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> |
9bffad1e | 40 | |
3dcf5451 | 41 | #include "ext4_jbd2.h" |
ac27a0ec DK |
42 | #include "xattr.h" |
43 | #include "acl.h" | |
9f125d64 | 44 | #include "truncate.h" |
ac27a0ec | 45 | |
9bffad1e TT |
46 | #include <trace/events/ext4.h> |
47 | ||
a1d6cc56 AK |
48 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
49 | ||
814525f4 DW |
50 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
51 | struct ext4_inode_info *ei) | |
52 | { | |
53 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
54 | __u16 csum_lo; | |
55 | __u16 csum_hi = 0; | |
56 | __u32 csum; | |
57 | ||
58 | csum_lo = raw->i_checksum_lo; | |
59 | raw->i_checksum_lo = 0; | |
60 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
61 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
62 | csum_hi = raw->i_checksum_hi; | |
63 | raw->i_checksum_hi = 0; | |
64 | } | |
65 | ||
66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
67 | EXT4_INODE_SIZE(inode->i_sb)); | |
68 | ||
69 | raw->i_checksum_lo = csum_lo; | |
70 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
71 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
72 | raw->i_checksum_hi = csum_hi; | |
73 | ||
74 | return csum; | |
75 | } | |
76 | ||
77 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
78 | struct ext4_inode_info *ei) | |
79 | { | |
80 | __u32 provided, calculated; | |
81 | ||
82 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
83 | cpu_to_le32(EXT4_OS_LINUX) || | |
84 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
85 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
86 | return 1; | |
87 | ||
88 | provided = le16_to_cpu(raw->i_checksum_lo); | |
89 | calculated = ext4_inode_csum(inode, raw, ei); | |
90 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
91 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
92 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
93 | else | |
94 | calculated &= 0xFFFF; | |
95 | ||
96 | return provided == calculated; | |
97 | } | |
98 | ||
99 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
100 | struct ext4_inode_info *ei) | |
101 | { | |
102 | __u32 csum; | |
103 | ||
104 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
105 | cpu_to_le32(EXT4_OS_LINUX) || | |
106 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
107 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
108 | return; | |
109 | ||
110 | csum = ext4_inode_csum(inode, raw, ei); | |
111 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
112 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
113 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
114 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
115 | } | |
116 | ||
678aaf48 JK |
117 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
118 | loff_t new_size) | |
119 | { | |
7ff9c073 | 120 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
121 | /* |
122 | * If jinode is zero, then we never opened the file for | |
123 | * writing, so there's no need to call | |
124 | * jbd2_journal_begin_ordered_truncate() since there's no | |
125 | * outstanding writes we need to flush. | |
126 | */ | |
127 | if (!EXT4_I(inode)->jinode) | |
128 | return 0; | |
129 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
130 | EXT4_I(inode)->jinode, | |
131 | new_size); | |
678aaf48 JK |
132 | } |
133 | ||
64769240 | 134 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
135 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
136 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
5f163cc7 ES |
137 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
138 | struct inode *inode, struct page *page, loff_t from, | |
139 | loff_t length, int flags); | |
64769240 | 140 | |
ac27a0ec DK |
141 | /* |
142 | * Test whether an inode is a fast symlink. | |
143 | */ | |
617ba13b | 144 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 145 | { |
617ba13b | 146 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
147 | (inode->i_sb->s_blocksize >> 9) : 0; |
148 | ||
149 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
150 | } | |
151 | ||
ac27a0ec DK |
152 | /* |
153 | * Restart the transaction associated with *handle. This does a commit, | |
154 | * so before we call here everything must be consistently dirtied against | |
155 | * this transaction. | |
156 | */ | |
fa5d1113 | 157 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 158 | int nblocks) |
ac27a0ec | 159 | { |
487caeef JK |
160 | int ret; |
161 | ||
162 | /* | |
e35fd660 | 163 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
164 | * moment, get_block can be called only for blocks inside i_size since |
165 | * page cache has been already dropped and writes are blocked by | |
166 | * i_mutex. So we can safely drop the i_data_sem here. | |
167 | */ | |
0390131b | 168 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 169 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 170 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 171 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 172 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 173 | ext4_discard_preallocations(inode); |
487caeef JK |
174 | |
175 | return ret; | |
ac27a0ec DK |
176 | } |
177 | ||
178 | /* | |
179 | * Called at the last iput() if i_nlink is zero. | |
180 | */ | |
0930fcc1 | 181 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
182 | { |
183 | handle_t *handle; | |
bc965ab3 | 184 | int err; |
ac27a0ec | 185 | |
7ff9c073 | 186 | trace_ext4_evict_inode(inode); |
2581fdc8 | 187 | |
0930fcc1 | 188 | if (inode->i_nlink) { |
2d859db3 JK |
189 | /* |
190 | * When journalling data dirty buffers are tracked only in the | |
191 | * journal. So although mm thinks everything is clean and | |
192 | * ready for reaping the inode might still have some pages to | |
193 | * write in the running transaction or waiting to be | |
194 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
195 | * (via truncate_inode_pages()) to discard these buffers can | |
196 | * cause data loss. Also even if we did not discard these | |
197 | * buffers, we would have no way to find them after the inode | |
198 | * is reaped and thus user could see stale data if he tries to | |
199 | * read them before the transaction is checkpointed. So be | |
200 | * careful and force everything to disk here... We use | |
201 | * ei->i_datasync_tid to store the newest transaction | |
202 | * containing inode's data. | |
203 | * | |
204 | * Note that directories do not have this problem because they | |
205 | * don't use page cache. | |
206 | */ | |
207 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
208 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
209 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
210 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
211 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
212 | ||
d76a3a77 | 213 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
214 | filemap_write_and_wait(&inode->i_data); |
215 | } | |
0930fcc1 | 216 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 217 | ext4_ioend_shutdown(inode); |
0930fcc1 AV |
218 | goto no_delete; |
219 | } | |
220 | ||
907f4554 | 221 | if (!is_bad_inode(inode)) |
871a2931 | 222 | dquot_initialize(inode); |
907f4554 | 223 | |
678aaf48 JK |
224 | if (ext4_should_order_data(inode)) |
225 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec | 226 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 227 | ext4_ioend_shutdown(inode); |
ac27a0ec DK |
228 | |
229 | if (is_bad_inode(inode)) | |
230 | goto no_delete; | |
231 | ||
8e8ad8a5 JK |
232 | /* |
233 | * Protect us against freezing - iput() caller didn't have to have any | |
234 | * protection against it | |
235 | */ | |
236 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
237 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
238 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 239 | if (IS_ERR(handle)) { |
bc965ab3 | 240 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
241 | /* |
242 | * If we're going to skip the normal cleanup, we still need to | |
243 | * make sure that the in-core orphan linked list is properly | |
244 | * cleaned up. | |
245 | */ | |
617ba13b | 246 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 247 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
248 | goto no_delete; |
249 | } | |
250 | ||
251 | if (IS_SYNC(inode)) | |
0390131b | 252 | ext4_handle_sync(handle); |
ac27a0ec | 253 | inode->i_size = 0; |
bc965ab3 TT |
254 | err = ext4_mark_inode_dirty(handle, inode); |
255 | if (err) { | |
12062ddd | 256 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
257 | "couldn't mark inode dirty (err %d)", err); |
258 | goto stop_handle; | |
259 | } | |
ac27a0ec | 260 | if (inode->i_blocks) |
617ba13b | 261 | ext4_truncate(inode); |
bc965ab3 TT |
262 | |
263 | /* | |
264 | * ext4_ext_truncate() doesn't reserve any slop when it | |
265 | * restarts journal transactions; therefore there may not be | |
266 | * enough credits left in the handle to remove the inode from | |
267 | * the orphan list and set the dtime field. | |
268 | */ | |
0390131b | 269 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
270 | err = ext4_journal_extend(handle, 3); |
271 | if (err > 0) | |
272 | err = ext4_journal_restart(handle, 3); | |
273 | if (err != 0) { | |
12062ddd | 274 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
275 | "couldn't extend journal (err %d)", err); |
276 | stop_handle: | |
277 | ext4_journal_stop(handle); | |
45388219 | 278 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 279 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
280 | goto no_delete; |
281 | } | |
282 | } | |
283 | ||
ac27a0ec | 284 | /* |
617ba13b | 285 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 286 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 287 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 288 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 289 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
290 | * (Well, we could do this if we need to, but heck - it works) |
291 | */ | |
617ba13b MC |
292 | ext4_orphan_del(handle, inode); |
293 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
294 | |
295 | /* | |
296 | * One subtle ordering requirement: if anything has gone wrong | |
297 | * (transaction abort, IO errors, whatever), then we can still | |
298 | * do these next steps (the fs will already have been marked as | |
299 | * having errors), but we can't free the inode if the mark_dirty | |
300 | * fails. | |
301 | */ | |
617ba13b | 302 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 303 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 304 | ext4_clear_inode(inode); |
ac27a0ec | 305 | else |
617ba13b MC |
306 | ext4_free_inode(handle, inode); |
307 | ext4_journal_stop(handle); | |
8e8ad8a5 | 308 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
309 | return; |
310 | no_delete: | |
0930fcc1 | 311 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
312 | } |
313 | ||
a9e7f447 DM |
314 | #ifdef CONFIG_QUOTA |
315 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 316 | { |
a9e7f447 | 317 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 318 | } |
a9e7f447 | 319 | #endif |
9d0be502 | 320 | |
12219aea AK |
321 | /* |
322 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 323 | * to allocate a block located at @lblock |
12219aea | 324 | */ |
01f49d0b | 325 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 326 | { |
12e9b892 | 327 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 328 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 329 | |
8bb2b247 | 330 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
331 | } |
332 | ||
0637c6f4 TT |
333 | /* |
334 | * Called with i_data_sem down, which is important since we can call | |
335 | * ext4_discard_preallocations() from here. | |
336 | */ | |
5f634d06 AK |
337 | void ext4_da_update_reserve_space(struct inode *inode, |
338 | int used, int quota_claim) | |
12219aea AK |
339 | { |
340 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 341 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
342 | |
343 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 344 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 345 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 346 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 347 | "with only %d reserved data blocks", |
0637c6f4 TT |
348 | __func__, inode->i_ino, used, |
349 | ei->i_reserved_data_blocks); | |
350 | WARN_ON(1); | |
351 | used = ei->i_reserved_data_blocks; | |
352 | } | |
12219aea | 353 | |
97795d2a | 354 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
01a523eb TT |
355 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
356 | "with only %d reserved metadata blocks " | |
357 | "(releasing %d blocks with reserved %d data blocks)", | |
358 | inode->i_ino, ei->i_allocated_meta_blocks, | |
359 | ei->i_reserved_meta_blocks, used, | |
360 | ei->i_reserved_data_blocks); | |
97795d2a BF |
361 | WARN_ON(1); |
362 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; | |
363 | } | |
364 | ||
0637c6f4 TT |
365 | /* Update per-inode reservations */ |
366 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 367 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 368 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 369 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 370 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 371 | |
0637c6f4 TT |
372 | if (ei->i_reserved_data_blocks == 0) { |
373 | /* | |
374 | * We can release all of the reserved metadata blocks | |
375 | * only when we have written all of the delayed | |
376 | * allocation blocks. | |
377 | */ | |
57042651 | 378 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 379 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 380 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 381 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 382 | } |
12219aea | 383 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 384 | |
72b8ab9d ES |
385 | /* Update quota subsystem for data blocks */ |
386 | if (quota_claim) | |
7b415bf6 | 387 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 388 | else { |
5f634d06 AK |
389 | /* |
390 | * We did fallocate with an offset that is already delayed | |
391 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 392 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 393 | */ |
7b415bf6 | 394 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 395 | } |
d6014301 AK |
396 | |
397 | /* | |
398 | * If we have done all the pending block allocations and if | |
399 | * there aren't any writers on the inode, we can discard the | |
400 | * inode's preallocations. | |
401 | */ | |
0637c6f4 TT |
402 | if ((ei->i_reserved_data_blocks == 0) && |
403 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 404 | ext4_discard_preallocations(inode); |
12219aea AK |
405 | } |
406 | ||
e29136f8 | 407 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
408 | unsigned int line, |
409 | struct ext4_map_blocks *map) | |
6fd058f7 | 410 | { |
24676da4 TT |
411 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
412 | map->m_len)) { | |
c398eda0 TT |
413 | ext4_error_inode(inode, func, line, map->m_pblk, |
414 | "lblock %lu mapped to illegal pblock " | |
415 | "(length %d)", (unsigned long) map->m_lblk, | |
416 | map->m_len); | |
6fd058f7 TT |
417 | return -EIO; |
418 | } | |
419 | return 0; | |
420 | } | |
421 | ||
e29136f8 | 422 | #define check_block_validity(inode, map) \ |
c398eda0 | 423 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 424 | |
55138e0b | 425 | /* |
1f94533d TT |
426 | * Return the number of contiguous dirty pages in a given inode |
427 | * starting at page frame idx. | |
55138e0b TT |
428 | */ |
429 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
430 | unsigned int max_pages) | |
431 | { | |
432 | struct address_space *mapping = inode->i_mapping; | |
433 | pgoff_t index; | |
434 | struct pagevec pvec; | |
435 | pgoff_t num = 0; | |
436 | int i, nr_pages, done = 0; | |
437 | ||
438 | if (max_pages == 0) | |
439 | return 0; | |
440 | pagevec_init(&pvec, 0); | |
441 | while (!done) { | |
442 | index = idx; | |
443 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
444 | PAGECACHE_TAG_DIRTY, | |
445 | (pgoff_t)PAGEVEC_SIZE); | |
446 | if (nr_pages == 0) | |
447 | break; | |
448 | for (i = 0; i < nr_pages; i++) { | |
449 | struct page *page = pvec.pages[i]; | |
450 | struct buffer_head *bh, *head; | |
451 | ||
452 | lock_page(page); | |
453 | if (unlikely(page->mapping != mapping) || | |
454 | !PageDirty(page) || | |
455 | PageWriteback(page) || | |
456 | page->index != idx) { | |
457 | done = 1; | |
458 | unlock_page(page); | |
459 | break; | |
460 | } | |
1f94533d TT |
461 | if (page_has_buffers(page)) { |
462 | bh = head = page_buffers(page); | |
463 | do { | |
464 | if (!buffer_delay(bh) && | |
465 | !buffer_unwritten(bh)) | |
466 | done = 1; | |
467 | bh = bh->b_this_page; | |
468 | } while (!done && (bh != head)); | |
469 | } | |
55138e0b TT |
470 | unlock_page(page); |
471 | if (done) | |
472 | break; | |
473 | idx++; | |
474 | num++; | |
659c6009 ES |
475 | if (num >= max_pages) { |
476 | done = 1; | |
55138e0b | 477 | break; |
659c6009 | 478 | } |
55138e0b TT |
479 | } |
480 | pagevec_release(&pvec); | |
481 | } | |
482 | return num; | |
483 | } | |
484 | ||
921f266b DM |
485 | #ifdef ES_AGGRESSIVE_TEST |
486 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
487 | struct inode *inode, | |
488 | struct ext4_map_blocks *es_map, | |
489 | struct ext4_map_blocks *map, | |
490 | int flags) | |
491 | { | |
492 | int retval; | |
493 | ||
494 | map->m_flags = 0; | |
495 | /* | |
496 | * There is a race window that the result is not the same. | |
497 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
498 | * is that we lookup a block mapping in extent status tree with | |
499 | * out taking i_data_sem. So at the time the unwritten extent | |
500 | * could be converted. | |
501 | */ | |
502 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
503 | down_read((&EXT4_I(inode)->i_data_sem)); | |
504 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
505 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
506 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
507 | } else { | |
508 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
509 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
510 | } | |
511 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
512 | up_read((&EXT4_I(inode)->i_data_sem)); | |
513 | /* | |
514 | * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag | |
515 | * because it shouldn't be marked in es_map->m_flags. | |
516 | */ | |
517 | map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); | |
518 | ||
519 | /* | |
520 | * We don't check m_len because extent will be collpased in status | |
521 | * tree. So the m_len might not equal. | |
522 | */ | |
523 | if (es_map->m_lblk != map->m_lblk || | |
524 | es_map->m_flags != map->m_flags || | |
525 | es_map->m_pblk != map->m_pblk) { | |
526 | printk("ES cache assertation failed for inode: %lu " | |
527 | "es_cached ex [%d/%d/%llu/%x] != " | |
528 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
529 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
530 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
531 | map->m_len, map->m_pblk, map->m_flags, | |
532 | retval, flags); | |
533 | } | |
534 | } | |
535 | #endif /* ES_AGGRESSIVE_TEST */ | |
536 | ||
f5ab0d1f | 537 | /* |
e35fd660 | 538 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 539 | * and returns if the blocks are already mapped. |
f5ab0d1f | 540 | * |
f5ab0d1f MC |
541 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
542 | * and store the allocated blocks in the result buffer head and mark it | |
543 | * mapped. | |
544 | * | |
e35fd660 TT |
545 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
546 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
547 | * based files |
548 | * | |
549 | * On success, it returns the number of blocks being mapped or allocate. | |
550 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
551 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
552 | * the buffer head is mapped. | |
553 | * | |
554 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 555 | * that case, buffer head is unmapped |
f5ab0d1f MC |
556 | * |
557 | * It returns the error in case of allocation failure. | |
558 | */ | |
e35fd660 TT |
559 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
560 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 561 | { |
d100eef2 | 562 | struct extent_status es; |
0e855ac8 | 563 | int retval; |
921f266b DM |
564 | #ifdef ES_AGGRESSIVE_TEST |
565 | struct ext4_map_blocks orig_map; | |
566 | ||
567 | memcpy(&orig_map, map, sizeof(*map)); | |
568 | #endif | |
f5ab0d1f | 569 | |
e35fd660 TT |
570 | map->m_flags = 0; |
571 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
572 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
573 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
574 | |
575 | /* Lookup extent status tree firstly */ | |
576 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
577 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
578 | map->m_pblk = ext4_es_pblock(&es) + | |
579 | map->m_lblk - es.es_lblk; | |
580 | map->m_flags |= ext4_es_is_written(&es) ? | |
581 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
582 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
583 | if (retval > map->m_len) | |
584 | retval = map->m_len; | |
585 | map->m_len = retval; | |
586 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
587 | retval = 0; | |
588 | } else { | |
589 | BUG_ON(1); | |
590 | } | |
921f266b DM |
591 | #ifdef ES_AGGRESSIVE_TEST |
592 | ext4_map_blocks_es_recheck(handle, inode, map, | |
593 | &orig_map, flags); | |
594 | #endif | |
d100eef2 ZL |
595 | goto found; |
596 | } | |
597 | ||
4df3d265 | 598 | /* |
b920c755 TT |
599 | * Try to see if we can get the block without requesting a new |
600 | * file system block. | |
4df3d265 | 601 | */ |
729f52c6 ZL |
602 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
603 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 604 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
605 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
606 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 607 | } else { |
a4e5d88b DM |
608 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
609 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 610 | } |
f7fec032 ZL |
611 | if (retval > 0) { |
612 | int ret; | |
613 | unsigned long long status; | |
614 | ||
921f266b DM |
615 | #ifdef ES_AGGRESSIVE_TEST |
616 | if (retval != map->m_len) { | |
617 | printk("ES len assertation failed for inode: %lu " | |
618 | "retval %d != map->m_len %d " | |
619 | "in %s (lookup)\n", inode->i_ino, retval, | |
620 | map->m_len, __func__); | |
621 | } | |
622 | #endif | |
623 | ||
f7fec032 ZL |
624 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
625 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
626 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
627 | ext4_find_delalloc_range(inode, map->m_lblk, | |
628 | map->m_lblk + map->m_len - 1)) | |
629 | status |= EXTENT_STATUS_DELAYED; | |
630 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
631 | map->m_len, map->m_pblk, status); | |
632 | if (ret < 0) | |
633 | retval = ret; | |
634 | } | |
729f52c6 ZL |
635 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
636 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 637 | |
d100eef2 | 638 | found: |
e35fd660 | 639 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
f7fec032 | 640 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
641 | if (ret != 0) |
642 | return ret; | |
643 | } | |
644 | ||
f5ab0d1f | 645 | /* If it is only a block(s) look up */ |
c2177057 | 646 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
647 | return retval; |
648 | ||
649 | /* | |
650 | * Returns if the blocks have already allocated | |
651 | * | |
652 | * Note that if blocks have been preallocated | |
df3ab170 | 653 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
654 | * with buffer head unmapped. |
655 | */ | |
e35fd660 | 656 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
657 | return retval; |
658 | ||
2a8964d6 | 659 | /* |
a25a4e1a ZL |
660 | * Here we clear m_flags because after allocating an new extent, |
661 | * it will be set again. | |
2a8964d6 | 662 | */ |
a25a4e1a | 663 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 664 | |
4df3d265 | 665 | /* |
f5ab0d1f MC |
666 | * New blocks allocate and/or writing to uninitialized extent |
667 | * will possibly result in updating i_data, so we take | |
668 | * the write lock of i_data_sem, and call get_blocks() | |
669 | * with create == 1 flag. | |
4df3d265 AK |
670 | */ |
671 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
672 | |
673 | /* | |
674 | * if the caller is from delayed allocation writeout path | |
675 | * we have already reserved fs blocks for allocation | |
676 | * let the underlying get_block() function know to | |
677 | * avoid double accounting | |
678 | */ | |
c2177057 | 679 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 680 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
681 | /* |
682 | * We need to check for EXT4 here because migrate | |
683 | * could have changed the inode type in between | |
684 | */ | |
12e9b892 | 685 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 686 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 687 | } else { |
e35fd660 | 688 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 689 | |
e35fd660 | 690 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
691 | /* |
692 | * We allocated new blocks which will result in | |
693 | * i_data's format changing. Force the migrate | |
694 | * to fail by clearing migrate flags | |
695 | */ | |
19f5fb7a | 696 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 697 | } |
d2a17637 | 698 | |
5f634d06 AK |
699 | /* |
700 | * Update reserved blocks/metadata blocks after successful | |
701 | * block allocation which had been deferred till now. We don't | |
702 | * support fallocate for non extent files. So we can update | |
703 | * reserve space here. | |
704 | */ | |
705 | if ((retval > 0) && | |
1296cc85 | 706 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
707 | ext4_da_update_reserve_space(inode, retval, 1); |
708 | } | |
f7fec032 | 709 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 710 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 711 | |
f7fec032 ZL |
712 | if (retval > 0) { |
713 | int ret; | |
714 | unsigned long long status; | |
715 | ||
921f266b DM |
716 | #ifdef ES_AGGRESSIVE_TEST |
717 | if (retval != map->m_len) { | |
718 | printk("ES len assertation failed for inode: %lu " | |
719 | "retval %d != map->m_len %d " | |
720 | "in %s (allocation)\n", inode->i_ino, retval, | |
721 | map->m_len, __func__); | |
722 | } | |
723 | #endif | |
724 | ||
adb23551 ZL |
725 | /* |
726 | * If the extent has been zeroed out, we don't need to update | |
727 | * extent status tree. | |
728 | */ | |
729 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
730 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
731 | if (ext4_es_is_written(&es)) | |
732 | goto has_zeroout; | |
733 | } | |
f7fec032 ZL |
734 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
735 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
736 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
737 | ext4_find_delalloc_range(inode, map->m_lblk, | |
738 | map->m_lblk + map->m_len - 1)) | |
739 | status |= EXTENT_STATUS_DELAYED; | |
740 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
741 | map->m_pblk, status); | |
742 | if (ret < 0) | |
743 | retval = ret; | |
5356f261 AK |
744 | } |
745 | ||
adb23551 | 746 | has_zeroout: |
4df3d265 | 747 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 748 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 749 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
750 | if (ret != 0) |
751 | return ret; | |
752 | } | |
0e855ac8 AK |
753 | return retval; |
754 | } | |
755 | ||
f3bd1f3f MC |
756 | /* Maximum number of blocks we map for direct IO at once. */ |
757 | #define DIO_MAX_BLOCKS 4096 | |
758 | ||
2ed88685 TT |
759 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
760 | struct buffer_head *bh, int flags) | |
ac27a0ec | 761 | { |
3e4fdaf8 | 762 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 763 | struct ext4_map_blocks map; |
7fb5409d | 764 | int ret = 0, started = 0; |
f3bd1f3f | 765 | int dio_credits; |
ac27a0ec | 766 | |
46c7f254 TM |
767 | if (ext4_has_inline_data(inode)) |
768 | return -ERANGE; | |
769 | ||
2ed88685 TT |
770 | map.m_lblk = iblock; |
771 | map.m_len = bh->b_size >> inode->i_blkbits; | |
772 | ||
8b0f165f | 773 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 774 | /* Direct IO write... */ |
2ed88685 TT |
775 | if (map.m_len > DIO_MAX_BLOCKS) |
776 | map.m_len = DIO_MAX_BLOCKS; | |
777 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
778 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
779 | dio_credits); | |
7fb5409d | 780 | if (IS_ERR(handle)) { |
ac27a0ec | 781 | ret = PTR_ERR(handle); |
2ed88685 | 782 | return ret; |
ac27a0ec | 783 | } |
7fb5409d | 784 | started = 1; |
ac27a0ec DK |
785 | } |
786 | ||
2ed88685 | 787 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 788 | if (ret > 0) { |
2ed88685 TT |
789 | map_bh(bh, inode->i_sb, map.m_pblk); |
790 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
791 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 792 | ret = 0; |
ac27a0ec | 793 | } |
7fb5409d JK |
794 | if (started) |
795 | ext4_journal_stop(handle); | |
ac27a0ec DK |
796 | return ret; |
797 | } | |
798 | ||
2ed88685 TT |
799 | int ext4_get_block(struct inode *inode, sector_t iblock, |
800 | struct buffer_head *bh, int create) | |
801 | { | |
802 | return _ext4_get_block(inode, iblock, bh, | |
803 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
804 | } | |
805 | ||
ac27a0ec DK |
806 | /* |
807 | * `handle' can be NULL if create is zero | |
808 | */ | |
617ba13b | 809 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 810 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 811 | { |
2ed88685 TT |
812 | struct ext4_map_blocks map; |
813 | struct buffer_head *bh; | |
ac27a0ec DK |
814 | int fatal = 0, err; |
815 | ||
816 | J_ASSERT(handle != NULL || create == 0); | |
817 | ||
2ed88685 TT |
818 | map.m_lblk = block; |
819 | map.m_len = 1; | |
820 | err = ext4_map_blocks(handle, inode, &map, | |
821 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 822 | |
90b0a973 CM |
823 | /* ensure we send some value back into *errp */ |
824 | *errp = 0; | |
825 | ||
0f70b406 TT |
826 | if (create && err == 0) |
827 | err = -ENOSPC; /* should never happen */ | |
2ed88685 TT |
828 | if (err < 0) |
829 | *errp = err; | |
830 | if (err <= 0) | |
831 | return NULL; | |
2ed88685 TT |
832 | |
833 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
aebf0243 | 834 | if (unlikely(!bh)) { |
860d21e2 | 835 | *errp = -ENOMEM; |
2ed88685 | 836 | return NULL; |
ac27a0ec | 837 | } |
2ed88685 TT |
838 | if (map.m_flags & EXT4_MAP_NEW) { |
839 | J_ASSERT(create != 0); | |
840 | J_ASSERT(handle != NULL); | |
ac27a0ec | 841 | |
2ed88685 TT |
842 | /* |
843 | * Now that we do not always journal data, we should | |
844 | * keep in mind whether this should always journal the | |
845 | * new buffer as metadata. For now, regular file | |
846 | * writes use ext4_get_block instead, so it's not a | |
847 | * problem. | |
848 | */ | |
849 | lock_buffer(bh); | |
850 | BUFFER_TRACE(bh, "call get_create_access"); | |
851 | fatal = ext4_journal_get_create_access(handle, bh); | |
852 | if (!fatal && !buffer_uptodate(bh)) { | |
853 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
854 | set_buffer_uptodate(bh); | |
ac27a0ec | 855 | } |
2ed88685 TT |
856 | unlock_buffer(bh); |
857 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
858 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
859 | if (!fatal) | |
860 | fatal = err; | |
861 | } else { | |
862 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 863 | } |
2ed88685 TT |
864 | if (fatal) { |
865 | *errp = fatal; | |
866 | brelse(bh); | |
867 | bh = NULL; | |
868 | } | |
869 | return bh; | |
ac27a0ec DK |
870 | } |
871 | ||
617ba13b | 872 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 873 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 874 | { |
af5bc92d | 875 | struct buffer_head *bh; |
ac27a0ec | 876 | |
617ba13b | 877 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
878 | if (!bh) |
879 | return bh; | |
880 | if (buffer_uptodate(bh)) | |
881 | return bh; | |
65299a3b | 882 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
883 | wait_on_buffer(bh); |
884 | if (buffer_uptodate(bh)) | |
885 | return bh; | |
886 | put_bh(bh); | |
887 | *err = -EIO; | |
888 | return NULL; | |
889 | } | |
890 | ||
f19d5870 TM |
891 | int ext4_walk_page_buffers(handle_t *handle, |
892 | struct buffer_head *head, | |
893 | unsigned from, | |
894 | unsigned to, | |
895 | int *partial, | |
896 | int (*fn)(handle_t *handle, | |
897 | struct buffer_head *bh)) | |
ac27a0ec DK |
898 | { |
899 | struct buffer_head *bh; | |
900 | unsigned block_start, block_end; | |
901 | unsigned blocksize = head->b_size; | |
902 | int err, ret = 0; | |
903 | struct buffer_head *next; | |
904 | ||
af5bc92d TT |
905 | for (bh = head, block_start = 0; |
906 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 907 | block_start = block_end, bh = next) { |
ac27a0ec DK |
908 | next = bh->b_this_page; |
909 | block_end = block_start + blocksize; | |
910 | if (block_end <= from || block_start >= to) { | |
911 | if (partial && !buffer_uptodate(bh)) | |
912 | *partial = 1; | |
913 | continue; | |
914 | } | |
915 | err = (*fn)(handle, bh); | |
916 | if (!ret) | |
917 | ret = err; | |
918 | } | |
919 | return ret; | |
920 | } | |
921 | ||
922 | /* | |
923 | * To preserve ordering, it is essential that the hole instantiation and | |
924 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 925 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 926 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
927 | * prepare_write() is the right place. |
928 | * | |
36ade451 JK |
929 | * Also, this function can nest inside ext4_writepage(). In that case, we |
930 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
931 | * whole page. So we won't block on the journal in that case, which is good, | |
932 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 933 | * |
617ba13b | 934 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
935 | * quota file writes. If we were to commit the transaction while thus |
936 | * reentered, there can be a deadlock - we would be holding a quota | |
937 | * lock, and the commit would never complete if another thread had a | |
938 | * transaction open and was blocking on the quota lock - a ranking | |
939 | * violation. | |
940 | * | |
dab291af | 941 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
942 | * will _not_ run commit under these circumstances because handle->h_ref |
943 | * is elevated. We'll still have enough credits for the tiny quotafile | |
944 | * write. | |
945 | */ | |
f19d5870 TM |
946 | int do_journal_get_write_access(handle_t *handle, |
947 | struct buffer_head *bh) | |
ac27a0ec | 948 | { |
56d35a4c JK |
949 | int dirty = buffer_dirty(bh); |
950 | int ret; | |
951 | ||
ac27a0ec DK |
952 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
953 | return 0; | |
56d35a4c | 954 | /* |
ebdec241 | 955 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
956 | * the dirty bit as jbd2_journal_get_write_access() could complain |
957 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 958 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
959 | * the bit before releasing a page lock and thus writeback cannot |
960 | * ever write the buffer. | |
961 | */ | |
962 | if (dirty) | |
963 | clear_buffer_dirty(bh); | |
964 | ret = ext4_journal_get_write_access(handle, bh); | |
965 | if (!ret && dirty) | |
966 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
967 | return ret; | |
ac27a0ec DK |
968 | } |
969 | ||
8b0f165f AP |
970 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
971 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 972 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
973 | loff_t pos, unsigned len, unsigned flags, |
974 | struct page **pagep, void **fsdata) | |
ac27a0ec | 975 | { |
af5bc92d | 976 | struct inode *inode = mapping->host; |
1938a150 | 977 | int ret, needed_blocks; |
ac27a0ec DK |
978 | handle_t *handle; |
979 | int retries = 0; | |
af5bc92d | 980 | struct page *page; |
de9a55b8 | 981 | pgoff_t index; |
af5bc92d | 982 | unsigned from, to; |
bfc1af65 | 983 | |
9bffad1e | 984 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
985 | /* |
986 | * Reserve one block more for addition to orphan list in case | |
987 | * we allocate blocks but write fails for some reason | |
988 | */ | |
989 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 990 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
991 | from = pos & (PAGE_CACHE_SIZE - 1); |
992 | to = from + len; | |
ac27a0ec | 993 | |
f19d5870 TM |
994 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
995 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
996 | flags, pagep); | |
997 | if (ret < 0) | |
47564bfb TT |
998 | return ret; |
999 | if (ret == 1) | |
1000 | return 0; | |
f19d5870 TM |
1001 | } |
1002 | ||
47564bfb TT |
1003 | /* |
1004 | * grab_cache_page_write_begin() can take a long time if the | |
1005 | * system is thrashing due to memory pressure, or if the page | |
1006 | * is being written back. So grab it first before we start | |
1007 | * the transaction handle. This also allows us to allocate | |
1008 | * the page (if needed) without using GFP_NOFS. | |
1009 | */ | |
1010 | retry_grab: | |
1011 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1012 | if (!page) | |
1013 | return -ENOMEM; | |
1014 | unlock_page(page); | |
1015 | ||
1016 | retry_journal: | |
9924a92a | 1017 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1018 | if (IS_ERR(handle)) { |
47564bfb TT |
1019 | page_cache_release(page); |
1020 | return PTR_ERR(handle); | |
7479d2b9 | 1021 | } |
ac27a0ec | 1022 | |
47564bfb TT |
1023 | lock_page(page); |
1024 | if (page->mapping != mapping) { | |
1025 | /* The page got truncated from under us */ | |
1026 | unlock_page(page); | |
1027 | page_cache_release(page); | |
cf108bca | 1028 | ext4_journal_stop(handle); |
47564bfb | 1029 | goto retry_grab; |
cf108bca | 1030 | } |
47564bfb | 1031 | wait_on_page_writeback(page); |
cf108bca | 1032 | |
744692dc | 1033 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1034 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1035 | else |
6e1db88d | 1036 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1037 | |
1038 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 TM |
1039 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1040 | from, to, NULL, | |
1041 | do_journal_get_write_access); | |
ac27a0ec | 1042 | } |
bfc1af65 NP |
1043 | |
1044 | if (ret) { | |
af5bc92d | 1045 | unlock_page(page); |
ae4d5372 | 1046 | /* |
6e1db88d | 1047 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1048 | * outside i_size. Trim these off again. Don't need |
1049 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1050 | * |
1051 | * Add inode to orphan list in case we crash before | |
1052 | * truncate finishes | |
ae4d5372 | 1053 | */ |
ffacfa7a | 1054 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1055 | ext4_orphan_add(handle, inode); |
1056 | ||
1057 | ext4_journal_stop(handle); | |
1058 | if (pos + len > inode->i_size) { | |
b9a4207d | 1059 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1060 | /* |
ffacfa7a | 1061 | * If truncate failed early the inode might |
1938a150 AK |
1062 | * still be on the orphan list; we need to |
1063 | * make sure the inode is removed from the | |
1064 | * orphan list in that case. | |
1065 | */ | |
1066 | if (inode->i_nlink) | |
1067 | ext4_orphan_del(NULL, inode); | |
1068 | } | |
bfc1af65 | 1069 | |
47564bfb TT |
1070 | if (ret == -ENOSPC && |
1071 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1072 | goto retry_journal; | |
1073 | page_cache_release(page); | |
1074 | return ret; | |
1075 | } | |
1076 | *pagep = page; | |
ac27a0ec DK |
1077 | return ret; |
1078 | } | |
1079 | ||
bfc1af65 NP |
1080 | /* For write_end() in data=journal mode */ |
1081 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1082 | { |
1083 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1084 | return 0; | |
1085 | set_buffer_uptodate(bh); | |
0390131b | 1086 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1087 | } |
1088 | ||
eed4333f ZL |
1089 | /* |
1090 | * We need to pick up the new inode size which generic_commit_write gave us | |
1091 | * `file' can be NULL - eg, when called from page_symlink(). | |
1092 | * | |
1093 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1094 | * buffers are managed internally. | |
1095 | */ | |
1096 | static int ext4_write_end(struct file *file, | |
1097 | struct address_space *mapping, | |
1098 | loff_t pos, unsigned len, unsigned copied, | |
1099 | struct page *page, void *fsdata) | |
f8514083 | 1100 | { |
f8514083 | 1101 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f ZL |
1102 | struct inode *inode = mapping->host; |
1103 | int ret = 0, ret2; | |
1104 | int i_size_changed = 0; | |
1105 | ||
1106 | trace_ext4_write_end(inode, pos, len, copied); | |
1107 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1108 | ret = ext4_jbd2_file_inode(handle, inode); | |
1109 | if (ret) { | |
1110 | unlock_page(page); | |
1111 | page_cache_release(page); | |
1112 | goto errout; | |
1113 | } | |
1114 | } | |
f8514083 | 1115 | |
f19d5870 TM |
1116 | if (ext4_has_inline_data(inode)) |
1117 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1118 | copied, page); | |
1119 | else | |
1120 | copied = block_write_end(file, mapping, pos, | |
1121 | len, copied, page, fsdata); | |
f8514083 AK |
1122 | |
1123 | /* | |
1124 | * No need to use i_size_read() here, the i_size | |
eed4333f | 1125 | * cannot change under us because we hole i_mutex. |
f8514083 AK |
1126 | * |
1127 | * But it's important to update i_size while still holding page lock: | |
1128 | * page writeout could otherwise come in and zero beyond i_size. | |
1129 | */ | |
1130 | if (pos + copied > inode->i_size) { | |
1131 | i_size_write(inode, pos + copied); | |
1132 | i_size_changed = 1; | |
1133 | } | |
1134 | ||
eed4333f | 1135 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
f8514083 AK |
1136 | /* We need to mark inode dirty even if |
1137 | * new_i_size is less that inode->i_size | |
eed4333f | 1138 | * but greater than i_disksize. (hint delalloc) |
f8514083 AK |
1139 | */ |
1140 | ext4_update_i_disksize(inode, (pos + copied)); | |
1141 | i_size_changed = 1; | |
1142 | } | |
1143 | unlock_page(page); | |
1144 | page_cache_release(page); | |
1145 | ||
1146 | /* | |
1147 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1148 | * makes the holding time of page lock longer. Second, it forces lock | |
1149 | * ordering of page lock and transaction start for journaling | |
1150 | * filesystems. | |
1151 | */ | |
1152 | if (i_size_changed) | |
1153 | ext4_mark_inode_dirty(handle, inode); | |
1154 | ||
74d553aa TT |
1155 | if (copied < 0) |
1156 | ret = copied; | |
ffacfa7a | 1157 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1158 | /* if we have allocated more blocks and copied |
1159 | * less. We will have blocks allocated outside | |
1160 | * inode->i_size. So truncate them | |
1161 | */ | |
1162 | ext4_orphan_add(handle, inode); | |
74d553aa | 1163 | errout: |
617ba13b | 1164 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1165 | if (!ret) |
1166 | ret = ret2; | |
bfc1af65 | 1167 | |
f8514083 | 1168 | if (pos + len > inode->i_size) { |
b9a4207d | 1169 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1170 | /* |
ffacfa7a | 1171 | * If truncate failed early the inode might still be |
f8514083 AK |
1172 | * on the orphan list; we need to make sure the inode |
1173 | * is removed from the orphan list in that case. | |
1174 | */ | |
1175 | if (inode->i_nlink) | |
1176 | ext4_orphan_del(NULL, inode); | |
1177 | } | |
1178 | ||
bfc1af65 | 1179 | return ret ? ret : copied; |
ac27a0ec DK |
1180 | } |
1181 | ||
bfc1af65 | 1182 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1183 | struct address_space *mapping, |
1184 | loff_t pos, unsigned len, unsigned copied, | |
1185 | struct page *page, void *fsdata) | |
ac27a0ec | 1186 | { |
617ba13b | 1187 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1188 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1189 | int ret = 0, ret2; |
1190 | int partial = 0; | |
bfc1af65 | 1191 | unsigned from, to; |
cf17fea6 | 1192 | loff_t new_i_size; |
ac27a0ec | 1193 | |
9bffad1e | 1194 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1195 | from = pos & (PAGE_CACHE_SIZE - 1); |
1196 | to = from + len; | |
1197 | ||
441c8508 CW |
1198 | BUG_ON(!ext4_handle_valid(handle)); |
1199 | ||
3fdcfb66 TM |
1200 | if (ext4_has_inline_data(inode)) |
1201 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1202 | copied, page); | |
1203 | else { | |
1204 | if (copied < len) { | |
1205 | if (!PageUptodate(page)) | |
1206 | copied = 0; | |
1207 | page_zero_new_buffers(page, from+copied, to); | |
1208 | } | |
ac27a0ec | 1209 | |
3fdcfb66 TM |
1210 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1211 | to, &partial, write_end_fn); | |
1212 | if (!partial) | |
1213 | SetPageUptodate(page); | |
1214 | } | |
cf17fea6 AK |
1215 | new_i_size = pos + copied; |
1216 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1217 | i_size_write(inode, pos+copied); |
19f5fb7a | 1218 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1219 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1220 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1221 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1222 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1223 | if (!ret) |
1224 | ret = ret2; | |
1225 | } | |
bfc1af65 | 1226 | |
cf108bca | 1227 | unlock_page(page); |
f8514083 | 1228 | page_cache_release(page); |
ffacfa7a | 1229 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1230 | /* if we have allocated more blocks and copied |
1231 | * less. We will have blocks allocated outside | |
1232 | * inode->i_size. So truncate them | |
1233 | */ | |
1234 | ext4_orphan_add(handle, inode); | |
1235 | ||
617ba13b | 1236 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1237 | if (!ret) |
1238 | ret = ret2; | |
f8514083 | 1239 | if (pos + len > inode->i_size) { |
b9a4207d | 1240 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1241 | /* |
ffacfa7a | 1242 | * If truncate failed early the inode might still be |
f8514083 AK |
1243 | * on the orphan list; we need to make sure the inode |
1244 | * is removed from the orphan list in that case. | |
1245 | */ | |
1246 | if (inode->i_nlink) | |
1247 | ext4_orphan_del(NULL, inode); | |
1248 | } | |
bfc1af65 NP |
1249 | |
1250 | return ret ? ret : copied; | |
ac27a0ec | 1251 | } |
d2a17637 | 1252 | |
386ad67c LC |
1253 | /* |
1254 | * Reserve a metadata for a single block located at lblock | |
1255 | */ | |
1256 | static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) | |
1257 | { | |
1258 | int retries = 0; | |
1259 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1260 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1261 | unsigned int md_needed; | |
1262 | ext4_lblk_t save_last_lblock; | |
1263 | int save_len; | |
1264 | ||
1265 | /* | |
1266 | * recalculate the amount of metadata blocks to reserve | |
1267 | * in order to allocate nrblocks | |
1268 | * worse case is one extent per block | |
1269 | */ | |
1270 | repeat: | |
1271 | spin_lock(&ei->i_block_reservation_lock); | |
1272 | /* | |
1273 | * ext4_calc_metadata_amount() has side effects, which we have | |
1274 | * to be prepared undo if we fail to claim space. | |
1275 | */ | |
1276 | save_len = ei->i_da_metadata_calc_len; | |
1277 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
1278 | md_needed = EXT4_NUM_B2C(sbi, | |
1279 | ext4_calc_metadata_amount(inode, lblock)); | |
1280 | trace_ext4_da_reserve_space(inode, md_needed); | |
1281 | ||
1282 | /* | |
1283 | * We do still charge estimated metadata to the sb though; | |
1284 | * we cannot afford to run out of free blocks. | |
1285 | */ | |
1286 | if (ext4_claim_free_clusters(sbi, md_needed, 0)) { | |
1287 | ei->i_da_metadata_calc_len = save_len; | |
1288 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1289 | spin_unlock(&ei->i_block_reservation_lock); | |
1290 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
1291 | cond_resched(); | |
1292 | goto repeat; | |
1293 | } | |
1294 | return -ENOSPC; | |
1295 | } | |
1296 | ei->i_reserved_meta_blocks += md_needed; | |
1297 | spin_unlock(&ei->i_block_reservation_lock); | |
1298 | ||
1299 | return 0; /* success */ | |
1300 | } | |
1301 | ||
9d0be502 | 1302 | /* |
7b415bf6 | 1303 | * Reserve a single cluster located at lblock |
9d0be502 | 1304 | */ |
01f49d0b | 1305 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1306 | { |
030ba6bc | 1307 | int retries = 0; |
60e58e0f | 1308 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1309 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1310 | unsigned int md_needed; |
5dd4056d | 1311 | int ret; |
03179fe9 TT |
1312 | ext4_lblk_t save_last_lblock; |
1313 | int save_len; | |
1314 | ||
1315 | /* | |
1316 | * We will charge metadata quota at writeout time; this saves | |
1317 | * us from metadata over-estimation, though we may go over by | |
1318 | * a small amount in the end. Here we just reserve for data. | |
1319 | */ | |
1320 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1321 | if (ret) | |
1322 | return ret; | |
d2a17637 MC |
1323 | |
1324 | /* | |
1325 | * recalculate the amount of metadata blocks to reserve | |
1326 | * in order to allocate nrblocks | |
1327 | * worse case is one extent per block | |
1328 | */ | |
030ba6bc | 1329 | repeat: |
0637c6f4 | 1330 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1331 | /* |
1332 | * ext4_calc_metadata_amount() has side effects, which we have | |
1333 | * to be prepared undo if we fail to claim space. | |
1334 | */ | |
1335 | save_len = ei->i_da_metadata_calc_len; | |
1336 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
7b415bf6 AK |
1337 | md_needed = EXT4_NUM_B2C(sbi, |
1338 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1339 | trace_ext4_da_reserve_space(inode, md_needed); |
d2a17637 | 1340 | |
72b8ab9d ES |
1341 | /* |
1342 | * We do still charge estimated metadata to the sb though; | |
1343 | * we cannot afford to run out of free blocks. | |
1344 | */ | |
e7d5f315 | 1345 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
03179fe9 TT |
1346 | ei->i_da_metadata_calc_len = save_len; |
1347 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1348 | spin_unlock(&ei->i_block_reservation_lock); | |
030ba6bc | 1349 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
bb8b20ed | 1350 | cond_resched(); |
030ba6bc AK |
1351 | goto repeat; |
1352 | } | |
03179fe9 | 1353 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1354 | return -ENOSPC; |
1355 | } | |
9d0be502 | 1356 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1357 | ei->i_reserved_meta_blocks += md_needed; |
1358 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1359 | |
d2a17637 MC |
1360 | return 0; /* success */ |
1361 | } | |
1362 | ||
12219aea | 1363 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1364 | { |
1365 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1366 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1367 | |
cd213226 MC |
1368 | if (!to_free) |
1369 | return; /* Nothing to release, exit */ | |
1370 | ||
d2a17637 | 1371 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1372 | |
5a58ec87 | 1373 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1374 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1375 | /* |
0637c6f4 TT |
1376 | * if there aren't enough reserved blocks, then the |
1377 | * counter is messed up somewhere. Since this | |
1378 | * function is called from invalidate page, it's | |
1379 | * harmless to return without any action. | |
cd213226 | 1380 | */ |
8de5c325 | 1381 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1382 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1383 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1384 | ei->i_reserved_data_blocks); |
1385 | WARN_ON(1); | |
1386 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1387 | } |
0637c6f4 | 1388 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1389 | |
0637c6f4 TT |
1390 | if (ei->i_reserved_data_blocks == 0) { |
1391 | /* | |
1392 | * We can release all of the reserved metadata blocks | |
1393 | * only when we have written all of the delayed | |
1394 | * allocation blocks. | |
7b415bf6 AK |
1395 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1396 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1397 | */ |
57042651 | 1398 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1399 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1400 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1401 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1402 | } |
d2a17637 | 1403 | |
72b8ab9d | 1404 | /* update fs dirty data blocks counter */ |
57042651 | 1405 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1406 | |
d2a17637 | 1407 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1408 | |
7b415bf6 | 1409 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1410 | } |
1411 | ||
1412 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1413 | unsigned long offset) |
d2a17637 MC |
1414 | { |
1415 | int to_release = 0; | |
1416 | struct buffer_head *head, *bh; | |
1417 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1418 | struct inode *inode = page->mapping->host; |
1419 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1420 | int num_clusters; | |
51865fda | 1421 | ext4_fsblk_t lblk; |
d2a17637 MC |
1422 | |
1423 | head = page_buffers(page); | |
1424 | bh = head; | |
1425 | do { | |
1426 | unsigned int next_off = curr_off + bh->b_size; | |
1427 | ||
1428 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1429 | to_release++; | |
1430 | clear_buffer_delay(bh); | |
1431 | } | |
1432 | curr_off = next_off; | |
1433 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1434 | |
51865fda ZL |
1435 | if (to_release) { |
1436 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1437 | ext4_es_remove_extent(inode, lblk, to_release); | |
1438 | } | |
1439 | ||
7b415bf6 AK |
1440 | /* If we have released all the blocks belonging to a cluster, then we |
1441 | * need to release the reserved space for that cluster. */ | |
1442 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1443 | while (num_clusters > 0) { | |
7b415bf6 AK |
1444 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1445 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1446 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1447 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1448 | ext4_da_release_space(inode, 1); |
1449 | ||
1450 | num_clusters--; | |
1451 | } | |
d2a17637 | 1452 | } |
ac27a0ec | 1453 | |
64769240 AT |
1454 | /* |
1455 | * Delayed allocation stuff | |
1456 | */ | |
1457 | ||
64769240 AT |
1458 | /* |
1459 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1460 | * them with writepage() call back |
64769240 AT |
1461 | * |
1462 | * @mpd->inode: inode | |
1463 | * @mpd->first_page: first page of the extent | |
1464 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1465 | * |
1466 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1467 | * to be allocated. this may be wrong if allocation failed. | |
1468 | * | |
1469 | * As pages are already locked by write_cache_pages(), we can't use it | |
1470 | */ | |
1de3e3df TT |
1471 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1472 | struct ext4_map_blocks *map) | |
64769240 | 1473 | { |
791b7f08 AK |
1474 | struct pagevec pvec; |
1475 | unsigned long index, end; | |
1476 | int ret = 0, err, nr_pages, i; | |
1477 | struct inode *inode = mpd->inode; | |
1478 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1479 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1480 | unsigned int len, block_start; |
1481 | struct buffer_head *bh, *page_bufs = NULL; | |
1de3e3df | 1482 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1483 | struct ext4_io_submit io_submit; |
64769240 AT |
1484 | |
1485 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 1486 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1487 | /* |
1488 | * We need to start from the first_page to the next_page - 1 | |
1489 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1490 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1491 | * at the currently mapped buffer_heads. |
1492 | */ | |
64769240 AT |
1493 | index = mpd->first_page; |
1494 | end = mpd->next_page - 1; | |
1495 | ||
791b7f08 | 1496 | pagevec_init(&pvec, 0); |
64769240 | 1497 | while (index <= end) { |
791b7f08 | 1498 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1499 | if (nr_pages == 0) |
1500 | break; | |
1501 | for (i = 0; i < nr_pages; i++) { | |
f8bec370 | 1502 | int skip_page = 0; |
64769240 AT |
1503 | struct page *page = pvec.pages[i]; |
1504 | ||
791b7f08 AK |
1505 | index = page->index; |
1506 | if (index > end) | |
1507 | break; | |
cb20d518 TT |
1508 | |
1509 | if (index == size >> PAGE_CACHE_SHIFT) | |
1510 | len = size & ~PAGE_CACHE_MASK; | |
1511 | else | |
1512 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1513 | if (map) { |
1514 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1515 | inode->i_blkbits); | |
1516 | pblock = map->m_pblk + (cur_logical - | |
1517 | map->m_lblk); | |
1518 | } | |
791b7f08 AK |
1519 | index++; |
1520 | ||
1521 | BUG_ON(!PageLocked(page)); | |
1522 | BUG_ON(PageWriteback(page)); | |
1523 | ||
3ecdb3a1 TT |
1524 | bh = page_bufs = page_buffers(page); |
1525 | block_start = 0; | |
64769240 | 1526 | do { |
1de3e3df TT |
1527 | if (map && (cur_logical >= map->m_lblk) && |
1528 | (cur_logical <= (map->m_lblk + | |
1529 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1530 | if (buffer_delay(bh)) { |
1531 | clear_buffer_delay(bh); | |
1532 | bh->b_blocknr = pblock; | |
29fa89d0 | 1533 | } |
1de3e3df TT |
1534 | if (buffer_unwritten(bh) || |
1535 | buffer_mapped(bh)) | |
1536 | BUG_ON(bh->b_blocknr != pblock); | |
1537 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1538 | set_buffer_uninit(bh); | |
1539 | clear_buffer_unwritten(bh); | |
1540 | } | |
29fa89d0 | 1541 | |
13a79a47 YY |
1542 | /* |
1543 | * skip page if block allocation undone and | |
1544 | * block is dirty | |
1545 | */ | |
1546 | if (ext4_bh_delay_or_unwritten(NULL, bh)) | |
97498956 | 1547 | skip_page = 1; |
3ecdb3a1 TT |
1548 | bh = bh->b_this_page; |
1549 | block_start += bh->b_size; | |
64769240 AT |
1550 | cur_logical++; |
1551 | pblock++; | |
1de3e3df TT |
1552 | } while (bh != page_bufs); |
1553 | ||
f8bec370 JK |
1554 | if (skip_page) { |
1555 | unlock_page(page); | |
1556 | continue; | |
1557 | } | |
cb20d518 | 1558 | |
97498956 | 1559 | clear_page_dirty_for_io(page); |
fe089c77 JK |
1560 | err = ext4_bio_write_page(&io_submit, page, len, |
1561 | mpd->wbc); | |
cb20d518 | 1562 | if (!err) |
a1d6cc56 | 1563 | mpd->pages_written++; |
64769240 AT |
1564 | /* |
1565 | * In error case, we have to continue because | |
1566 | * remaining pages are still locked | |
64769240 AT |
1567 | */ |
1568 | if (ret == 0) | |
1569 | ret = err; | |
64769240 AT |
1570 | } |
1571 | pagevec_release(&pvec); | |
1572 | } | |
bd2d0210 | 1573 | ext4_io_submit(&io_submit); |
64769240 | 1574 | return ret; |
64769240 AT |
1575 | } |
1576 | ||
c7f5938a | 1577 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1578 | { |
1579 | int nr_pages, i; | |
1580 | pgoff_t index, end; | |
1581 | struct pagevec pvec; | |
1582 | struct inode *inode = mpd->inode; | |
1583 | struct address_space *mapping = inode->i_mapping; | |
51865fda | 1584 | ext4_lblk_t start, last; |
c4a0c46e | 1585 | |
c7f5938a CW |
1586 | index = mpd->first_page; |
1587 | end = mpd->next_page - 1; | |
51865fda ZL |
1588 | |
1589 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1590 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1591 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1592 | ||
66bea92c | 1593 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1594 | while (index <= end) { |
1595 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1596 | if (nr_pages == 0) | |
1597 | break; | |
1598 | for (i = 0; i < nr_pages; i++) { | |
1599 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1600 | if (page->index > end) |
c4a0c46e | 1601 | break; |
c4a0c46e AK |
1602 | BUG_ON(!PageLocked(page)); |
1603 | BUG_ON(PageWriteback(page)); | |
1604 | block_invalidatepage(page, 0); | |
1605 | ClearPageUptodate(page); | |
1606 | unlock_page(page); | |
1607 | } | |
9b1d0998 JK |
1608 | index = pvec.pages[nr_pages - 1]->index + 1; |
1609 | pagevec_release(&pvec); | |
c4a0c46e AK |
1610 | } |
1611 | return; | |
1612 | } | |
1613 | ||
df22291f AK |
1614 | static void ext4_print_free_blocks(struct inode *inode) |
1615 | { | |
1616 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 TT |
1617 | struct super_block *sb = inode->i_sb; |
1618 | ||
1619 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 TT |
1620 | EXT4_C2B(EXT4_SB(inode->i_sb), |
1621 | ext4_count_free_clusters(inode->i_sb))); | |
92b97816 TT |
1622 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1623 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
57042651 TT |
1624 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1625 | percpu_counter_sum(&sbi->s_freeclusters_counter))); | |
92b97816 | 1626 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
7b415bf6 AK |
1627 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1628 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); | |
92b97816 TT |
1629 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1630 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
1631 | EXT4_I(inode)->i_reserved_data_blocks); | |
1632 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", | |
1693918e | 1633 | EXT4_I(inode)->i_reserved_meta_blocks); |
df22291f AK |
1634 | return; |
1635 | } | |
1636 | ||
64769240 | 1637 | /* |
5a87b7a5 TT |
1638 | * mpage_da_map_and_submit - go through given space, map them |
1639 | * if necessary, and then submit them for I/O | |
64769240 | 1640 | * |
8dc207c0 | 1641 | * @mpd - bh describing space |
64769240 AT |
1642 | * |
1643 | * The function skips space we know is already mapped to disk blocks. | |
1644 | * | |
64769240 | 1645 | */ |
5a87b7a5 | 1646 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1647 | { |
2ac3b6e0 | 1648 | int err, blks, get_blocks_flags; |
1de3e3df | 1649 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1650 | sector_t next = mpd->b_blocknr; |
1651 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1652 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1653 | handle_t *handle = NULL; | |
64769240 AT |
1654 | |
1655 | /* | |
5a87b7a5 TT |
1656 | * If the blocks are mapped already, or we couldn't accumulate |
1657 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1658 | */ |
5a87b7a5 TT |
1659 | if ((mpd->b_size == 0) || |
1660 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1661 | !(mpd->b_state & (1 << BH_Delay)) && | |
1662 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1663 | goto submit_io; | |
2fa3cdfb TT |
1664 | |
1665 | handle = ext4_journal_current_handle(); | |
1666 | BUG_ON(!handle); | |
1667 | ||
79ffab34 | 1668 | /* |
79e83036 | 1669 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1670 | * blocks, or to convert an uninitialized extent to be |
1671 | * initialized (in the case where we have written into | |
1672 | * one or more preallocated blocks). | |
1673 | * | |
1674 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1675 | * indicate that we are on the delayed allocation path. This | |
1676 | * affects functions in many different parts of the allocation | |
1677 | * call path. This flag exists primarily because we don't | |
79e83036 | 1678 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1679 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1680 | * inode's allocation semaphore is taken. |
1681 | * | |
1682 | * If the blocks in questions were delalloc blocks, set | |
1683 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1684 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1685 | */ |
2ed88685 TT |
1686 | map.m_lblk = next; |
1687 | map.m_len = max_blocks; | |
1296cc85 | 1688 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
1689 | if (ext4_should_dioread_nolock(mpd->inode)) |
1690 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1691 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1692 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1693 | ||
2ed88685 | 1694 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1695 | if (blks < 0) { |
e3570639 ES |
1696 | struct super_block *sb = mpd->inode->i_sb; |
1697 | ||
2fa3cdfb | 1698 | err = blks; |
ed5bde0b | 1699 | /* |
5a87b7a5 | 1700 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1701 | * appears to be free blocks we will just let |
1702 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1703 | */ |
1704 | if (err == -EAGAIN) | |
5a87b7a5 | 1705 | goto submit_io; |
df22291f | 1706 | |
5dee5437 | 1707 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1708 | mpd->retval = err; |
5a87b7a5 | 1709 | goto submit_io; |
df22291f AK |
1710 | } |
1711 | ||
c4a0c46e | 1712 | /* |
ed5bde0b TT |
1713 | * get block failure will cause us to loop in |
1714 | * writepages, because a_ops->writepage won't be able | |
1715 | * to make progress. The page will be redirtied by | |
1716 | * writepage and writepages will again try to write | |
1717 | * the same. | |
c4a0c46e | 1718 | */ |
e3570639 ES |
1719 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1720 | ext4_msg(sb, KERN_CRIT, | |
1721 | "delayed block allocation failed for inode %lu " | |
1722 | "at logical offset %llu with max blocks %zd " | |
1723 | "with error %d", mpd->inode->i_ino, | |
1724 | (unsigned long long) next, | |
1725 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1726 | ext4_msg(sb, KERN_CRIT, | |
01a523eb | 1727 | "This should not happen!! Data will be lost"); |
e3570639 ES |
1728 | if (err == -ENOSPC) |
1729 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1730 | } |
2fa3cdfb | 1731 | /* invalidate all the pages */ |
c7f5938a | 1732 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1733 | |
1734 | /* Mark this page range as having been completed */ | |
1735 | mpd->io_done = 1; | |
5a87b7a5 | 1736 | return; |
c4a0c46e | 1737 | } |
2fa3cdfb TT |
1738 | BUG_ON(blks == 0); |
1739 | ||
1de3e3df | 1740 | mapp = ↦ |
2ed88685 TT |
1741 | if (map.m_flags & EXT4_MAP_NEW) { |
1742 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1743 | int i; | |
64769240 | 1744 | |
2ed88685 TT |
1745 | for (i = 0; i < map.m_len; i++) |
1746 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2fa3cdfb TT |
1747 | } |
1748 | ||
1749 | /* | |
03f5d8bc | 1750 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1751 | */ |
1752 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1753 | if (disksize > i_size_read(mpd->inode)) | |
1754 | disksize = i_size_read(mpd->inode); | |
1755 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1756 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1757 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1758 | if (err) | |
1759 | ext4_error(mpd->inode->i_sb, | |
1760 | "Failed to mark inode %lu dirty", | |
1761 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1762 | } |
1763 | ||
5a87b7a5 | 1764 | submit_io: |
1de3e3df | 1765 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1766 | mpd->io_done = 1; |
64769240 AT |
1767 | } |
1768 | ||
bf068ee2 AK |
1769 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1770 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1771 | |
1772 | /* | |
1773 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1774 | * | |
1775 | * @mpd->lbh - extent of blocks | |
1776 | * @logical - logical number of the block in the file | |
b6a8e62f | 1777 | * @b_state - b_state of the buffer head added |
64769240 AT |
1778 | * |
1779 | * the function is used to collect contig. blocks in same state | |
1780 | */ | |
b6a8e62f | 1781 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical, |
8dc207c0 | 1782 | unsigned long b_state) |
64769240 | 1783 | { |
64769240 | 1784 | sector_t next; |
b6a8e62f JK |
1785 | int blkbits = mpd->inode->i_blkbits; |
1786 | int nrblocks = mpd->b_size >> blkbits; | |
64769240 | 1787 | |
c445e3e0 ES |
1788 | /* |
1789 | * XXX Don't go larger than mballoc is willing to allocate | |
1790 | * This is a stopgap solution. We eventually need to fold | |
1791 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1792 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 | 1793 | */ |
b6a8e62f | 1794 | if (nrblocks >= (8*1024*1024 >> blkbits)) |
c445e3e0 ES |
1795 | goto flush_it; |
1796 | ||
b6a8e62f JK |
1797 | /* check if the reserved journal credits might overflow */ |
1798 | if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) { | |
525f4ed8 MC |
1799 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1800 | /* | |
1801 | * With non-extent format we are limited by the journal | |
1802 | * credit available. Total credit needed to insert | |
1803 | * nrblocks contiguous blocks is dependent on the | |
1804 | * nrblocks. So limit nrblocks. | |
1805 | */ | |
1806 | goto flush_it; | |
525f4ed8 MC |
1807 | } |
1808 | } | |
64769240 AT |
1809 | /* |
1810 | * First block in the extent | |
1811 | */ | |
8dc207c0 TT |
1812 | if (mpd->b_size == 0) { |
1813 | mpd->b_blocknr = logical; | |
b6a8e62f | 1814 | mpd->b_size = 1 << blkbits; |
8dc207c0 | 1815 | mpd->b_state = b_state & BH_FLAGS; |
64769240 AT |
1816 | return; |
1817 | } | |
1818 | ||
8dc207c0 | 1819 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1820 | /* |
1821 | * Can we merge the block to our big extent? | |
1822 | */ | |
8dc207c0 | 1823 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
b6a8e62f | 1824 | mpd->b_size += 1 << blkbits; |
64769240 AT |
1825 | return; |
1826 | } | |
1827 | ||
525f4ed8 | 1828 | flush_it: |
64769240 AT |
1829 | /* |
1830 | * We couldn't merge the block to our extent, so we | |
1831 | * need to flush current extent and start new one | |
1832 | */ | |
5a87b7a5 | 1833 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1834 | return; |
64769240 AT |
1835 | } |
1836 | ||
c364b22c | 1837 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1838 | { |
c364b22c | 1839 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1840 | } |
1841 | ||
5356f261 AK |
1842 | /* |
1843 | * This function is grabs code from the very beginning of | |
1844 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1845 | * time. This function looks up the requested blocks and sets the | |
1846 | * buffer delay bit under the protection of i_data_sem. | |
1847 | */ | |
1848 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1849 | struct ext4_map_blocks *map, | |
1850 | struct buffer_head *bh) | |
1851 | { | |
d100eef2 | 1852 | struct extent_status es; |
5356f261 AK |
1853 | int retval; |
1854 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1855 | #ifdef ES_AGGRESSIVE_TEST |
1856 | struct ext4_map_blocks orig_map; | |
1857 | ||
1858 | memcpy(&orig_map, map, sizeof(*map)); | |
1859 | #endif | |
5356f261 AK |
1860 | |
1861 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1862 | invalid_block = ~0; | |
1863 | ||
1864 | map->m_flags = 0; | |
1865 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1866 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1867 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1868 | |
1869 | /* Lookup extent status tree firstly */ | |
1870 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
1871 | ||
1872 | if (ext4_es_is_hole(&es)) { | |
1873 | retval = 0; | |
1874 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1875 | goto add_delayed; | |
1876 | } | |
1877 | ||
1878 | /* | |
1879 | * Delayed extent could be allocated by fallocate. | |
1880 | * So we need to check it. | |
1881 | */ | |
1882 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1883 | map_bh(bh, inode->i_sb, invalid_block); | |
1884 | set_buffer_new(bh); | |
1885 | set_buffer_delay(bh); | |
1886 | return 0; | |
1887 | } | |
1888 | ||
1889 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1890 | retval = es.es_len - (iblock - es.es_lblk); | |
1891 | if (retval > map->m_len) | |
1892 | retval = map->m_len; | |
1893 | map->m_len = retval; | |
1894 | if (ext4_es_is_written(&es)) | |
1895 | map->m_flags |= EXT4_MAP_MAPPED; | |
1896 | else if (ext4_es_is_unwritten(&es)) | |
1897 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1898 | else | |
1899 | BUG_ON(1); | |
1900 | ||
921f266b DM |
1901 | #ifdef ES_AGGRESSIVE_TEST |
1902 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1903 | #endif | |
d100eef2 ZL |
1904 | return retval; |
1905 | } | |
1906 | ||
5356f261 AK |
1907 | /* |
1908 | * Try to see if we can get the block without requesting a new | |
1909 | * file system block. | |
1910 | */ | |
1911 | down_read((&EXT4_I(inode)->i_data_sem)); | |
9c3569b5 TM |
1912 | if (ext4_has_inline_data(inode)) { |
1913 | /* | |
1914 | * We will soon create blocks for this page, and let | |
1915 | * us pretend as if the blocks aren't allocated yet. | |
1916 | * In case of clusters, we have to handle the work | |
1917 | * of mapping from cluster so that the reserved space | |
1918 | * is calculated properly. | |
1919 | */ | |
1920 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && | |
1921 | ext4_find_delalloc_cluster(inode, map->m_lblk)) | |
1922 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
1923 | retval = 0; | |
1924 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
d100eef2 ZL |
1925 | retval = ext4_ext_map_blocks(NULL, inode, map, |
1926 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1927 | else |
d100eef2 ZL |
1928 | retval = ext4_ind_map_blocks(NULL, inode, map, |
1929 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1930 | |
d100eef2 | 1931 | add_delayed: |
5356f261 | 1932 | if (retval == 0) { |
f7fec032 | 1933 | int ret; |
5356f261 AK |
1934 | /* |
1935 | * XXX: __block_prepare_write() unmaps passed block, | |
1936 | * is it OK? | |
1937 | */ | |
386ad67c LC |
1938 | /* |
1939 | * If the block was allocated from previously allocated cluster, | |
1940 | * then we don't need to reserve it again. However we still need | |
1941 | * to reserve metadata for every block we're going to write. | |
1942 | */ | |
5356f261 | 1943 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
f7fec032 ZL |
1944 | ret = ext4_da_reserve_space(inode, iblock); |
1945 | if (ret) { | |
5356f261 | 1946 | /* not enough space to reserve */ |
f7fec032 | 1947 | retval = ret; |
5356f261 | 1948 | goto out_unlock; |
f7fec032 | 1949 | } |
386ad67c LC |
1950 | } else { |
1951 | ret = ext4_da_reserve_metadata(inode, iblock); | |
1952 | if (ret) { | |
1953 | /* not enough space to reserve */ | |
1954 | retval = ret; | |
1955 | goto out_unlock; | |
1956 | } | |
5356f261 AK |
1957 | } |
1958 | ||
f7fec032 ZL |
1959 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1960 | ~0, EXTENT_STATUS_DELAYED); | |
1961 | if (ret) { | |
1962 | retval = ret; | |
51865fda | 1963 | goto out_unlock; |
f7fec032 | 1964 | } |
51865fda | 1965 | |
5356f261 AK |
1966 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
1967 | * and it should not appear on the bh->b_state. | |
1968 | */ | |
1969 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1970 | ||
1971 | map_bh(bh, inode->i_sb, invalid_block); | |
1972 | set_buffer_new(bh); | |
1973 | set_buffer_delay(bh); | |
f7fec032 ZL |
1974 | } else if (retval > 0) { |
1975 | int ret; | |
1976 | unsigned long long status; | |
1977 | ||
921f266b DM |
1978 | #ifdef ES_AGGRESSIVE_TEST |
1979 | if (retval != map->m_len) { | |
1980 | printk("ES len assertation failed for inode: %lu " | |
1981 | "retval %d != map->m_len %d " | |
1982 | "in %s (lookup)\n", inode->i_ino, retval, | |
1983 | map->m_len, __func__); | |
1984 | } | |
1985 | #endif | |
1986 | ||
f7fec032 ZL |
1987 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1988 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1989 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1990 | map->m_pblk, status); | |
1991 | if (ret != 0) | |
1992 | retval = ret; | |
5356f261 AK |
1993 | } |
1994 | ||
1995 | out_unlock: | |
1996 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1997 | ||
1998 | return retval; | |
1999 | } | |
2000 | ||
64769240 | 2001 | /* |
b920c755 TT |
2002 | * This is a special get_blocks_t callback which is used by |
2003 | * ext4_da_write_begin(). It will either return mapped block or | |
2004 | * reserve space for a single block. | |
29fa89d0 AK |
2005 | * |
2006 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2007 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2008 | * | |
2009 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2010 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2011 | * initialized properly. | |
64769240 | 2012 | */ |
9c3569b5 TM |
2013 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
2014 | struct buffer_head *bh, int create) | |
64769240 | 2015 | { |
2ed88685 | 2016 | struct ext4_map_blocks map; |
64769240 AT |
2017 | int ret = 0; |
2018 | ||
2019 | BUG_ON(create == 0); | |
2ed88685 TT |
2020 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2021 | ||
2022 | map.m_lblk = iblock; | |
2023 | map.m_len = 1; | |
64769240 AT |
2024 | |
2025 | /* | |
2026 | * first, we need to know whether the block is allocated already | |
2027 | * preallocated blocks are unmapped but should treated | |
2028 | * the same as allocated blocks. | |
2029 | */ | |
5356f261 AK |
2030 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
2031 | if (ret <= 0) | |
2ed88685 | 2032 | return ret; |
64769240 | 2033 | |
2ed88685 TT |
2034 | map_bh(bh, inode->i_sb, map.m_pblk); |
2035 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2036 | ||
2037 | if (buffer_unwritten(bh)) { | |
2038 | /* A delayed write to unwritten bh should be marked | |
2039 | * new and mapped. Mapped ensures that we don't do | |
2040 | * get_block multiple times when we write to the same | |
2041 | * offset and new ensures that we do proper zero out | |
2042 | * for partial write. | |
2043 | */ | |
2044 | set_buffer_new(bh); | |
c8205636 | 2045 | set_buffer_mapped(bh); |
2ed88685 TT |
2046 | } |
2047 | return 0; | |
64769240 | 2048 | } |
61628a3f | 2049 | |
62e086be AK |
2050 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2051 | { | |
2052 | get_bh(bh); | |
2053 | return 0; | |
2054 | } | |
2055 | ||
2056 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2057 | { | |
2058 | put_bh(bh); | |
2059 | return 0; | |
2060 | } | |
2061 | ||
2062 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2063 | unsigned int len) |
2064 | { | |
2065 | struct address_space *mapping = page->mapping; | |
2066 | struct inode *inode = mapping->host; | |
3fdcfb66 | 2067 | struct buffer_head *page_bufs = NULL; |
62e086be | 2068 | handle_t *handle = NULL; |
3fdcfb66 TM |
2069 | int ret = 0, err = 0; |
2070 | int inline_data = ext4_has_inline_data(inode); | |
2071 | struct buffer_head *inode_bh = NULL; | |
62e086be | 2072 | |
cb20d518 | 2073 | ClearPageChecked(page); |
3fdcfb66 TM |
2074 | |
2075 | if (inline_data) { | |
2076 | BUG_ON(page->index != 0); | |
2077 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
2078 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
2079 | if (inode_bh == NULL) | |
2080 | goto out; | |
2081 | } else { | |
2082 | page_bufs = page_buffers(page); | |
2083 | if (!page_bufs) { | |
2084 | BUG(); | |
2085 | goto out; | |
2086 | } | |
2087 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2088 | NULL, bget_one); | |
2089 | } | |
62e086be AK |
2090 | /* As soon as we unlock the page, it can go away, but we have |
2091 | * references to buffers so we are safe */ | |
2092 | unlock_page(page); | |
2093 | ||
9924a92a TT |
2094 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2095 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2096 | if (IS_ERR(handle)) { |
2097 | ret = PTR_ERR(handle); | |
2098 | goto out; | |
2099 | } | |
2100 | ||
441c8508 CW |
2101 | BUG_ON(!ext4_handle_valid(handle)); |
2102 | ||
3fdcfb66 TM |
2103 | if (inline_data) { |
2104 | ret = ext4_journal_get_write_access(handle, inode_bh); | |
62e086be | 2105 | |
3fdcfb66 TM |
2106 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
2107 | ||
2108 | } else { | |
2109 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2110 | do_journal_get_write_access); | |
2111 | ||
2112 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2113 | write_end_fn); | |
2114 | } | |
62e086be AK |
2115 | if (ret == 0) |
2116 | ret = err; | |
2d859db3 | 2117 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2118 | err = ext4_journal_stop(handle); |
2119 | if (!ret) | |
2120 | ret = err; | |
2121 | ||
3fdcfb66 TM |
2122 | if (!ext4_has_inline_data(inode)) |
2123 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2124 | NULL, bput_one); | |
19f5fb7a | 2125 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2126 | out: |
3fdcfb66 | 2127 | brelse(inode_bh); |
62e086be AK |
2128 | return ret; |
2129 | } | |
2130 | ||
61628a3f | 2131 | /* |
43ce1d23 AK |
2132 | * Note that we don't need to start a transaction unless we're journaling data |
2133 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2134 | * need to file the inode to the transaction's list in ordered mode because if | |
2135 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2136 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2137 | * transaction the data will hit the disk. In case we are journaling data, we |
2138 | * cannot start transaction directly because transaction start ranks above page | |
2139 | * lock so we have to do some magic. | |
2140 | * | |
b920c755 TT |
2141 | * This function can get called via... |
2142 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2143 | * - journal_submit_inode_data_buffers (no journal handle) | |
f6463b0d | 2144 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2145 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2146 | * |
2147 | * We don't do any block allocation in this function. If we have page with | |
2148 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2149 | * is important for mmaped based write. So if we do with blocksize 1K | |
2150 | * truncate(f, 1024); | |
2151 | * a = mmap(f, 0, 4096); | |
2152 | * a[0] = 'a'; | |
2153 | * truncate(f, 4096); | |
2154 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2155 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2156 | * do_wp_page). So writepage should write the first block. If we modify |
2157 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2158 | * page_mkwrite callback will do the block allocation and mark the | |
2159 | * buffer_heads mapped. | |
2160 | * | |
2161 | * We redirty the page if we have any buffer_heads that is either delay or | |
2162 | * unwritten in the page. | |
2163 | * | |
2164 | * We can get recursively called as show below. | |
2165 | * | |
2166 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2167 | * ext4_writepage() | |
2168 | * | |
2169 | * But since we don't do any block allocation we should not deadlock. | |
2170 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2171 | */ |
43ce1d23 | 2172 | static int ext4_writepage(struct page *page, |
62e086be | 2173 | struct writeback_control *wbc) |
64769240 | 2174 | { |
f8bec370 | 2175 | int ret = 0; |
61628a3f | 2176 | loff_t size; |
498e5f24 | 2177 | unsigned int len; |
744692dc | 2178 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2179 | struct inode *inode = page->mapping->host; |
36ade451 | 2180 | struct ext4_io_submit io_submit; |
61628a3f | 2181 | |
a9c667f8 | 2182 | trace_ext4_writepage(page); |
f0e6c985 AK |
2183 | size = i_size_read(inode); |
2184 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2185 | len = size & ~PAGE_CACHE_MASK; | |
2186 | else | |
2187 | len = PAGE_CACHE_SIZE; | |
64769240 | 2188 | |
a42afc5f | 2189 | page_bufs = page_buffers(page); |
a42afc5f | 2190 | /* |
fe386132 JK |
2191 | * We cannot do block allocation or other extent handling in this |
2192 | * function. If there are buffers needing that, we have to redirty | |
2193 | * the page. But we may reach here when we do a journal commit via | |
2194 | * journal_submit_inode_data_buffers() and in that case we must write | |
2195 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 2196 | */ |
f19d5870 TM |
2197 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2198 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2199 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
2200 | if (current->flags & PF_MEMALLOC) { |
2201 | /* | |
2202 | * For memory cleaning there's no point in writing only | |
2203 | * some buffers. So just bail out. Warn if we came here | |
2204 | * from direct reclaim. | |
2205 | */ | |
2206 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2207 | == PF_MEMALLOC); | |
f0e6c985 AK |
2208 | unlock_page(page); |
2209 | return 0; | |
2210 | } | |
a42afc5f | 2211 | } |
64769240 | 2212 | |
cb20d518 | 2213 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2214 | /* |
2215 | * It's mmapped pagecache. Add buffers and journal it. There | |
2216 | * doesn't seem much point in redirtying the page here. | |
2217 | */ | |
3f0ca309 | 2218 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2219 | |
36ade451 JK |
2220 | memset(&io_submit, 0, sizeof(io_submit)); |
2221 | ret = ext4_bio_write_page(&io_submit, page, len, wbc); | |
2222 | ext4_io_submit(&io_submit); | |
64769240 AT |
2223 | return ret; |
2224 | } | |
2225 | ||
61628a3f | 2226 | /* |
525f4ed8 | 2227 | * This is called via ext4_da_writepages() to |
25985edc | 2228 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2229 | * a single extent allocation into a single transaction, |
2230 | * ext4_da_writpeages() will loop calling this before | |
2231 | * the block allocation. | |
61628a3f | 2232 | */ |
525f4ed8 MC |
2233 | |
2234 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2235 | { | |
2236 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2237 | ||
2238 | /* | |
2239 | * With non-extent format the journal credit needed to | |
2240 | * insert nrblocks contiguous block is dependent on | |
2241 | * number of contiguous block. So we will limit | |
2242 | * number of contiguous block to a sane value | |
2243 | */ | |
12e9b892 | 2244 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2245 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2246 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2247 | ||
2248 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2249 | } | |
61628a3f | 2250 | |
8e48dcfb TT |
2251 | /* |
2252 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2253 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2254 | * mpage_da_map_and_submit to map a single contiguous memory region |
2255 | * and then write them. | |
8e48dcfb | 2256 | */ |
9c3569b5 TM |
2257 | static int write_cache_pages_da(handle_t *handle, |
2258 | struct address_space *mapping, | |
8e48dcfb | 2259 | struct writeback_control *wbc, |
72f84e65 ES |
2260 | struct mpage_da_data *mpd, |
2261 | pgoff_t *done_index) | |
8e48dcfb | 2262 | { |
4f01b02c | 2263 | struct buffer_head *bh, *head; |
168fc022 | 2264 | struct inode *inode = mapping->host; |
4f01b02c TT |
2265 | struct pagevec pvec; |
2266 | unsigned int nr_pages; | |
2267 | sector_t logical; | |
2268 | pgoff_t index, end; | |
2269 | long nr_to_write = wbc->nr_to_write; | |
2270 | int i, tag, ret = 0; | |
8e48dcfb | 2271 | |
168fc022 TT |
2272 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2273 | mpd->wbc = wbc; | |
2274 | mpd->inode = inode; | |
8e48dcfb TT |
2275 | pagevec_init(&pvec, 0); |
2276 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2277 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2278 | ||
6e6938b6 | 2279 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2280 | tag = PAGECACHE_TAG_TOWRITE; |
2281 | else | |
2282 | tag = PAGECACHE_TAG_DIRTY; | |
2283 | ||
72f84e65 | 2284 | *done_index = index; |
4f01b02c | 2285 | while (index <= end) { |
5b41d924 | 2286 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2287 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2288 | if (nr_pages == 0) | |
4f01b02c | 2289 | return 0; |
8e48dcfb TT |
2290 | |
2291 | for (i = 0; i < nr_pages; i++) { | |
2292 | struct page *page = pvec.pages[i]; | |
2293 | ||
2294 | /* | |
2295 | * At this point, the page may be truncated or | |
2296 | * invalidated (changing page->mapping to NULL), or | |
2297 | * even swizzled back from swapper_space to tmpfs file | |
2298 | * mapping. However, page->index will not change | |
2299 | * because we have a reference on the page. | |
2300 | */ | |
4f01b02c TT |
2301 | if (page->index > end) |
2302 | goto out; | |
8e48dcfb | 2303 | |
72f84e65 ES |
2304 | *done_index = page->index + 1; |
2305 | ||
78aaced3 TT |
2306 | /* |
2307 | * If we can't merge this page, and we have | |
2308 | * accumulated an contiguous region, write it | |
2309 | */ | |
2310 | if ((mpd->next_page != page->index) && | |
2311 | (mpd->next_page != mpd->first_page)) { | |
2312 | mpage_da_map_and_submit(mpd); | |
2313 | goto ret_extent_tail; | |
2314 | } | |
2315 | ||
8e48dcfb TT |
2316 | lock_page(page); |
2317 | ||
2318 | /* | |
4f01b02c TT |
2319 | * If the page is no longer dirty, or its |
2320 | * mapping no longer corresponds to inode we | |
2321 | * are writing (which means it has been | |
2322 | * truncated or invalidated), or the page is | |
2323 | * already under writeback and we are not | |
2324 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2325 | */ |
4f01b02c TT |
2326 | if (!PageDirty(page) || |
2327 | (PageWriteback(page) && | |
2328 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2329 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2330 | unlock_page(page); |
2331 | continue; | |
2332 | } | |
2333 | ||
7cb1a535 | 2334 | wait_on_page_writeback(page); |
8e48dcfb | 2335 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2336 | |
9c3569b5 TM |
2337 | /* |
2338 | * If we have inline data and arrive here, it means that | |
2339 | * we will soon create the block for the 1st page, so | |
2340 | * we'd better clear the inline data here. | |
2341 | */ | |
2342 | if (ext4_has_inline_data(inode)) { | |
2343 | BUG_ON(ext4_test_inode_state(inode, | |
2344 | EXT4_STATE_MAY_INLINE_DATA)); | |
2345 | ext4_destroy_inline_data(handle, inode); | |
2346 | } | |
2347 | ||
168fc022 | 2348 | if (mpd->next_page != page->index) |
8eb9e5ce | 2349 | mpd->first_page = page->index; |
8eb9e5ce TT |
2350 | mpd->next_page = page->index + 1; |
2351 | logical = (sector_t) page->index << | |
2352 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2353 | ||
f8bec370 JK |
2354 | /* Add all dirty buffers to mpd */ |
2355 | head = page_buffers(page); | |
2356 | bh = head; | |
2357 | do { | |
2358 | BUG_ON(buffer_locked(bh)); | |
8eb9e5ce | 2359 | /* |
f8bec370 JK |
2360 | * We need to try to allocate unmapped blocks |
2361 | * in the same page. Otherwise we won't make | |
2362 | * progress with the page in ext4_writepage | |
8eb9e5ce | 2363 | */ |
f8bec370 JK |
2364 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
2365 | mpage_add_bh_to_extent(mpd, logical, | |
f8bec370 JK |
2366 | bh->b_state); |
2367 | if (mpd->io_done) | |
2368 | goto ret_extent_tail; | |
2369 | } else if (buffer_dirty(bh) && | |
2370 | buffer_mapped(bh)) { | |
8eb9e5ce | 2371 | /* |
f8bec370 JK |
2372 | * mapped dirty buffer. We need to |
2373 | * update the b_state because we look | |
2374 | * at b_state in mpage_da_map_blocks. | |
2375 | * We don't update b_size because if we | |
2376 | * find an unmapped buffer_head later | |
2377 | * we need to use the b_state flag of | |
2378 | * that buffer_head. | |
8eb9e5ce | 2379 | */ |
f8bec370 JK |
2380 | if (mpd->b_size == 0) |
2381 | mpd->b_state = | |
2382 | bh->b_state & BH_FLAGS; | |
2383 | } | |
2384 | logical++; | |
2385 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2386 | |
2387 | if (nr_to_write > 0) { | |
2388 | nr_to_write--; | |
2389 | if (nr_to_write == 0 && | |
4f01b02c | 2390 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2391 | /* |
2392 | * We stop writing back only if we are | |
2393 | * not doing integrity sync. In case of | |
2394 | * integrity sync we have to keep going | |
2395 | * because someone may be concurrently | |
2396 | * dirtying pages, and we might have | |
2397 | * synced a lot of newly appeared dirty | |
2398 | * pages, but have not synced all of the | |
2399 | * old dirty pages. | |
2400 | */ | |
4f01b02c | 2401 | goto out; |
8e48dcfb TT |
2402 | } |
2403 | } | |
2404 | pagevec_release(&pvec); | |
2405 | cond_resched(); | |
2406 | } | |
4f01b02c TT |
2407 | return 0; |
2408 | ret_extent_tail: | |
2409 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2410 | out: |
2411 | pagevec_release(&pvec); | |
2412 | cond_resched(); | |
8e48dcfb TT |
2413 | return ret; |
2414 | } | |
2415 | ||
2416 | ||
64769240 | 2417 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2418 | struct writeback_control *wbc) |
64769240 | 2419 | { |
22208ded AK |
2420 | pgoff_t index; |
2421 | int range_whole = 0; | |
61628a3f | 2422 | handle_t *handle = NULL; |
df22291f | 2423 | struct mpage_da_data mpd; |
5e745b04 | 2424 | struct inode *inode = mapping->host; |
498e5f24 | 2425 | int pages_written = 0; |
55138e0b | 2426 | unsigned int max_pages; |
2acf2c26 | 2427 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2428 | int needed_blocks, ret = 0; |
2429 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2430 | loff_t range_start = wbc->range_start; |
5e745b04 | 2431 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2432 | pgoff_t done_index = 0; |
5b41d924 | 2433 | pgoff_t end; |
1bce63d1 | 2434 | struct blk_plug plug; |
61628a3f | 2435 | |
9bffad1e | 2436 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2437 | |
61628a3f MC |
2438 | /* |
2439 | * No pages to write? This is mainly a kludge to avoid starting | |
2440 | * a transaction for special inodes like journal inode on last iput() | |
2441 | * because that could violate lock ordering on umount | |
2442 | */ | |
a1d6cc56 | 2443 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2444 | return 0; |
2a21e37e TT |
2445 | |
2446 | /* | |
2447 | * If the filesystem has aborted, it is read-only, so return | |
2448 | * right away instead of dumping stack traces later on that | |
2449 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2450 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2451 | * the latter could be true if the filesystem is mounted |
2452 | * read-only, and in that case, ext4_da_writepages should | |
2453 | * *never* be called, so if that ever happens, we would want | |
2454 | * the stack trace. | |
2455 | */ | |
4ab2f15b | 2456 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2457 | return -EROFS; |
2458 | ||
22208ded AK |
2459 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2460 | range_whole = 1; | |
61628a3f | 2461 | |
2acf2c26 AK |
2462 | range_cyclic = wbc->range_cyclic; |
2463 | if (wbc->range_cyclic) { | |
22208ded | 2464 | index = mapping->writeback_index; |
2acf2c26 AK |
2465 | if (index) |
2466 | cycled = 0; | |
2467 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2468 | wbc->range_end = LLONG_MAX; | |
2469 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2470 | end = -1; |
2471 | } else { | |
22208ded | 2472 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2473 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2474 | } | |
a1d6cc56 | 2475 | |
55138e0b TT |
2476 | /* |
2477 | * This works around two forms of stupidity. The first is in | |
2478 | * the writeback code, which caps the maximum number of pages | |
2479 | * written to be 1024 pages. This is wrong on multiple | |
2480 | * levels; different architectues have a different page size, | |
2481 | * which changes the maximum amount of data which gets | |
2482 | * written. Secondly, 4 megabytes is way too small. XFS | |
2483 | * forces this value to be 16 megabytes by multiplying | |
2484 | * nr_to_write parameter by four, and then relies on its | |
2485 | * allocator to allocate larger extents to make them | |
2486 | * contiguous. Unfortunately this brings us to the second | |
2487 | * stupidity, which is that ext4's mballoc code only allocates | |
2488 | * at most 2048 blocks. So we force contiguous writes up to | |
2489 | * the number of dirty blocks in the inode, or | |
2490 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2491 | */ | |
2492 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2493 | if (!range_cyclic && range_whole) { |
2494 | if (wbc->nr_to_write == LONG_MAX) | |
2495 | desired_nr_to_write = wbc->nr_to_write; | |
2496 | else | |
2497 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2498 | } else | |
55138e0b TT |
2499 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2500 | max_pages); | |
2501 | if (desired_nr_to_write > max_pages) | |
2502 | desired_nr_to_write = max_pages; | |
2503 | ||
2504 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2505 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2506 | wbc->nr_to_write = desired_nr_to_write; | |
2507 | } | |
2508 | ||
2acf2c26 | 2509 | retry: |
6e6938b6 | 2510 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2511 | tag_pages_for_writeback(mapping, index, end); |
2512 | ||
1bce63d1 | 2513 | blk_start_plug(&plug); |
22208ded | 2514 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2515 | |
2516 | /* | |
2517 | * we insert one extent at a time. So we need | |
2518 | * credit needed for single extent allocation. | |
2519 | * journalled mode is currently not supported | |
2520 | * by delalloc | |
2521 | */ | |
2522 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2523 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2524 | |
61628a3f | 2525 | /* start a new transaction*/ |
9924a92a TT |
2526 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2527 | needed_blocks); | |
61628a3f MC |
2528 | if (IS_ERR(handle)) { |
2529 | ret = PTR_ERR(handle); | |
1693918e | 2530 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2531 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2532 | wbc->nr_to_write, inode->i_ino, ret); |
3c1fcb2c | 2533 | blk_finish_plug(&plug); |
61628a3f MC |
2534 | goto out_writepages; |
2535 | } | |
f63e6005 TT |
2536 | |
2537 | /* | |
8eb9e5ce | 2538 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2539 | * contiguous region of logical blocks that need |
8eb9e5ce | 2540 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2541 | */ |
9c3569b5 TM |
2542 | ret = write_cache_pages_da(handle, mapping, |
2543 | wbc, &mpd, &done_index); | |
f63e6005 | 2544 | /* |
af901ca1 | 2545 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2546 | * haven't done the I/O yet, map the blocks and submit |
2547 | * them for I/O. | |
2548 | */ | |
2549 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2550 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2551 | ret = MPAGE_DA_EXTENT_TAIL; |
2552 | } | |
b3a3ca8c | 2553 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2554 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2555 | |
61628a3f | 2556 | ext4_journal_stop(handle); |
df22291f | 2557 | |
8f64b32e | 2558 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2559 | /* commit the transaction which would |
2560 | * free blocks released in the transaction | |
2561 | * and try again | |
2562 | */ | |
df22291f | 2563 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2564 | ret = 0; |
2565 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 | 2566 | /* |
8de49e67 KM |
2567 | * Got one extent now try with rest of the pages. |
2568 | * If mpd.retval is set -EIO, journal is aborted. | |
2569 | * So we don't need to write any more. | |
a1d6cc56 | 2570 | */ |
22208ded | 2571 | pages_written += mpd.pages_written; |
8de49e67 | 2572 | ret = mpd.retval; |
2acf2c26 | 2573 | io_done = 1; |
22208ded | 2574 | } else if (wbc->nr_to_write) |
61628a3f MC |
2575 | /* |
2576 | * There is no more writeout needed | |
2577 | * or we requested for a noblocking writeout | |
2578 | * and we found the device congested | |
2579 | */ | |
61628a3f | 2580 | break; |
a1d6cc56 | 2581 | } |
1bce63d1 | 2582 | blk_finish_plug(&plug); |
2acf2c26 AK |
2583 | if (!io_done && !cycled) { |
2584 | cycled = 1; | |
2585 | index = 0; | |
2586 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2587 | wbc->range_end = mapping->writeback_index - 1; | |
2588 | goto retry; | |
2589 | } | |
22208ded AK |
2590 | |
2591 | /* Update index */ | |
2acf2c26 | 2592 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2593 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2594 | /* | |
2595 | * set the writeback_index so that range_cyclic | |
2596 | * mode will write it back later | |
2597 | */ | |
72f84e65 | 2598 | mapping->writeback_index = done_index; |
a1d6cc56 | 2599 | |
61628a3f | 2600 | out_writepages: |
2faf2e19 | 2601 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2602 | wbc->range_start = range_start; |
9bffad1e | 2603 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2604 | return ret; |
64769240 AT |
2605 | } |
2606 | ||
79f0be8d AK |
2607 | static int ext4_nonda_switch(struct super_block *sb) |
2608 | { | |
2609 | s64 free_blocks, dirty_blocks; | |
2610 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2611 | ||
2612 | /* | |
2613 | * switch to non delalloc mode if we are running low | |
2614 | * on free block. The free block accounting via percpu | |
179f7ebf | 2615 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2616 | * accumulated on each CPU without updating global counters |
2617 | * Delalloc need an accurate free block accounting. So switch | |
2618 | * to non delalloc when we are near to error range. | |
2619 | */ | |
57042651 TT |
2620 | free_blocks = EXT4_C2B(sbi, |
2621 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
2622 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2623 | /* |
2624 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2625 | */ | |
10ee27a0 MX |
2626 | if (dirty_blocks && (free_blocks < 2 * dirty_blocks)) |
2627 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); | |
00d4e736 | 2628 | |
79f0be8d | 2629 | if (2 * free_blocks < 3 * dirty_blocks || |
df55c99d | 2630 | free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) { |
79f0be8d | 2631 | /* |
c8afb446 ES |
2632 | * free block count is less than 150% of dirty blocks |
2633 | * or free blocks is less than watermark | |
79f0be8d AK |
2634 | */ |
2635 | return 1; | |
2636 | } | |
2637 | return 0; | |
2638 | } | |
2639 | ||
64769240 | 2640 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2641 | loff_t pos, unsigned len, unsigned flags, |
2642 | struct page **pagep, void **fsdata) | |
64769240 | 2643 | { |
72b8ab9d | 2644 | int ret, retries = 0; |
64769240 AT |
2645 | struct page *page; |
2646 | pgoff_t index; | |
64769240 AT |
2647 | struct inode *inode = mapping->host; |
2648 | handle_t *handle; | |
2649 | ||
2650 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2651 | |
2652 | if (ext4_nonda_switch(inode->i_sb)) { | |
2653 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2654 | return ext4_write_begin(file, mapping, pos, | |
2655 | len, flags, pagep, fsdata); | |
2656 | } | |
2657 | *fsdata = (void *)0; | |
9bffad1e | 2658 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2659 | |
2660 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2661 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2662 | pos, len, flags, | |
2663 | pagep, fsdata); | |
2664 | if (ret < 0) | |
47564bfb TT |
2665 | return ret; |
2666 | if (ret == 1) | |
2667 | return 0; | |
9c3569b5 TM |
2668 | } |
2669 | ||
47564bfb TT |
2670 | /* |
2671 | * grab_cache_page_write_begin() can take a long time if the | |
2672 | * system is thrashing due to memory pressure, or if the page | |
2673 | * is being written back. So grab it first before we start | |
2674 | * the transaction handle. This also allows us to allocate | |
2675 | * the page (if needed) without using GFP_NOFS. | |
2676 | */ | |
2677 | retry_grab: | |
2678 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2679 | if (!page) | |
2680 | return -ENOMEM; | |
2681 | unlock_page(page); | |
2682 | ||
64769240 AT |
2683 | /* |
2684 | * With delayed allocation, we don't log the i_disksize update | |
2685 | * if there is delayed block allocation. But we still need | |
2686 | * to journalling the i_disksize update if writes to the end | |
2687 | * of file which has an already mapped buffer. | |
2688 | */ | |
47564bfb | 2689 | retry_journal: |
9924a92a | 2690 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
64769240 | 2691 | if (IS_ERR(handle)) { |
47564bfb TT |
2692 | page_cache_release(page); |
2693 | return PTR_ERR(handle); | |
64769240 AT |
2694 | } |
2695 | ||
47564bfb TT |
2696 | lock_page(page); |
2697 | if (page->mapping != mapping) { | |
2698 | /* The page got truncated from under us */ | |
2699 | unlock_page(page); | |
2700 | page_cache_release(page); | |
d5a0d4f7 | 2701 | ext4_journal_stop(handle); |
47564bfb | 2702 | goto retry_grab; |
d5a0d4f7 | 2703 | } |
47564bfb TT |
2704 | /* In case writeback began while the page was unlocked */ |
2705 | wait_on_page_writeback(page); | |
64769240 | 2706 | |
6e1db88d | 2707 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2708 | if (ret < 0) { |
2709 | unlock_page(page); | |
2710 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2711 | /* |
2712 | * block_write_begin may have instantiated a few blocks | |
2713 | * outside i_size. Trim these off again. Don't need | |
2714 | * i_size_read because we hold i_mutex. | |
2715 | */ | |
2716 | if (pos + len > inode->i_size) | |
b9a4207d | 2717 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2718 | |
2719 | if (ret == -ENOSPC && | |
2720 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2721 | goto retry_journal; | |
2722 | ||
2723 | page_cache_release(page); | |
2724 | return ret; | |
64769240 AT |
2725 | } |
2726 | ||
47564bfb | 2727 | *pagep = page; |
64769240 AT |
2728 | return ret; |
2729 | } | |
2730 | ||
632eaeab MC |
2731 | /* |
2732 | * Check if we should update i_disksize | |
2733 | * when write to the end of file but not require block allocation | |
2734 | */ | |
2735 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2736 | unsigned long offset) |
632eaeab MC |
2737 | { |
2738 | struct buffer_head *bh; | |
2739 | struct inode *inode = page->mapping->host; | |
2740 | unsigned int idx; | |
2741 | int i; | |
2742 | ||
2743 | bh = page_buffers(page); | |
2744 | idx = offset >> inode->i_blkbits; | |
2745 | ||
af5bc92d | 2746 | for (i = 0; i < idx; i++) |
632eaeab MC |
2747 | bh = bh->b_this_page; |
2748 | ||
29fa89d0 | 2749 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2750 | return 0; |
2751 | return 1; | |
2752 | } | |
2753 | ||
64769240 | 2754 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2755 | struct address_space *mapping, |
2756 | loff_t pos, unsigned len, unsigned copied, | |
2757 | struct page *page, void *fsdata) | |
64769240 AT |
2758 | { |
2759 | struct inode *inode = mapping->host; | |
2760 | int ret = 0, ret2; | |
2761 | handle_t *handle = ext4_journal_current_handle(); | |
2762 | loff_t new_i_size; | |
632eaeab | 2763 | unsigned long start, end; |
79f0be8d AK |
2764 | int write_mode = (int)(unsigned long)fsdata; |
2765 | ||
74d553aa TT |
2766 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2767 | return ext4_write_end(file, mapping, pos, | |
2768 | len, copied, page, fsdata); | |
632eaeab | 2769 | |
9bffad1e | 2770 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2771 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2772 | end = start + copied - 1; |
64769240 AT |
2773 | |
2774 | /* | |
2775 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2776 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2777 | * into that. | |
2778 | */ | |
64769240 | 2779 | new_i_size = pos + copied; |
ea51d132 | 2780 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2781 | if (ext4_has_inline_data(inode) || |
2782 | ext4_da_should_update_i_disksize(page, end)) { | |
632eaeab | 2783 | down_write(&EXT4_I(inode)->i_data_sem); |
f3b59291 | 2784 | if (new_i_size > EXT4_I(inode)->i_disksize) |
632eaeab | 2785 | EXT4_I(inode)->i_disksize = new_i_size; |
632eaeab | 2786 | up_write(&EXT4_I(inode)->i_data_sem); |
cf17fea6 AK |
2787 | /* We need to mark inode dirty even if |
2788 | * new_i_size is less that inode->i_size | |
2789 | * bu greater than i_disksize.(hint delalloc) | |
2790 | */ | |
2791 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2792 | } |
632eaeab | 2793 | } |
9c3569b5 TM |
2794 | |
2795 | if (write_mode != CONVERT_INLINE_DATA && | |
2796 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2797 | ext4_has_inline_data(inode)) | |
2798 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2799 | page); | |
2800 | else | |
2801 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2802 | page, fsdata); |
9c3569b5 | 2803 | |
64769240 AT |
2804 | copied = ret2; |
2805 | if (ret2 < 0) | |
2806 | ret = ret2; | |
2807 | ret2 = ext4_journal_stop(handle); | |
2808 | if (!ret) | |
2809 | ret = ret2; | |
2810 | ||
2811 | return ret ? ret : copied; | |
2812 | } | |
2813 | ||
2814 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2815 | { | |
64769240 AT |
2816 | /* |
2817 | * Drop reserved blocks | |
2818 | */ | |
2819 | BUG_ON(!PageLocked(page)); | |
2820 | if (!page_has_buffers(page)) | |
2821 | goto out; | |
2822 | ||
d2a17637 | 2823 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2824 | |
2825 | out: | |
2826 | ext4_invalidatepage(page, offset); | |
2827 | ||
2828 | return; | |
2829 | } | |
2830 | ||
ccd2506b TT |
2831 | /* |
2832 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2833 | */ | |
2834 | int ext4_alloc_da_blocks(struct inode *inode) | |
2835 | { | |
fb40ba0d TT |
2836 | trace_ext4_alloc_da_blocks(inode); |
2837 | ||
ccd2506b TT |
2838 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2839 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2840 | return 0; | |
2841 | ||
2842 | /* | |
2843 | * We do something simple for now. The filemap_flush() will | |
2844 | * also start triggering a write of the data blocks, which is | |
2845 | * not strictly speaking necessary (and for users of | |
2846 | * laptop_mode, not even desirable). However, to do otherwise | |
2847 | * would require replicating code paths in: | |
de9a55b8 | 2848 | * |
ccd2506b TT |
2849 | * ext4_da_writepages() -> |
2850 | * write_cache_pages() ---> (via passed in callback function) | |
2851 | * __mpage_da_writepage() --> | |
2852 | * mpage_add_bh_to_extent() | |
2853 | * mpage_da_map_blocks() | |
2854 | * | |
2855 | * The problem is that write_cache_pages(), located in | |
2856 | * mm/page-writeback.c, marks pages clean in preparation for | |
2857 | * doing I/O, which is not desirable if we're not planning on | |
2858 | * doing I/O at all. | |
2859 | * | |
2860 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2861 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2862 | * would be ugly in the extreme. So instead we would need to |
2863 | * replicate parts of the code in the above functions, | |
25985edc | 2864 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2865 | * write out the pages, but rather only collect contiguous |
2866 | * logical block extents, call the multi-block allocator, and | |
2867 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2868 | * |
ccd2506b TT |
2869 | * For now, though, we'll cheat by calling filemap_flush(), |
2870 | * which will map the blocks, and start the I/O, but not | |
2871 | * actually wait for the I/O to complete. | |
2872 | */ | |
2873 | return filemap_flush(inode->i_mapping); | |
2874 | } | |
64769240 | 2875 | |
ac27a0ec DK |
2876 | /* |
2877 | * bmap() is special. It gets used by applications such as lilo and by | |
2878 | * the swapper to find the on-disk block of a specific piece of data. | |
2879 | * | |
2880 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2881 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2882 | * filesystem and enables swap, then they may get a nasty shock when the |
2883 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2884 | * the original zero's written out previously to the journal and | |
2885 | * awaiting writeback in the kernel's buffer cache. | |
2886 | * | |
2887 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2888 | * take extra steps to flush any blocks which might be in the cache. | |
2889 | */ | |
617ba13b | 2890 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2891 | { |
2892 | struct inode *inode = mapping->host; | |
2893 | journal_t *journal; | |
2894 | int err; | |
2895 | ||
46c7f254 TM |
2896 | /* |
2897 | * We can get here for an inline file via the FIBMAP ioctl | |
2898 | */ | |
2899 | if (ext4_has_inline_data(inode)) | |
2900 | return 0; | |
2901 | ||
64769240 AT |
2902 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2903 | test_opt(inode->i_sb, DELALLOC)) { | |
2904 | /* | |
2905 | * With delalloc we want to sync the file | |
2906 | * so that we can make sure we allocate | |
2907 | * blocks for file | |
2908 | */ | |
2909 | filemap_write_and_wait(mapping); | |
2910 | } | |
2911 | ||
19f5fb7a TT |
2912 | if (EXT4_JOURNAL(inode) && |
2913 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2914 | /* |
2915 | * This is a REALLY heavyweight approach, but the use of | |
2916 | * bmap on dirty files is expected to be extremely rare: | |
2917 | * only if we run lilo or swapon on a freshly made file | |
2918 | * do we expect this to happen. | |
2919 | * | |
2920 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2921 | * represent an unprivileged user DOS attack --- we'd be | |
2922 | * in trouble if mortal users could trigger this path at | |
2923 | * will.) | |
2924 | * | |
617ba13b | 2925 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2926 | * regular files. If somebody wants to bmap a directory |
2927 | * or symlink and gets confused because the buffer | |
2928 | * hasn't yet been flushed to disk, they deserve | |
2929 | * everything they get. | |
2930 | */ | |
2931 | ||
19f5fb7a | 2932 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2933 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2934 | jbd2_journal_lock_updates(journal); |
2935 | err = jbd2_journal_flush(journal); | |
2936 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2937 | |
2938 | if (err) | |
2939 | return 0; | |
2940 | } | |
2941 | ||
af5bc92d | 2942 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2943 | } |
2944 | ||
617ba13b | 2945 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2946 | { |
46c7f254 TM |
2947 | int ret = -EAGAIN; |
2948 | struct inode *inode = page->mapping->host; | |
2949 | ||
0562e0ba | 2950 | trace_ext4_readpage(page); |
46c7f254 TM |
2951 | |
2952 | if (ext4_has_inline_data(inode)) | |
2953 | ret = ext4_readpage_inline(inode, page); | |
2954 | ||
2955 | if (ret == -EAGAIN) | |
2956 | return mpage_readpage(page, ext4_get_block); | |
2957 | ||
2958 | return ret; | |
ac27a0ec DK |
2959 | } |
2960 | ||
2961 | static int | |
617ba13b | 2962 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2963 | struct list_head *pages, unsigned nr_pages) |
2964 | { | |
46c7f254 TM |
2965 | struct inode *inode = mapping->host; |
2966 | ||
2967 | /* If the file has inline data, no need to do readpages. */ | |
2968 | if (ext4_has_inline_data(inode)) | |
2969 | return 0; | |
2970 | ||
617ba13b | 2971 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2972 | } |
2973 | ||
617ba13b | 2974 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2975 | { |
0562e0ba JZ |
2976 | trace_ext4_invalidatepage(page, offset); |
2977 | ||
4520fb3c JK |
2978 | /* No journalling happens on data buffers when this function is used */ |
2979 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
2980 | ||
2981 | block_invalidatepage(page, offset); | |
2982 | } | |
2983 | ||
53e87268 JK |
2984 | static int __ext4_journalled_invalidatepage(struct page *page, |
2985 | unsigned long offset) | |
4520fb3c JK |
2986 | { |
2987 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
2988 | ||
2989 | trace_ext4_journalled_invalidatepage(page, offset); | |
2990 | ||
ac27a0ec DK |
2991 | /* |
2992 | * If it's a full truncate we just forget about the pending dirtying | |
2993 | */ | |
2994 | if (offset == 0) | |
2995 | ClearPageChecked(page); | |
2996 | ||
53e87268 JK |
2997 | return jbd2_journal_invalidatepage(journal, page, offset); |
2998 | } | |
2999 | ||
3000 | /* Wrapper for aops... */ | |
3001 | static void ext4_journalled_invalidatepage(struct page *page, | |
3002 | unsigned long offset) | |
3003 | { | |
3004 | WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0); | |
ac27a0ec DK |
3005 | } |
3006 | ||
617ba13b | 3007 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3008 | { |
617ba13b | 3009 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3010 | |
0562e0ba JZ |
3011 | trace_ext4_releasepage(page); |
3012 | ||
e1c36595 JK |
3013 | /* Page has dirty journalled data -> cannot release */ |
3014 | if (PageChecked(page)) | |
ac27a0ec | 3015 | return 0; |
0390131b FM |
3016 | if (journal) |
3017 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3018 | else | |
3019 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3020 | } |
3021 | ||
2ed88685 TT |
3022 | /* |
3023 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3024 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3025 | * The extent will be converted to initialized after the IO is complete. | |
3026 | */ | |
f19d5870 | 3027 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3028 | struct buffer_head *bh_result, int create) |
3029 | { | |
c7064ef1 | 3030 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3031 | inode->i_ino, create); |
2ed88685 TT |
3032 | return _ext4_get_block(inode, iblock, bh_result, |
3033 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3034 | } |
3035 | ||
729f52c6 | 3036 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 3037 | struct buffer_head *bh_result, int create) |
729f52c6 | 3038 | { |
8b0f165f AP |
3039 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
3040 | inode->i_ino, create); | |
3041 | return _ext4_get_block(inode, iblock, bh_result, | |
3042 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
3043 | } |
3044 | ||
4c0425ff | 3045 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
3046 | ssize_t size, void *private, int ret, |
3047 | bool is_async) | |
4c0425ff | 3048 | { |
496ad9aa | 3049 | struct inode *inode = file_inode(iocb->ki_filp); |
4c0425ff | 3050 | ext4_io_end_t *io_end = iocb->private; |
4c0425ff | 3051 | |
4b70df18 M |
3052 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3053 | if (!io_end || !size) | |
552ef802 | 3054 | goto out; |
4b70df18 | 3055 | |
88635ca2 | 3056 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3057 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3058 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3059 | size); | |
8d5d02e6 | 3060 | |
b5a7e970 TT |
3061 | iocb->private = NULL; |
3062 | ||
8d5d02e6 | 3063 | /* if not aio dio with unwritten extents, just free io and return */ |
bd2d0210 | 3064 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 | 3065 | ext4_free_io_end(io_end); |
5b3ff237 | 3066 | out: |
091e26df | 3067 | inode_dio_done(inode); |
5b3ff237 JZ |
3068 | if (is_async) |
3069 | aio_complete(iocb, ret, 0); | |
3070 | return; | |
8d5d02e6 MC |
3071 | } |
3072 | ||
4c0425ff MC |
3073 | io_end->offset = offset; |
3074 | io_end->size = size; | |
5b3ff237 JZ |
3075 | if (is_async) { |
3076 | io_end->iocb = iocb; | |
3077 | io_end->result = ret; | |
3078 | } | |
4c0425ff | 3079 | |
28a535f9 | 3080 | ext4_add_complete_io(io_end); |
4c0425ff | 3081 | } |
c7064ef1 | 3082 | |
4c0425ff MC |
3083 | /* |
3084 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3085 | * preallocated extents, and those write extend the file, no need to | |
3086 | * fall back to buffered IO. | |
3087 | * | |
b595076a | 3088 | * For holes, we fallocate those blocks, mark them as uninitialized |
69c499d1 | 3089 | * If those blocks were preallocated, we mark sure they are split, but |
b595076a | 3090 | * still keep the range to write as uninitialized. |
4c0425ff | 3091 | * |
69c499d1 | 3092 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3093 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3094 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3095 | * when async direct IO completed. |
4c0425ff MC |
3096 | * |
3097 | * If the O_DIRECT write will extend the file then add this inode to the | |
3098 | * orphan list. So recovery will truncate it back to the original size | |
3099 | * if the machine crashes during the write. | |
3100 | * | |
3101 | */ | |
3102 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3103 | const struct iovec *iov, loff_t offset, | |
3104 | unsigned long nr_segs) | |
3105 | { | |
3106 | struct file *file = iocb->ki_filp; | |
3107 | struct inode *inode = file->f_mapping->host; | |
3108 | ssize_t ret; | |
3109 | size_t count = iov_length(iov, nr_segs); | |
69c499d1 TT |
3110 | int overwrite = 0; |
3111 | get_block_t *get_block_func = NULL; | |
3112 | int dio_flags = 0; | |
4c0425ff | 3113 | loff_t final_size = offset + count; |
729f52c6 | 3114 | |
69c499d1 TT |
3115 | /* Use the old path for reads and writes beyond i_size. */ |
3116 | if (rw != WRITE || final_size > inode->i_size) | |
3117 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
4bd809db | 3118 | |
69c499d1 | 3119 | BUG_ON(iocb->private == NULL); |
4bd809db | 3120 | |
69c499d1 TT |
3121 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3122 | overwrite = *((int *)iocb->private); | |
4bd809db | 3123 | |
69c499d1 TT |
3124 | if (overwrite) { |
3125 | atomic_inc(&inode->i_dio_count); | |
3126 | down_read(&EXT4_I(inode)->i_data_sem); | |
3127 | mutex_unlock(&inode->i_mutex); | |
3128 | } | |
8d5d02e6 | 3129 | |
69c499d1 TT |
3130 | /* |
3131 | * We could direct write to holes and fallocate. | |
3132 | * | |
3133 | * Allocated blocks to fill the hole are marked as | |
3134 | * uninitialized to prevent parallel buffered read to expose | |
3135 | * the stale data before DIO complete the data IO. | |
3136 | * | |
3137 | * As to previously fallocated extents, ext4 get_block will | |
3138 | * just simply mark the buffer mapped but still keep the | |
3139 | * extents uninitialized. | |
3140 | * | |
3141 | * For non AIO case, we will convert those unwritten extents | |
3142 | * to written after return back from blockdev_direct_IO. | |
3143 | * | |
3144 | * For async DIO, the conversion needs to be deferred when the | |
3145 | * IO is completed. The ext4 end_io callback function will be | |
3146 | * called to take care of the conversion work. Here for async | |
3147 | * case, we allocate an io_end structure to hook to the iocb. | |
3148 | */ | |
3149 | iocb->private = NULL; | |
3150 | ext4_inode_aio_set(inode, NULL); | |
3151 | if (!is_sync_kiocb(iocb)) { | |
3152 | ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS); | |
3153 | if (!io_end) { | |
3154 | ret = -ENOMEM; | |
3155 | goto retake_lock; | |
8b0f165f | 3156 | } |
69c499d1 TT |
3157 | io_end->flag |= EXT4_IO_END_DIRECT; |
3158 | iocb->private = io_end; | |
8d5d02e6 | 3159 | /* |
69c499d1 TT |
3160 | * we save the io structure for current async direct |
3161 | * IO, so that later ext4_map_blocks() could flag the | |
3162 | * io structure whether there is a unwritten extents | |
3163 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3164 | */ |
69c499d1 TT |
3165 | ext4_inode_aio_set(inode, io_end); |
3166 | } | |
4bd809db | 3167 | |
69c499d1 TT |
3168 | if (overwrite) { |
3169 | get_block_func = ext4_get_block_write_nolock; | |
3170 | } else { | |
3171 | get_block_func = ext4_get_block_write; | |
3172 | dio_flags = DIO_LOCKING; | |
3173 | } | |
3174 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
3175 | inode->i_sb->s_bdev, iov, | |
3176 | offset, nr_segs, | |
3177 | get_block_func, | |
3178 | ext4_end_io_dio, | |
3179 | NULL, | |
3180 | dio_flags); | |
3181 | ||
3182 | if (iocb->private) | |
3183 | ext4_inode_aio_set(inode, NULL); | |
3184 | /* | |
3185 | * The io_end structure takes a reference to the inode, that | |
3186 | * structure needs to be destroyed and the reference to the | |
3187 | * inode need to be dropped, when IO is complete, even with 0 | |
3188 | * byte write, or failed. | |
3189 | * | |
3190 | * In the successful AIO DIO case, the io_end structure will | |
3191 | * be destroyed and the reference to the inode will be dropped | |
3192 | * after the end_io call back function is called. | |
3193 | * | |
3194 | * In the case there is 0 byte write, or error case, since VFS | |
3195 | * direct IO won't invoke the end_io call back function, we | |
3196 | * need to free the end_io structure here. | |
3197 | */ | |
3198 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3199 | ext4_free_io_end(iocb->private); | |
3200 | iocb->private = NULL; | |
3201 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
3202 | EXT4_STATE_DIO_UNWRITTEN)) { | |
3203 | int err; | |
3204 | /* | |
3205 | * for non AIO case, since the IO is already | |
3206 | * completed, we could do the conversion right here | |
3207 | */ | |
3208 | err = ext4_convert_unwritten_extents(inode, | |
3209 | offset, ret); | |
3210 | if (err < 0) | |
3211 | ret = err; | |
3212 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3213 | } | |
4bd809db | 3214 | |
69c499d1 TT |
3215 | retake_lock: |
3216 | /* take i_mutex locking again if we do a ovewrite dio */ | |
3217 | if (overwrite) { | |
3218 | inode_dio_done(inode); | |
3219 | up_read(&EXT4_I(inode)->i_data_sem); | |
3220 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3221 | } |
8d5d02e6 | 3222 | |
69c499d1 | 3223 | return ret; |
4c0425ff MC |
3224 | } |
3225 | ||
3226 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3227 | const struct iovec *iov, loff_t offset, | |
3228 | unsigned long nr_segs) | |
3229 | { | |
3230 | struct file *file = iocb->ki_filp; | |
3231 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3232 | ssize_t ret; |
4c0425ff | 3233 | |
84ebd795 TT |
3234 | /* |
3235 | * If we are doing data journalling we don't support O_DIRECT | |
3236 | */ | |
3237 | if (ext4_should_journal_data(inode)) | |
3238 | return 0; | |
3239 | ||
46c7f254 TM |
3240 | /* Let buffer I/O handle the inline data case. */ |
3241 | if (ext4_has_inline_data(inode)) | |
3242 | return 0; | |
3243 | ||
0562e0ba | 3244 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3245 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3246 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3247 | else | |
3248 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3249 | trace_ext4_direct_IO_exit(inode, offset, | |
3250 | iov_length(iov, nr_segs), rw, ret); | |
3251 | return ret; | |
4c0425ff MC |
3252 | } |
3253 | ||
ac27a0ec | 3254 | /* |
617ba13b | 3255 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3256 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3257 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3258 | * not necessarily locked. | |
3259 | * | |
3260 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3261 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3262 | * or jbddirty because all the journalling code will explode. | |
3263 | * | |
3264 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3265 | * is called, propagate that into the buffers appropriately. | |
3266 | */ | |
617ba13b | 3267 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3268 | { |
3269 | SetPageChecked(page); | |
3270 | return __set_page_dirty_nobuffers(page); | |
3271 | } | |
3272 | ||
74d553aa | 3273 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3274 | .readpage = ext4_readpage, |
3275 | .readpages = ext4_readpages, | |
43ce1d23 | 3276 | .writepage = ext4_writepage, |
8ab22b9a | 3277 | .write_begin = ext4_write_begin, |
74d553aa | 3278 | .write_end = ext4_write_end, |
8ab22b9a HH |
3279 | .bmap = ext4_bmap, |
3280 | .invalidatepage = ext4_invalidatepage, | |
3281 | .releasepage = ext4_releasepage, | |
3282 | .direct_IO = ext4_direct_IO, | |
3283 | .migratepage = buffer_migrate_page, | |
3284 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3285 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3286 | }; |
3287 | ||
617ba13b | 3288 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3289 | .readpage = ext4_readpage, |
3290 | .readpages = ext4_readpages, | |
43ce1d23 | 3291 | .writepage = ext4_writepage, |
8ab22b9a HH |
3292 | .write_begin = ext4_write_begin, |
3293 | .write_end = ext4_journalled_write_end, | |
3294 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3295 | .bmap = ext4_bmap, | |
4520fb3c | 3296 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3297 | .releasepage = ext4_releasepage, |
84ebd795 | 3298 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3299 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3300 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3301 | }; |
3302 | ||
64769240 | 3303 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3304 | .readpage = ext4_readpage, |
3305 | .readpages = ext4_readpages, | |
43ce1d23 | 3306 | .writepage = ext4_writepage, |
8ab22b9a | 3307 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3308 | .write_begin = ext4_da_write_begin, |
3309 | .write_end = ext4_da_write_end, | |
3310 | .bmap = ext4_bmap, | |
3311 | .invalidatepage = ext4_da_invalidatepage, | |
3312 | .releasepage = ext4_releasepage, | |
3313 | .direct_IO = ext4_direct_IO, | |
3314 | .migratepage = buffer_migrate_page, | |
3315 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3316 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3317 | }; |
3318 | ||
617ba13b | 3319 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3320 | { |
3d2b1582 LC |
3321 | switch (ext4_inode_journal_mode(inode)) { |
3322 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3323 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3324 | break; |
3325 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3326 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3327 | break; |
3328 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3329 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3330 | return; |
3d2b1582 LC |
3331 | default: |
3332 | BUG(); | |
3333 | } | |
74d553aa TT |
3334 | if (test_opt(inode->i_sb, DELALLOC)) |
3335 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3336 | else | |
3337 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3338 | } |
3339 | ||
4e96b2db AH |
3340 | |
3341 | /* | |
3342 | * ext4_discard_partial_page_buffers() | |
3343 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3344 | * This function finds and locks the page containing the offset | |
3345 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3346 | * Calling functions that already have the page locked should call | |
3347 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3348 | */ | |
3349 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3350 | struct address_space *mapping, loff_t from, | |
3351 | loff_t length, int flags) | |
3352 | { | |
3353 | struct inode *inode = mapping->host; | |
3354 | struct page *page; | |
3355 | int err = 0; | |
3356 | ||
3357 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3358 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3359 | if (!page) | |
5129d05f | 3360 | return -ENOMEM; |
4e96b2db AH |
3361 | |
3362 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3363 | from, length, flags); | |
3364 | ||
3365 | unlock_page(page); | |
3366 | page_cache_release(page); | |
3367 | return err; | |
3368 | } | |
3369 | ||
3370 | /* | |
3371 | * ext4_discard_partial_page_buffers_no_lock() | |
3372 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3373 | * Buffer heads that correspond to the block aligned regions of the | |
3374 | * zeroed range will be unmapped. Unblock aligned regions | |
3375 | * will have the corresponding buffer head mapped if needed so that | |
3376 | * that region of the page can be updated with the partial zero out. | |
3377 | * | |
3378 | * This function assumes that the page has already been locked. The | |
3379 | * The range to be discarded must be contained with in the given page. | |
3380 | * If the specified range exceeds the end of the page it will be shortened | |
3381 | * to the end of the page that corresponds to 'from'. This function is | |
3382 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3383 | * zeroed for blocks that have been either released, or are going to be | |
3384 | * released. | |
3385 | * | |
3386 | * handle: The journal handle | |
3387 | * inode: The files inode | |
3388 | * page: A locked page that contains the offset "from" | |
4907cb7b | 3389 | * from: The starting byte offset (from the beginning of the file) |
4e96b2db AH |
3390 | * to begin discarding |
3391 | * len: The length of bytes to discard | |
3392 | * flags: Optional flags that may be used: | |
3393 | * | |
3394 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3395 | * Only zero the regions of the page whose buffer heads | |
3396 | * have already been unmapped. This flag is appropriate | |
4907cb7b | 3397 | * for updating the contents of a page whose blocks may |
4e96b2db AH |
3398 | * have already been released, and we only want to zero |
3399 | * out the regions that correspond to those released blocks. | |
3400 | * | |
4907cb7b | 3401 | * Returns zero on success or negative on failure. |
4e96b2db | 3402 | */ |
5f163cc7 | 3403 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
4e96b2db AH |
3404 | struct inode *inode, struct page *page, loff_t from, |
3405 | loff_t length, int flags) | |
3406 | { | |
3407 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3408 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3409 | unsigned int blocksize, max, pos; | |
4e96b2db AH |
3410 | ext4_lblk_t iblock; |
3411 | struct buffer_head *bh; | |
3412 | int err = 0; | |
3413 | ||
3414 | blocksize = inode->i_sb->s_blocksize; | |
3415 | max = PAGE_CACHE_SIZE - offset; | |
3416 | ||
3417 | if (index != page->index) | |
3418 | return -EINVAL; | |
3419 | ||
3420 | /* | |
3421 | * correct length if it does not fall between | |
3422 | * 'from' and the end of the page | |
3423 | */ | |
3424 | if (length > max || length < 0) | |
3425 | length = max; | |
3426 | ||
3427 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3428 | ||
093e6e36 YY |
3429 | if (!page_has_buffers(page)) |
3430 | create_empty_buffers(page, blocksize, 0); | |
4e96b2db AH |
3431 | |
3432 | /* Find the buffer that contains "offset" */ | |
3433 | bh = page_buffers(page); | |
3434 | pos = blocksize; | |
3435 | while (offset >= pos) { | |
3436 | bh = bh->b_this_page; | |
3437 | iblock++; | |
3438 | pos += blocksize; | |
3439 | } | |
3440 | ||
3441 | pos = offset; | |
3442 | while (pos < offset + length) { | |
e260daf2 YY |
3443 | unsigned int end_of_block, range_to_discard; |
3444 | ||
4e96b2db AH |
3445 | err = 0; |
3446 | ||
3447 | /* The length of space left to zero and unmap */ | |
3448 | range_to_discard = offset + length - pos; | |
3449 | ||
3450 | /* The length of space until the end of the block */ | |
3451 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3452 | ||
3453 | /* | |
3454 | * Do not unmap or zero past end of block | |
3455 | * for this buffer head | |
3456 | */ | |
3457 | if (range_to_discard > end_of_block) | |
3458 | range_to_discard = end_of_block; | |
3459 | ||
3460 | ||
3461 | /* | |
3462 | * Skip this buffer head if we are only zeroing unampped | |
3463 | * regions of the page | |
3464 | */ | |
3465 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3466 | buffer_mapped(bh)) | |
3467 | goto next; | |
3468 | ||
3469 | /* If the range is block aligned, unmap */ | |
3470 | if (range_to_discard == blocksize) { | |
3471 | clear_buffer_dirty(bh); | |
3472 | bh->b_bdev = NULL; | |
3473 | clear_buffer_mapped(bh); | |
3474 | clear_buffer_req(bh); | |
3475 | clear_buffer_new(bh); | |
3476 | clear_buffer_delay(bh); | |
3477 | clear_buffer_unwritten(bh); | |
3478 | clear_buffer_uptodate(bh); | |
3479 | zero_user(page, pos, range_to_discard); | |
3480 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3481 | goto next; | |
3482 | } | |
3483 | ||
3484 | /* | |
3485 | * If this block is not completely contained in the range | |
3486 | * to be discarded, then it is not going to be released. Because | |
3487 | * we need to keep this block, we need to make sure this part | |
3488 | * of the page is uptodate before we modify it by writeing | |
3489 | * partial zeros on it. | |
3490 | */ | |
3491 | if (!buffer_mapped(bh)) { | |
3492 | /* | |
3493 | * Buffer head must be mapped before we can read | |
3494 | * from the block | |
3495 | */ | |
3496 | BUFFER_TRACE(bh, "unmapped"); | |
3497 | ext4_get_block(inode, iblock, bh, 0); | |
3498 | /* unmapped? It's a hole - nothing to do */ | |
3499 | if (!buffer_mapped(bh)) { | |
3500 | BUFFER_TRACE(bh, "still unmapped"); | |
3501 | goto next; | |
3502 | } | |
3503 | } | |
3504 | ||
3505 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3506 | if (PageUptodate(page)) | |
3507 | set_buffer_uptodate(bh); | |
3508 | ||
3509 | if (!buffer_uptodate(bh)) { | |
3510 | err = -EIO; | |
3511 | ll_rw_block(READ, 1, &bh); | |
3512 | wait_on_buffer(bh); | |
3513 | /* Uhhuh. Read error. Complain and punt.*/ | |
3514 | if (!buffer_uptodate(bh)) | |
3515 | goto next; | |
3516 | } | |
3517 | ||
3518 | if (ext4_should_journal_data(inode)) { | |
3519 | BUFFER_TRACE(bh, "get write access"); | |
3520 | err = ext4_journal_get_write_access(handle, bh); | |
3521 | if (err) | |
3522 | goto next; | |
3523 | } | |
3524 | ||
3525 | zero_user(page, pos, range_to_discard); | |
3526 | ||
3527 | err = 0; | |
3528 | if (ext4_should_journal_data(inode)) { | |
3529 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3530 | } else |
4e96b2db | 3531 | mark_buffer_dirty(bh); |
4e96b2db AH |
3532 | |
3533 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3534 | next: | |
3535 | bh = bh->b_this_page; | |
3536 | iblock++; | |
3537 | pos += range_to_discard; | |
3538 | } | |
3539 | ||
3540 | return err; | |
3541 | } | |
3542 | ||
91ef4caf DG |
3543 | int ext4_can_truncate(struct inode *inode) |
3544 | { | |
91ef4caf DG |
3545 | if (S_ISREG(inode->i_mode)) |
3546 | return 1; | |
3547 | if (S_ISDIR(inode->i_mode)) | |
3548 | return 1; | |
3549 | if (S_ISLNK(inode->i_mode)) | |
3550 | return !ext4_inode_is_fast_symlink(inode); | |
3551 | return 0; | |
3552 | } | |
3553 | ||
a4bb6b64 AH |
3554 | /* |
3555 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3556 | * associated with the given offset and length | |
3557 | * | |
3558 | * @inode: File inode | |
3559 | * @offset: The offset where the hole will begin | |
3560 | * @len: The length of the hole | |
3561 | * | |
4907cb7b | 3562 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3563 | */ |
3564 | ||
3565 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3566 | { | |
496ad9aa | 3567 | struct inode *inode = file_inode(file); |
26a4c0c6 TT |
3568 | struct super_block *sb = inode->i_sb; |
3569 | ext4_lblk_t first_block, stop_block; | |
3570 | struct address_space *mapping = inode->i_mapping; | |
3571 | loff_t first_page, last_page, page_len; | |
3572 | loff_t first_page_offset, last_page_offset; | |
3573 | handle_t *handle; | |
3574 | unsigned int credits; | |
3575 | int ret = 0; | |
3576 | ||
a4bb6b64 | 3577 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3578 | return -EOPNOTSUPP; |
a4bb6b64 | 3579 | |
26a4c0c6 | 3580 | if (EXT4_SB(sb)->s_cluster_ratio > 1) { |
bab08ab9 | 3581 | /* TODO: Add support for bigalloc file systems */ |
73355192 | 3582 | return -EOPNOTSUPP; |
bab08ab9 TT |
3583 | } |
3584 | ||
aaddea81 ZL |
3585 | trace_ext4_punch_hole(inode, offset, length); |
3586 | ||
26a4c0c6 TT |
3587 | /* |
3588 | * Write out all dirty pages to avoid race conditions | |
3589 | * Then release them. | |
3590 | */ | |
3591 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3592 | ret = filemap_write_and_wait_range(mapping, offset, | |
3593 | offset + length - 1); | |
3594 | if (ret) | |
3595 | return ret; | |
3596 | } | |
3597 | ||
3598 | mutex_lock(&inode->i_mutex); | |
3599 | /* It's not possible punch hole on append only file */ | |
3600 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
3601 | ret = -EPERM; | |
3602 | goto out_mutex; | |
3603 | } | |
3604 | if (IS_SWAPFILE(inode)) { | |
3605 | ret = -ETXTBSY; | |
3606 | goto out_mutex; | |
3607 | } | |
3608 | ||
3609 | /* No need to punch hole beyond i_size */ | |
3610 | if (offset >= inode->i_size) | |
3611 | goto out_mutex; | |
3612 | ||
3613 | /* | |
3614 | * If the hole extends beyond i_size, set the hole | |
3615 | * to end after the page that contains i_size | |
3616 | */ | |
3617 | if (offset + length > inode->i_size) { | |
3618 | length = inode->i_size + | |
3619 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3620 | offset; | |
3621 | } | |
3622 | ||
3623 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
3624 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
3625 | ||
3626 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
3627 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
3628 | ||
3629 | /* Now release the pages */ | |
3630 | if (last_page_offset > first_page_offset) { | |
3631 | truncate_pagecache_range(inode, first_page_offset, | |
3632 | last_page_offset - 1); | |
3633 | } | |
3634 | ||
3635 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3636 | ext4_inode_block_unlocked_dio(inode); | |
3637 | ret = ext4_flush_unwritten_io(inode); | |
3638 | if (ret) | |
3639 | goto out_dio; | |
3640 | inode_dio_wait(inode); | |
3641 | ||
3642 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3643 | credits = ext4_writepage_trans_blocks(inode); | |
3644 | else | |
3645 | credits = ext4_blocks_for_truncate(inode); | |
3646 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3647 | if (IS_ERR(handle)) { | |
3648 | ret = PTR_ERR(handle); | |
3649 | ext4_std_error(sb, ret); | |
3650 | goto out_dio; | |
3651 | } | |
3652 | ||
3653 | /* | |
3654 | * Now we need to zero out the non-page-aligned data in the | |
3655 | * pages at the start and tail of the hole, and unmap the | |
3656 | * buffer heads for the block aligned regions of the page that | |
3657 | * were completely zeroed. | |
3658 | */ | |
3659 | if (first_page > last_page) { | |
3660 | /* | |
3661 | * If the file space being truncated is contained | |
3662 | * within a page just zero out and unmap the middle of | |
3663 | * that page | |
3664 | */ | |
3665 | ret = ext4_discard_partial_page_buffers(handle, | |
3666 | mapping, offset, length, 0); | |
3667 | ||
3668 | if (ret) | |
3669 | goto out_stop; | |
3670 | } else { | |
3671 | /* | |
3672 | * zero out and unmap the partial page that contains | |
3673 | * the start of the hole | |
3674 | */ | |
3675 | page_len = first_page_offset - offset; | |
3676 | if (page_len > 0) { | |
3677 | ret = ext4_discard_partial_page_buffers(handle, mapping, | |
3678 | offset, page_len, 0); | |
3679 | if (ret) | |
3680 | goto out_stop; | |
3681 | } | |
3682 | ||
3683 | /* | |
3684 | * zero out and unmap the partial page that contains | |
3685 | * the end of the hole | |
3686 | */ | |
3687 | page_len = offset + length - last_page_offset; | |
3688 | if (page_len > 0) { | |
3689 | ret = ext4_discard_partial_page_buffers(handle, mapping, | |
3690 | last_page_offset, page_len, 0); | |
3691 | if (ret) | |
3692 | goto out_stop; | |
3693 | } | |
3694 | } | |
3695 | ||
3696 | /* | |
3697 | * If i_size is contained in the last page, we need to | |
3698 | * unmap and zero the partial page after i_size | |
3699 | */ | |
3700 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
3701 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
3702 | page_len = PAGE_CACHE_SIZE - | |
3703 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3704 | ||
3705 | if (page_len > 0) { | |
3706 | ret = ext4_discard_partial_page_buffers(handle, | |
3707 | mapping, inode->i_size, page_len, 0); | |
3708 | ||
3709 | if (ret) | |
3710 | goto out_stop; | |
3711 | } | |
3712 | } | |
3713 | ||
3714 | first_block = (offset + sb->s_blocksize - 1) >> | |
3715 | EXT4_BLOCK_SIZE_BITS(sb); | |
3716 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3717 | ||
3718 | /* If there are no blocks to remove, return now */ | |
3719 | if (first_block >= stop_block) | |
3720 | goto out_stop; | |
3721 | ||
3722 | down_write(&EXT4_I(inode)->i_data_sem); | |
3723 | ext4_discard_preallocations(inode); | |
3724 | ||
3725 | ret = ext4_es_remove_extent(inode, first_block, | |
3726 | stop_block - first_block); | |
3727 | if (ret) { | |
3728 | up_write(&EXT4_I(inode)->i_data_sem); | |
3729 | goto out_stop; | |
3730 | } | |
3731 | ||
3732 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3733 | ret = ext4_ext_remove_space(inode, first_block, | |
3734 | stop_block - 1); | |
3735 | else | |
3736 | ret = ext4_free_hole_blocks(handle, inode, first_block, | |
3737 | stop_block); | |
3738 | ||
3739 | ext4_discard_preallocations(inode); | |
819c4920 | 3740 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3741 | if (IS_SYNC(inode)) |
3742 | ext4_handle_sync(handle); | |
26a4c0c6 TT |
3743 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3744 | ext4_mark_inode_dirty(handle, inode); | |
3745 | out_stop: | |
3746 | ext4_journal_stop(handle); | |
3747 | out_dio: | |
3748 | ext4_inode_resume_unlocked_dio(inode); | |
3749 | out_mutex: | |
3750 | mutex_unlock(&inode->i_mutex); | |
3751 | return ret; | |
a4bb6b64 AH |
3752 | } |
3753 | ||
ac27a0ec | 3754 | /* |
617ba13b | 3755 | * ext4_truncate() |
ac27a0ec | 3756 | * |
617ba13b MC |
3757 | * We block out ext4_get_block() block instantiations across the entire |
3758 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3759 | * simultaneously on behalf of the same inode. |
3760 | * | |
42b2aa86 | 3761 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3762 | * is one core, guiding principle: the file's tree must always be consistent on |
3763 | * disk. We must be able to restart the truncate after a crash. | |
3764 | * | |
3765 | * The file's tree may be transiently inconsistent in memory (although it | |
3766 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3767 | * the contents of (the filesystem + the journal) must be consistent and | |
3768 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3769 | * left-to-right works OK too). | |
3770 | * | |
3771 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3772 | * truncate against the orphan inode list. | |
3773 | * | |
3774 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3775 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3776 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3777 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3778 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3779 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3780 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3781 | */ |
617ba13b | 3782 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3783 | { |
819c4920 TT |
3784 | struct ext4_inode_info *ei = EXT4_I(inode); |
3785 | unsigned int credits; | |
3786 | handle_t *handle; | |
3787 | struct address_space *mapping = inode->i_mapping; | |
3788 | loff_t page_len; | |
3789 | ||
19b5ef61 TT |
3790 | /* |
3791 | * There is a possibility that we're either freeing the inode | |
3792 | * or it completely new indode. In those cases we might not | |
3793 | * have i_mutex locked because it's not necessary. | |
3794 | */ | |
3795 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3796 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3797 | trace_ext4_truncate_enter(inode); |
3798 | ||
91ef4caf | 3799 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3800 | return; |
3801 | ||
12e9b892 | 3802 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3803 | |
5534fb5b | 3804 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3805 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3806 | |
aef1c851 TM |
3807 | if (ext4_has_inline_data(inode)) { |
3808 | int has_inline = 1; | |
3809 | ||
3810 | ext4_inline_data_truncate(inode, &has_inline); | |
3811 | if (has_inline) | |
3812 | return; | |
3813 | } | |
3814 | ||
819c4920 TT |
3815 | /* |
3816 | * finish any pending end_io work so we won't run the risk of | |
3817 | * converting any truncated blocks to initialized later | |
3818 | */ | |
3819 | ext4_flush_unwritten_io(inode); | |
3820 | ||
3821 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3822 | credits = ext4_writepage_trans_blocks(inode); | |
3823 | else | |
3824 | credits = ext4_blocks_for_truncate(inode); | |
3825 | ||
3826 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3827 | if (IS_ERR(handle)) { | |
3828 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3829 | return; | |
3830 | } | |
3831 | ||
3832 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { | |
3833 | page_len = PAGE_CACHE_SIZE - | |
3834 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3835 | ||
3836 | if (ext4_discard_partial_page_buffers(handle, | |
3837 | mapping, inode->i_size, page_len, 0)) | |
3838 | goto out_stop; | |
3839 | } | |
3840 | ||
3841 | /* | |
3842 | * We add the inode to the orphan list, so that if this | |
3843 | * truncate spans multiple transactions, and we crash, we will | |
3844 | * resume the truncate when the filesystem recovers. It also | |
3845 | * marks the inode dirty, to catch the new size. | |
3846 | * | |
3847 | * Implication: the file must always be in a sane, consistent | |
3848 | * truncatable state while each transaction commits. | |
3849 | */ | |
3850 | if (ext4_orphan_add(handle, inode)) | |
3851 | goto out_stop; | |
3852 | ||
3853 | down_write(&EXT4_I(inode)->i_data_sem); | |
3854 | ||
3855 | ext4_discard_preallocations(inode); | |
3856 | ||
ff9893dc | 3857 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3858 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3859 | else |
819c4920 TT |
3860 | ext4_ind_truncate(handle, inode); |
3861 | ||
3862 | up_write(&ei->i_data_sem); | |
3863 | ||
3864 | if (IS_SYNC(inode)) | |
3865 | ext4_handle_sync(handle); | |
3866 | ||
3867 | out_stop: | |
3868 | /* | |
3869 | * If this was a simple ftruncate() and the file will remain alive, | |
3870 | * then we need to clear up the orphan record which we created above. | |
3871 | * However, if this was a real unlink then we were called by | |
3872 | * ext4_delete_inode(), and we allow that function to clean up the | |
3873 | * orphan info for us. | |
3874 | */ | |
3875 | if (inode->i_nlink) | |
3876 | ext4_orphan_del(handle, inode); | |
3877 | ||
3878 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3879 | ext4_mark_inode_dirty(handle, inode); | |
3880 | ext4_journal_stop(handle); | |
ac27a0ec | 3881 | |
0562e0ba | 3882 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3883 | } |
3884 | ||
ac27a0ec | 3885 | /* |
617ba13b | 3886 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3887 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3888 | * data in memory that is needed to recreate the on-disk version of this | |
3889 | * inode. | |
3890 | */ | |
617ba13b MC |
3891 | static int __ext4_get_inode_loc(struct inode *inode, |
3892 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3893 | { |
240799cd TT |
3894 | struct ext4_group_desc *gdp; |
3895 | struct buffer_head *bh; | |
3896 | struct super_block *sb = inode->i_sb; | |
3897 | ext4_fsblk_t block; | |
3898 | int inodes_per_block, inode_offset; | |
3899 | ||
3a06d778 | 3900 | iloc->bh = NULL; |
240799cd TT |
3901 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3902 | return -EIO; | |
ac27a0ec | 3903 | |
240799cd TT |
3904 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3905 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3906 | if (!gdp) | |
ac27a0ec DK |
3907 | return -EIO; |
3908 | ||
240799cd TT |
3909 | /* |
3910 | * Figure out the offset within the block group inode table | |
3911 | */ | |
00d09882 | 3912 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3913 | inode_offset = ((inode->i_ino - 1) % |
3914 | EXT4_INODES_PER_GROUP(sb)); | |
3915 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3916 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3917 | ||
3918 | bh = sb_getblk(sb, block); | |
aebf0243 | 3919 | if (unlikely(!bh)) |
860d21e2 | 3920 | return -ENOMEM; |
ac27a0ec DK |
3921 | if (!buffer_uptodate(bh)) { |
3922 | lock_buffer(bh); | |
9c83a923 HK |
3923 | |
3924 | /* | |
3925 | * If the buffer has the write error flag, we have failed | |
3926 | * to write out another inode in the same block. In this | |
3927 | * case, we don't have to read the block because we may | |
3928 | * read the old inode data successfully. | |
3929 | */ | |
3930 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3931 | set_buffer_uptodate(bh); | |
3932 | ||
ac27a0ec DK |
3933 | if (buffer_uptodate(bh)) { |
3934 | /* someone brought it uptodate while we waited */ | |
3935 | unlock_buffer(bh); | |
3936 | goto has_buffer; | |
3937 | } | |
3938 | ||
3939 | /* | |
3940 | * If we have all information of the inode in memory and this | |
3941 | * is the only valid inode in the block, we need not read the | |
3942 | * block. | |
3943 | */ | |
3944 | if (in_mem) { | |
3945 | struct buffer_head *bitmap_bh; | |
240799cd | 3946 | int i, start; |
ac27a0ec | 3947 | |
240799cd | 3948 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3949 | |
240799cd TT |
3950 | /* Is the inode bitmap in cache? */ |
3951 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3952 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3953 | goto make_io; |
3954 | ||
3955 | /* | |
3956 | * If the inode bitmap isn't in cache then the | |
3957 | * optimisation may end up performing two reads instead | |
3958 | * of one, so skip it. | |
3959 | */ | |
3960 | if (!buffer_uptodate(bitmap_bh)) { | |
3961 | brelse(bitmap_bh); | |
3962 | goto make_io; | |
3963 | } | |
240799cd | 3964 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3965 | if (i == inode_offset) |
3966 | continue; | |
617ba13b | 3967 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3968 | break; |
3969 | } | |
3970 | brelse(bitmap_bh); | |
240799cd | 3971 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3972 | /* all other inodes are free, so skip I/O */ |
3973 | memset(bh->b_data, 0, bh->b_size); | |
3974 | set_buffer_uptodate(bh); | |
3975 | unlock_buffer(bh); | |
3976 | goto has_buffer; | |
3977 | } | |
3978 | } | |
3979 | ||
3980 | make_io: | |
240799cd TT |
3981 | /* |
3982 | * If we need to do any I/O, try to pre-readahead extra | |
3983 | * blocks from the inode table. | |
3984 | */ | |
3985 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3986 | ext4_fsblk_t b, end, table; | |
3987 | unsigned num; | |
3988 | ||
3989 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3990 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
3991 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
3992 | if (table > b) | |
3993 | b = table; | |
3994 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
3995 | num = EXT4_INODES_PER_GROUP(sb); | |
feb0ab32 | 3996 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 3997 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3998 | table += num / inodes_per_block; |
3999 | if (end > table) | |
4000 | end = table; | |
4001 | while (b <= end) | |
4002 | sb_breadahead(sb, b++); | |
4003 | } | |
4004 | ||
ac27a0ec DK |
4005 | /* |
4006 | * There are other valid inodes in the buffer, this inode | |
4007 | * has in-inode xattrs, or we don't have this inode in memory. | |
4008 | * Read the block from disk. | |
4009 | */ | |
0562e0ba | 4010 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4011 | get_bh(bh); |
4012 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 4013 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4014 | wait_on_buffer(bh); |
4015 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4016 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4017 | "unable to read itable block"); | |
ac27a0ec DK |
4018 | brelse(bh); |
4019 | return -EIO; | |
4020 | } | |
4021 | } | |
4022 | has_buffer: | |
4023 | iloc->bh = bh; | |
4024 | return 0; | |
4025 | } | |
4026 | ||
617ba13b | 4027 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4028 | { |
4029 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4030 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4031 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4032 | } |
4033 | ||
617ba13b | 4034 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4035 | { |
617ba13b | 4036 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4037 | |
4038 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4039 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4040 | inode->i_flags |= S_SYNC; |
617ba13b | 4041 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4042 | inode->i_flags |= S_APPEND; |
617ba13b | 4043 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4044 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4045 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4046 | inode->i_flags |= S_NOATIME; |
617ba13b | 4047 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4048 | inode->i_flags |= S_DIRSYNC; |
4049 | } | |
4050 | ||
ff9ddf7e JK |
4051 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4052 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4053 | { | |
84a8dce2 DM |
4054 | unsigned int vfs_fl; |
4055 | unsigned long old_fl, new_fl; | |
4056 | ||
4057 | do { | |
4058 | vfs_fl = ei->vfs_inode.i_flags; | |
4059 | old_fl = ei->i_flags; | |
4060 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4061 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4062 | EXT4_DIRSYNC_FL); | |
4063 | if (vfs_fl & S_SYNC) | |
4064 | new_fl |= EXT4_SYNC_FL; | |
4065 | if (vfs_fl & S_APPEND) | |
4066 | new_fl |= EXT4_APPEND_FL; | |
4067 | if (vfs_fl & S_IMMUTABLE) | |
4068 | new_fl |= EXT4_IMMUTABLE_FL; | |
4069 | if (vfs_fl & S_NOATIME) | |
4070 | new_fl |= EXT4_NOATIME_FL; | |
4071 | if (vfs_fl & S_DIRSYNC) | |
4072 | new_fl |= EXT4_DIRSYNC_FL; | |
4073 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4074 | } |
de9a55b8 | 4075 | |
0fc1b451 | 4076 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4077 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4078 | { |
4079 | blkcnt_t i_blocks ; | |
8180a562 AK |
4080 | struct inode *inode = &(ei->vfs_inode); |
4081 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4082 | |
4083 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4084 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4085 | /* we are using combined 48 bit field */ | |
4086 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4087 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4088 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4089 | /* i_blocks represent file system block size */ |
4090 | return i_blocks << (inode->i_blkbits - 9); | |
4091 | } else { | |
4092 | return i_blocks; | |
4093 | } | |
0fc1b451 AK |
4094 | } else { |
4095 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4096 | } | |
4097 | } | |
ff9ddf7e | 4098 | |
152a7b0a TM |
4099 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4100 | struct ext4_inode *raw_inode, | |
4101 | struct ext4_inode_info *ei) | |
4102 | { | |
4103 | __le32 *magic = (void *)raw_inode + | |
4104 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4105 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4106 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4107 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4108 | } else |
4109 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4110 | } |
4111 | ||
1d1fe1ee | 4112 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4113 | { |
617ba13b MC |
4114 | struct ext4_iloc iloc; |
4115 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4116 | struct ext4_inode_info *ei; |
1d1fe1ee | 4117 | struct inode *inode; |
b436b9be | 4118 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4119 | long ret; |
ac27a0ec | 4120 | int block; |
08cefc7a EB |
4121 | uid_t i_uid; |
4122 | gid_t i_gid; | |
ac27a0ec | 4123 | |
1d1fe1ee DH |
4124 | inode = iget_locked(sb, ino); |
4125 | if (!inode) | |
4126 | return ERR_PTR(-ENOMEM); | |
4127 | if (!(inode->i_state & I_NEW)) | |
4128 | return inode; | |
4129 | ||
4130 | ei = EXT4_I(inode); | |
7dc57615 | 4131 | iloc.bh = NULL; |
ac27a0ec | 4132 | |
1d1fe1ee DH |
4133 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4134 | if (ret < 0) | |
ac27a0ec | 4135 | goto bad_inode; |
617ba13b | 4136 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4137 | |
4138 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4139 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4140 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4141 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4142 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4143 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4144 | EXT4_INODE_SIZE(inode->i_sb)); | |
4145 | ret = -EIO; | |
4146 | goto bad_inode; | |
4147 | } | |
4148 | } else | |
4149 | ei->i_extra_isize = 0; | |
4150 | ||
4151 | /* Precompute checksum seed for inode metadata */ | |
4152 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4153 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
4154 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4155 | __u32 csum; | |
4156 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4157 | __le32 gen = raw_inode->i_generation; | |
4158 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4159 | sizeof(inum)); | |
4160 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4161 | sizeof(gen)); | |
4162 | } | |
4163 | ||
4164 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4165 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
4166 | ret = -EIO; | |
4167 | goto bad_inode; | |
4168 | } | |
4169 | ||
ac27a0ec | 4170 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4171 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4172 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4173 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4174 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4175 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4176 | } |
08cefc7a EB |
4177 | i_uid_write(inode, i_uid); |
4178 | i_gid_write(inode, i_gid); | |
bfe86848 | 4179 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4180 | |
353eb83c | 4181 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4182 | ei->i_inline_off = 0; |
ac27a0ec DK |
4183 | ei->i_dir_start_lookup = 0; |
4184 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4185 | /* We now have enough fields to check if the inode was active or not. | |
4186 | * This is needed because nfsd might try to access dead inodes | |
4187 | * the test is that same one that e2fsck uses | |
4188 | * NeilBrown 1999oct15 | |
4189 | */ | |
4190 | if (inode->i_nlink == 0) { | |
4191 | if (inode->i_mode == 0 || | |
617ba13b | 4192 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 4193 | /* this inode is deleted */ |
1d1fe1ee | 4194 | ret = -ESTALE; |
ac27a0ec DK |
4195 | goto bad_inode; |
4196 | } | |
4197 | /* The only unlinked inodes we let through here have | |
4198 | * valid i_mode and are being read by the orphan | |
4199 | * recovery code: that's fine, we're about to complete | |
4200 | * the process of deleting those. */ | |
4201 | } | |
ac27a0ec | 4202 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4203 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4204 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4205 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4206 | ei->i_file_acl |= |
4207 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4208 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4209 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4210 | #ifdef CONFIG_QUOTA |
4211 | ei->i_reserved_quota = 0; | |
4212 | #endif | |
ac27a0ec DK |
4213 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4214 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4215 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4216 | /* |
4217 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4218 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4219 | */ | |
617ba13b | 4220 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4221 | ei->i_data[block] = raw_inode->i_block[block]; |
4222 | INIT_LIST_HEAD(&ei->i_orphan); | |
4223 | ||
b436b9be JK |
4224 | /* |
4225 | * Set transaction id's of transactions that have to be committed | |
4226 | * to finish f[data]sync. We set them to currently running transaction | |
4227 | * as we cannot be sure that the inode or some of its metadata isn't | |
4228 | * part of the transaction - the inode could have been reclaimed and | |
4229 | * now it is reread from disk. | |
4230 | */ | |
4231 | if (journal) { | |
4232 | transaction_t *transaction; | |
4233 | tid_t tid; | |
4234 | ||
a931da6a | 4235 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4236 | if (journal->j_running_transaction) |
4237 | transaction = journal->j_running_transaction; | |
4238 | else | |
4239 | transaction = journal->j_committing_transaction; | |
4240 | if (transaction) | |
4241 | tid = transaction->t_tid; | |
4242 | else | |
4243 | tid = journal->j_commit_sequence; | |
a931da6a | 4244 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4245 | ei->i_sync_tid = tid; |
4246 | ei->i_datasync_tid = tid; | |
4247 | } | |
4248 | ||
0040d987 | 4249 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4250 | if (ei->i_extra_isize == 0) { |
4251 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4252 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4253 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4254 | } else { |
152a7b0a | 4255 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4256 | } |
814525f4 | 4257 | } |
ac27a0ec | 4258 | |
ef7f3835 KS |
4259 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4260 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4261 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4262 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4263 | ||
25ec56b5 JNC |
4264 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4265 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4266 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4267 | inode->i_version |= | |
4268 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4269 | } | |
4270 | ||
c4b5a614 | 4271 | ret = 0; |
485c26ec | 4272 | if (ei->i_file_acl && |
1032988c | 4273 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4274 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4275 | ei->i_file_acl); | |
485c26ec TT |
4276 | ret = -EIO; |
4277 | goto bad_inode; | |
f19d5870 TM |
4278 | } else if (!ext4_has_inline_data(inode)) { |
4279 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4280 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4281 | (S_ISLNK(inode->i_mode) && | |
4282 | !ext4_inode_is_fast_symlink(inode)))) | |
4283 | /* Validate extent which is part of inode */ | |
4284 | ret = ext4_ext_check_inode(inode); | |
4285 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4286 | (S_ISLNK(inode->i_mode) && | |
4287 | !ext4_inode_is_fast_symlink(inode))) { | |
4288 | /* Validate block references which are part of inode */ | |
4289 | ret = ext4_ind_check_inode(inode); | |
4290 | } | |
fe2c8191 | 4291 | } |
567f3e9a | 4292 | if (ret) |
de9a55b8 | 4293 | goto bad_inode; |
7a262f7c | 4294 | |
ac27a0ec | 4295 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4296 | inode->i_op = &ext4_file_inode_operations; |
4297 | inode->i_fop = &ext4_file_operations; | |
4298 | ext4_set_aops(inode); | |
ac27a0ec | 4299 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4300 | inode->i_op = &ext4_dir_inode_operations; |
4301 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4302 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4303 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4304 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4305 | nd_terminate_link(ei->i_data, inode->i_size, |
4306 | sizeof(ei->i_data) - 1); | |
4307 | } else { | |
617ba13b MC |
4308 | inode->i_op = &ext4_symlink_inode_operations; |
4309 | ext4_set_aops(inode); | |
ac27a0ec | 4310 | } |
563bdd61 TT |
4311 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4312 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4313 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4314 | if (raw_inode->i_block[0]) |
4315 | init_special_inode(inode, inode->i_mode, | |
4316 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4317 | else | |
4318 | init_special_inode(inode, inode->i_mode, | |
4319 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 4320 | } else { |
563bdd61 | 4321 | ret = -EIO; |
24676da4 | 4322 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4323 | goto bad_inode; |
ac27a0ec | 4324 | } |
af5bc92d | 4325 | brelse(iloc.bh); |
617ba13b | 4326 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4327 | unlock_new_inode(inode); |
4328 | return inode; | |
ac27a0ec DK |
4329 | |
4330 | bad_inode: | |
567f3e9a | 4331 | brelse(iloc.bh); |
1d1fe1ee DH |
4332 | iget_failed(inode); |
4333 | return ERR_PTR(ret); | |
ac27a0ec DK |
4334 | } |
4335 | ||
0fc1b451 AK |
4336 | static int ext4_inode_blocks_set(handle_t *handle, |
4337 | struct ext4_inode *raw_inode, | |
4338 | struct ext4_inode_info *ei) | |
4339 | { | |
4340 | struct inode *inode = &(ei->vfs_inode); | |
4341 | u64 i_blocks = inode->i_blocks; | |
4342 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4343 | |
4344 | if (i_blocks <= ~0U) { | |
4345 | /* | |
4907cb7b | 4346 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4347 | * as multiple of 512 bytes |
4348 | */ | |
8180a562 | 4349 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4350 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4351 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4352 | return 0; |
4353 | } | |
4354 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4355 | return -EFBIG; | |
4356 | ||
4357 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4358 | /* |
4359 | * i_blocks can be represented in a 48 bit variable | |
4360 | * as multiple of 512 bytes | |
4361 | */ | |
8180a562 | 4362 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4363 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4364 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4365 | } else { |
84a8dce2 | 4366 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4367 | /* i_block is stored in file system block size */ |
4368 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4369 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4370 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4371 | } |
f287a1a5 | 4372 | return 0; |
0fc1b451 AK |
4373 | } |
4374 | ||
ac27a0ec DK |
4375 | /* |
4376 | * Post the struct inode info into an on-disk inode location in the | |
4377 | * buffer-cache. This gobbles the caller's reference to the | |
4378 | * buffer_head in the inode location struct. | |
4379 | * | |
4380 | * The caller must have write access to iloc->bh. | |
4381 | */ | |
617ba13b | 4382 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4383 | struct inode *inode, |
830156c7 | 4384 | struct ext4_iloc *iloc) |
ac27a0ec | 4385 | { |
617ba13b MC |
4386 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4387 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4388 | struct buffer_head *bh = iloc->bh; |
4389 | int err = 0, rc, block; | |
b71fc079 | 4390 | int need_datasync = 0; |
08cefc7a EB |
4391 | uid_t i_uid; |
4392 | gid_t i_gid; | |
ac27a0ec DK |
4393 | |
4394 | /* For fields not not tracking in the in-memory inode, | |
4395 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 4396 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4397 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4398 | |
ff9ddf7e | 4399 | ext4_get_inode_flags(ei); |
ac27a0ec | 4400 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4401 | i_uid = i_uid_read(inode); |
4402 | i_gid = i_gid_read(inode); | |
af5bc92d | 4403 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4404 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4405 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4406 | /* |
4407 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4408 | * re-used with the upper 16 bits of the uid/gid intact | |
4409 | */ | |
af5bc92d | 4410 | if (!ei->i_dtime) { |
ac27a0ec | 4411 | raw_inode->i_uid_high = |
08cefc7a | 4412 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4413 | raw_inode->i_gid_high = |
08cefc7a | 4414 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4415 | } else { |
4416 | raw_inode->i_uid_high = 0; | |
4417 | raw_inode->i_gid_high = 0; | |
4418 | } | |
4419 | } else { | |
08cefc7a EB |
4420 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4421 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4422 | raw_inode->i_uid_high = 0; |
4423 | raw_inode->i_gid_high = 0; | |
4424 | } | |
4425 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4426 | |
4427 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4428 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4429 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4430 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4431 | ||
0fc1b451 AK |
4432 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4433 | goto out_brelse; | |
ac27a0ec | 4434 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4435 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4436 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4437 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4438 | raw_inode->i_file_acl_high = |
4439 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4440 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4441 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4442 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4443 | need_datasync = 1; | |
4444 | } | |
a48380f7 AK |
4445 | if (ei->i_disksize > 0x7fffffffULL) { |
4446 | struct super_block *sb = inode->i_sb; | |
4447 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4448 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4449 | EXT4_SB(sb)->s_es->s_rev_level == | |
4450 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4451 | /* If this is the first large file | |
4452 | * created, add a flag to the superblock. | |
4453 | */ | |
4454 | err = ext4_journal_get_write_access(handle, | |
4455 | EXT4_SB(sb)->s_sbh); | |
4456 | if (err) | |
4457 | goto out_brelse; | |
4458 | ext4_update_dynamic_rev(sb); | |
4459 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4460 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
0390131b | 4461 | ext4_handle_sync(handle); |
b50924c2 | 4462 | err = ext4_handle_dirty_super(handle, sb); |
ac27a0ec DK |
4463 | } |
4464 | } | |
4465 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4466 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4467 | if (old_valid_dev(inode->i_rdev)) { | |
4468 | raw_inode->i_block[0] = | |
4469 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4470 | raw_inode->i_block[1] = 0; | |
4471 | } else { | |
4472 | raw_inode->i_block[0] = 0; | |
4473 | raw_inode->i_block[1] = | |
4474 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4475 | raw_inode->i_block[2] = 0; | |
4476 | } | |
f19d5870 | 4477 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4478 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4479 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4480 | } |
ac27a0ec | 4481 | |
25ec56b5 JNC |
4482 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4483 | if (ei->i_extra_isize) { | |
4484 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4485 | raw_inode->i_version_hi = | |
4486 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4487 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4488 | } |
4489 | ||
814525f4 DW |
4490 | ext4_inode_csum_set(inode, raw_inode, ei); |
4491 | ||
830156c7 | 4492 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4493 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4494 | if (!err) |
4495 | err = rc; | |
19f5fb7a | 4496 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4497 | |
b71fc079 | 4498 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4499 | out_brelse: |
af5bc92d | 4500 | brelse(bh); |
617ba13b | 4501 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4502 | return err; |
4503 | } | |
4504 | ||
4505 | /* | |
617ba13b | 4506 | * ext4_write_inode() |
ac27a0ec DK |
4507 | * |
4508 | * We are called from a few places: | |
4509 | * | |
4510 | * - Within generic_file_write() for O_SYNC files. | |
4511 | * Here, there will be no transaction running. We wait for any running | |
4907cb7b | 4512 | * transaction to commit. |
ac27a0ec DK |
4513 | * |
4514 | * - Within sys_sync(), kupdate and such. | |
4515 | * We wait on commit, if tol to. | |
4516 | * | |
4517 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4518 | * Here we simply return. We can't afford to block kswapd on the | |
4519 | * journal commit. | |
4520 | * | |
4521 | * In all cases it is actually safe for us to return without doing anything, | |
4522 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4523 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4524 | * knfsd. |
4525 | * | |
4526 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4527 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4528 | * which we are interested. | |
4529 | * | |
4530 | * It would be a bug for them to not do this. The code: | |
4531 | * | |
4532 | * mark_inode_dirty(inode) | |
4533 | * stuff(); | |
4534 | * inode->i_size = expr; | |
4535 | * | |
4536 | * is in error because a kswapd-driven write_inode() could occur while | |
4537 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4538 | * will no longer be on the superblock's dirty inode list. | |
4539 | */ | |
a9185b41 | 4540 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4541 | { |
91ac6f43 FM |
4542 | int err; |
4543 | ||
ac27a0ec DK |
4544 | if (current->flags & PF_MEMALLOC) |
4545 | return 0; | |
4546 | ||
91ac6f43 FM |
4547 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4548 | if (ext4_journal_current_handle()) { | |
4549 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4550 | dump_stack(); | |
4551 | return -EIO; | |
4552 | } | |
ac27a0ec | 4553 | |
a9185b41 | 4554 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4555 | return 0; |
4556 | ||
4557 | err = ext4_force_commit(inode->i_sb); | |
4558 | } else { | |
4559 | struct ext4_iloc iloc; | |
ac27a0ec | 4560 | |
8b472d73 | 4561 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4562 | if (err) |
4563 | return err; | |
a9185b41 | 4564 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4565 | sync_dirty_buffer(iloc.bh); |
4566 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4567 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4568 | "IO error syncing inode"); | |
830156c7 FM |
4569 | err = -EIO; |
4570 | } | |
fd2dd9fb | 4571 | brelse(iloc.bh); |
91ac6f43 FM |
4572 | } |
4573 | return err; | |
ac27a0ec DK |
4574 | } |
4575 | ||
53e87268 JK |
4576 | /* |
4577 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4578 | * buffers that are attached to a page stradding i_size and are undergoing | |
4579 | * commit. In that case we have to wait for commit to finish and try again. | |
4580 | */ | |
4581 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4582 | { | |
4583 | struct page *page; | |
4584 | unsigned offset; | |
4585 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4586 | tid_t commit_tid = 0; | |
4587 | int ret; | |
4588 | ||
4589 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4590 | /* | |
4591 | * All buffers in the last page remain valid? Then there's nothing to | |
4592 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4593 | * blocksize case | |
4594 | */ | |
4595 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4596 | return; | |
4597 | while (1) { | |
4598 | page = find_lock_page(inode->i_mapping, | |
4599 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4600 | if (!page) | |
4601 | return; | |
4602 | ret = __ext4_journalled_invalidatepage(page, offset); | |
4603 | unlock_page(page); | |
4604 | page_cache_release(page); | |
4605 | if (ret != -EBUSY) | |
4606 | return; | |
4607 | commit_tid = 0; | |
4608 | read_lock(&journal->j_state_lock); | |
4609 | if (journal->j_committing_transaction) | |
4610 | commit_tid = journal->j_committing_transaction->t_tid; | |
4611 | read_unlock(&journal->j_state_lock); | |
4612 | if (commit_tid) | |
4613 | jbd2_log_wait_commit(journal, commit_tid); | |
4614 | } | |
4615 | } | |
4616 | ||
ac27a0ec | 4617 | /* |
617ba13b | 4618 | * ext4_setattr() |
ac27a0ec DK |
4619 | * |
4620 | * Called from notify_change. | |
4621 | * | |
4622 | * We want to trap VFS attempts to truncate the file as soon as | |
4623 | * possible. In particular, we want to make sure that when the VFS | |
4624 | * shrinks i_size, we put the inode on the orphan list and modify | |
4625 | * i_disksize immediately, so that during the subsequent flushing of | |
4626 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4627 | * commit will leave the blocks being flushed in an unused state on | |
4628 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4629 | * be freed, so we have a strong guarantee that no future commit will | |
4630 | * leave these blocks visible to the user.) | |
4631 | * | |
678aaf48 JK |
4632 | * Another thing we have to assure is that if we are in ordered mode |
4633 | * and inode is still attached to the committing transaction, we must | |
4634 | * we start writeout of all the dirty pages which are being truncated. | |
4635 | * This way we are sure that all the data written in the previous | |
4636 | * transaction are already on disk (truncate waits for pages under | |
4637 | * writeback). | |
4638 | * | |
4639 | * Called with inode->i_mutex down. | |
ac27a0ec | 4640 | */ |
617ba13b | 4641 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4642 | { |
4643 | struct inode *inode = dentry->d_inode; | |
4644 | int error, rc = 0; | |
3d287de3 | 4645 | int orphan = 0; |
ac27a0ec DK |
4646 | const unsigned int ia_valid = attr->ia_valid; |
4647 | ||
4648 | error = inode_change_ok(inode, attr); | |
4649 | if (error) | |
4650 | return error; | |
4651 | ||
12755627 | 4652 | if (is_quota_modification(inode, attr)) |
871a2931 | 4653 | dquot_initialize(inode); |
08cefc7a EB |
4654 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4655 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4656 | handle_t *handle; |
4657 | ||
4658 | /* (user+group)*(old+new) structure, inode write (sb, | |
4659 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4660 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4661 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4662 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4663 | if (IS_ERR(handle)) { |
4664 | error = PTR_ERR(handle); | |
4665 | goto err_out; | |
4666 | } | |
b43fa828 | 4667 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4668 | if (error) { |
617ba13b | 4669 | ext4_journal_stop(handle); |
ac27a0ec DK |
4670 | return error; |
4671 | } | |
4672 | /* Update corresponding info in inode so that everything is in | |
4673 | * one transaction */ | |
4674 | if (attr->ia_valid & ATTR_UID) | |
4675 | inode->i_uid = attr->ia_uid; | |
4676 | if (attr->ia_valid & ATTR_GID) | |
4677 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4678 | error = ext4_mark_inode_dirty(handle, inode); |
4679 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4680 | } |
4681 | ||
e2b46574 | 4682 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa | 4683 | |
12e9b892 | 4684 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4685 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4686 | ||
0c095c7f TT |
4687 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4688 | return -EFBIG; | |
e2b46574 ES |
4689 | } |
4690 | } | |
4691 | ||
ac27a0ec | 4692 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4693 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4694 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4695 | handle_t *handle; |
4696 | ||
9924a92a | 4697 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
ac27a0ec DK |
4698 | if (IS_ERR(handle)) { |
4699 | error = PTR_ERR(handle); | |
4700 | goto err_out; | |
4701 | } | |
3d287de3 DM |
4702 | if (ext4_handle_valid(handle)) { |
4703 | error = ext4_orphan_add(handle, inode); | |
4704 | orphan = 1; | |
4705 | } | |
617ba13b MC |
4706 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4707 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4708 | if (!error) |
4709 | error = rc; | |
617ba13b | 4710 | ext4_journal_stop(handle); |
678aaf48 JK |
4711 | |
4712 | if (ext4_should_order_data(inode)) { | |
4713 | error = ext4_begin_ordered_truncate(inode, | |
4714 | attr->ia_size); | |
4715 | if (error) { | |
4716 | /* Do as much error cleanup as possible */ | |
9924a92a TT |
4717 | handle = ext4_journal_start(inode, |
4718 | EXT4_HT_INODE, 3); | |
678aaf48 JK |
4719 | if (IS_ERR(handle)) { |
4720 | ext4_orphan_del(NULL, inode); | |
4721 | goto err_out; | |
4722 | } | |
4723 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4724 | orphan = 0; |
678aaf48 JK |
4725 | ext4_journal_stop(handle); |
4726 | goto err_out; | |
4727 | } | |
4728 | } | |
ac27a0ec DK |
4729 | } |
4730 | ||
072bd7ea | 4731 | if (attr->ia_valid & ATTR_SIZE) { |
53e87268 JK |
4732 | if (attr->ia_size != inode->i_size) { |
4733 | loff_t oldsize = inode->i_size; | |
4734 | ||
4735 | i_size_write(inode, attr->ia_size); | |
4736 | /* | |
4737 | * Blocks are going to be removed from the inode. Wait | |
4738 | * for dio in flight. Temporarily disable | |
4739 | * dioread_nolock to prevent livelock. | |
4740 | */ | |
1b65007e | 4741 | if (orphan) { |
53e87268 JK |
4742 | if (!ext4_should_journal_data(inode)) { |
4743 | ext4_inode_block_unlocked_dio(inode); | |
4744 | inode_dio_wait(inode); | |
4745 | ext4_inode_resume_unlocked_dio(inode); | |
4746 | } else | |
4747 | ext4_wait_for_tail_page_commit(inode); | |
1b65007e | 4748 | } |
53e87268 JK |
4749 | /* |
4750 | * Truncate pagecache after we've waited for commit | |
4751 | * in data=journal mode to make pages freeable. | |
4752 | */ | |
4753 | truncate_pagecache(inode, oldsize, inode->i_size); | |
1c9114f9 | 4754 | } |
afcff5d8 | 4755 | ext4_truncate(inode); |
072bd7ea | 4756 | } |
ac27a0ec | 4757 | |
1025774c CH |
4758 | if (!rc) { |
4759 | setattr_copy(inode, attr); | |
4760 | mark_inode_dirty(inode); | |
4761 | } | |
4762 | ||
4763 | /* | |
4764 | * If the call to ext4_truncate failed to get a transaction handle at | |
4765 | * all, we need to clean up the in-core orphan list manually. | |
4766 | */ | |
3d287de3 | 4767 | if (orphan && inode->i_nlink) |
617ba13b | 4768 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4769 | |
4770 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4771 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4772 | |
4773 | err_out: | |
617ba13b | 4774 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4775 | if (!error) |
4776 | error = rc; | |
4777 | return error; | |
4778 | } | |
4779 | ||
3e3398a0 MC |
4780 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4781 | struct kstat *stat) | |
4782 | { | |
4783 | struct inode *inode; | |
4784 | unsigned long delalloc_blocks; | |
4785 | ||
4786 | inode = dentry->d_inode; | |
4787 | generic_fillattr(inode, stat); | |
4788 | ||
4789 | /* | |
4790 | * We can't update i_blocks if the block allocation is delayed | |
4791 | * otherwise in the case of system crash before the real block | |
4792 | * allocation is done, we will have i_blocks inconsistent with | |
4793 | * on-disk file blocks. | |
4794 | * We always keep i_blocks updated together with real | |
4795 | * allocation. But to not confuse with user, stat | |
4796 | * will return the blocks that include the delayed allocation | |
4797 | * blocks for this file. | |
4798 | */ | |
96607551 TM |
4799 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
4800 | EXT4_I(inode)->i_reserved_data_blocks); | |
3e3398a0 MC |
4801 | |
4802 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4803 | return 0; | |
4804 | } | |
ac27a0ec | 4805 | |
a02908f1 MC |
4806 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4807 | { | |
12e9b892 | 4808 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4809 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4810 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4811 | } |
ac51d837 | 4812 | |
ac27a0ec | 4813 | /* |
a02908f1 MC |
4814 | * Account for index blocks, block groups bitmaps and block group |
4815 | * descriptor blocks if modify datablocks and index blocks | |
4816 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4817 | * |
a02908f1 | 4818 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4819 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4820 | * they could still across block group boundary. |
ac27a0ec | 4821 | * |
a02908f1 MC |
4822 | * Also account for superblock, inode, quota and xattr blocks |
4823 | */ | |
1f109d5a | 4824 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4825 | { |
8df9675f TT |
4826 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4827 | int gdpblocks; | |
a02908f1 MC |
4828 | int idxblocks; |
4829 | int ret = 0; | |
4830 | ||
4831 | /* | |
4832 | * How many index blocks need to touch to modify nrblocks? | |
4833 | * The "Chunk" flag indicating whether the nrblocks is | |
4834 | * physically contiguous on disk | |
4835 | * | |
4836 | * For Direct IO and fallocate, they calls get_block to allocate | |
4837 | * one single extent at a time, so they could set the "Chunk" flag | |
4838 | */ | |
4839 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4840 | ||
4841 | ret = idxblocks; | |
4842 | ||
4843 | /* | |
4844 | * Now let's see how many group bitmaps and group descriptors need | |
4845 | * to account | |
4846 | */ | |
4847 | groups = idxblocks; | |
4848 | if (chunk) | |
4849 | groups += 1; | |
4850 | else | |
4851 | groups += nrblocks; | |
4852 | ||
4853 | gdpblocks = groups; | |
8df9675f TT |
4854 | if (groups > ngroups) |
4855 | groups = ngroups; | |
a02908f1 MC |
4856 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4857 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4858 | ||
4859 | /* bitmaps and block group descriptor blocks */ | |
4860 | ret += groups + gdpblocks; | |
4861 | ||
4862 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4863 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4864 | ||
4865 | return ret; | |
4866 | } | |
4867 | ||
4868 | /* | |
25985edc | 4869 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4870 | * the modification of a single pages into a single transaction, |
4871 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4872 | * |
525f4ed8 | 4873 | * This could be called via ext4_write_begin() |
ac27a0ec | 4874 | * |
525f4ed8 | 4875 | * We need to consider the worse case, when |
a02908f1 | 4876 | * one new block per extent. |
ac27a0ec | 4877 | */ |
a86c6181 | 4878 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4879 | { |
617ba13b | 4880 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4881 | int ret; |
4882 | ||
a02908f1 | 4883 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4884 | |
a02908f1 | 4885 | /* Account for data blocks for journalled mode */ |
617ba13b | 4886 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4887 | ret += bpp; |
ac27a0ec DK |
4888 | return ret; |
4889 | } | |
f3bd1f3f MC |
4890 | |
4891 | /* | |
4892 | * Calculate the journal credits for a chunk of data modification. | |
4893 | * | |
4894 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4895 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4896 | * |
4897 | * journal buffers for data blocks are not included here, as DIO | |
4898 | * and fallocate do no need to journal data buffers. | |
4899 | */ | |
4900 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4901 | { | |
4902 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4903 | } | |
4904 | ||
ac27a0ec | 4905 | /* |
617ba13b | 4906 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4907 | * Give this, we know that the caller already has write access to iloc->bh. |
4908 | */ | |
617ba13b | 4909 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4910 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4911 | { |
4912 | int err = 0; | |
4913 | ||
c64db50e | 4914 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4915 | inode_inc_iversion(inode); |
4916 | ||
ac27a0ec DK |
4917 | /* the do_update_inode consumes one bh->b_count */ |
4918 | get_bh(iloc->bh); | |
4919 | ||
dab291af | 4920 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4921 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4922 | put_bh(iloc->bh); |
4923 | return err; | |
4924 | } | |
4925 | ||
4926 | /* | |
4927 | * On success, We end up with an outstanding reference count against | |
4928 | * iloc->bh. This _must_ be cleaned up later. | |
4929 | */ | |
4930 | ||
4931 | int | |
617ba13b MC |
4932 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4933 | struct ext4_iloc *iloc) | |
ac27a0ec | 4934 | { |
0390131b FM |
4935 | int err; |
4936 | ||
4937 | err = ext4_get_inode_loc(inode, iloc); | |
4938 | if (!err) { | |
4939 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4940 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4941 | if (err) { | |
4942 | brelse(iloc->bh); | |
4943 | iloc->bh = NULL; | |
ac27a0ec DK |
4944 | } |
4945 | } | |
617ba13b | 4946 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4947 | return err; |
4948 | } | |
4949 | ||
6dd4ee7c KS |
4950 | /* |
4951 | * Expand an inode by new_extra_isize bytes. | |
4952 | * Returns 0 on success or negative error number on failure. | |
4953 | */ | |
1d03ec98 AK |
4954 | static int ext4_expand_extra_isize(struct inode *inode, |
4955 | unsigned int new_extra_isize, | |
4956 | struct ext4_iloc iloc, | |
4957 | handle_t *handle) | |
6dd4ee7c KS |
4958 | { |
4959 | struct ext4_inode *raw_inode; | |
4960 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4961 | |
4962 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4963 | return 0; | |
4964 | ||
4965 | raw_inode = ext4_raw_inode(&iloc); | |
4966 | ||
4967 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4968 | |
4969 | /* No extended attributes present */ | |
19f5fb7a TT |
4970 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4971 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4972 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4973 | new_extra_isize); | |
4974 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4975 | return 0; | |
4976 | } | |
4977 | ||
4978 | /* try to expand with EAs present */ | |
4979 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4980 | raw_inode, handle); | |
4981 | } | |
4982 | ||
ac27a0ec DK |
4983 | /* |
4984 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4985 | * dirtiness (it may still be data-dirty). | |
4986 | * This means that the in-core inode may be reaped by prune_icache | |
4987 | * without having to perform any I/O. This is a very good thing, | |
4988 | * because *any* task may call prune_icache - even ones which | |
4989 | * have a transaction open against a different journal. | |
4990 | * | |
4991 | * Is this cheating? Not really. Sure, we haven't written the | |
4992 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4993 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4994 | * we start and wait on commits. | |
ac27a0ec | 4995 | */ |
617ba13b | 4996 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4997 | { |
617ba13b | 4998 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4999 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5000 | static unsigned int mnt_count; | |
5001 | int err, ret; | |
ac27a0ec DK |
5002 | |
5003 | might_sleep(); | |
7ff9c073 | 5004 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5005 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5006 | if (ext4_handle_valid(handle) && |
5007 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5008 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5009 | /* |
5010 | * We need extra buffer credits since we may write into EA block | |
5011 | * with this same handle. If journal_extend fails, then it will | |
5012 | * only result in a minor loss of functionality for that inode. | |
5013 | * If this is felt to be critical, then e2fsck should be run to | |
5014 | * force a large enough s_min_extra_isize. | |
5015 | */ | |
5016 | if ((jbd2_journal_extend(handle, | |
5017 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5018 | ret = ext4_expand_extra_isize(inode, | |
5019 | sbi->s_want_extra_isize, | |
5020 | iloc, handle); | |
5021 | if (ret) { | |
19f5fb7a TT |
5022 | ext4_set_inode_state(inode, |
5023 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5024 | if (mnt_count != |
5025 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5026 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5027 | "Unable to expand inode %lu. Delete" |
5028 | " some EAs or run e2fsck.", | |
5029 | inode->i_ino); | |
c1bddad9 AK |
5030 | mnt_count = |
5031 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5032 | } |
5033 | } | |
5034 | } | |
5035 | } | |
ac27a0ec | 5036 | if (!err) |
617ba13b | 5037 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5038 | return err; |
5039 | } | |
5040 | ||
5041 | /* | |
617ba13b | 5042 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5043 | * |
5044 | * We're really interested in the case where a file is being extended. | |
5045 | * i_size has been changed by generic_commit_write() and we thus need | |
5046 | * to include the updated inode in the current transaction. | |
5047 | * | |
5dd4056d | 5048 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5049 | * are allocated to the file. |
5050 | * | |
5051 | * If the inode is marked synchronous, we don't honour that here - doing | |
5052 | * so would cause a commit on atime updates, which we don't bother doing. | |
5053 | * We handle synchronous inodes at the highest possible level. | |
5054 | */ | |
aa385729 | 5055 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5056 | { |
ac27a0ec DK |
5057 | handle_t *handle; |
5058 | ||
9924a92a | 5059 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5060 | if (IS_ERR(handle)) |
5061 | goto out; | |
f3dc272f | 5062 | |
f3dc272f CW |
5063 | ext4_mark_inode_dirty(handle, inode); |
5064 | ||
617ba13b | 5065 | ext4_journal_stop(handle); |
ac27a0ec DK |
5066 | out: |
5067 | return; | |
5068 | } | |
5069 | ||
5070 | #if 0 | |
5071 | /* | |
5072 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5073 | * it from being flushed to disk early. Unlike | |
617ba13b | 5074 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5075 | * returns no iloc structure, so the caller needs to repeat the iloc |
5076 | * lookup to mark the inode dirty later. | |
5077 | */ | |
617ba13b | 5078 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5079 | { |
617ba13b | 5080 | struct ext4_iloc iloc; |
ac27a0ec DK |
5081 | |
5082 | int err = 0; | |
5083 | if (handle) { | |
617ba13b | 5084 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5085 | if (!err) { |
5086 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5087 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5088 | if (!err) |
0390131b | 5089 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5090 | NULL, |
0390131b | 5091 | iloc.bh); |
ac27a0ec DK |
5092 | brelse(iloc.bh); |
5093 | } | |
5094 | } | |
617ba13b | 5095 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5096 | return err; |
5097 | } | |
5098 | #endif | |
5099 | ||
617ba13b | 5100 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5101 | { |
5102 | journal_t *journal; | |
5103 | handle_t *handle; | |
5104 | int err; | |
5105 | ||
5106 | /* | |
5107 | * We have to be very careful here: changing a data block's | |
5108 | * journaling status dynamically is dangerous. If we write a | |
5109 | * data block to the journal, change the status and then delete | |
5110 | * that block, we risk forgetting to revoke the old log record | |
5111 | * from the journal and so a subsequent replay can corrupt data. | |
5112 | * So, first we make sure that the journal is empty and that | |
5113 | * nobody is changing anything. | |
5114 | */ | |
5115 | ||
617ba13b | 5116 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5117 | if (!journal) |
5118 | return 0; | |
d699594d | 5119 | if (is_journal_aborted(journal)) |
ac27a0ec | 5120 | return -EROFS; |
2aff57b0 YY |
5121 | /* We have to allocate physical blocks for delalloc blocks |
5122 | * before flushing journal. otherwise delalloc blocks can not | |
5123 | * be allocated any more. even more truncate on delalloc blocks | |
5124 | * could trigger BUG by flushing delalloc blocks in journal. | |
5125 | * There is no delalloc block in non-journal data mode. | |
5126 | */ | |
5127 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5128 | err = ext4_alloc_da_blocks(inode); | |
5129 | if (err < 0) | |
5130 | return err; | |
5131 | } | |
ac27a0ec | 5132 | |
17335dcc DM |
5133 | /* Wait for all existing dio workers */ |
5134 | ext4_inode_block_unlocked_dio(inode); | |
5135 | inode_dio_wait(inode); | |
5136 | ||
dab291af | 5137 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5138 | |
5139 | /* | |
5140 | * OK, there are no updates running now, and all cached data is | |
5141 | * synced to disk. We are now in a completely consistent state | |
5142 | * which doesn't have anything in the journal, and we know that | |
5143 | * no filesystem updates are running, so it is safe to modify | |
5144 | * the inode's in-core data-journaling state flag now. | |
5145 | */ | |
5146 | ||
5147 | if (val) | |
12e9b892 | 5148 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
5149 | else { |
5150 | jbd2_journal_flush(journal); | |
12e9b892 | 5151 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5152 | } |
617ba13b | 5153 | ext4_set_aops(inode); |
ac27a0ec | 5154 | |
dab291af | 5155 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5156 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5157 | |
5158 | /* Finally we can mark the inode as dirty. */ | |
5159 | ||
9924a92a | 5160 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5161 | if (IS_ERR(handle)) |
5162 | return PTR_ERR(handle); | |
5163 | ||
617ba13b | 5164 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5165 | ext4_handle_sync(handle); |
617ba13b MC |
5166 | ext4_journal_stop(handle); |
5167 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5168 | |
5169 | return err; | |
5170 | } | |
2e9ee850 AK |
5171 | |
5172 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5173 | { | |
5174 | return !buffer_mapped(bh); | |
5175 | } | |
5176 | ||
c2ec175c | 5177 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5178 | { |
c2ec175c | 5179 | struct page *page = vmf->page; |
2e9ee850 AK |
5180 | loff_t size; |
5181 | unsigned long len; | |
9ea7df53 | 5182 | int ret; |
2e9ee850 | 5183 | struct file *file = vma->vm_file; |
496ad9aa | 5184 | struct inode *inode = file_inode(file); |
2e9ee850 | 5185 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5186 | handle_t *handle; |
5187 | get_block_t *get_block; | |
5188 | int retries = 0; | |
2e9ee850 | 5189 | |
8e8ad8a5 | 5190 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5191 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5192 | /* Delalloc case is easy... */ |
5193 | if (test_opt(inode->i_sb, DELALLOC) && | |
5194 | !ext4_should_journal_data(inode) && | |
5195 | !ext4_nonda_switch(inode->i_sb)) { | |
5196 | do { | |
5197 | ret = __block_page_mkwrite(vma, vmf, | |
5198 | ext4_da_get_block_prep); | |
5199 | } while (ret == -ENOSPC && | |
5200 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5201 | goto out_ret; | |
2e9ee850 | 5202 | } |
0e499890 DW |
5203 | |
5204 | lock_page(page); | |
9ea7df53 JK |
5205 | size = i_size_read(inode); |
5206 | /* Page got truncated from under us? */ | |
5207 | if (page->mapping != mapping || page_offset(page) > size) { | |
5208 | unlock_page(page); | |
5209 | ret = VM_FAULT_NOPAGE; | |
5210 | goto out; | |
0e499890 | 5211 | } |
2e9ee850 AK |
5212 | |
5213 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5214 | len = size & ~PAGE_CACHE_MASK; | |
5215 | else | |
5216 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5217 | /* |
9ea7df53 JK |
5218 | * Return if we have all the buffers mapped. This avoids the need to do |
5219 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5220 | */ |
2e9ee850 | 5221 | if (page_has_buffers(page)) { |
f19d5870 TM |
5222 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5223 | 0, len, NULL, | |
5224 | ext4_bh_unmapped)) { | |
9ea7df53 | 5225 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5226 | wait_for_stable_page(page); |
9ea7df53 JK |
5227 | ret = VM_FAULT_LOCKED; |
5228 | goto out; | |
a827eaff | 5229 | } |
2e9ee850 | 5230 | } |
a827eaff | 5231 | unlock_page(page); |
9ea7df53 JK |
5232 | /* OK, we need to fill the hole... */ |
5233 | if (ext4_should_dioread_nolock(inode)) | |
5234 | get_block = ext4_get_block_write; | |
5235 | else | |
5236 | get_block = ext4_get_block; | |
5237 | retry_alloc: | |
9924a92a TT |
5238 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5239 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5240 | if (IS_ERR(handle)) { |
c2ec175c | 5241 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5242 | goto out; |
5243 | } | |
5244 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5245 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5246 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5247 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5248 | unlock_page(page); | |
5249 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5250 | ext4_journal_stop(handle); |
9ea7df53 JK |
5251 | goto out; |
5252 | } | |
5253 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5254 | } | |
5255 | ext4_journal_stop(handle); | |
5256 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5257 | goto retry_alloc; | |
5258 | out_ret: | |
5259 | ret = block_page_mkwrite_return(ret); | |
5260 | out: | |
8e8ad8a5 | 5261 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5262 | return ret; |
5263 | } |