMEM); | |
4146 | if (!(inode->i_state & I_NEW)) | |
4147 | return inode; | |
4148 | ||
4149 | ei = EXT4_I(inode); | |
7dc57615 | 4150 | iloc.bh = NULL; |
ac27a0ec | 4151 | |
1d1fe1ee DH |
4152 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4153 | if (ret < 0) | |
ac27a0ec | 4154 | goto bad_inode; |
617ba13b | 4155 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4156 | |
4157 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4158 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4159 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4160 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4161 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4162 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4163 | EXT4_INODE_SIZE(inode->i_sb)); | |
4164 | ret = -EIO; | |
4165 | goto bad_inode; | |
4166 | } | |
4167 | } else | |
4168 | ei->i_extra_isize = 0; | |
4169 | ||
4170 | /* Precompute checksum seed for inode metadata */ | |
4171 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4172 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
4173 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4174 | __u32 csum; | |
4175 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4176 | __le32 gen = raw_inode->i_generation; | |
4177 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4178 | sizeof(inum)); | |
4179 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4180 | sizeof(gen)); | |
4181 | } | |
4182 | ||
4183 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4184 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
4185 | ret = -EIO; | |
4186 | goto bad_inode; | |
4187 | } | |
4188 | ||
ac27a0ec | 4189 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4190 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4191 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4192 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4193 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4194 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4195 | } |
08cefc7a EB |
4196 | i_uid_write(inode, i_uid); |
4197 | i_gid_write(inode, i_gid); | |
bfe86848 | 4198 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4199 | |
353eb83c | 4200 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4201 | ei->i_inline_off = 0; |
ac27a0ec DK |
4202 | ei->i_dir_start_lookup = 0; |
4203 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4204 | /* We now have enough fields to check if the inode was active or not. | |
4205 | * This is needed because nfsd might try to access dead inodes | |
4206 | * the test is that same one that e2fsck uses | |
4207 | * NeilBrown 1999oct15 | |
4208 | */ | |
4209 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4210 | if ((inode->i_mode == 0 || |
4211 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4212 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4213 | /* this inode is deleted */ |
1d1fe1ee | 4214 | ret = -ESTALE; |
ac27a0ec DK |
4215 | goto bad_inode; |
4216 | } | |
4217 | /* The only unlinked inodes we let through here have | |
4218 | * valid i_mode and are being read by the orphan | |
4219 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4220 | * the process of deleting those. |
4221 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4222 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4223 | } |
ac27a0ec | 4224 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4225 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4226 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4227 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4228 | ei->i_file_acl |= |
4229 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4230 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4231 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4232 | #ifdef CONFIG_QUOTA |
4233 | ei->i_reserved_quota = 0; | |
4234 | #endif | |
ac27a0ec DK |
4235 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4236 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4237 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4238 | /* |
4239 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4240 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4241 | */ | |
617ba13b | 4242 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4243 | ei->i_data[block] = raw_inode->i_block[block]; |
4244 | INIT_LIST_HEAD(&ei->i_orphan); | |
4245 | ||
b436b9be JK |
4246 | /* |
4247 | * Set transaction id's of transactions that have to be committed | |
4248 | * to finish f[data]sync. We set them to currently running transaction | |
4249 | * as we cannot be sure that the inode or some of its metadata isn't | |
4250 | * part of the transaction - the inode could have been reclaimed and | |
4251 | * now it is reread from disk. | |
4252 | */ | |
4253 | if (journal) { | |
4254 | transaction_t *transaction; | |
4255 | tid_t tid; | |
4256 | ||
a931da6a | 4257 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4258 | if (journal->j_running_transaction) |
4259 | transaction = journal->j_running_transaction; | |
4260 | else | |
4261 | transaction = journal->j_committing_transaction; | |
4262 | if (transaction) | |
4263 | tid = transaction->t_tid; | |
4264 | else | |
4265 | tid = journal->j_commit_sequence; | |
a931da6a | 4266 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4267 | ei->i_sync_tid = tid; |
4268 | ei->i_datasync_tid = tid; | |
4269 | } | |
4270 | ||
0040d987 | 4271 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4272 | if (ei->i_extra_isize == 0) { |
4273 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4274 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4275 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4276 | } else { |
152a7b0a | 4277 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4278 | } |
814525f4 | 4279 | } |
ac27a0ec | 4280 | |
ef7f3835 KS |
4281 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4282 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4283 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4284 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4285 | ||
25ec56b5 JNC |
4286 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4287 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4288 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4289 | inode->i_version |= | |
4290 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4291 | } | |
4292 | ||
c4b5a614 | 4293 | ret = 0; |
485c26ec | 4294 | if (ei->i_file_acl && |
1032988c | 4295 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4296 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4297 | ei->i_file_acl); | |
485c26ec TT |
4298 | ret = -EIO; |
4299 | goto bad_inode; | |
f19d5870 TM |
4300 | } else if (!ext4_has_inline_data(inode)) { |
4301 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4302 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4303 | (S_ISLNK(inode->i_mode) && | |
4304 | !ext4_inode_is_fast_symlink(inode)))) | |
4305 | /* Validate extent which is part of inode */ | |
4306 | ret = ext4_ext_check_inode(inode); | |
4307 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4308 | (S_ISLNK(inode->i_mode) && | |
4309 | !ext4_inode_is_fast_symlink(inode))) { | |
4310 | /* Validate block references which are part of inode */ | |
4311 | ret = ext4_ind_check_inode(inode); | |
4312 | } | |
fe2c8191 | 4313 | } |
567f3e9a | 4314 | if (ret) |
de9a55b8 | 4315 | goto bad_inode; |
7a262f7c | 4316 | |
ac27a0ec | 4317 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4318 | inode->i_op = &ext4_file_inode_operations; |
4319 | inode->i_fop = &ext4_file_operations; | |
4320 | ext4_set_aops(inode); | |
ac27a0ec | 4321 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4322 | inode->i_op = &ext4_dir_inode_operations; |
4323 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4324 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4325 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4326 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4327 | nd_terminate_link(ei->i_data, inode->i_size, |
4328 | sizeof(ei->i_data) - 1); | |
4329 | } else { | |
617ba13b MC |
4330 | inode->i_op = &ext4_symlink_inode_operations; |
4331 | ext4_set_aops(inode); | |
ac27a0ec | 4332 | } |
563bdd61 TT |
4333 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4334 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4335 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4336 | if (raw_inode->i_block[0]) |
4337 | init_special_inode(inode, inode->i_mode, | |
4338 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4339 | else | |
4340 | init_special_inode(inode, inode->i_mode, | |
4341 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4342 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4343 | make_bad_inode(inode); | |
563bdd61 | 4344 | } else { |
563bdd61 | 4345 | ret = -EIO; |
24676da4 | 4346 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4347 | goto bad_inode; |
ac27a0ec | 4348 | } |
af5bc92d | 4349 | brelse(iloc.bh); |
617ba13b | 4350 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4351 | unlock_new_inode(inode); |
4352 | return inode; | |
ac27a0ec DK |
4353 | |
4354 | bad_inode: | |
567f3e9a | 4355 | brelse(iloc.bh); |
1d1fe1ee DH |
4356 | iget_failed(inode); |
4357 | return ERR_PTR(ret); | |
ac27a0ec DK |
4358 | } |
4359 | ||
0fc1b451 AK |
4360 | static int ext4_inode_blocks_set(handle_t *handle, |
4361 | struct ext4_inode *raw_inode, | |
4362 | struct ext4_inode_info *ei) | |
4363 | { | |
4364 | struct inode *inode = &(ei->vfs_inode); | |
4365 | u64 i_blocks = inode->i_blocks; | |
4366 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4367 | |
4368 | if (i_blocks <= ~0U) { | |
4369 | /* | |
4907cb7b | 4370 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4371 | * as multiple of 512 bytes |
4372 | */ | |
8180a562 | 4373 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4374 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4375 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4376 | return 0; |
4377 | } | |
4378 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4379 | return -EFBIG; | |
4380 | ||
4381 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4382 | /* |
4383 | * i_blocks can be represented in a 48 bit variable | |
4384 | * as multiple of 512 bytes | |
4385 | */ | |
8180a562 | 4386 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4387 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4388 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4389 | } else { |
84a8dce2 | 4390 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4391 | /* i_block is stored in file system block size */ |
4392 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4393 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4394 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4395 | } |
f287a1a5 | 4396 | return 0; |
0fc1b451 AK |
4397 | } |
4398 | ||
ac27a0ec DK |
4399 | /* |
4400 | * Post the struct inode info into an on-disk inode location in the | |
4401 | * buffer-cache. This gobbles the caller's reference to the | |
4402 | * buffer_head in the inode location struct. | |
4403 | * | |
4404 | * The caller must have write access to iloc->bh. | |
4405 | */ | |
617ba13b | 4406 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4407 | struct inode *inode, |
830156c7 | 4408 | struct ext4_iloc *iloc) |
ac27a0ec | 4409 | { |
617ba13b MC |
4410 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4411 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4412 | struct buffer_head *bh = iloc->bh; |
4413 | int err = 0, rc, block; | |
b71fc079 | 4414 | int need_datasync = 0; |
08cefc7a EB |
4415 | uid_t i_uid; |
4416 | gid_t i_gid; | |
ac27a0ec DK |
4417 | |
4418 | /* For fields not not tracking in the in-memory inode, | |
4419 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 4420 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4421 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4422 | |
ff9ddf7e | 4423 | ext4_get_inode_flags(ei); |
ac27a0ec | 4424 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4425 | i_uid = i_uid_read(inode); |
4426 | i_gid = i_gid_read(inode); | |
af5bc92d | 4427 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4428 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4429 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4430 | /* |
4431 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4432 | * re-used with the upper 16 bits of the uid/gid intact | |
4433 | */ | |
af5bc92d | 4434 | if (!ei->i_dtime) { |
ac27a0ec | 4435 | raw_inode->i_uid_high = |
08cefc7a | 4436 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4437 | raw_inode->i_gid_high = |
08cefc7a | 4438 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4439 | } else { |
4440 | raw_inode->i_uid_high = 0; | |
4441 | raw_inode->i_gid_high = 0; | |
4442 | } | |
4443 | } else { | |
08cefc7a EB |
4444 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4445 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4446 | raw_inode->i_uid_high = 0; |
4447 | raw_inode->i_gid_high = 0; | |
4448 | } | |
4449 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4450 | |
4451 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4452 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4453 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4454 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4455 | ||
0fc1b451 AK |
4456 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4457 | goto out_brelse; | |
ac27a0ec | 4458 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4459 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4460 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4461 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4462 | raw_inode->i_file_acl_high = |
4463 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4464 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4465 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4466 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4467 | need_datasync = 1; | |
4468 | } | |
a48380f7 AK |
4469 | if (ei->i_disksize > 0x7fffffffULL) { |
4470 | struct super_block *sb = inode->i_sb; | |
4471 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4472 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4473 | EXT4_SB(sb)->s_es->s_rev_level == | |
4474 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4475 | /* If this is the first large file | |
4476 | * created, add a flag to the superblock. | |
4477 | */ | |
4478 | err = ext4_journal_get_write_access(handle, | |
4479 | EXT4_SB(sb)->s_sbh); | |
4480 | if (err) | |
4481 | goto out_brelse; | |
4482 | ext4_update_dynamic_rev(sb); | |
4483 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4484 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
0390131b | 4485 | ext4_handle_sync(handle); |
b50924c2 | 4486 | err = ext4_handle_dirty_super(handle, sb); |
ac27a0ec DK |
4487 | } |
4488 | } | |
4489 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4490 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4491 | if (old_valid_dev(inode->i_rdev)) { | |
4492 | raw_inode->i_block[0] = | |
4493 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4494 | raw_inode->i_block[1] = 0; | |
4495 | } else { | |
4496 | raw_inode->i_block[0] = 0; | |
4497 | raw_inode->i_block[1] = | |
4498 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4499 | raw_inode->i_block[2] = 0; | |
4500 | } | |
f19d5870 | 4501 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4502 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4503 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4504 | } |
ac27a0ec | 4505 | |
25ec56b5 JNC |
4506 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4507 | if (ei->i_extra_isize) { | |
4508 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4509 | raw_inode->i_version_hi = | |
4510 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4511 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4512 | } |
4513 | ||
814525f4 DW |
4514 | ext4_inode_csum_set(inode, raw_inode, ei); |
4515 | ||
830156c7 | 4516 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4517 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4518 | if (!err) |
4519 | err = rc; | |
19f5fb7a | 4520 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4521 | |
b71fc079 | 4522 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4523 | out_brelse: |
af5bc92d | 4524 | brelse(bh); |
617ba13b | 4525 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4526 | return err; |
4527 | } | |
4528 | ||
4529 | /* | |
617ba13b | 4530 | * ext4_write_inode() |
ac27a0ec DK |
4531 | * |
4532 | * We are called from a few places: | |
4533 | * | |
4534 | * - Within generic_file_write() for O_SYNC files. | |
4535 | * Here, there will be no transaction running. We wait for any running | |
4907cb7b | 4536 | * transaction to commit. |
ac27a0ec DK |
4537 | * |
4538 | * - Within sys_sync(), kupdate and such. | |
4539 | * We wait on commit, if tol to. | |
4540 | * | |
4541 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4542 | * Here we simply return. We can't afford to block kswapd on the | |
4543 | * journal commit. | |
4544 | * | |
4545 | * In all cases it is actually safe for us to return without doing anything, | |
4546 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4547 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4548 | * knfsd. |
4549 | * | |
4550 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4551 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4552 | * which we are interested. | |
4553 | * | |
4554 | * It would be a bug for them to not do this. The code: | |
4555 | * | |
4556 | * mark_inode_dirty(inode) | |
4557 | * stuff(); | |
4558 | * inode->i_size = expr; | |
4559 | * | |
4560 | * is in error because a kswapd-driven write_inode() could occur while | |
4561 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4562 | * will no longer be on the superblock's dirty inode list. | |
4563 | */ | |
a9185b41 | 4564 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4565 | { |
91ac6f43 FM |
4566 | int err; |
4567 | ||
ac27a0ec DK |
4568 | if (current->flags & PF_MEMALLOC) |
4569 | return 0; | |
4570 | ||
91ac6f43 FM |
4571 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4572 | if (ext4_journal_current_handle()) { | |
4573 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4574 | dump_stack(); | |
4575 | return -EIO; | |
4576 | } | |
ac27a0ec | 4577 | |
a9185b41 | 4578 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4579 | return 0; |
4580 | ||
4581 | err = ext4_force_commit(inode->i_sb); | |
4582 | } else { | |
4583 | struct ext4_iloc iloc; | |
ac27a0ec | 4584 | |
8b472d73 | 4585 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4586 | if (err) |
4587 | return err; | |
a9185b41 | 4588 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4589 | sync_dirty_buffer(iloc.bh); |
4590 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4591 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4592 | "IO error syncing inode"); | |
830156c7 FM |
4593 | err = -EIO; |
4594 | } | |
fd2dd9fb | 4595 | brelse(iloc.bh); |
91ac6f43 FM |
4596 | } |
4597 | return err; | |
ac27a0ec DK |
4598 | } |
4599 | ||
53e87268 JK |
4600 | /* |
4601 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4602 | * buffers that are attached to a page stradding i_size and are undergoing | |
4603 | * commit. In that case we have to wait for commit to finish and try again. | |
4604 | */ | |
4605 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4606 | { | |
4607 | struct page *page; | |
4608 | unsigned offset; | |
4609 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4610 | tid_t commit_tid = 0; | |
4611 | int ret; | |
4612 | ||
4613 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4614 | /* | |
4615 | * All buffers in the last page remain valid? Then there's nothing to | |
4616 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4617 | * blocksize case | |
4618 | */ | |
4619 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4620 | return; | |
4621 | while (1) { | |
4622 | page = find_lock_page(inode->i_mapping, | |
4623 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4624 | if (!page) | |
4625 | return; | |
4626 | ret = __ext4_journalled_invalidatepage(page, offset); | |
4627 | unlock_page(page); | |
4628 | page_cache_release(page); | |
4629 | if (ret != -EBUSY) | |
4630 | return; | |
4631 | commit_tid = 0; | |
4632 | read_lock(&journal->j_state_lock); | |
4633 | if (journal->j_committing_transaction) | |
4634 | commit_tid = journal->j_committing_transaction->t_tid; | |
4635 | read_unlock(&journal->j_state_lock); | |
4636 | if (commit_tid) | |
4637 | jbd2_log_wait_commit(journal, commit_tid); | |
4638 | } | |
4639 | } | |
4640 | ||
ac27a0ec | 4641 | /* |
617ba13b | 4642 | * ext4_setattr() |
ac27a0ec DK |
4643 | * |
4644 | * Called from notify_change. | |
4645 | * | |
4646 | * We want to trap VFS attempts to truncate the file as soon as | |
4647 | * possible. In particular, we want to make sure that when the VFS | |
4648 | * shrinks i_size, we put the inode on the orphan list and modify | |
4649 | * i_disksize immediately, so that during the subsequent flushing of | |
4650 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4651 | * commit will leave the blocks being flushed in an unused state on | |
4652 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4653 | * be freed, so we have a strong guarantee that no future commit will | |
4654 | * leave these blocks visible to the user.) | |
4655 | * | |
678aaf48 JK |
4656 | * Another thing we have to assure is that if we are in ordered mode |
4657 | * and inode is still attached to the committing transaction, we must | |
4658 | * we start writeout of all the dirty pages which are being truncated. | |
4659 | * This way we are sure that all the data written in the previous | |
4660 | * transaction are already on disk (truncate waits for pages under | |
4661 | * writeback). | |
4662 | * | |
4663 | * Called with inode->i_mutex down. | |
ac27a0ec | 4664 | */ |
617ba13b | 4665 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4666 | { |
4667 | struct inode *inode = dentry->d_inode; | |
4668 | int error, rc = 0; | |
3d287de3 | 4669 | int orphan = 0; |
ac27a0ec DK |
4670 | const unsigned int ia_valid = attr->ia_valid; |
4671 | ||
4672 | error = inode_change_ok(inode, attr); | |
4673 | if (error) | |
4674 | return error; | |
4675 | ||
12755627 | 4676 | if (is_quota_modification(inode, attr)) |
871a2931 | 4677 | dquot_initialize(inode); |
08cefc7a EB |
4678 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4679 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4680 | handle_t *handle; |
4681 | ||
4682 | /* (user+group)*(old+new) structure, inode write (sb, | |
4683 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4684 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4685 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4686 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4687 | if (IS_ERR(handle)) { |
4688 | error = PTR_ERR(handle); | |
4689 | goto err_out; | |
4690 | } | |
b43fa828 | 4691 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4692 | if (error) { |
617ba13b | 4693 | ext4_journal_stop(handle); |
ac27a0ec DK |
4694 | return error; |
4695 | } | |
4696 | /* Update corresponding info in inode so that everything is in | |
4697 | * one transaction */ | |
4698 | if (attr->ia_valid & ATTR_UID) | |
4699 | inode->i_uid = attr->ia_uid; | |
4700 | if (attr->ia_valid & ATTR_GID) | |
4701 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4702 | error = ext4_mark_inode_dirty(handle, inode); |
4703 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4704 | } |
4705 | ||
e2b46574 | 4706 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa | 4707 | |
12e9b892 | 4708 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4709 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4710 | ||
0c095c7f TT |
4711 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4712 | return -EFBIG; | |
e2b46574 ES |
4713 | } |
4714 | } | |
4715 | ||
ac27a0ec | 4716 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4717 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4718 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4719 | handle_t *handle; |
4720 | ||
9924a92a | 4721 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
ac27a0ec DK |
4722 | if (IS_ERR(handle)) { |
4723 | error = PTR_ERR(handle); | |
4724 | goto err_out; | |
4725 | } | |
3d287de3 DM |
4726 | if (ext4_handle_valid(handle)) { |
4727 | error = ext4_orphan_add(handle, inode); | |
4728 | orphan = 1; | |
4729 | } | |
617ba13b MC |
4730 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4731 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4732 | if (!error) |
4733 | error = rc; | |
617ba13b | 4734 | ext4_journal_stop(handle); |
678aaf48 JK |
4735 | |
4736 | if (ext4_should_order_data(inode)) { | |
4737 | error = ext4_begin_ordered_truncate(inode, | |
4738 | attr->ia_size); | |
4739 | if (error) { | |
4740 | /* Do as much error cleanup as possible */ | |
9924a92a TT |
4741 | handle = ext4_journal_start(inode, |
4742 | EXT4_HT_INODE, 3); | |
678aaf48 JK |
4743 | if (IS_ERR(handle)) { |
4744 | ext4_orphan_del(NULL, inode); | |
4745 | goto err_out; | |
4746 | } | |
4747 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4748 | orphan = 0; |
678aaf48 JK |
4749 | ext4_journal_stop(handle); |
4750 | goto err_out; | |
4751 | } | |
4752 | } | |
ac27a0ec DK |
4753 | } |
4754 | ||
072bd7ea | 4755 | if (attr->ia_valid & ATTR_SIZE) { |
53e87268 JK |
4756 | if (attr->ia_size != inode->i_size) { |
4757 | loff_t oldsize = inode->i_size; | |
4758 | ||
4759 | i_size_write(inode, attr->ia_size); | |
4760 | /* | |
4761 | * Blocks are going to be removed from the inode. Wait | |
4762 | * for dio in flight. Temporarily disable | |
4763 | * dioread_nolock to prevent livelock. | |
4764 | */ | |
1b65007e | 4765 | if (orphan) { |
53e87268 JK |
4766 | if (!ext4_should_journal_data(inode)) { |
4767 | ext4_inode_block_unlocked_dio(inode); | |
4768 | inode_dio_wait(inode); | |
4769 | ext4_inode_resume_unlocked_dio(inode); | |
4770 | } else | |
4771 | ext4_wait_for_tail_page_commit(inode); | |
1b65007e | 4772 | } |
53e87268 JK |
4773 | /* |
4774 | * Truncate pagecache after we've waited for commit | |
4775 | * in data=journal mode to make pages freeable. | |
4776 | */ | |
4777 | truncate_pagecache(inode, oldsize, inode->i_size); | |
1c9114f9 | 4778 | } |
afcff5d8 | 4779 | ext4_truncate(inode); |
072bd7ea | 4780 | } |
ac27a0ec | 4781 | |
1025774c CH |
4782 | if (!rc) { |
4783 | setattr_copy(inode, attr); | |
4784 | mark_inode_dirty(inode); | |
4785 | } | |
4786 | ||
4787 | /* | |
4788 | * If the call to ext4_truncate failed to get a transaction handle at | |
4789 | * all, we need to clean up the in-core orphan list manually. | |
4790 | */ | |
3d287de3 | 4791 | if (orphan && inode->i_nlink) |
617ba13b | 4792 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4793 | |
4794 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4795 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4796 | |
4797 | err_out: | |
617ba13b | 4798 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4799 | if (!error) |
4800 | error = rc; | |
4801 | return error; | |
4802 | } | |
4803 | ||
3e3398a0 MC |
4804 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4805 | struct kstat *stat) | |
4806 | { | |
4807 | struct inode *inode; | |
4808 | unsigned long delalloc_blocks; | |
4809 | ||
4810 | inode = dentry->d_inode; | |
4811 | generic_fillattr(inode, stat); | |
4812 | ||
4813 | /* | |
4814 | * We can't update i_blocks if the block allocation is delayed | |
4815 | * otherwise in the case of system crash before the real block | |
4816 | * allocation is done, we will have i_blocks inconsistent with | |
4817 | * on-disk file blocks. | |
4818 | * We always keep i_blocks updated together with real | |
4819 | * allocation. But to not confuse with user, stat | |
4820 | * will return the blocks that include the delayed allocation | |
4821 | * blocks for this file. | |
4822 | */ | |
96607551 TM |
4823 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
4824 | EXT4_I(inode)->i_reserved_data_blocks); | |
3e3398a0 MC |
4825 | |
4826 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4827 | return 0; | |
4828 | } | |
ac27a0ec | 4829 | |
a02908f1 MC |
4830 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4831 | { | |
12e9b892 | 4832 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4833 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4834 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4835 | } |
ac51d837 | 4836 | |
ac27a0ec | 4837 | /* |
a02908f1 MC |
4838 | * Account for index blocks, block groups bitmaps and block group |
4839 | * descriptor blocks if modify datablocks and index blocks | |
4840 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4841 | * |
a02908f1 | 4842 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4843 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4844 | * they could still across block group boundary. |
ac27a0ec | 4845 | * |
a02908f1 MC |
4846 | * Also account for superblock, inode, quota and xattr blocks |
4847 | */ | |
1f109d5a | 4848 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4849 | { |
8df9675f TT |
4850 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4851 | int gdpblocks; | |
a02908f1 MC |
4852 | int idxblocks; |
4853 | int ret = 0; | |
4854 | ||
4855 | /* | |
4856 | * How many index blocks need to touch to modify nrblocks? | |
4857 | * The "Chunk" flag indicating whether the nrblocks is | |
4858 | * physically contiguous on disk | |
4859 | * | |
4860 | * For Direct IO and fallocate, they calls get_block to allocate | |
4861 | * one single extent at a time, so they could set the "Chunk" flag | |
4862 | */ | |
4863 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4864 | ||
4865 | ret = idxblocks; | |
4866 | ||
4867 | /* | |
4868 | * Now let's see how many group bitmaps and group descriptors need | |
4869 | * to account | |
4870 | */ | |
4871 | groups = idxblocks; | |
4872 | if (chunk) | |
4873 | groups += 1; | |
4874 | else | |
4875 | groups += nrblocks; | |
4876 | ||
4877 | gdpblocks = groups; | |
8df9675f TT |
4878 | if (groups > ngroups) |
4879 | groups = ngroups; | |
a02908f1 MC |
4880 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4881 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4882 | ||
4883 | /* bitmaps and block group descriptor blocks */ | |
4884 | ret += groups + gdpblocks; | |
4885 | ||
4886 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4887 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4888 | ||
4889 | return ret; | |
4890 | } | |
4891 | ||
4892 | /* | |
25985edc | 4893 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4894 | * the modification of a single pages into a single transaction, |
4895 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4896 | * |
525f4ed8 | 4897 | * This could be called via ext4_write_begin() |
ac27a0ec | 4898 | * |
525f4ed8 | 4899 | * We need to consider the worse case, when |
a02908f1 | 4900 | * one new block per extent. |
ac27a0ec | 4901 | */ |
a86c6181 | 4902 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4903 | { |
617ba13b | 4904 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4905 | int ret; |
4906 | ||
a02908f1 | 4907 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4908 | |
a02908f1 | 4909 | /* Account for data blocks for journalled mode */ |
617ba13b | 4910 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4911 | ret += bpp; |
ac27a0ec DK |
4912 | return ret; |
4913 | } | |
f3bd1f3f MC |
4914 | |
4915 | /* | |
4916 | * Calculate the journal credits for a chunk of data modification. | |
4917 | * | |
4918 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4919 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4920 | * |
4921 | * journal buffers for data blocks are not included here, as DIO | |
4922 | * and fallocate do no need to journal data buffers. | |
4923 | */ | |
4924 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4925 | { | |
4926 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4927 | } | |
4928 | ||
ac27a0ec | 4929 | /* |
617ba13b | 4930 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4931 | * Give this, we know that the caller already has write access to iloc->bh. |
4932 | */ | |
617ba13b | 4933 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4934 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4935 | { |
4936 | int err = 0; | |
4937 | ||
c64db50e | 4938 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4939 | inode_inc_iversion(inode); |
4940 | ||
ac27a0ec DK |
4941 | /* the do_update_inode consumes one bh->b_count */ |
4942 | get_bh(iloc->bh); | |
4943 | ||
dab291af | 4944 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4945 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4946 | put_bh(iloc->bh); |
4947 | return err; | |
4948 | } | |
4949 | ||
4950 | /* | |
4951 | * On success, We end up with an outstanding reference count against | |
4952 | * iloc->bh. This _must_ be cleaned up later. | |
4953 | */ | |
4954 | ||
4955 | int | |
617ba13b MC |
4956 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4957 | struct ext4_iloc *iloc) | |
ac27a0ec | 4958 | { |
0390131b FM |
4959 | int err; |
4960 | ||
4961 | err = ext4_get_inode_loc(inode, iloc); | |
4962 | if (!err) { | |
4963 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4964 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4965 | if (err) { | |
4966 | brelse(iloc->bh); | |
4967 | iloc->bh = NULL; | |
ac27a0ec DK |
4968 | } |
4969 | } | |
617ba13b | 4970 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4971 | return err; |
4972 | } | |
4973 | ||
6dd4ee7c KS |
4974 | /* |
4975 | * Expand an inode by new_extra_isize bytes. | |
4976 | * Returns 0 on success or negative error number on failure. | |
4977 | */ | |
1d03ec98 AK |
4978 | static int ext4_expand_extra_isize(struct inode *inode, |
4979 | unsigned int new_extra_isize, | |
4980 | struct ext4_iloc iloc, | |
4981 | handle_t *handle) | |
6dd4ee7c KS |
4982 | { |
4983 | struct ext4_inode *raw_inode; | |
4984 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4985 | |
4986 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4987 | return 0; | |
4988 | ||
4989 | raw_inode = ext4_raw_inode(&iloc); | |
4990 | ||
4991 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4992 | |
4993 | /* No extended attributes present */ | |
19f5fb7a TT |
4994 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4995 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4996 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4997 | new_extra_isize); | |
4998 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4999 | return 0; | |
5000 | } | |
5001 | ||
5002 | /* try to expand with EAs present */ | |
5003 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5004 | raw_inode, handle); | |
5005 | } | |
5006 | ||
ac27a0ec DK |
5007 | /* |
5008 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5009 | * dirtiness (it may still be data-dirty). | |
5010 | * This means that the in-core inode may be reaped by prune_icache | |
5011 | * without having to perform any I/O. This is a very good thing, | |
5012 | * because *any* task may call prune_icache - even ones which | |
5013 | * have a transaction open against a different journal. | |
5014 | * | |
5015 | * Is this cheating? Not really. Sure, we haven't written the | |
5016 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5017 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5018 | * we start and wait on commits. | |
ac27a0ec | 5019 | */ |
617ba13b | 5020 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5021 | { |
617ba13b | 5022 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5023 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5024 | static unsigned int mnt_count; | |
5025 | int err, ret; | |
ac27a0ec DK |
5026 | |
5027 | might_sleep(); | |
7ff9c073 | 5028 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5029 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5030 | if (ext4_handle_valid(handle) && |
5031 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5032 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5033 | /* |
5034 | * We need extra buffer credits since we may write into EA block | |
5035 | * with this same handle. If journal_extend fails, then it will | |
5036 | * only result in a minor loss of functionality for that inode. | |
5037 | * If this is felt to be critical, then e2fsck should be run to | |
5038 | * force a large enough s_min_extra_isize. | |
5039 | */ | |
5040 | if ((jbd2_journal_extend(handle, | |
5041 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5042 | ret = ext4_expand_extra_isize(inode, | |
5043 | sbi->s_want_extra_isize, | |
5044 | iloc, handle); | |
5045 | if (ret) { | |
19f5fb7a TT |
5046 | ext4_set_inode_state(inode, |
5047 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5048 | if (mnt_count != |
5049 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5050 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5051 | "Unable to expand inode %lu. Delete" |
5052 | " some EAs or run e2fsck.", | |
5053 | inode->i_ino); | |
c1bddad9 AK |
5054 | mnt_count = |
5055 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5056 | } |
5057 | } | |
5058 | } | |
5059 | } | |
ac27a0ec | 5060 | if (!err) |
617ba13b | 5061 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5062 | return err; |
5063 | } | |
5064 | ||
5065 | /* | |
617ba13b | 5066 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5067 | * |
5068 | * We're really interested in the case where a file is being extended. | |
5069 | * i_size has been changed by generic_commit_write() and we thus need | |
5070 | * to include the updated inode in the current transaction. | |
5071 | * | |
5dd4056d | 5072 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5073 | * are allocated to the file. |
5074 | * | |
5075 | * If the inode is marked synchronous, we don't honour that here - doing | |
5076 | * so would cause a commit on atime updates, which we don't bother doing. | |
5077 | * We handle synchronous inodes at the highest possible level. | |
5078 | */ | |
aa385729 | 5079 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5080 | { |
ac27a0ec DK |
5081 | handle_t *handle; |
5082 | ||
9924a92a | 5083 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5084 | if (IS_ERR(handle)) |
5085 | goto out; | |
f3dc272f | 5086 | |
f3dc272f CW |
5087 | ext4_mark_inode_dirty(handle, inode); |
5088 | ||
617ba13b | 5089 | ext4_journal_stop(handle); |
ac27a0ec DK |
5090 | out: |
5091 | return; | |
5092 | } | |
5093 | ||
5094 | #if 0 | |
5095 | /* | |
5096 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5097 | * it from being flushed to disk early. Unlike | |
617ba13b | 5098 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5099 | * returns no iloc structure, so the caller needs to repeat the iloc |
5100 | * lookup to mark the inode dirty later. | |
5101 | */ | |
617ba13b | 5102 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5103 | { |
617ba13b | 5104 | struct ext4_iloc iloc; |
ac27a0ec DK |
5105 | |
5106 | int err = 0; | |
5107 | if (handle) { | |
617ba13b | 5108 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5109 | if (!err) { |
5110 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5111 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5112 | if (!err) |
0390131b | 5113 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5114 | NULL, |
0390131b | 5115 | iloc.bh); |
ac27a0ec DK |
5116 | brelse(iloc.bh); |
5117 | } | |
5118 | } | |
617ba13b | 5119 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5120 | return err; |
5121 | } | |
5122 | #endif | |
5123 | ||
617ba13b | 5124 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5125 | { |
5126 | journal_t *journal; | |
5127 | handle_t *handle; | |
5128 | int err; | |
5129 | ||
5130 | /* | |
5131 | * We have to be very careful here: changing a data block's | |
5132 | * journaling status dynamically is dangerous. If we write a | |
5133 | * data block to the journal, change the status and then delete | |
5134 | * that block, we risk forgetting to revoke the old log record | |
5135 | * from the journal and so a subsequent replay can corrupt data. | |
5136 | * So, first we make sure that the journal is empty and that | |
5137 | * nobody is changing anything. | |
5138 | */ | |
5139 | ||
617ba13b | 5140 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5141 | if (!journal) |
5142 | return 0; | |
d699594d | 5143 | if (is_journal_aborted(journal)) |
ac27a0ec | 5144 | return -EROFS; |
2aff57b0 YY |
5145 | /* We have to allocate physical blocks for delalloc blocks |
5146 | * before flushing journal. otherwise delalloc blocks can not | |
5147 | * be allocated any more. even more truncate on delalloc blocks | |
5148 | * could trigger BUG by flushing delalloc blocks in journal. | |
5149 | * There is no delalloc block in non-journal data mode. | |
5150 | */ | |
5151 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5152 | err = ext4_alloc_da_blocks(inode); | |
5153 | if (err < 0) | |
5154 | return err; | |
5155 | } | |
ac27a0ec | 5156 | |
17335dcc DM |
5157 | /* Wait for all existing dio workers */ |
5158 | ext4_inode_block_unlocked_dio(inode); | |
5159 | inode_dio_wait(inode); | |
5160 | ||
dab291af | 5161 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5162 | |
5163 | /* | |
5164 | * OK, there are no updates running now, and all cached data is | |
5165 | * synced to disk. We are now in a completely consistent state | |
5166 | * which doesn't have anything in the journal, and we know that | |
5167 | * no filesystem updates are running, so it is safe to modify | |
5168 | * the inode's in-core data-journaling state flag now. | |
5169 | */ | |
5170 | ||
5171 | if (val) | |
12e9b892 | 5172 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
5173 | else { |
5174 | jbd2_journal_flush(journal); | |
12e9b892 | 5175 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5176 | } |
617ba13b | 5177 | ext4_set_aops(inode); |
ac27a0ec | 5178 | |
dab291af | 5179 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5180 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5181 | |
5182 | /* Finally we can mark the inode as dirty. */ | |
5183 | ||
9924a92a | 5184 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5185 | if (IS_ERR(handle)) |
5186 | return PTR_ERR(handle); | |
5187 | ||
617ba13b | 5188 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5189 | ext4_handle_sync(handle); |
617ba13b MC |
5190 | ext4_journal_stop(handle); |
5191 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5192 | |
5193 | return err; | |
5194 | } | |
2e9ee850 AK |
5195 | |
5196 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5197 | { | |
5198 | return !buffer_mapped(bh); | |
5199 | } | |
5200 | ||
c2ec175c | 5201 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5202 | { |
c2ec175c | 5203 | struct page *page = vmf->page; |
2e9ee850 AK |
5204 | loff_t size; |
5205 | unsigned long len; | |
9ea7df53 | 5206 | int ret; |
2e9ee850 | 5207 | struct file *file = vma->vm_file; |
496ad9aa | 5208 | struct inode *inode = file_inode(file); |
2e9ee850 | 5209 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5210 | handle_t *handle; |
5211 | get_block_t *get_block; | |
5212 | int retries = 0; | |
2e9ee850 | 5213 | |
8e8ad8a5 | 5214 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5215 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5216 | /* Delalloc case is easy... */ |
5217 | if (test_opt(inode->i_sb, DELALLOC) && | |
5218 | !ext4_should_journal_data(inode) && | |
5219 | !ext4_nonda_switch(inode->i_sb)) { | |
5220 | do { | |
5221 | ret = __block_page_mkwrite(vma, vmf, | |
5222 | ext4_da_get_block_prep); | |
5223 | } while (ret == -ENOSPC && | |
5224 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5225 | goto out_ret; | |
2e9ee850 | 5226 | } |
0e499890 DW |
5227 | |
5228 | lock_page(page); | |
9ea7df53 JK |
5229 | size = i_size_read(inode); |
5230 | /* Page got truncated from under us? */ | |
5231 | if (page->mapping != mapping || page_offset(page) > size) { | |
5232 | unlock_page(page); | |
5233 | ret = VM_FAULT_NOPAGE; | |
5234 | goto out; | |
0e499890 | 5235 | } |
2e9ee850 AK |
5236 | |
5237 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5238 | len = size & ~PAGE_CACHE_MASK; | |
5239 | else | |
5240 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5241 | /* |
9ea7df53 JK |
5242 | * Return if we have all the buffers mapped. This avoids the need to do |
5243 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5244 | */ |
2e9ee850 | 5245 | if (page_has_buffers(page)) { |
f19d5870 TM |
5246 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5247 | 0, len, NULL, | |
5248 | ext4_bh_unmapped)) { | |
9ea7df53 | 5249 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5250 | wait_for_stable_page(page); |
9ea7df53 JK |
5251 | ret = VM_FAULT_LOCKED; |
5252 | goto out; | |
a827eaff | 5253 | } |
2e9ee850 | 5254 | } |
a827eaff | 5255 | unlock_page(page); |
9ea7df53 JK |
5256 | /* OK, we need to fill the hole... */ |
5257 | if (ext4_should_dioread_nolock(inode)) | |
5258 | get_block = ext4_get_block_write; | |
5259 | else | |
5260 | get_block = ext4_get_block; | |
5261 | retry_alloc: | |
9924a92a TT |
5262 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5263 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5264 | if (IS_ERR(handle)) { |
c2ec175c | 5265 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5266 | goto out; |
5267 | } | |
5268 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5269 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5270 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5271 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5272 | unlock_page(page); | |
5273 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5274 | ext4_journal_stop(handle); |
9ea7df53 JK |
5275 | goto out; |
5276 | } | |
5277 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5278 | } | |
5279 | ext4_journal_stop(handle); | |
5280 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5281 | goto retry_alloc; | |
5282 | out_ret: | |
5283 | ret = block_page_mkwrite_return(ret); | |
5284 | out: | |
8e8ad8a5 | 5285 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5286 | return ret; |
5287 | } |