]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
ac27a0ec DK |
15 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
16 | * (jj@sunsite.ms.mff.cuni.cz) | |
17 | * | |
617ba13b | 18 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
19 | */ |
20 | ||
ac27a0ec DK |
21 | #include <linux/fs.h> |
22 | #include <linux/time.h> | |
dab291af | 23 | #include <linux/jbd2.h> |
ac27a0ec DK |
24 | #include <linux/highuid.h> |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/quotaops.h> | |
27 | #include <linux/string.h> | |
28 | #include <linux/buffer_head.h> | |
29 | #include <linux/writeback.h> | |
64769240 | 30 | #include <linux/pagevec.h> |
ac27a0ec | 31 | #include <linux/mpage.h> |
e83c1397 | 32 | #include <linux/namei.h> |
ac27a0ec DK |
33 | #include <linux/uio.h> |
34 | #include <linux/bio.h> | |
4c0425ff | 35 | #include <linux/workqueue.h> |
744692dc | 36 | #include <linux/kernel.h> |
6db26ffc | 37 | #include <linux/printk.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
a8901d34 | 39 | #include <linux/ratelimit.h> |
a27bb332 | 40 | #include <linux/aio.h> |
00a1a053 | 41 | #include <linux/bitops.h> |
9bffad1e | 42 | |
3dcf5451 | 43 | #include "ext4_jbd2.h" |
ac27a0ec DK |
44 | #include "xattr.h" |
45 | #include "acl.h" | |
9f125d64 | 46 | #include "truncate.h" |
ac27a0ec | 47 | |
9bffad1e TT |
48 | #include <trace/events/ext4.h> |
49 | ||
a1d6cc56 AK |
50 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
51 | ||
814525f4 DW |
52 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
53 | struct ext4_inode_info *ei) | |
54 | { | |
55 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
56 | __u16 csum_lo; | |
57 | __u16 csum_hi = 0; | |
58 | __u32 csum; | |
59 | ||
171a7f21 | 60 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
61 | raw->i_checksum_lo = 0; |
62 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
63 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 64 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
65 | raw->i_checksum_hi = 0; |
66 | } | |
67 | ||
68 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
69 | EXT4_INODE_SIZE(inode->i_sb)); | |
70 | ||
171a7f21 | 71 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
72 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
73 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 74 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
75 | |
76 | return csum; | |
77 | } | |
78 | ||
79 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
80 | struct ext4_inode_info *ei) | |
81 | { | |
82 | __u32 provided, calculated; | |
83 | ||
84 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
85 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 86 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
87 | return 1; |
88 | ||
89 | provided = le16_to_cpu(raw->i_checksum_lo); | |
90 | calculated = ext4_inode_csum(inode, raw, ei); | |
91 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
92 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
93 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
94 | else | |
95 | calculated &= 0xFFFF; | |
96 | ||
97 | return provided == calculated; | |
98 | } | |
99 | ||
100 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
101 | struct ext4_inode_info *ei) | |
102 | { | |
103 | __u32 csum; | |
104 | ||
105 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
106 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 107 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
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 | ||
d47992f8 LC |
134 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
135 | unsigned int length); | |
cb20d518 TT |
136 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
137 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
fffb2739 JK |
138 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
139 | int pextents); | |
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 | { |
65eddb56 YY |
146 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
147 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
ac27a0ec | 148 | |
bd9db175 ZL |
149 | if (ext4_has_inline_data(inode)) |
150 | return 0; | |
151 | ||
ac27a0ec DK |
152 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
153 | } | |
154 | ||
ac27a0ec DK |
155 | /* |
156 | * Restart the transaction associated with *handle. This does a commit, | |
157 | * so before we call here everything must be consistently dirtied against | |
158 | * this transaction. | |
159 | */ | |
fa5d1113 | 160 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 161 | int nblocks) |
ac27a0ec | 162 | { |
487caeef JK |
163 | int ret; |
164 | ||
165 | /* | |
e35fd660 | 166 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
167 | * moment, get_block can be called only for blocks inside i_size since |
168 | * page cache has been already dropped and writes are blocked by | |
169 | * i_mutex. So we can safely drop the i_data_sem here. | |
170 | */ | |
0390131b | 171 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 172 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 173 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 174 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 175 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 176 | ext4_discard_preallocations(inode); |
487caeef JK |
177 | |
178 | return ret; | |
ac27a0ec DK |
179 | } |
180 | ||
181 | /* | |
182 | * Called at the last iput() if i_nlink is zero. | |
183 | */ | |
0930fcc1 | 184 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
185 | { |
186 | handle_t *handle; | |
bc965ab3 | 187 | int err; |
ac27a0ec | 188 | |
7ff9c073 | 189 | trace_ext4_evict_inode(inode); |
2581fdc8 | 190 | |
0930fcc1 | 191 | if (inode->i_nlink) { |
2d859db3 JK |
192 | /* |
193 | * When journalling data dirty buffers are tracked only in the | |
194 | * journal. So although mm thinks everything is clean and | |
195 | * ready for reaping the inode might still have some pages to | |
196 | * write in the running transaction or waiting to be | |
197 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
198 | * (via truncate_inode_pages()) to discard these buffers can | |
199 | * cause data loss. Also even if we did not discard these | |
200 | * buffers, we would have no way to find them after the inode | |
201 | * is reaped and thus user could see stale data if he tries to | |
202 | * read them before the transaction is checkpointed. So be | |
203 | * careful and force everything to disk here... We use | |
204 | * ei->i_datasync_tid to store the newest transaction | |
205 | * containing inode's data. | |
206 | * | |
207 | * Note that directories do not have this problem because they | |
208 | * don't use page cache. | |
209 | */ | |
210 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
211 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
212 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
213 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
214 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
215 | ||
d76a3a77 | 216 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
217 | filemap_write_and_wait(&inode->i_data); |
218 | } | |
91b0abe3 | 219 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd JK |
220 | |
221 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); | |
0930fcc1 AV |
222 | goto no_delete; |
223 | } | |
224 | ||
e2bfb088 TT |
225 | if (is_bad_inode(inode)) |
226 | goto no_delete; | |
227 | dquot_initialize(inode); | |
907f4554 | 228 | |
678aaf48 JK |
229 | if (ext4_should_order_data(inode)) |
230 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 231 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 232 | |
5dc23bdd | 233 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); |
ac27a0ec | 234 | |
8e8ad8a5 JK |
235 | /* |
236 | * Protect us against freezing - iput() caller didn't have to have any | |
237 | * protection against it | |
238 | */ | |
239 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
240 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
241 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 242 | if (IS_ERR(handle)) { |
bc965ab3 | 243 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
244 | /* |
245 | * If we're going to skip the normal cleanup, we still need to | |
246 | * make sure that the in-core orphan linked list is properly | |
247 | * cleaned up. | |
248 | */ | |
617ba13b | 249 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 250 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
251 | goto no_delete; |
252 | } | |
253 | ||
254 | if (IS_SYNC(inode)) | |
0390131b | 255 | ext4_handle_sync(handle); |
ac27a0ec | 256 | inode->i_size = 0; |
bc965ab3 TT |
257 | err = ext4_mark_inode_dirty(handle, inode); |
258 | if (err) { | |
12062ddd | 259 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
260 | "couldn't mark inode dirty (err %d)", err); |
261 | goto stop_handle; | |
262 | } | |
ac27a0ec | 263 | if (inode->i_blocks) |
617ba13b | 264 | ext4_truncate(inode); |
bc965ab3 TT |
265 | |
266 | /* | |
267 | * ext4_ext_truncate() doesn't reserve any slop when it | |
268 | * restarts journal transactions; therefore there may not be | |
269 | * enough credits left in the handle to remove the inode from | |
270 | * the orphan list and set the dtime field. | |
271 | */ | |
0390131b | 272 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
273 | err = ext4_journal_extend(handle, 3); |
274 | if (err > 0) | |
275 | err = ext4_journal_restart(handle, 3); | |
276 | if (err != 0) { | |
12062ddd | 277 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
278 | "couldn't extend journal (err %d)", err); |
279 | stop_handle: | |
280 | ext4_journal_stop(handle); | |
45388219 | 281 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 282 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
283 | goto no_delete; |
284 | } | |
285 | } | |
286 | ||
ac27a0ec | 287 | /* |
617ba13b | 288 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 289 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 290 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 291 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 292 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
293 | * (Well, we could do this if we need to, but heck - it works) |
294 | */ | |
617ba13b MC |
295 | ext4_orphan_del(handle, inode); |
296 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
297 | |
298 | /* | |
299 | * One subtle ordering requirement: if anything has gone wrong | |
300 | * (transaction abort, IO errors, whatever), then we can still | |
301 | * do these next steps (the fs will already have been marked as | |
302 | * having errors), but we can't free the inode if the mark_dirty | |
303 | * fails. | |
304 | */ | |
617ba13b | 305 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 306 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 307 | ext4_clear_inode(inode); |
ac27a0ec | 308 | else |
617ba13b MC |
309 | ext4_free_inode(handle, inode); |
310 | ext4_journal_stop(handle); | |
8e8ad8a5 | 311 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
312 | return; |
313 | no_delete: | |
0930fcc1 | 314 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
315 | } |
316 | ||
a9e7f447 DM |
317 | #ifdef CONFIG_QUOTA |
318 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 319 | { |
a9e7f447 | 320 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 321 | } |
a9e7f447 | 322 | #endif |
9d0be502 | 323 | |
0637c6f4 TT |
324 | /* |
325 | * Called with i_data_sem down, which is important since we can call | |
326 | * ext4_discard_preallocations() from here. | |
327 | */ | |
5f634d06 AK |
328 | void ext4_da_update_reserve_space(struct inode *inode, |
329 | int used, int quota_claim) | |
12219aea AK |
330 | { |
331 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 332 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
333 | |
334 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 335 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 336 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 337 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 338 | "with only %d reserved data blocks", |
0637c6f4 TT |
339 | __func__, inode->i_ino, used, |
340 | ei->i_reserved_data_blocks); | |
341 | WARN_ON(1); | |
342 | used = ei->i_reserved_data_blocks; | |
343 | } | |
12219aea | 344 | |
0637c6f4 TT |
345 | /* Update per-inode reservations */ |
346 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 347 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 348 | |
12219aea | 349 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 350 | |
72b8ab9d ES |
351 | /* Update quota subsystem for data blocks */ |
352 | if (quota_claim) | |
7b415bf6 | 353 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 354 | else { |
5f634d06 AK |
355 | /* |
356 | * We did fallocate with an offset that is already delayed | |
357 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 358 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 359 | */ |
7b415bf6 | 360 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 361 | } |
d6014301 AK |
362 | |
363 | /* | |
364 | * If we have done all the pending block allocations and if | |
365 | * there aren't any writers on the inode, we can discard the | |
366 | * inode's preallocations. | |
367 | */ | |
0637c6f4 TT |
368 | if ((ei->i_reserved_data_blocks == 0) && |
369 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 370 | ext4_discard_preallocations(inode); |
12219aea AK |
371 | } |
372 | ||
e29136f8 | 373 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
374 | unsigned int line, |
375 | struct ext4_map_blocks *map) | |
6fd058f7 | 376 | { |
24676da4 TT |
377 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
378 | map->m_len)) { | |
c398eda0 TT |
379 | ext4_error_inode(inode, func, line, map->m_pblk, |
380 | "lblock %lu mapped to illegal pblock " | |
381 | "(length %d)", (unsigned long) map->m_lblk, | |
382 | map->m_len); | |
6fd058f7 TT |
383 | return -EIO; |
384 | } | |
385 | return 0; | |
386 | } | |
387 | ||
e29136f8 | 388 | #define check_block_validity(inode, map) \ |
c398eda0 | 389 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 390 | |
921f266b DM |
391 | #ifdef ES_AGGRESSIVE_TEST |
392 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
393 | struct inode *inode, | |
394 | struct ext4_map_blocks *es_map, | |
395 | struct ext4_map_blocks *map, | |
396 | int flags) | |
397 | { | |
398 | int retval; | |
399 | ||
400 | map->m_flags = 0; | |
401 | /* | |
402 | * There is a race window that the result is not the same. | |
403 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
404 | * is that we lookup a block mapping in extent status tree with | |
405 | * out taking i_data_sem. So at the time the unwritten extent | |
406 | * could be converted. | |
407 | */ | |
408 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
c8b459f4 | 409 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
410 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
411 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
412 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
413 | } else { | |
414 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
415 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
416 | } | |
417 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
418 | up_read((&EXT4_I(inode)->i_data_sem)); | |
921f266b DM |
419 | |
420 | /* | |
421 | * We don't check m_len because extent will be collpased in status | |
422 | * tree. So the m_len might not equal. | |
423 | */ | |
424 | if (es_map->m_lblk != map->m_lblk || | |
425 | es_map->m_flags != map->m_flags || | |
426 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 427 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
428 | "es_cached ex [%d/%d/%llu/%x] != " |
429 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
430 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
431 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
432 | map->m_len, map->m_pblk, map->m_flags, | |
433 | retval, flags); | |
434 | } | |
435 | } | |
436 | #endif /* ES_AGGRESSIVE_TEST */ | |
437 | ||
f5ab0d1f | 438 | /* |
e35fd660 | 439 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 440 | * and returns if the blocks are already mapped. |
f5ab0d1f | 441 | * |
f5ab0d1f MC |
442 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
443 | * and store the allocated blocks in the result buffer head and mark it | |
444 | * mapped. | |
445 | * | |
e35fd660 TT |
446 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
447 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
448 | * based files |
449 | * | |
556615dc LC |
450 | * On success, it returns the number of blocks being mapped or allocated. |
451 | * if create==0 and the blocks are pre-allocated and unwritten block, | |
f5ab0d1f MC |
452 | * the result buffer head is unmapped. If the create ==1, it will make sure |
453 | * the buffer head is mapped. | |
454 | * | |
455 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 456 | * that case, buffer head is unmapped |
f5ab0d1f MC |
457 | * |
458 | * It returns the error in case of allocation failure. | |
459 | */ | |
e35fd660 TT |
460 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
461 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 462 | { |
d100eef2 | 463 | struct extent_status es; |
0e855ac8 | 464 | int retval; |
b8a86845 | 465 | int ret = 0; |
921f266b DM |
466 | #ifdef ES_AGGRESSIVE_TEST |
467 | struct ext4_map_blocks orig_map; | |
468 | ||
469 | memcpy(&orig_map, map, sizeof(*map)); | |
470 | #endif | |
f5ab0d1f | 471 | |
e35fd660 TT |
472 | map->m_flags = 0; |
473 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
474 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
475 | (unsigned long) map->m_lblk); | |
d100eef2 | 476 | |
e861b5e9 TT |
477 | /* |
478 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
479 | */ | |
480 | if (unlikely(map->m_len > INT_MAX)) | |
481 | map->m_len = INT_MAX; | |
482 | ||
4adb6ab3 KM |
483 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
484 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
485 | return -EIO; | |
486 | ||
d100eef2 ZL |
487 | /* Lookup extent status tree firstly */ |
488 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
489 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
490 | map->m_pblk = ext4_es_pblock(&es) + | |
491 | map->m_lblk - es.es_lblk; | |
492 | map->m_flags |= ext4_es_is_written(&es) ? | |
493 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
494 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
495 | if (retval > map->m_len) | |
496 | retval = map->m_len; | |
497 | map->m_len = retval; | |
498 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
499 | retval = 0; | |
500 | } else { | |
501 | BUG_ON(1); | |
502 | } | |
921f266b DM |
503 | #ifdef ES_AGGRESSIVE_TEST |
504 | ext4_map_blocks_es_recheck(handle, inode, map, | |
505 | &orig_map, flags); | |
506 | #endif | |
d100eef2 ZL |
507 | goto found; |
508 | } | |
509 | ||
4df3d265 | 510 | /* |
b920c755 TT |
511 | * Try to see if we can get the block without requesting a new |
512 | * file system block. | |
4df3d265 | 513 | */ |
729f52c6 | 514 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
c8b459f4 | 515 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 516 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
517 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
518 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 519 | } else { |
a4e5d88b DM |
520 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
521 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 522 | } |
f7fec032 | 523 | if (retval > 0) { |
3be78c73 | 524 | unsigned int status; |
f7fec032 | 525 | |
44fb851d ZL |
526 | if (unlikely(retval != map->m_len)) { |
527 | ext4_warning(inode->i_sb, | |
528 | "ES len assertion failed for inode " | |
529 | "%lu: retval %d != map->m_len %d", | |
530 | inode->i_ino, retval, map->m_len); | |
531 | WARN_ON(1); | |
921f266b | 532 | } |
921f266b | 533 | |
f7fec032 ZL |
534 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
535 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
536 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
537 | ext4_find_delalloc_range(inode, map->m_lblk, | |
538 | map->m_lblk + map->m_len - 1)) | |
539 | status |= EXTENT_STATUS_DELAYED; | |
540 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
541 | map->m_len, map->m_pblk, status); | |
542 | if (ret < 0) | |
543 | retval = ret; | |
544 | } | |
729f52c6 ZL |
545 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
546 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 547 | |
d100eef2 | 548 | found: |
e35fd660 | 549 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 550 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
551 | if (ret != 0) |
552 | return ret; | |
553 | } | |
554 | ||
f5ab0d1f | 555 | /* If it is only a block(s) look up */ |
c2177057 | 556 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
557 | return retval; |
558 | ||
559 | /* | |
560 | * Returns if the blocks have already allocated | |
561 | * | |
562 | * Note that if blocks have been preallocated | |
df3ab170 | 563 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
564 | * with buffer head unmapped. |
565 | */ | |
e35fd660 | 566 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
567 | /* |
568 | * If we need to convert extent to unwritten | |
569 | * we continue and do the actual work in | |
570 | * ext4_ext_map_blocks() | |
571 | */ | |
572 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
573 | return retval; | |
4df3d265 | 574 | |
2a8964d6 | 575 | /* |
a25a4e1a ZL |
576 | * Here we clear m_flags because after allocating an new extent, |
577 | * it will be set again. | |
2a8964d6 | 578 | */ |
a25a4e1a | 579 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 580 | |
4df3d265 | 581 | /* |
556615dc | 582 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 583 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 584 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 585 | * with create == 1 flag. |
4df3d265 | 586 | */ |
c8b459f4 | 587 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 588 | |
4df3d265 AK |
589 | /* |
590 | * We need to check for EXT4 here because migrate | |
591 | * could have changed the inode type in between | |
592 | */ | |
12e9b892 | 593 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 594 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 595 | } else { |
e35fd660 | 596 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 597 | |
e35fd660 | 598 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
599 | /* |
600 | * We allocated new blocks which will result in | |
601 | * i_data's format changing. Force the migrate | |
602 | * to fail by clearing migrate flags | |
603 | */ | |
19f5fb7a | 604 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 605 | } |
d2a17637 | 606 | |
5f634d06 AK |
607 | /* |
608 | * Update reserved blocks/metadata blocks after successful | |
609 | * block allocation which had been deferred till now. We don't | |
610 | * support fallocate for non extent files. So we can update | |
611 | * reserve space here. | |
612 | */ | |
613 | if ((retval > 0) && | |
1296cc85 | 614 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
615 | ext4_da_update_reserve_space(inode, retval, 1); |
616 | } | |
2ac3b6e0 | 617 | |
f7fec032 | 618 | if (retval > 0) { |
3be78c73 | 619 | unsigned int status; |
f7fec032 | 620 | |
44fb851d ZL |
621 | if (unlikely(retval != map->m_len)) { |
622 | ext4_warning(inode->i_sb, | |
623 | "ES len assertion failed for inode " | |
624 | "%lu: retval %d != map->m_len %d", | |
625 | inode->i_ino, retval, map->m_len); | |
626 | WARN_ON(1); | |
921f266b | 627 | } |
921f266b | 628 | |
adb23551 ZL |
629 | /* |
630 | * If the extent has been zeroed out, we don't need to update | |
631 | * extent status tree. | |
632 | */ | |
633 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
634 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
635 | if (ext4_es_is_written(&es)) | |
636 | goto has_zeroout; | |
637 | } | |
f7fec032 ZL |
638 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
639 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
640 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
641 | ext4_find_delalloc_range(inode, map->m_lblk, | |
642 | map->m_lblk + map->m_len - 1)) | |
643 | status |= EXTENT_STATUS_DELAYED; | |
644 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
645 | map->m_pblk, status); | |
646 | if (ret < 0) | |
647 | retval = ret; | |
5356f261 AK |
648 | } |
649 | ||
adb23551 | 650 | has_zeroout: |
4df3d265 | 651 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 652 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 653 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
654 | if (ret != 0) |
655 | return ret; | |
656 | } | |
0e855ac8 AK |
657 | return retval; |
658 | } | |
659 | ||
923ae0ff RZ |
660 | static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate) |
661 | { | |
662 | struct inode *inode = bh->b_assoc_map->host; | |
663 | /* XXX: breaks on 32-bit > 16GB. Is that even supported? */ | |
664 | loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits; | |
665 | int err; | |
666 | if (!uptodate) | |
667 | return; | |
668 | WARN_ON(!buffer_unwritten(bh)); | |
669 | err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size); | |
670 | } | |
671 | ||
f3bd1f3f MC |
672 | /* Maximum number of blocks we map for direct IO at once. */ |
673 | #define DIO_MAX_BLOCKS 4096 | |
674 | ||
2ed88685 TT |
675 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
676 | struct buffer_head *bh, int flags) | |
ac27a0ec | 677 | { |
3e4fdaf8 | 678 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 679 | struct ext4_map_blocks map; |
7fb5409d | 680 | int ret = 0, started = 0; |
f3bd1f3f | 681 | int dio_credits; |
ac27a0ec | 682 | |
46c7f254 TM |
683 | if (ext4_has_inline_data(inode)) |
684 | return -ERANGE; | |
685 | ||
2ed88685 TT |
686 | map.m_lblk = iblock; |
687 | map.m_len = bh->b_size >> inode->i_blkbits; | |
688 | ||
8b0f165f | 689 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 690 | /* Direct IO write... */ |
2ed88685 TT |
691 | if (map.m_len > DIO_MAX_BLOCKS) |
692 | map.m_len = DIO_MAX_BLOCKS; | |
693 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
694 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
695 | dio_credits); | |
7fb5409d | 696 | if (IS_ERR(handle)) { |
ac27a0ec | 697 | ret = PTR_ERR(handle); |
2ed88685 | 698 | return ret; |
ac27a0ec | 699 | } |
7fb5409d | 700 | started = 1; |
ac27a0ec DK |
701 | } |
702 | ||
2ed88685 | 703 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 704 | if (ret > 0) { |
7b7a8665 CH |
705 | ext4_io_end_t *io_end = ext4_inode_aio(inode); |
706 | ||
2ed88685 TT |
707 | map_bh(bh, inode->i_sb, map.m_pblk); |
708 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
923ae0ff RZ |
709 | if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) { |
710 | bh->b_assoc_map = inode->i_mapping; | |
711 | bh->b_private = (void *)(unsigned long)iblock; | |
712 | bh->b_end_io = ext4_end_io_unwritten; | |
713 | } | |
7b7a8665 CH |
714 | if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) |
715 | set_buffer_defer_completion(bh); | |
2ed88685 | 716 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 717 | ret = 0; |
ac27a0ec | 718 | } |
7fb5409d JK |
719 | if (started) |
720 | ext4_journal_stop(handle); | |
ac27a0ec DK |
721 | return ret; |
722 | } | |
723 | ||
2ed88685 TT |
724 | int ext4_get_block(struct inode *inode, sector_t iblock, |
725 | struct buffer_head *bh, int create) | |
726 | { | |
727 | return _ext4_get_block(inode, iblock, bh, | |
728 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
729 | } | |
730 | ||
ac27a0ec DK |
731 | /* |
732 | * `handle' can be NULL if create is zero | |
733 | */ | |
617ba13b | 734 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
10560082 | 735 | ext4_lblk_t block, int create) |
ac27a0ec | 736 | { |
2ed88685 TT |
737 | struct ext4_map_blocks map; |
738 | struct buffer_head *bh; | |
10560082 | 739 | int err; |
ac27a0ec DK |
740 | |
741 | J_ASSERT(handle != NULL || create == 0); | |
742 | ||
2ed88685 TT |
743 | map.m_lblk = block; |
744 | map.m_len = 1; | |
745 | err = ext4_map_blocks(handle, inode, &map, | |
746 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 747 | |
10560082 TT |
748 | if (err == 0) |
749 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 750 | if (err < 0) |
10560082 | 751 | return ERR_PTR(err); |
2ed88685 TT |
752 | |
753 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
754 | if (unlikely(!bh)) |
755 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
756 | if (map.m_flags & EXT4_MAP_NEW) { |
757 | J_ASSERT(create != 0); | |
758 | J_ASSERT(handle != NULL); | |
ac27a0ec | 759 | |
2ed88685 TT |
760 | /* |
761 | * Now that we do not always journal data, we should | |
762 | * keep in mind whether this should always journal the | |
763 | * new buffer as metadata. For now, regular file | |
764 | * writes use ext4_get_block instead, so it's not a | |
765 | * problem. | |
766 | */ | |
767 | lock_buffer(bh); | |
768 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
769 | err = ext4_journal_get_create_access(handle, bh); |
770 | if (unlikely(err)) { | |
771 | unlock_buffer(bh); | |
772 | goto errout; | |
773 | } | |
774 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
775 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
776 | set_buffer_uptodate(bh); | |
ac27a0ec | 777 | } |
2ed88685 TT |
778 | unlock_buffer(bh); |
779 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
780 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
781 | if (unlikely(err)) |
782 | goto errout; | |
783 | } else | |
2ed88685 | 784 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 785 | return bh; |
10560082 TT |
786 | errout: |
787 | brelse(bh); | |
788 | return ERR_PTR(err); | |
ac27a0ec DK |
789 | } |
790 | ||
617ba13b | 791 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
1c215028 | 792 | ext4_lblk_t block, int create) |
ac27a0ec | 793 | { |
af5bc92d | 794 | struct buffer_head *bh; |
ac27a0ec | 795 | |
10560082 | 796 | bh = ext4_getblk(handle, inode, block, create); |
1c215028 | 797 | if (IS_ERR(bh)) |
ac27a0ec | 798 | return bh; |
1c215028 | 799 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 800 | return bh; |
65299a3b | 801 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
802 | wait_on_buffer(bh); |
803 | if (buffer_uptodate(bh)) | |
804 | return bh; | |
805 | put_bh(bh); | |
1c215028 | 806 | return ERR_PTR(-EIO); |
ac27a0ec DK |
807 | } |
808 | ||
f19d5870 TM |
809 | int ext4_walk_page_buffers(handle_t *handle, |
810 | struct buffer_head *head, | |
811 | unsigned from, | |
812 | unsigned to, | |
813 | int *partial, | |
814 | int (*fn)(handle_t *handle, | |
815 | struct buffer_head *bh)) | |
ac27a0ec DK |
816 | { |
817 | struct buffer_head *bh; | |
818 | unsigned block_start, block_end; | |
819 | unsigned blocksize = head->b_size; | |
820 | int err, ret = 0; | |
821 | struct buffer_head *next; | |
822 | ||
af5bc92d TT |
823 | for (bh = head, block_start = 0; |
824 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 825 | block_start = block_end, bh = next) { |
ac27a0ec DK |
826 | next = bh->b_this_page; |
827 | block_end = block_start + blocksize; | |
828 | if (block_end <= from || block_start >= to) { | |
829 | if (partial && !buffer_uptodate(bh)) | |
830 | *partial = 1; | |
831 | continue; | |
832 | } | |
833 | err = (*fn)(handle, bh); | |
834 | if (!ret) | |
835 | ret = err; | |
836 | } | |
837 | return ret; | |
838 | } | |
839 | ||
840 | /* | |
841 | * To preserve ordering, it is essential that the hole instantiation and | |
842 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 843 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 844 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
845 | * prepare_write() is the right place. |
846 | * | |
36ade451 JK |
847 | * Also, this function can nest inside ext4_writepage(). In that case, we |
848 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
849 | * whole page. So we won't block on the journal in that case, which is good, | |
850 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 851 | * |
617ba13b | 852 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
853 | * quota file writes. If we were to commit the transaction while thus |
854 | * reentered, there can be a deadlock - we would be holding a quota | |
855 | * lock, and the commit would never complete if another thread had a | |
856 | * transaction open and was blocking on the quota lock - a ranking | |
857 | * violation. | |
858 | * | |
dab291af | 859 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
860 | * will _not_ run commit under these circumstances because handle->h_ref |
861 | * is elevated. We'll still have enough credits for the tiny quotafile | |
862 | * write. | |
863 | */ | |
f19d5870 TM |
864 | int do_journal_get_write_access(handle_t *handle, |
865 | struct buffer_head *bh) | |
ac27a0ec | 866 | { |
56d35a4c JK |
867 | int dirty = buffer_dirty(bh); |
868 | int ret; | |
869 | ||
ac27a0ec DK |
870 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
871 | return 0; | |
56d35a4c | 872 | /* |
ebdec241 | 873 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
874 | * the dirty bit as jbd2_journal_get_write_access() could complain |
875 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 876 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
877 | * the bit before releasing a page lock and thus writeback cannot |
878 | * ever write the buffer. | |
879 | */ | |
880 | if (dirty) | |
881 | clear_buffer_dirty(bh); | |
5d601255 | 882 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
883 | ret = ext4_journal_get_write_access(handle, bh); |
884 | if (!ret && dirty) | |
885 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
886 | return ret; | |
ac27a0ec DK |
887 | } |
888 | ||
8b0f165f AP |
889 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
890 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 891 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
892 | loff_t pos, unsigned len, unsigned flags, |
893 | struct page **pagep, void **fsdata) | |
ac27a0ec | 894 | { |
af5bc92d | 895 | struct inode *inode = mapping->host; |
1938a150 | 896 | int ret, needed_blocks; |
ac27a0ec DK |
897 | handle_t *handle; |
898 | int retries = 0; | |
af5bc92d | 899 | struct page *page; |
de9a55b8 | 900 | pgoff_t index; |
af5bc92d | 901 | unsigned from, to; |
bfc1af65 | 902 | |
9bffad1e | 903 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
904 | /* |
905 | * Reserve one block more for addition to orphan list in case | |
906 | * we allocate blocks but write fails for some reason | |
907 | */ | |
908 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 909 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
910 | from = pos & (PAGE_CACHE_SIZE - 1); |
911 | to = from + len; | |
ac27a0ec | 912 | |
f19d5870 TM |
913 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
914 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
915 | flags, pagep); | |
916 | if (ret < 0) | |
47564bfb TT |
917 | return ret; |
918 | if (ret == 1) | |
919 | return 0; | |
f19d5870 TM |
920 | } |
921 | ||
47564bfb TT |
922 | /* |
923 | * grab_cache_page_write_begin() can take a long time if the | |
924 | * system is thrashing due to memory pressure, or if the page | |
925 | * is being written back. So grab it first before we start | |
926 | * the transaction handle. This also allows us to allocate | |
927 | * the page (if needed) without using GFP_NOFS. | |
928 | */ | |
929 | retry_grab: | |
930 | page = grab_cache_page_write_begin(mapping, index, flags); | |
931 | if (!page) | |
932 | return -ENOMEM; | |
933 | unlock_page(page); | |
934 | ||
935 | retry_journal: | |
9924a92a | 936 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 937 | if (IS_ERR(handle)) { |
47564bfb TT |
938 | page_cache_release(page); |
939 | return PTR_ERR(handle); | |
7479d2b9 | 940 | } |
ac27a0ec | 941 | |
47564bfb TT |
942 | lock_page(page); |
943 | if (page->mapping != mapping) { | |
944 | /* The page got truncated from under us */ | |
945 | unlock_page(page); | |
946 | page_cache_release(page); | |
cf108bca | 947 | ext4_journal_stop(handle); |
47564bfb | 948 | goto retry_grab; |
cf108bca | 949 | } |
7afe5aa5 DM |
950 | /* In case writeback began while the page was unlocked */ |
951 | wait_for_stable_page(page); | |
cf108bca | 952 | |
744692dc | 953 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 954 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 955 | else |
6e1db88d | 956 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
957 | |
958 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 TM |
959 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
960 | from, to, NULL, | |
961 | do_journal_get_write_access); | |
ac27a0ec | 962 | } |
bfc1af65 NP |
963 | |
964 | if (ret) { | |
af5bc92d | 965 | unlock_page(page); |
ae4d5372 | 966 | /* |
6e1db88d | 967 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
968 | * outside i_size. Trim these off again. Don't need |
969 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
970 | * |
971 | * Add inode to orphan list in case we crash before | |
972 | * truncate finishes | |
ae4d5372 | 973 | */ |
ffacfa7a | 974 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
975 | ext4_orphan_add(handle, inode); |
976 | ||
977 | ext4_journal_stop(handle); | |
978 | if (pos + len > inode->i_size) { | |
b9a4207d | 979 | ext4_truncate_failed_write(inode); |
de9a55b8 | 980 | /* |
ffacfa7a | 981 | * If truncate failed early the inode might |
1938a150 AK |
982 | * still be on the orphan list; we need to |
983 | * make sure the inode is removed from the | |
984 | * orphan list in that case. | |
985 | */ | |
986 | if (inode->i_nlink) | |
987 | ext4_orphan_del(NULL, inode); | |
988 | } | |
bfc1af65 | 989 | |
47564bfb TT |
990 | if (ret == -ENOSPC && |
991 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
992 | goto retry_journal; | |
993 | page_cache_release(page); | |
994 | return ret; | |
995 | } | |
996 | *pagep = page; | |
ac27a0ec DK |
997 | return ret; |
998 | } | |
999 | ||
bfc1af65 NP |
1000 | /* For write_end() in data=journal mode */ |
1001 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1002 | { |
13fca323 | 1003 | int ret; |
ac27a0ec DK |
1004 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1005 | return 0; | |
1006 | set_buffer_uptodate(bh); | |
13fca323 TT |
1007 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1008 | clear_buffer_meta(bh); | |
1009 | clear_buffer_prio(bh); | |
1010 | return ret; | |
ac27a0ec DK |
1011 | } |
1012 | ||
eed4333f ZL |
1013 | /* |
1014 | * We need to pick up the new inode size which generic_commit_write gave us | |
1015 | * `file' can be NULL - eg, when called from page_symlink(). | |
1016 | * | |
1017 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1018 | * buffers are managed internally. | |
1019 | */ | |
1020 | static int ext4_write_end(struct file *file, | |
1021 | struct address_space *mapping, | |
1022 | loff_t pos, unsigned len, unsigned copied, | |
1023 | struct page *page, void *fsdata) | |
f8514083 | 1024 | { |
f8514083 | 1025 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1026 | struct inode *inode = mapping->host; |
0572639f | 1027 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1028 | int ret = 0, ret2; |
1029 | int i_size_changed = 0; | |
1030 | ||
1031 | trace_ext4_write_end(inode, pos, len, copied); | |
1032 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1033 | ret = ext4_jbd2_file_inode(handle, inode); | |
1034 | if (ret) { | |
1035 | unlock_page(page); | |
1036 | page_cache_release(page); | |
1037 | goto errout; | |
1038 | } | |
1039 | } | |
f8514083 | 1040 | |
42c832de TT |
1041 | if (ext4_has_inline_data(inode)) { |
1042 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1043 | copied, page); | |
1044 | if (ret < 0) | |
1045 | goto errout; | |
1046 | copied = ret; | |
1047 | } else | |
f19d5870 TM |
1048 | copied = block_write_end(file, mapping, pos, |
1049 | len, copied, page, fsdata); | |
f8514083 | 1050 | /* |
4631dbf6 | 1051 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1052 | * page writeout could otherwise come in and zero beyond i_size. |
1053 | */ | |
4631dbf6 | 1054 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 AK |
1055 | unlock_page(page); |
1056 | page_cache_release(page); | |
1057 | ||
0572639f XW |
1058 | if (old_size < pos) |
1059 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1060 | /* |
1061 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1062 | * makes the holding time of page lock longer. Second, it forces lock | |
1063 | * ordering of page lock and transaction start for journaling | |
1064 | * filesystems. | |
1065 | */ | |
1066 | if (i_size_changed) | |
1067 | ext4_mark_inode_dirty(handle, inode); | |
1068 | ||
ffacfa7a | 1069 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1070 | /* if we have allocated more blocks and copied |
1071 | * less. We will have blocks allocated outside | |
1072 | * inode->i_size. So truncate them | |
1073 | */ | |
1074 | ext4_orphan_add(handle, inode); | |
74d553aa | 1075 | errout: |
617ba13b | 1076 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1077 | if (!ret) |
1078 | ret = ret2; | |
bfc1af65 | 1079 | |
f8514083 | 1080 | if (pos + len > inode->i_size) { |
b9a4207d | 1081 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1082 | /* |
ffacfa7a | 1083 | * If truncate failed early the inode might still be |
f8514083 AK |
1084 | * on the orphan list; we need to make sure the inode |
1085 | * is removed from the orphan list in that case. | |
1086 | */ | |
1087 | if (inode->i_nlink) | |
1088 | ext4_orphan_del(NULL, inode); | |
1089 | } | |
1090 | ||
bfc1af65 | 1091 | return ret ? ret : copied; |
ac27a0ec DK |
1092 | } |
1093 | ||
bfc1af65 | 1094 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1095 | struct address_space *mapping, |
1096 | loff_t pos, unsigned len, unsigned copied, | |
1097 | struct page *page, void *fsdata) | |
ac27a0ec | 1098 | { |
617ba13b | 1099 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1100 | struct inode *inode = mapping->host; |
0572639f | 1101 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1102 | int ret = 0, ret2; |
1103 | int partial = 0; | |
bfc1af65 | 1104 | unsigned from, to; |
4631dbf6 | 1105 | int size_changed = 0; |
ac27a0ec | 1106 | |
9bffad1e | 1107 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1108 | from = pos & (PAGE_CACHE_SIZE - 1); |
1109 | to = from + len; | |
1110 | ||
441c8508 CW |
1111 | BUG_ON(!ext4_handle_valid(handle)); |
1112 | ||
3fdcfb66 TM |
1113 | if (ext4_has_inline_data(inode)) |
1114 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1115 | copied, page); | |
1116 | else { | |
1117 | if (copied < len) { | |
1118 | if (!PageUptodate(page)) | |
1119 | copied = 0; | |
1120 | page_zero_new_buffers(page, from+copied, to); | |
1121 | } | |
ac27a0ec | 1122 | |
3fdcfb66 TM |
1123 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1124 | to, &partial, write_end_fn); | |
1125 | if (!partial) | |
1126 | SetPageUptodate(page); | |
1127 | } | |
4631dbf6 | 1128 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1129 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1130 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 DM |
1131 | unlock_page(page); |
1132 | page_cache_release(page); | |
1133 | ||
0572639f XW |
1134 | if (old_size < pos) |
1135 | pagecache_isize_extended(inode, old_size, pos); | |
1136 | ||
4631dbf6 | 1137 | if (size_changed) { |
617ba13b | 1138 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1139 | if (!ret) |
1140 | ret = ret2; | |
1141 | } | |
bfc1af65 | 1142 | |
ffacfa7a | 1143 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1144 | /* if we have allocated more blocks and copied |
1145 | * less. We will have blocks allocated outside | |
1146 | * inode->i_size. So truncate them | |
1147 | */ | |
1148 | ext4_orphan_add(handle, inode); | |
1149 | ||
617ba13b | 1150 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1151 | if (!ret) |
1152 | ret = ret2; | |
f8514083 | 1153 | if (pos + len > inode->i_size) { |
b9a4207d | 1154 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1155 | /* |
ffacfa7a | 1156 | * If truncate failed early the inode might still be |
f8514083 AK |
1157 | * on the orphan list; we need to make sure the inode |
1158 | * is removed from the orphan list in that case. | |
1159 | */ | |
1160 | if (inode->i_nlink) | |
1161 | ext4_orphan_del(NULL, inode); | |
1162 | } | |
bfc1af65 NP |
1163 | |
1164 | return ret ? ret : copied; | |
ac27a0ec | 1165 | } |
d2a17637 | 1166 | |
9d0be502 | 1167 | /* |
7b415bf6 | 1168 | * Reserve a single cluster located at lblock |
9d0be502 | 1169 | */ |
01f49d0b | 1170 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1171 | { |
60e58e0f | 1172 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1173 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1174 | unsigned int md_needed; |
5dd4056d | 1175 | int ret; |
03179fe9 TT |
1176 | |
1177 | /* | |
1178 | * We will charge metadata quota at writeout time; this saves | |
1179 | * us from metadata over-estimation, though we may go over by | |
1180 | * a small amount in the end. Here we just reserve for data. | |
1181 | */ | |
1182 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1183 | if (ret) | |
1184 | return ret; | |
d2a17637 MC |
1185 | |
1186 | /* | |
1187 | * recalculate the amount of metadata blocks to reserve | |
1188 | * in order to allocate nrblocks | |
1189 | * worse case is one extent per block | |
1190 | */ | |
0637c6f4 | 1191 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1192 | /* |
1193 | * ext4_calc_metadata_amount() has side effects, which we have | |
1194 | * to be prepared undo if we fail to claim space. | |
1195 | */ | |
71d4f7d0 TT |
1196 | md_needed = 0; |
1197 | trace_ext4_da_reserve_space(inode, 0); | |
d2a17637 | 1198 | |
71d4f7d0 | 1199 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1200 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1201 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1202 | return -ENOSPC; |
1203 | } | |
9d0be502 | 1204 | ei->i_reserved_data_blocks++; |
0637c6f4 | 1205 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1206 | |
d2a17637 MC |
1207 | return 0; /* success */ |
1208 | } | |
1209 | ||
12219aea | 1210 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1211 | { |
1212 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1213 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1214 | |
cd213226 MC |
1215 | if (!to_free) |
1216 | return; /* Nothing to release, exit */ | |
1217 | ||
d2a17637 | 1218 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1219 | |
5a58ec87 | 1220 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1221 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1222 | /* |
0637c6f4 TT |
1223 | * if there aren't enough reserved blocks, then the |
1224 | * counter is messed up somewhere. Since this | |
1225 | * function is called from invalidate page, it's | |
1226 | * harmless to return without any action. | |
cd213226 | 1227 | */ |
8de5c325 | 1228 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1229 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1230 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1231 | ei->i_reserved_data_blocks); |
1232 | WARN_ON(1); | |
1233 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1234 | } |
0637c6f4 | 1235 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1236 | |
72b8ab9d | 1237 | /* update fs dirty data blocks counter */ |
57042651 | 1238 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1239 | |
d2a17637 | 1240 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1241 | |
7b415bf6 | 1242 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1243 | } |
1244 | ||
1245 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1246 | unsigned int offset, |
1247 | unsigned int length) | |
d2a17637 MC |
1248 | { |
1249 | int to_release = 0; | |
1250 | struct buffer_head *head, *bh; | |
1251 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1252 | struct inode *inode = page->mapping->host; |
1253 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1254 | unsigned int stop = offset + length; |
7b415bf6 | 1255 | int num_clusters; |
51865fda | 1256 | ext4_fsblk_t lblk; |
d2a17637 | 1257 | |
ca99fdd2 LC |
1258 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
1259 | ||
d2a17637 MC |
1260 | head = page_buffers(page); |
1261 | bh = head; | |
1262 | do { | |
1263 | unsigned int next_off = curr_off + bh->b_size; | |
1264 | ||
ca99fdd2 LC |
1265 | if (next_off > stop) |
1266 | break; | |
1267 | ||
d2a17637 MC |
1268 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1269 | to_release++; | |
1270 | clear_buffer_delay(bh); | |
1271 | } | |
1272 | curr_off = next_off; | |
1273 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1274 | |
51865fda ZL |
1275 | if (to_release) { |
1276 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1277 | ext4_es_remove_extent(inode, lblk, to_release); | |
1278 | } | |
1279 | ||
7b415bf6 AK |
1280 | /* If we have released all the blocks belonging to a cluster, then we |
1281 | * need to release the reserved space for that cluster. */ | |
1282 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1283 | while (num_clusters > 0) { | |
7b415bf6 AK |
1284 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1285 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1286 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1287 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1288 | ext4_da_release_space(inode, 1); |
1289 | ||
1290 | num_clusters--; | |
1291 | } | |
d2a17637 | 1292 | } |
ac27a0ec | 1293 | |
64769240 AT |
1294 | /* |
1295 | * Delayed allocation stuff | |
1296 | */ | |
1297 | ||
4e7ea81d JK |
1298 | struct mpage_da_data { |
1299 | struct inode *inode; | |
1300 | struct writeback_control *wbc; | |
6b523df4 | 1301 | |
4e7ea81d JK |
1302 | pgoff_t first_page; /* The first page to write */ |
1303 | pgoff_t next_page; /* Current page to examine */ | |
1304 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1305 | /* |
4e7ea81d JK |
1306 | * Extent to map - this can be after first_page because that can be |
1307 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1308 | * is delalloc or unwritten. | |
791b7f08 | 1309 | */ |
4e7ea81d JK |
1310 | struct ext4_map_blocks map; |
1311 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1312 | }; | |
64769240 | 1313 | |
4e7ea81d JK |
1314 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1315 | bool invalidate) | |
c4a0c46e AK |
1316 | { |
1317 | int nr_pages, i; | |
1318 | pgoff_t index, end; | |
1319 | struct pagevec pvec; | |
1320 | struct inode *inode = mpd->inode; | |
1321 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1322 | |
1323 | /* This is necessary when next_page == 0. */ | |
1324 | if (mpd->first_page >= mpd->next_page) | |
1325 | return; | |
c4a0c46e | 1326 | |
c7f5938a CW |
1327 | index = mpd->first_page; |
1328 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1329 | if (invalidate) { |
1330 | ext4_lblk_t start, last; | |
1331 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1332 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1333 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1334 | } | |
51865fda | 1335 | |
66bea92c | 1336 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1337 | while (index <= end) { |
1338 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1339 | if (nr_pages == 0) | |
1340 | break; | |
1341 | for (i = 0; i < nr_pages; i++) { | |
1342 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1343 | if (page->index > end) |
c4a0c46e | 1344 | break; |
c4a0c46e AK |
1345 | BUG_ON(!PageLocked(page)); |
1346 | BUG_ON(PageWriteback(page)); | |
4e7ea81d JK |
1347 | if (invalidate) { |
1348 | block_invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
1349 | ClearPageUptodate(page); | |
1350 | } | |
c4a0c46e AK |
1351 | unlock_page(page); |
1352 | } | |
9b1d0998 JK |
1353 | index = pvec.pages[nr_pages - 1]->index + 1; |
1354 | pagevec_release(&pvec); | |
c4a0c46e | 1355 | } |
c4a0c46e AK |
1356 | } |
1357 | ||
df22291f AK |
1358 | static void ext4_print_free_blocks(struct inode *inode) |
1359 | { | |
1360 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1361 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1362 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1363 | |
1364 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1365 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1366 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1367 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1368 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1369 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1370 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1371 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1372 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1373 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1374 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1375 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1376 | ei->i_reserved_data_blocks); |
df22291f AK |
1377 | return; |
1378 | } | |
1379 | ||
c364b22c | 1380 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1381 | { |
c364b22c | 1382 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1383 | } |
1384 | ||
5356f261 AK |
1385 | /* |
1386 | * This function is grabs code from the very beginning of | |
1387 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1388 | * time. This function looks up the requested blocks and sets the | |
1389 | * buffer delay bit under the protection of i_data_sem. | |
1390 | */ | |
1391 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1392 | struct ext4_map_blocks *map, | |
1393 | struct buffer_head *bh) | |
1394 | { | |
d100eef2 | 1395 | struct extent_status es; |
5356f261 AK |
1396 | int retval; |
1397 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1398 | #ifdef ES_AGGRESSIVE_TEST |
1399 | struct ext4_map_blocks orig_map; | |
1400 | ||
1401 | memcpy(&orig_map, map, sizeof(*map)); | |
1402 | #endif | |
5356f261 AK |
1403 | |
1404 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1405 | invalid_block = ~0; | |
1406 | ||
1407 | map->m_flags = 0; | |
1408 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1409 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1410 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1411 | |
1412 | /* Lookup extent status tree firstly */ | |
1413 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1414 | if (ext4_es_is_hole(&es)) { |
1415 | retval = 0; | |
c8b459f4 | 1416 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1417 | goto add_delayed; |
1418 | } | |
1419 | ||
1420 | /* | |
1421 | * Delayed extent could be allocated by fallocate. | |
1422 | * So we need to check it. | |
1423 | */ | |
1424 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1425 | map_bh(bh, inode->i_sb, invalid_block); | |
1426 | set_buffer_new(bh); | |
1427 | set_buffer_delay(bh); | |
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1432 | retval = es.es_len - (iblock - es.es_lblk); | |
1433 | if (retval > map->m_len) | |
1434 | retval = map->m_len; | |
1435 | map->m_len = retval; | |
1436 | if (ext4_es_is_written(&es)) | |
1437 | map->m_flags |= EXT4_MAP_MAPPED; | |
1438 | else if (ext4_es_is_unwritten(&es)) | |
1439 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1440 | else | |
1441 | BUG_ON(1); | |
1442 | ||
921f266b DM |
1443 | #ifdef ES_AGGRESSIVE_TEST |
1444 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1445 | #endif | |
d100eef2 ZL |
1446 | return retval; |
1447 | } | |
1448 | ||
5356f261 AK |
1449 | /* |
1450 | * Try to see if we can get the block without requesting a new | |
1451 | * file system block. | |
1452 | */ | |
c8b459f4 | 1453 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1454 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1455 | retval = 0; |
cbd7584e | 1456 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1457 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1458 | else |
2f8e0a7c | 1459 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1460 | |
d100eef2 | 1461 | add_delayed: |
5356f261 | 1462 | if (retval == 0) { |
f7fec032 | 1463 | int ret; |
5356f261 AK |
1464 | /* |
1465 | * XXX: __block_prepare_write() unmaps passed block, | |
1466 | * is it OK? | |
1467 | */ | |
386ad67c LC |
1468 | /* |
1469 | * If the block was allocated from previously allocated cluster, | |
1470 | * then we don't need to reserve it again. However we still need | |
1471 | * to reserve metadata for every block we're going to write. | |
1472 | */ | |
cbd7584e JK |
1473 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio <= 1 || |
1474 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { | |
f7fec032 ZL |
1475 | ret = ext4_da_reserve_space(inode, iblock); |
1476 | if (ret) { | |
5356f261 | 1477 | /* not enough space to reserve */ |
f7fec032 | 1478 | retval = ret; |
5356f261 | 1479 | goto out_unlock; |
f7fec032 | 1480 | } |
5356f261 AK |
1481 | } |
1482 | ||
f7fec032 ZL |
1483 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1484 | ~0, EXTENT_STATUS_DELAYED); | |
1485 | if (ret) { | |
1486 | retval = ret; | |
51865fda | 1487 | goto out_unlock; |
f7fec032 | 1488 | } |
51865fda | 1489 | |
5356f261 AK |
1490 | map_bh(bh, inode->i_sb, invalid_block); |
1491 | set_buffer_new(bh); | |
1492 | set_buffer_delay(bh); | |
f7fec032 ZL |
1493 | } else if (retval > 0) { |
1494 | int ret; | |
3be78c73 | 1495 | unsigned int status; |
f7fec032 | 1496 | |
44fb851d ZL |
1497 | if (unlikely(retval != map->m_len)) { |
1498 | ext4_warning(inode->i_sb, | |
1499 | "ES len assertion failed for inode " | |
1500 | "%lu: retval %d != map->m_len %d", | |
1501 | inode->i_ino, retval, map->m_len); | |
1502 | WARN_ON(1); | |
921f266b | 1503 | } |
921f266b | 1504 | |
f7fec032 ZL |
1505 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1506 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1507 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1508 | map->m_pblk, status); | |
1509 | if (ret != 0) | |
1510 | retval = ret; | |
5356f261 AK |
1511 | } |
1512 | ||
1513 | out_unlock: | |
1514 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1515 | ||
1516 | return retval; | |
1517 | } | |
1518 | ||
64769240 | 1519 | /* |
d91bd2c1 | 1520 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1521 | * ext4_da_write_begin(). It will either return mapped block or |
1522 | * reserve space for a single block. | |
29fa89d0 AK |
1523 | * |
1524 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1525 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1526 | * | |
1527 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1528 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1529 | * initialized properly. | |
64769240 | 1530 | */ |
9c3569b5 TM |
1531 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1532 | struct buffer_head *bh, int create) | |
64769240 | 1533 | { |
2ed88685 | 1534 | struct ext4_map_blocks map; |
64769240 AT |
1535 | int ret = 0; |
1536 | ||
1537 | BUG_ON(create == 0); | |
2ed88685 TT |
1538 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1539 | ||
1540 | map.m_lblk = iblock; | |
1541 | map.m_len = 1; | |
64769240 AT |
1542 | |
1543 | /* | |
1544 | * first, we need to know whether the block is allocated already | |
1545 | * preallocated blocks are unmapped but should treated | |
1546 | * the same as allocated blocks. | |
1547 | */ | |
5356f261 AK |
1548 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1549 | if (ret <= 0) | |
2ed88685 | 1550 | return ret; |
64769240 | 1551 | |
2ed88685 TT |
1552 | map_bh(bh, inode->i_sb, map.m_pblk); |
1553 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1554 | ||
1555 | if (buffer_unwritten(bh)) { | |
1556 | /* A delayed write to unwritten bh should be marked | |
1557 | * new and mapped. Mapped ensures that we don't do | |
1558 | * get_block multiple times when we write to the same | |
1559 | * offset and new ensures that we do proper zero out | |
1560 | * for partial write. | |
1561 | */ | |
1562 | set_buffer_new(bh); | |
c8205636 | 1563 | set_buffer_mapped(bh); |
2ed88685 TT |
1564 | } |
1565 | return 0; | |
64769240 | 1566 | } |
61628a3f | 1567 | |
62e086be AK |
1568 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1569 | { | |
1570 | get_bh(bh); | |
1571 | return 0; | |
1572 | } | |
1573 | ||
1574 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1575 | { | |
1576 | put_bh(bh); | |
1577 | return 0; | |
1578 | } | |
1579 | ||
1580 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1581 | unsigned int len) |
1582 | { | |
1583 | struct address_space *mapping = page->mapping; | |
1584 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1585 | struct buffer_head *page_bufs = NULL; |
62e086be | 1586 | handle_t *handle = NULL; |
3fdcfb66 TM |
1587 | int ret = 0, err = 0; |
1588 | int inline_data = ext4_has_inline_data(inode); | |
1589 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1590 | |
cb20d518 | 1591 | ClearPageChecked(page); |
3fdcfb66 TM |
1592 | |
1593 | if (inline_data) { | |
1594 | BUG_ON(page->index != 0); | |
1595 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1596 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1597 | if (inode_bh == NULL) | |
1598 | goto out; | |
1599 | } else { | |
1600 | page_bufs = page_buffers(page); | |
1601 | if (!page_bufs) { | |
1602 | BUG(); | |
1603 | goto out; | |
1604 | } | |
1605 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1606 | NULL, bget_one); | |
1607 | } | |
62e086be AK |
1608 | /* As soon as we unlock the page, it can go away, but we have |
1609 | * references to buffers so we are safe */ | |
1610 | unlock_page(page); | |
1611 | ||
9924a92a TT |
1612 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1613 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1614 | if (IS_ERR(handle)) { |
1615 | ret = PTR_ERR(handle); | |
1616 | goto out; | |
1617 | } | |
1618 | ||
441c8508 CW |
1619 | BUG_ON(!ext4_handle_valid(handle)); |
1620 | ||
3fdcfb66 | 1621 | if (inline_data) { |
5d601255 | 1622 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1623 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1624 | |
3fdcfb66 TM |
1625 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1626 | ||
1627 | } else { | |
1628 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1629 | do_journal_get_write_access); | |
1630 | ||
1631 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1632 | write_end_fn); | |
1633 | } | |
62e086be AK |
1634 | if (ret == 0) |
1635 | ret = err; | |
2d859db3 | 1636 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1637 | err = ext4_journal_stop(handle); |
1638 | if (!ret) | |
1639 | ret = err; | |
1640 | ||
3fdcfb66 | 1641 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1642 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1643 | NULL, bput_one); |
19f5fb7a | 1644 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1645 | out: |
3fdcfb66 | 1646 | brelse(inode_bh); |
62e086be AK |
1647 | return ret; |
1648 | } | |
1649 | ||
61628a3f | 1650 | /* |
43ce1d23 AK |
1651 | * Note that we don't need to start a transaction unless we're journaling data |
1652 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1653 | * need to file the inode to the transaction's list in ordered mode because if | |
1654 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1655 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1656 | * transaction the data will hit the disk. In case we are journaling data, we |
1657 | * cannot start transaction directly because transaction start ranks above page | |
1658 | * lock so we have to do some magic. | |
1659 | * | |
b920c755 | 1660 | * This function can get called via... |
20970ba6 | 1661 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1662 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1663 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1664 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1665 | * |
1666 | * We don't do any block allocation in this function. If we have page with | |
1667 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1668 | * is important for mmaped based write. So if we do with blocksize 1K | |
1669 | * truncate(f, 1024); | |
1670 | * a = mmap(f, 0, 4096); | |
1671 | * a[0] = 'a'; | |
1672 | * truncate(f, 4096); | |
1673 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1674 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1675 | * do_wp_page). So writepage should write the first block. If we modify |
1676 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1677 | * page_mkwrite callback will do the block allocation and mark the | |
1678 | * buffer_heads mapped. | |
1679 | * | |
1680 | * We redirty the page if we have any buffer_heads that is either delay or | |
1681 | * unwritten in the page. | |
1682 | * | |
1683 | * We can get recursively called as show below. | |
1684 | * | |
1685 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1686 | * ext4_writepage() | |
1687 | * | |
1688 | * But since we don't do any block allocation we should not deadlock. | |
1689 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1690 | */ |
43ce1d23 | 1691 | static int ext4_writepage(struct page *page, |
62e086be | 1692 | struct writeback_control *wbc) |
64769240 | 1693 | { |
f8bec370 | 1694 | int ret = 0; |
61628a3f | 1695 | loff_t size; |
498e5f24 | 1696 | unsigned int len; |
744692dc | 1697 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1698 | struct inode *inode = page->mapping->host; |
36ade451 | 1699 | struct ext4_io_submit io_submit; |
1c8349a1 | 1700 | bool keep_towrite = false; |
61628a3f | 1701 | |
a9c667f8 | 1702 | trace_ext4_writepage(page); |
f0e6c985 AK |
1703 | size = i_size_read(inode); |
1704 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1705 | len = size & ~PAGE_CACHE_MASK; | |
1706 | else | |
1707 | len = PAGE_CACHE_SIZE; | |
64769240 | 1708 | |
a42afc5f | 1709 | page_bufs = page_buffers(page); |
a42afc5f | 1710 | /* |
fe386132 JK |
1711 | * We cannot do block allocation or other extent handling in this |
1712 | * function. If there are buffers needing that, we have to redirty | |
1713 | * the page. But we may reach here when we do a journal commit via | |
1714 | * journal_submit_inode_data_buffers() and in that case we must write | |
1715 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 1716 | */ |
f19d5870 TM |
1717 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
1718 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 1719 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
1720 | if (current->flags & PF_MEMALLOC) { |
1721 | /* | |
1722 | * For memory cleaning there's no point in writing only | |
1723 | * some buffers. So just bail out. Warn if we came here | |
1724 | * from direct reclaim. | |
1725 | */ | |
1726 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
1727 | == PF_MEMALLOC); | |
f0e6c985 AK |
1728 | unlock_page(page); |
1729 | return 0; | |
1730 | } | |
1c8349a1 | 1731 | keep_towrite = true; |
a42afc5f | 1732 | } |
64769240 | 1733 | |
cb20d518 | 1734 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1735 | /* |
1736 | * It's mmapped pagecache. Add buffers and journal it. There | |
1737 | * doesn't seem much point in redirtying the page here. | |
1738 | */ | |
3f0ca309 | 1739 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1740 | |
97a851ed JK |
1741 | ext4_io_submit_init(&io_submit, wbc); |
1742 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
1743 | if (!io_submit.io_end) { | |
1744 | redirty_page_for_writepage(wbc, page); | |
1745 | unlock_page(page); | |
1746 | return -ENOMEM; | |
1747 | } | |
1c8349a1 | 1748 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 1749 | ext4_io_submit(&io_submit); |
97a851ed JK |
1750 | /* Drop io_end reference we got from init */ |
1751 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
1752 | return ret; |
1753 | } | |
1754 | ||
5f1132b2 JK |
1755 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
1756 | { | |
1757 | int len; | |
1758 | loff_t size = i_size_read(mpd->inode); | |
1759 | int err; | |
1760 | ||
1761 | BUG_ON(page->index != mpd->first_page); | |
1762 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1763 | len = size & ~PAGE_CACHE_MASK; | |
1764 | else | |
1765 | len = PAGE_CACHE_SIZE; | |
1766 | clear_page_dirty_for_io(page); | |
1c8349a1 | 1767 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
1768 | if (!err) |
1769 | mpd->wbc->nr_to_write--; | |
1770 | mpd->first_page++; | |
1771 | ||
1772 | return err; | |
1773 | } | |
1774 | ||
4e7ea81d JK |
1775 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
1776 | ||
61628a3f | 1777 | /* |
fffb2739 JK |
1778 | * mballoc gives us at most this number of blocks... |
1779 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 1780 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 1781 | */ |
fffb2739 | 1782 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 1783 | |
4e7ea81d JK |
1784 | /* |
1785 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
1786 | * | |
1787 | * @mpd - extent of blocks | |
1788 | * @lblk - logical number of the block in the file | |
09930042 | 1789 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 1790 | * |
09930042 JK |
1791 | * The function is used to collect contig. blocks in the same state. If the |
1792 | * buffer doesn't require mapping for writeback and we haven't started the | |
1793 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
1794 | * caller can write the buffer right away. Otherwise the function returns true | |
1795 | * if the block has been added to the extent, false if the block couldn't be | |
1796 | * added. | |
4e7ea81d | 1797 | */ |
09930042 JK |
1798 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
1799 | struct buffer_head *bh) | |
4e7ea81d JK |
1800 | { |
1801 | struct ext4_map_blocks *map = &mpd->map; | |
1802 | ||
09930042 JK |
1803 | /* Buffer that doesn't need mapping for writeback? */ |
1804 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
1805 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
1806 | /* So far no extent to map => we write the buffer right away */ | |
1807 | if (map->m_len == 0) | |
1808 | return true; | |
1809 | return false; | |
1810 | } | |
4e7ea81d JK |
1811 | |
1812 | /* First block in the extent? */ | |
1813 | if (map->m_len == 0) { | |
1814 | map->m_lblk = lblk; | |
1815 | map->m_len = 1; | |
09930042 JK |
1816 | map->m_flags = bh->b_state & BH_FLAGS; |
1817 | return true; | |
4e7ea81d JK |
1818 | } |
1819 | ||
09930042 JK |
1820 | /* Don't go larger than mballoc is willing to allocate */ |
1821 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
1822 | return false; | |
1823 | ||
4e7ea81d JK |
1824 | /* Can we merge the block to our big extent? */ |
1825 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 1826 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 1827 | map->m_len++; |
09930042 | 1828 | return true; |
4e7ea81d | 1829 | } |
09930042 | 1830 | return false; |
4e7ea81d JK |
1831 | } |
1832 | ||
5f1132b2 JK |
1833 | /* |
1834 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
1835 | * | |
1836 | * @mpd - extent of blocks for mapping | |
1837 | * @head - the first buffer in the page | |
1838 | * @bh - buffer we should start processing from | |
1839 | * @lblk - logical number of the block in the file corresponding to @bh | |
1840 | * | |
1841 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
1842 | * the page for IO if all buffers in this page were mapped and there's no | |
1843 | * accumulated extent of buffers to map or add buffers in the page to the | |
1844 | * extent of buffers to map. The function returns 1 if the caller can continue | |
1845 | * by processing the next page, 0 if it should stop adding buffers to the | |
1846 | * extent to map because we cannot extend it anymore. It can also return value | |
1847 | * < 0 in case of error during IO submission. | |
1848 | */ | |
1849 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
1850 | struct buffer_head *head, | |
1851 | struct buffer_head *bh, | |
1852 | ext4_lblk_t lblk) | |
4e7ea81d JK |
1853 | { |
1854 | struct inode *inode = mpd->inode; | |
5f1132b2 | 1855 | int err; |
4e7ea81d JK |
1856 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
1857 | >> inode->i_blkbits; | |
1858 | ||
1859 | do { | |
1860 | BUG_ON(buffer_locked(bh)); | |
1861 | ||
09930042 | 1862 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
1863 | /* Found extent to map? */ |
1864 | if (mpd->map.m_len) | |
5f1132b2 | 1865 | return 0; |
09930042 | 1866 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 1867 | break; |
4e7ea81d | 1868 | } |
4e7ea81d | 1869 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
1870 | /* So far everything mapped? Submit the page for IO. */ |
1871 | if (mpd->map.m_len == 0) { | |
1872 | err = mpage_submit_page(mpd, head->b_page); | |
1873 | if (err < 0) | |
1874 | return err; | |
1875 | } | |
1876 | return lblk < blocks; | |
4e7ea81d JK |
1877 | } |
1878 | ||
1879 | /* | |
1880 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
1881 | * submit fully mapped pages for IO | |
1882 | * | |
1883 | * @mpd - description of extent to map, on return next extent to map | |
1884 | * | |
1885 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
1886 | * to be already locked) and update buffer state according to new extent state. | |
1887 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 1888 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
1889 | * and do extent conversion after IO is finished. If the last page is not fully |
1890 | * mapped, we update @map to the next extent in the last page that needs | |
1891 | * mapping. Otherwise we submit the page for IO. | |
1892 | */ | |
1893 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
1894 | { | |
1895 | struct pagevec pvec; | |
1896 | int nr_pages, i; | |
1897 | struct inode *inode = mpd->inode; | |
1898 | struct buffer_head *head, *bh; | |
1899 | int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; | |
4e7ea81d JK |
1900 | pgoff_t start, end; |
1901 | ext4_lblk_t lblk; | |
1902 | sector_t pblock; | |
1903 | int err; | |
1904 | ||
1905 | start = mpd->map.m_lblk >> bpp_bits; | |
1906 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
1907 | lblk = start << bpp_bits; | |
1908 | pblock = mpd->map.m_pblk; | |
1909 | ||
1910 | pagevec_init(&pvec, 0); | |
1911 | while (start <= end) { | |
1912 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
1913 | PAGEVEC_SIZE); | |
1914 | if (nr_pages == 0) | |
1915 | break; | |
1916 | for (i = 0; i < nr_pages; i++) { | |
1917 | struct page *page = pvec.pages[i]; | |
1918 | ||
1919 | if (page->index > end) | |
1920 | break; | |
70261f56 | 1921 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
1922 | BUG_ON(page->index != start); |
1923 | bh = head = page_buffers(page); | |
1924 | do { | |
1925 | if (lblk < mpd->map.m_lblk) | |
1926 | continue; | |
1927 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
1928 | /* | |
1929 | * Buffer after end of mapped extent. | |
1930 | * Find next buffer in the page to map. | |
1931 | */ | |
1932 | mpd->map.m_len = 0; | |
1933 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
1934 | /* |
1935 | * FIXME: If dioread_nolock supports | |
1936 | * blocksize < pagesize, we need to make | |
1937 | * sure we add size mapped so far to | |
1938 | * io_end->size as the following call | |
1939 | * can submit the page for IO. | |
1940 | */ | |
1941 | err = mpage_process_page_bufs(mpd, head, | |
1942 | bh, lblk); | |
4e7ea81d | 1943 | pagevec_release(&pvec); |
5f1132b2 JK |
1944 | if (err > 0) |
1945 | err = 0; | |
1946 | return err; | |
4e7ea81d JK |
1947 | } |
1948 | if (buffer_delay(bh)) { | |
1949 | clear_buffer_delay(bh); | |
1950 | bh->b_blocknr = pblock++; | |
1951 | } | |
4e7ea81d | 1952 | clear_buffer_unwritten(bh); |
5f1132b2 | 1953 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
1954 | |
1955 | /* | |
1956 | * FIXME: This is going to break if dioread_nolock | |
1957 | * supports blocksize < pagesize as we will try to | |
1958 | * convert potentially unmapped parts of inode. | |
1959 | */ | |
1960 | mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; | |
1961 | /* Page fully mapped - let IO run! */ | |
1962 | err = mpage_submit_page(mpd, page); | |
1963 | if (err < 0) { | |
1964 | pagevec_release(&pvec); | |
1965 | return err; | |
1966 | } | |
1967 | start++; | |
1968 | } | |
1969 | pagevec_release(&pvec); | |
1970 | } | |
1971 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
1972 | mpd->map.m_len = 0; | |
1973 | mpd->map.m_flags = 0; | |
1974 | return 0; | |
1975 | } | |
1976 | ||
1977 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
1978 | { | |
1979 | struct inode *inode = mpd->inode; | |
1980 | struct ext4_map_blocks *map = &mpd->map; | |
1981 | int get_blocks_flags; | |
090f32ee | 1982 | int err, dioread_nolock; |
4e7ea81d JK |
1983 | |
1984 | trace_ext4_da_write_pages_extent(inode, map); | |
1985 | /* | |
1986 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 1987 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
1988 | * where we have written into one or more preallocated blocks). It is |
1989 | * possible that we're going to need more metadata blocks than | |
1990 | * previously reserved. However we must not fail because we're in | |
1991 | * writeback and there is nothing we can do about it so it might result | |
1992 | * in data loss. So use reserved blocks to allocate metadata if | |
1993 | * possible. | |
1994 | * | |
754cfed6 TT |
1995 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
1996 | * the blocks in question are delalloc blocks. This indicates | |
1997 | * that the blocks and quotas has already been checked when | |
1998 | * the data was copied into the page cache. | |
4e7ea81d JK |
1999 | */ |
2000 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
2001 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
090f32ee LC |
2002 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2003 | if (dioread_nolock) | |
4e7ea81d JK |
2004 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2005 | if (map->m_flags & (1 << BH_Delay)) | |
2006 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2007 | ||
2008 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2009 | if (err < 0) | |
2010 | return err; | |
090f32ee | 2011 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2012 | if (!mpd->io_submit.io_end->handle && |
2013 | ext4_handle_valid(handle)) { | |
2014 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2015 | handle->h_rsv_handle = NULL; | |
2016 | } | |
3613d228 | 2017 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2018 | } |
4e7ea81d JK |
2019 | |
2020 | BUG_ON(map->m_len == 0); | |
2021 | if (map->m_flags & EXT4_MAP_NEW) { | |
2022 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2023 | int i; | |
2024 | ||
2025 | for (i = 0; i < map->m_len; i++) | |
2026 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2027 | } | |
2028 | return 0; | |
2029 | } | |
2030 | ||
2031 | /* | |
2032 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2033 | * mpd->len and submit pages underlying it for IO | |
2034 | * | |
2035 | * @handle - handle for journal operations | |
2036 | * @mpd - extent to map | |
7534e854 JK |
2037 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2038 | * is no hope of writing the data. The caller should discard | |
2039 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2040 | * |
2041 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2042 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2043 | * them to initialized or split the described range from larger unwritten | |
2044 | * extent. Note that we need not map all the described range since allocation | |
2045 | * can return less blocks or the range is covered by more unwritten extents. We | |
2046 | * cannot map more because we are limited by reserved transaction credits. On | |
2047 | * the other hand we always make sure that the last touched page is fully | |
2048 | * mapped so that it can be written out (and thus forward progress is | |
2049 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2050 | */ | |
2051 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2052 | struct mpage_da_data *mpd, |
2053 | bool *give_up_on_write) | |
4e7ea81d JK |
2054 | { |
2055 | struct inode *inode = mpd->inode; | |
2056 | struct ext4_map_blocks *map = &mpd->map; | |
2057 | int err; | |
2058 | loff_t disksize; | |
6603120e | 2059 | int progress = 0; |
4e7ea81d JK |
2060 | |
2061 | mpd->io_submit.io_end->offset = | |
2062 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2063 | do { |
4e7ea81d JK |
2064 | err = mpage_map_one_extent(handle, mpd); |
2065 | if (err < 0) { | |
2066 | struct super_block *sb = inode->i_sb; | |
2067 | ||
cb530541 TT |
2068 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2069 | goto invalidate_dirty_pages; | |
4e7ea81d | 2070 | /* |
cb530541 TT |
2071 | * Let the uper layers retry transient errors. |
2072 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2073 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2074 | */ |
cb530541 | 2075 | if ((err == -ENOMEM) || |
6603120e DM |
2076 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2077 | if (progress) | |
2078 | goto update_disksize; | |
cb530541 | 2079 | return err; |
6603120e | 2080 | } |
cb530541 TT |
2081 | ext4_msg(sb, KERN_CRIT, |
2082 | "Delayed block allocation failed for " | |
2083 | "inode %lu at logical offset %llu with" | |
2084 | " max blocks %u with error %d", | |
2085 | inode->i_ino, | |
2086 | (unsigned long long)map->m_lblk, | |
2087 | (unsigned)map->m_len, -err); | |
2088 | ext4_msg(sb, KERN_CRIT, | |
2089 | "This should not happen!! Data will " | |
2090 | "be lost\n"); | |
2091 | if (err == -ENOSPC) | |
2092 | ext4_print_free_blocks(inode); | |
2093 | invalidate_dirty_pages: | |
2094 | *give_up_on_write = true; | |
4e7ea81d JK |
2095 | return err; |
2096 | } | |
6603120e | 2097 | progress = 1; |
4e7ea81d JK |
2098 | /* |
2099 | * Update buffer state, submit mapped pages, and get us new | |
2100 | * extent to map | |
2101 | */ | |
2102 | err = mpage_map_and_submit_buffers(mpd); | |
2103 | if (err < 0) | |
6603120e | 2104 | goto update_disksize; |
27d7c4ed | 2105 | } while (map->m_len); |
4e7ea81d | 2106 | |
6603120e | 2107 | update_disksize: |
622cad13 TT |
2108 | /* |
2109 | * Update on-disk size after IO is submitted. Races with | |
2110 | * truncate are avoided by checking i_size under i_data_sem. | |
2111 | */ | |
4e7ea81d | 2112 | disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; |
4e7ea81d JK |
2113 | if (disksize > EXT4_I(inode)->i_disksize) { |
2114 | int err2; | |
622cad13 TT |
2115 | loff_t i_size; |
2116 | ||
2117 | down_write(&EXT4_I(inode)->i_data_sem); | |
2118 | i_size = i_size_read(inode); | |
2119 | if (disksize > i_size) | |
2120 | disksize = i_size; | |
2121 | if (disksize > EXT4_I(inode)->i_disksize) | |
2122 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2123 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2124 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2125 | if (err2) |
2126 | ext4_error(inode->i_sb, | |
2127 | "Failed to mark inode %lu dirty", | |
2128 | inode->i_ino); | |
2129 | if (!err) | |
2130 | err = err2; | |
2131 | } | |
2132 | return err; | |
2133 | } | |
2134 | ||
fffb2739 JK |
2135 | /* |
2136 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2137 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2138 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2139 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2140 | * bpp - 1 blocks in bpp different extents. | |
2141 | */ | |
525f4ed8 MC |
2142 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2143 | { | |
fffb2739 | 2144 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2145 | |
fffb2739 JK |
2146 | return ext4_meta_trans_blocks(inode, |
2147 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2148 | } |
61628a3f | 2149 | |
8e48dcfb | 2150 | /* |
4e7ea81d JK |
2151 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2152 | * and underlying extent to map | |
2153 | * | |
2154 | * @mpd - where to look for pages | |
2155 | * | |
2156 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2157 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2158 | * extent of buffers underlying these pages that needs mapping (formed by | |
2159 | * either delayed or unwritten buffers). We also lock the pages containing | |
2160 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2161 | * mpd->lblk with length mpd->len blocks). | |
2162 | * | |
2163 | * Note that this function can attach bios to one io_end structure which are | |
2164 | * neither logically nor physically contiguous. Although it may seem as an | |
2165 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2166 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2167 | */ |
4e7ea81d | 2168 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2169 | { |
4e7ea81d JK |
2170 | struct address_space *mapping = mpd->inode->i_mapping; |
2171 | struct pagevec pvec; | |
2172 | unsigned int nr_pages; | |
aeac589a | 2173 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2174 | pgoff_t index = mpd->first_page; |
2175 | pgoff_t end = mpd->last_page; | |
2176 | int tag; | |
2177 | int i, err = 0; | |
2178 | int blkbits = mpd->inode->i_blkbits; | |
2179 | ext4_lblk_t lblk; | |
2180 | struct buffer_head *head; | |
8e48dcfb | 2181 | |
4e7ea81d | 2182 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2183 | tag = PAGECACHE_TAG_TOWRITE; |
2184 | else | |
2185 | tag = PAGECACHE_TAG_DIRTY; | |
2186 | ||
4e7ea81d JK |
2187 | pagevec_init(&pvec, 0); |
2188 | mpd->map.m_len = 0; | |
2189 | mpd->next_page = index; | |
4f01b02c | 2190 | while (index <= end) { |
5b41d924 | 2191 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2192 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2193 | if (nr_pages == 0) | |
4e7ea81d | 2194 | goto out; |
8e48dcfb TT |
2195 | |
2196 | for (i = 0; i < nr_pages; i++) { | |
2197 | struct page *page = pvec.pages[i]; | |
2198 | ||
2199 | /* | |
2200 | * At this point, the page may be truncated or | |
2201 | * invalidated (changing page->mapping to NULL), or | |
2202 | * even swizzled back from swapper_space to tmpfs file | |
2203 | * mapping. However, page->index will not change | |
2204 | * because we have a reference on the page. | |
2205 | */ | |
4f01b02c TT |
2206 | if (page->index > end) |
2207 | goto out; | |
8e48dcfb | 2208 | |
aeac589a ML |
2209 | /* |
2210 | * Accumulated enough dirty pages? This doesn't apply | |
2211 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2212 | * keep going because someone may be concurrently | |
2213 | * dirtying pages, and we might have synced a lot of | |
2214 | * newly appeared dirty pages, but have not synced all | |
2215 | * of the old dirty pages. | |
2216 | */ | |
2217 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2218 | goto out; | |
2219 | ||
4e7ea81d JK |
2220 | /* If we can't merge this page, we are done. */ |
2221 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2222 | goto out; | |
78aaced3 | 2223 | |
8e48dcfb | 2224 | lock_page(page); |
8e48dcfb | 2225 | /* |
4e7ea81d JK |
2226 | * If the page is no longer dirty, or its mapping no |
2227 | * longer corresponds to inode we are writing (which | |
2228 | * means it has been truncated or invalidated), or the | |
2229 | * page is already under writeback and we are not doing | |
2230 | * a data integrity writeback, skip the page | |
8e48dcfb | 2231 | */ |
4f01b02c TT |
2232 | if (!PageDirty(page) || |
2233 | (PageWriteback(page) && | |
4e7ea81d | 2234 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2235 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2236 | unlock_page(page); |
2237 | continue; | |
2238 | } | |
2239 | ||
7cb1a535 | 2240 | wait_on_page_writeback(page); |
8e48dcfb | 2241 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2242 | |
4e7ea81d | 2243 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2244 | mpd->first_page = page->index; |
8eb9e5ce | 2245 | mpd->next_page = page->index + 1; |
f8bec370 | 2246 | /* Add all dirty buffers to mpd */ |
4e7ea81d JK |
2247 | lblk = ((ext4_lblk_t)page->index) << |
2248 | (PAGE_CACHE_SHIFT - blkbits); | |
f8bec370 | 2249 | head = page_buffers(page); |
5f1132b2 JK |
2250 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2251 | if (err <= 0) | |
4e7ea81d | 2252 | goto out; |
5f1132b2 | 2253 | err = 0; |
aeac589a | 2254 | left--; |
8e48dcfb TT |
2255 | } |
2256 | pagevec_release(&pvec); | |
2257 | cond_resched(); | |
2258 | } | |
4f01b02c | 2259 | return 0; |
8eb9e5ce TT |
2260 | out: |
2261 | pagevec_release(&pvec); | |
4e7ea81d | 2262 | return err; |
8e48dcfb TT |
2263 | } |
2264 | ||
20970ba6 TT |
2265 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2266 | void *data) | |
2267 | { | |
2268 | struct address_space *mapping = data; | |
2269 | int ret = ext4_writepage(page, wbc); | |
2270 | mapping_set_error(mapping, ret); | |
2271 | return ret; | |
2272 | } | |
2273 | ||
2274 | static int ext4_writepages(struct address_space *mapping, | |
2275 | struct writeback_control *wbc) | |
64769240 | 2276 | { |
4e7ea81d JK |
2277 | pgoff_t writeback_index = 0; |
2278 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2279 | int range_whole = 0; |
4e7ea81d | 2280 | int cycled = 1; |
61628a3f | 2281 | handle_t *handle = NULL; |
df22291f | 2282 | struct mpage_da_data mpd; |
5e745b04 | 2283 | struct inode *inode = mapping->host; |
6b523df4 | 2284 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2285 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2286 | bool done; |
1bce63d1 | 2287 | struct blk_plug plug; |
cb530541 | 2288 | bool give_up_on_write = false; |
61628a3f | 2289 | |
20970ba6 | 2290 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2291 | |
61628a3f MC |
2292 | /* |
2293 | * No pages to write? This is mainly a kludge to avoid starting | |
2294 | * a transaction for special inodes like journal inode on last iput() | |
2295 | * because that could violate lock ordering on umount | |
2296 | */ | |
a1d6cc56 | 2297 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2298 | goto out_writepages; |
2a21e37e | 2299 | |
20970ba6 TT |
2300 | if (ext4_should_journal_data(inode)) { |
2301 | struct blk_plug plug; | |
20970ba6 TT |
2302 | |
2303 | blk_start_plug(&plug); | |
2304 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2305 | blk_finish_plug(&plug); | |
bbf023c7 | 2306 | goto out_writepages; |
20970ba6 TT |
2307 | } |
2308 | ||
2a21e37e TT |
2309 | /* |
2310 | * If the filesystem has aborted, it is read-only, so return | |
2311 | * right away instead of dumping stack traces later on that | |
2312 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2313 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2314 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2315 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2316 | * *never* be called, so if that ever happens, we would want |
2317 | * the stack trace. | |
2318 | */ | |
bbf023c7 ML |
2319 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2320 | ret = -EROFS; | |
2321 | goto out_writepages; | |
2322 | } | |
2a21e37e | 2323 | |
6b523df4 JK |
2324 | if (ext4_should_dioread_nolock(inode)) { |
2325 | /* | |
70261f56 | 2326 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2327 | * the page and we may dirty the inode. |
2328 | */ | |
2329 | rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); | |
2330 | } | |
2331 | ||
4e7ea81d JK |
2332 | /* |
2333 | * If we have inline data and arrive here, it means that | |
2334 | * we will soon create the block for the 1st page, so | |
2335 | * we'd better clear the inline data here. | |
2336 | */ | |
2337 | if (ext4_has_inline_data(inode)) { | |
2338 | /* Just inode will be modified... */ | |
2339 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2340 | if (IS_ERR(handle)) { | |
2341 | ret = PTR_ERR(handle); | |
2342 | goto out_writepages; | |
2343 | } | |
2344 | BUG_ON(ext4_test_inode_state(inode, | |
2345 | EXT4_STATE_MAY_INLINE_DATA)); | |
2346 | ext4_destroy_inline_data(handle, inode); | |
2347 | ext4_journal_stop(handle); | |
2348 | } | |
2349 | ||
22208ded AK |
2350 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2351 | range_whole = 1; | |
61628a3f | 2352 | |
2acf2c26 | 2353 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2354 | writeback_index = mapping->writeback_index; |
2355 | if (writeback_index) | |
2acf2c26 | 2356 | cycled = 0; |
4e7ea81d JK |
2357 | mpd.first_page = writeback_index; |
2358 | mpd.last_page = -1; | |
5b41d924 | 2359 | } else { |
4e7ea81d JK |
2360 | mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; |
2361 | mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; | |
5b41d924 | 2362 | } |
a1d6cc56 | 2363 | |
4e7ea81d JK |
2364 | mpd.inode = inode; |
2365 | mpd.wbc = wbc; | |
2366 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2367 | retry: |
6e6938b6 | 2368 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2369 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2370 | done = false; | |
1bce63d1 | 2371 | blk_start_plug(&plug); |
4e7ea81d JK |
2372 | while (!done && mpd.first_page <= mpd.last_page) { |
2373 | /* For each extent of pages we use new io_end */ | |
2374 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2375 | if (!mpd.io_submit.io_end) { | |
2376 | ret = -ENOMEM; | |
2377 | break; | |
2378 | } | |
a1d6cc56 AK |
2379 | |
2380 | /* | |
4e7ea81d JK |
2381 | * We have two constraints: We find one extent to map and we |
2382 | * must always write out whole page (makes a difference when | |
2383 | * blocksize < pagesize) so that we don't block on IO when we | |
2384 | * try to write out the rest of the page. Journalled mode is | |
2385 | * not supported by delalloc. | |
a1d6cc56 AK |
2386 | */ |
2387 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2388 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2389 | |
4e7ea81d | 2390 | /* start a new transaction */ |
6b523df4 JK |
2391 | handle = ext4_journal_start_with_reserve(inode, |
2392 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2393 | if (IS_ERR(handle)) { |
2394 | ret = PTR_ERR(handle); | |
1693918e | 2395 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2396 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2397 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2398 | /* Release allocated io_end */ |
2399 | ext4_put_io_end(mpd.io_submit.io_end); | |
2400 | break; | |
61628a3f | 2401 | } |
f63e6005 | 2402 | |
4e7ea81d JK |
2403 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2404 | ret = mpage_prepare_extent_to_map(&mpd); | |
2405 | if (!ret) { | |
2406 | if (mpd.map.m_len) | |
cb530541 TT |
2407 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2408 | &give_up_on_write); | |
4e7ea81d JK |
2409 | else { |
2410 | /* | |
2411 | * We scanned the whole range (or exhausted | |
2412 | * nr_to_write), submitted what was mapped and | |
2413 | * didn't find anything needing mapping. We are | |
2414 | * done. | |
2415 | */ | |
2416 | done = true; | |
2417 | } | |
f63e6005 | 2418 | } |
61628a3f | 2419 | ext4_journal_stop(handle); |
4e7ea81d JK |
2420 | /* Submit prepared bio */ |
2421 | ext4_io_submit(&mpd.io_submit); | |
2422 | /* Unlock pages we didn't use */ | |
cb530541 | 2423 | mpage_release_unused_pages(&mpd, give_up_on_write); |
4e7ea81d JK |
2424 | /* Drop our io_end reference we got from init */ |
2425 | ext4_put_io_end(mpd.io_submit.io_end); | |
2426 | ||
2427 | if (ret == -ENOSPC && sbi->s_journal) { | |
2428 | /* | |
2429 | * Commit the transaction which would | |
22208ded AK |
2430 | * free blocks released in the transaction |
2431 | * and try again | |
2432 | */ | |
df22291f | 2433 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2434 | ret = 0; |
4e7ea81d JK |
2435 | continue; |
2436 | } | |
2437 | /* Fatal error - ENOMEM, EIO... */ | |
2438 | if (ret) | |
61628a3f | 2439 | break; |
a1d6cc56 | 2440 | } |
1bce63d1 | 2441 | blk_finish_plug(&plug); |
9c12a831 | 2442 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2443 | cycled = 1; |
4e7ea81d JK |
2444 | mpd.last_page = writeback_index - 1; |
2445 | mpd.first_page = 0; | |
2acf2c26 AK |
2446 | goto retry; |
2447 | } | |
22208ded AK |
2448 | |
2449 | /* Update index */ | |
22208ded AK |
2450 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2451 | /* | |
4e7ea81d | 2452 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2453 | * mode will write it back later |
2454 | */ | |
4e7ea81d | 2455 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2456 | |
61628a3f | 2457 | out_writepages: |
20970ba6 TT |
2458 | trace_ext4_writepages_result(inode, wbc, ret, |
2459 | nr_to_write - wbc->nr_to_write); | |
61628a3f | 2460 | return ret; |
64769240 AT |
2461 | } |
2462 | ||
79f0be8d AK |
2463 | static int ext4_nonda_switch(struct super_block *sb) |
2464 | { | |
5c1ff336 | 2465 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2466 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2467 | ||
2468 | /* | |
2469 | * switch to non delalloc mode if we are running low | |
2470 | * on free block. The free block accounting via percpu | |
179f7ebf | 2471 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2472 | * accumulated on each CPU without updating global counters |
2473 | * Delalloc need an accurate free block accounting. So switch | |
2474 | * to non delalloc when we are near to error range. | |
2475 | */ | |
5c1ff336 EW |
2476 | free_clusters = |
2477 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2478 | dirty_clusters = | |
2479 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2480 | /* |
2481 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2482 | */ | |
5c1ff336 | 2483 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2484 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2485 | |
5c1ff336 EW |
2486 | if (2 * free_clusters < 3 * dirty_clusters || |
2487 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2488 | /* |
c8afb446 ES |
2489 | * free block count is less than 150% of dirty blocks |
2490 | * or free blocks is less than watermark | |
79f0be8d AK |
2491 | */ |
2492 | return 1; | |
2493 | } | |
2494 | return 0; | |
2495 | } | |
2496 | ||
0ff8947f ES |
2497 | /* We always reserve for an inode update; the superblock could be there too */ |
2498 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2499 | { | |
2500 | if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
2501 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE))) | |
2502 | return 1; | |
2503 | ||
2504 | if (pos + len <= 0x7fffffffULL) | |
2505 | return 1; | |
2506 | ||
2507 | /* We might need to update the superblock to set LARGE_FILE */ | |
2508 | return 2; | |
2509 | } | |
2510 | ||
64769240 | 2511 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2512 | loff_t pos, unsigned len, unsigned flags, |
2513 | struct page **pagep, void **fsdata) | |
64769240 | 2514 | { |
72b8ab9d | 2515 | int ret, retries = 0; |
64769240 AT |
2516 | struct page *page; |
2517 | pgoff_t index; | |
64769240 AT |
2518 | struct inode *inode = mapping->host; |
2519 | handle_t *handle; | |
2520 | ||
2521 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2522 | |
2523 | if (ext4_nonda_switch(inode->i_sb)) { | |
2524 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2525 | return ext4_write_begin(file, mapping, pos, | |
2526 | len, flags, pagep, fsdata); | |
2527 | } | |
2528 | *fsdata = (void *)0; | |
9bffad1e | 2529 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2530 | |
2531 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2532 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2533 | pos, len, flags, | |
2534 | pagep, fsdata); | |
2535 | if (ret < 0) | |
47564bfb TT |
2536 | return ret; |
2537 | if (ret == 1) | |
2538 | return 0; | |
9c3569b5 TM |
2539 | } |
2540 | ||
47564bfb TT |
2541 | /* |
2542 | * grab_cache_page_write_begin() can take a long time if the | |
2543 | * system is thrashing due to memory pressure, or if the page | |
2544 | * is being written back. So grab it first before we start | |
2545 | * the transaction handle. This also allows us to allocate | |
2546 | * the page (if needed) without using GFP_NOFS. | |
2547 | */ | |
2548 | retry_grab: | |
2549 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2550 | if (!page) | |
2551 | return -ENOMEM; | |
2552 | unlock_page(page); | |
2553 | ||
64769240 AT |
2554 | /* |
2555 | * With delayed allocation, we don't log the i_disksize update | |
2556 | * if there is delayed block allocation. But we still need | |
2557 | * to journalling the i_disksize update if writes to the end | |
2558 | * of file which has an already mapped buffer. | |
2559 | */ | |
47564bfb | 2560 | retry_journal: |
0ff8947f ES |
2561 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2562 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2563 | if (IS_ERR(handle)) { |
47564bfb TT |
2564 | page_cache_release(page); |
2565 | return PTR_ERR(handle); | |
64769240 AT |
2566 | } |
2567 | ||
47564bfb TT |
2568 | lock_page(page); |
2569 | if (page->mapping != mapping) { | |
2570 | /* The page got truncated from under us */ | |
2571 | unlock_page(page); | |
2572 | page_cache_release(page); | |
d5a0d4f7 | 2573 | ext4_journal_stop(handle); |
47564bfb | 2574 | goto retry_grab; |
d5a0d4f7 | 2575 | } |
47564bfb | 2576 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2577 | wait_for_stable_page(page); |
64769240 | 2578 | |
6e1db88d | 2579 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2580 | if (ret < 0) { |
2581 | unlock_page(page); | |
2582 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2583 | /* |
2584 | * block_write_begin may have instantiated a few blocks | |
2585 | * outside i_size. Trim these off again. Don't need | |
2586 | * i_size_read because we hold i_mutex. | |
2587 | */ | |
2588 | if (pos + len > inode->i_size) | |
b9a4207d | 2589 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2590 | |
2591 | if (ret == -ENOSPC && | |
2592 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2593 | goto retry_journal; | |
2594 | ||
2595 | page_cache_release(page); | |
2596 | return ret; | |
64769240 AT |
2597 | } |
2598 | ||
47564bfb | 2599 | *pagep = page; |
64769240 AT |
2600 | return ret; |
2601 | } | |
2602 | ||
632eaeab MC |
2603 | /* |
2604 | * Check if we should update i_disksize | |
2605 | * when write to the end of file but not require block allocation | |
2606 | */ | |
2607 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2608 | unsigned long offset) |
632eaeab MC |
2609 | { |
2610 | struct buffer_head *bh; | |
2611 | struct inode *inode = page->mapping->host; | |
2612 | unsigned int idx; | |
2613 | int i; | |
2614 | ||
2615 | bh = page_buffers(page); | |
2616 | idx = offset >> inode->i_blkbits; | |
2617 | ||
af5bc92d | 2618 | for (i = 0; i < idx; i++) |
632eaeab MC |
2619 | bh = bh->b_this_page; |
2620 | ||
29fa89d0 | 2621 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2622 | return 0; |
2623 | return 1; | |
2624 | } | |
2625 | ||
64769240 | 2626 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2627 | struct address_space *mapping, |
2628 | loff_t pos, unsigned len, unsigned copied, | |
2629 | struct page *page, void *fsdata) | |
64769240 AT |
2630 | { |
2631 | struct inode *inode = mapping->host; | |
2632 | int ret = 0, ret2; | |
2633 | handle_t *handle = ext4_journal_current_handle(); | |
2634 | loff_t new_i_size; | |
632eaeab | 2635 | unsigned long start, end; |
79f0be8d AK |
2636 | int write_mode = (int)(unsigned long)fsdata; |
2637 | ||
74d553aa TT |
2638 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2639 | return ext4_write_end(file, mapping, pos, | |
2640 | len, copied, page, fsdata); | |
632eaeab | 2641 | |
9bffad1e | 2642 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2643 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2644 | end = start + copied - 1; |
64769240 AT |
2645 | |
2646 | /* | |
2647 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2648 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2649 | * into that. | |
2650 | */ | |
64769240 | 2651 | new_i_size = pos + copied; |
ea51d132 | 2652 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2653 | if (ext4_has_inline_data(inode) || |
2654 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 2655 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
2656 | /* We need to mark inode dirty even if |
2657 | * new_i_size is less that inode->i_size | |
2658 | * bu greater than i_disksize.(hint delalloc) | |
2659 | */ | |
2660 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2661 | } |
632eaeab | 2662 | } |
9c3569b5 TM |
2663 | |
2664 | if (write_mode != CONVERT_INLINE_DATA && | |
2665 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2666 | ext4_has_inline_data(inode)) | |
2667 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2668 | page); | |
2669 | else | |
2670 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2671 | page, fsdata); |
9c3569b5 | 2672 | |
64769240 AT |
2673 | copied = ret2; |
2674 | if (ret2 < 0) | |
2675 | ret = ret2; | |
2676 | ret2 = ext4_journal_stop(handle); | |
2677 | if (!ret) | |
2678 | ret = ret2; | |
2679 | ||
2680 | return ret ? ret : copied; | |
2681 | } | |
2682 | ||
d47992f8 LC |
2683 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
2684 | unsigned int length) | |
64769240 | 2685 | { |
64769240 AT |
2686 | /* |
2687 | * Drop reserved blocks | |
2688 | */ | |
2689 | BUG_ON(!PageLocked(page)); | |
2690 | if (!page_has_buffers(page)) | |
2691 | goto out; | |
2692 | ||
ca99fdd2 | 2693 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
2694 | |
2695 | out: | |
d47992f8 | 2696 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
2697 | |
2698 | return; | |
2699 | } | |
2700 | ||
ccd2506b TT |
2701 | /* |
2702 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2703 | */ | |
2704 | int ext4_alloc_da_blocks(struct inode *inode) | |
2705 | { | |
fb40ba0d TT |
2706 | trace_ext4_alloc_da_blocks(inode); |
2707 | ||
71d4f7d0 | 2708 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
2709 | return 0; |
2710 | ||
2711 | /* | |
2712 | * We do something simple for now. The filemap_flush() will | |
2713 | * also start triggering a write of the data blocks, which is | |
2714 | * not strictly speaking necessary (and for users of | |
2715 | * laptop_mode, not even desirable). However, to do otherwise | |
2716 | * would require replicating code paths in: | |
de9a55b8 | 2717 | * |
20970ba6 | 2718 | * ext4_writepages() -> |
ccd2506b TT |
2719 | * write_cache_pages() ---> (via passed in callback function) |
2720 | * __mpage_da_writepage() --> | |
2721 | * mpage_add_bh_to_extent() | |
2722 | * mpage_da_map_blocks() | |
2723 | * | |
2724 | * The problem is that write_cache_pages(), located in | |
2725 | * mm/page-writeback.c, marks pages clean in preparation for | |
2726 | * doing I/O, which is not desirable if we're not planning on | |
2727 | * doing I/O at all. | |
2728 | * | |
2729 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2730 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2731 | * would be ugly in the extreme. So instead we would need to |
2732 | * replicate parts of the code in the above functions, | |
25985edc | 2733 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2734 | * write out the pages, but rather only collect contiguous |
2735 | * logical block extents, call the multi-block allocator, and | |
2736 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2737 | * |
ccd2506b TT |
2738 | * For now, though, we'll cheat by calling filemap_flush(), |
2739 | * which will map the blocks, and start the I/O, but not | |
2740 | * actually wait for the I/O to complete. | |
2741 | */ | |
2742 | return filemap_flush(inode->i_mapping); | |
2743 | } | |
64769240 | 2744 | |
ac27a0ec DK |
2745 | /* |
2746 | * bmap() is special. It gets used by applications such as lilo and by | |
2747 | * the swapper to find the on-disk block of a specific piece of data. | |
2748 | * | |
2749 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2750 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2751 | * filesystem and enables swap, then they may get a nasty shock when the |
2752 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2753 | * the original zero's written out previously to the journal and | |
2754 | * awaiting writeback in the kernel's buffer cache. | |
2755 | * | |
2756 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2757 | * take extra steps to flush any blocks which might be in the cache. | |
2758 | */ | |
617ba13b | 2759 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2760 | { |
2761 | struct inode *inode = mapping->host; | |
2762 | journal_t *journal; | |
2763 | int err; | |
2764 | ||
46c7f254 TM |
2765 | /* |
2766 | * We can get here for an inline file via the FIBMAP ioctl | |
2767 | */ | |
2768 | if (ext4_has_inline_data(inode)) | |
2769 | return 0; | |
2770 | ||
64769240 AT |
2771 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2772 | test_opt(inode->i_sb, DELALLOC)) { | |
2773 | /* | |
2774 | * With delalloc we want to sync the file | |
2775 | * so that we can make sure we allocate | |
2776 | * blocks for file | |
2777 | */ | |
2778 | filemap_write_and_wait(mapping); | |
2779 | } | |
2780 | ||
19f5fb7a TT |
2781 | if (EXT4_JOURNAL(inode) && |
2782 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2783 | /* |
2784 | * This is a REALLY heavyweight approach, but the use of | |
2785 | * bmap on dirty files is expected to be extremely rare: | |
2786 | * only if we run lilo or swapon on a freshly made file | |
2787 | * do we expect this to happen. | |
2788 | * | |
2789 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2790 | * represent an unprivileged user DOS attack --- we'd be | |
2791 | * in trouble if mortal users could trigger this path at | |
2792 | * will.) | |
2793 | * | |
617ba13b | 2794 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2795 | * regular files. If somebody wants to bmap a directory |
2796 | * or symlink and gets confused because the buffer | |
2797 | * hasn't yet been flushed to disk, they deserve | |
2798 | * everything they get. | |
2799 | */ | |
2800 | ||
19f5fb7a | 2801 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2802 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2803 | jbd2_journal_lock_updates(journal); |
2804 | err = jbd2_journal_flush(journal); | |
2805 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2806 | |
2807 | if (err) | |
2808 | return 0; | |
2809 | } | |
2810 | ||
af5bc92d | 2811 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2812 | } |
2813 | ||
617ba13b | 2814 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2815 | { |
46c7f254 TM |
2816 | int ret = -EAGAIN; |
2817 | struct inode *inode = page->mapping->host; | |
2818 | ||
0562e0ba | 2819 | trace_ext4_readpage(page); |
46c7f254 TM |
2820 | |
2821 | if (ext4_has_inline_data(inode)) | |
2822 | ret = ext4_readpage_inline(inode, page); | |
2823 | ||
2824 | if (ret == -EAGAIN) | |
2825 | return mpage_readpage(page, ext4_get_block); | |
2826 | ||
2827 | return ret; | |
ac27a0ec DK |
2828 | } |
2829 | ||
2830 | static int | |
617ba13b | 2831 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2832 | struct list_head *pages, unsigned nr_pages) |
2833 | { | |
46c7f254 TM |
2834 | struct inode *inode = mapping->host; |
2835 | ||
2836 | /* If the file has inline data, no need to do readpages. */ | |
2837 | if (ext4_has_inline_data(inode)) | |
2838 | return 0; | |
2839 | ||
617ba13b | 2840 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2841 | } |
2842 | ||
d47992f8 LC |
2843 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
2844 | unsigned int length) | |
ac27a0ec | 2845 | { |
ca99fdd2 | 2846 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 2847 | |
4520fb3c JK |
2848 | /* No journalling happens on data buffers when this function is used */ |
2849 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
2850 | ||
ca99fdd2 | 2851 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
2852 | } |
2853 | ||
53e87268 | 2854 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
2855 | unsigned int offset, |
2856 | unsigned int length) | |
4520fb3c JK |
2857 | { |
2858 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
2859 | ||
ca99fdd2 | 2860 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 2861 | |
ac27a0ec DK |
2862 | /* |
2863 | * If it's a full truncate we just forget about the pending dirtying | |
2864 | */ | |
ca99fdd2 | 2865 | if (offset == 0 && length == PAGE_CACHE_SIZE) |
ac27a0ec DK |
2866 | ClearPageChecked(page); |
2867 | ||
ca99fdd2 | 2868 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
2869 | } |
2870 | ||
2871 | /* Wrapper for aops... */ | |
2872 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
2873 | unsigned int offset, |
2874 | unsigned int length) | |
53e87268 | 2875 | { |
ca99fdd2 | 2876 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
2877 | } |
2878 | ||
617ba13b | 2879 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2880 | { |
617ba13b | 2881 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2882 | |
0562e0ba JZ |
2883 | trace_ext4_releasepage(page); |
2884 | ||
e1c36595 JK |
2885 | /* Page has dirty journalled data -> cannot release */ |
2886 | if (PageChecked(page)) | |
ac27a0ec | 2887 | return 0; |
0390131b FM |
2888 | if (journal) |
2889 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
2890 | else | |
2891 | return try_to_free_buffers(page); | |
ac27a0ec DK |
2892 | } |
2893 | ||
2ed88685 TT |
2894 | /* |
2895 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
2896 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
2897 | * The extent will be converted to initialized after the IO is complete. | |
2898 | */ | |
f19d5870 | 2899 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
2900 | struct buffer_head *bh_result, int create) |
2901 | { | |
c7064ef1 | 2902 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 2903 | inode->i_ino, create); |
2ed88685 TT |
2904 | return _ext4_get_block(inode, iblock, bh_result, |
2905 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
2906 | } |
2907 | ||
729f52c6 | 2908 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 2909 | struct buffer_head *bh_result, int create) |
729f52c6 | 2910 | { |
8b0f165f AP |
2911 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
2912 | inode->i_ino, create); | |
2913 | return _ext4_get_block(inode, iblock, bh_result, | |
2914 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
2915 | } |
2916 | ||
4c0425ff | 2917 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 2918 | ssize_t size, void *private) |
4c0425ff MC |
2919 | { |
2920 | ext4_io_end_t *io_end = iocb->private; | |
4c0425ff | 2921 | |
97a851ed | 2922 | /* if not async direct IO just return */ |
7b7a8665 | 2923 | if (!io_end) |
97a851ed | 2924 | return; |
4b70df18 | 2925 | |
88635ca2 | 2926 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 2927 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
2928 | iocb->private, io_end->inode->i_ino, iocb, offset, |
2929 | size); | |
8d5d02e6 | 2930 | |
b5a7e970 | 2931 | iocb->private = NULL; |
4c0425ff MC |
2932 | io_end->offset = offset; |
2933 | io_end->size = size; | |
7b7a8665 | 2934 | ext4_put_io_end(io_end); |
4c0425ff | 2935 | } |
c7064ef1 | 2936 | |
4c0425ff MC |
2937 | /* |
2938 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
2939 | * preallocated extents, and those write extend the file, no need to | |
2940 | * fall back to buffered IO. | |
2941 | * | |
556615dc | 2942 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 2943 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 2944 | * still keep the range to write as unwritten. |
4c0425ff | 2945 | * |
69c499d1 | 2946 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 2947 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 2948 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 2949 | * when async direct IO completed. |
4c0425ff MC |
2950 | * |
2951 | * If the O_DIRECT write will extend the file then add this inode to the | |
2952 | * orphan list. So recovery will truncate it back to the original size | |
2953 | * if the machine crashes during the write. | |
2954 | * | |
2955 | */ | |
2956 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
16b1f05d | 2957 | struct iov_iter *iter, loff_t offset) |
4c0425ff MC |
2958 | { |
2959 | struct file *file = iocb->ki_filp; | |
2960 | struct inode *inode = file->f_mapping->host; | |
2961 | ssize_t ret; | |
a6cbcd4a | 2962 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
2963 | int overwrite = 0; |
2964 | get_block_t *get_block_func = NULL; | |
2965 | int dio_flags = 0; | |
4c0425ff | 2966 | loff_t final_size = offset + count; |
97a851ed | 2967 | ext4_io_end_t *io_end = NULL; |
729f52c6 | 2968 | |
69c499d1 TT |
2969 | /* Use the old path for reads and writes beyond i_size. */ |
2970 | if (rw != WRITE || final_size > inode->i_size) | |
16b1f05d | 2971 | return ext4_ind_direct_IO(rw, iocb, iter, offset); |
4bd809db | 2972 | |
69c499d1 | 2973 | BUG_ON(iocb->private == NULL); |
4bd809db | 2974 | |
e8340395 JK |
2975 | /* |
2976 | * Make all waiters for direct IO properly wait also for extent | |
2977 | * conversion. This also disallows race between truncate() and | |
2978 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
2979 | */ | |
2980 | if (rw == WRITE) | |
2981 | atomic_inc(&inode->i_dio_count); | |
2982 | ||
69c499d1 TT |
2983 | /* If we do a overwrite dio, i_mutex locking can be released */ |
2984 | overwrite = *((int *)iocb->private); | |
4bd809db | 2985 | |
69c499d1 | 2986 | if (overwrite) { |
69c499d1 TT |
2987 | down_read(&EXT4_I(inode)->i_data_sem); |
2988 | mutex_unlock(&inode->i_mutex); | |
2989 | } | |
8d5d02e6 | 2990 | |
69c499d1 TT |
2991 | /* |
2992 | * We could direct write to holes and fallocate. | |
2993 | * | |
2994 | * Allocated blocks to fill the hole are marked as | |
556615dc | 2995 | * unwritten to prevent parallel buffered read to expose |
69c499d1 TT |
2996 | * the stale data before DIO complete the data IO. |
2997 | * | |
2998 | * As to previously fallocated extents, ext4 get_block will | |
2999 | * just simply mark the buffer mapped but still keep the | |
556615dc | 3000 | * extents unwritten. |
69c499d1 TT |
3001 | * |
3002 | * For non AIO case, we will convert those unwritten extents | |
3003 | * to written after return back from blockdev_direct_IO. | |
3004 | * | |
3005 | * For async DIO, the conversion needs to be deferred when the | |
3006 | * IO is completed. The ext4 end_io callback function will be | |
3007 | * called to take care of the conversion work. Here for async | |
3008 | * case, we allocate an io_end structure to hook to the iocb. | |
3009 | */ | |
3010 | iocb->private = NULL; | |
3011 | ext4_inode_aio_set(inode, NULL); | |
3012 | if (!is_sync_kiocb(iocb)) { | |
97a851ed | 3013 | io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3014 | if (!io_end) { |
3015 | ret = -ENOMEM; | |
3016 | goto retake_lock; | |
8b0f165f | 3017 | } |
97a851ed JK |
3018 | /* |
3019 | * Grab reference for DIO. Will be dropped in ext4_end_io_dio() | |
3020 | */ | |
3021 | iocb->private = ext4_get_io_end(io_end); | |
8d5d02e6 | 3022 | /* |
69c499d1 TT |
3023 | * we save the io structure for current async direct |
3024 | * IO, so that later ext4_map_blocks() could flag the | |
3025 | * io structure whether there is a unwritten extents | |
3026 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3027 | */ |
69c499d1 TT |
3028 | ext4_inode_aio_set(inode, io_end); |
3029 | } | |
4bd809db | 3030 | |
69c499d1 TT |
3031 | if (overwrite) { |
3032 | get_block_func = ext4_get_block_write_nolock; | |
3033 | } else { | |
3034 | get_block_func = ext4_get_block_write; | |
3035 | dio_flags = DIO_LOCKING; | |
3036 | } | |
923ae0ff RZ |
3037 | if (IS_DAX(inode)) |
3038 | ret = dax_do_io(rw, iocb, inode, iter, offset, get_block_func, | |
3039 | ext4_end_io_dio, dio_flags); | |
3040 | else | |
3041 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
3042 | inode->i_sb->s_bdev, iter, offset, | |
3043 | get_block_func, | |
3044 | ext4_end_io_dio, NULL, dio_flags); | |
69c499d1 | 3045 | |
69c499d1 | 3046 | /* |
97a851ed JK |
3047 | * Put our reference to io_end. This can free the io_end structure e.g. |
3048 | * in sync IO case or in case of error. It can even perform extent | |
3049 | * conversion if all bios we submitted finished before we got here. | |
3050 | * Note that in that case iocb->private can be already set to NULL | |
3051 | * here. | |
69c499d1 | 3052 | */ |
97a851ed JK |
3053 | if (io_end) { |
3054 | ext4_inode_aio_set(inode, NULL); | |
3055 | ext4_put_io_end(io_end); | |
3056 | /* | |
3057 | * When no IO was submitted ext4_end_io_dio() was not | |
3058 | * called so we have to put iocb's reference. | |
3059 | */ | |
3060 | if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { | |
3061 | WARN_ON(iocb->private != io_end); | |
3062 | WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
97a851ed JK |
3063 | ext4_put_io_end(io_end); |
3064 | iocb->private = NULL; | |
3065 | } | |
3066 | } | |
3067 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3068 | EXT4_STATE_DIO_UNWRITTEN)) { |
3069 | int err; | |
3070 | /* | |
3071 | * for non AIO case, since the IO is already | |
3072 | * completed, we could do the conversion right here | |
3073 | */ | |
6b523df4 | 3074 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3075 | offset, ret); |
3076 | if (err < 0) | |
3077 | ret = err; | |
3078 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3079 | } | |
4bd809db | 3080 | |
69c499d1 | 3081 | retake_lock: |
e8340395 JK |
3082 | if (rw == WRITE) |
3083 | inode_dio_done(inode); | |
69c499d1 TT |
3084 | /* take i_mutex locking again if we do a ovewrite dio */ |
3085 | if (overwrite) { | |
69c499d1 TT |
3086 | up_read(&EXT4_I(inode)->i_data_sem); |
3087 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3088 | } |
8d5d02e6 | 3089 | |
69c499d1 | 3090 | return ret; |
4c0425ff MC |
3091 | } |
3092 | ||
3093 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
d8d3d94b | 3094 | struct iov_iter *iter, loff_t offset) |
4c0425ff MC |
3095 | { |
3096 | struct file *file = iocb->ki_filp; | |
3097 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3098 | size_t count = iov_iter_count(iter); |
0562e0ba | 3099 | ssize_t ret; |
4c0425ff | 3100 | |
84ebd795 TT |
3101 | /* |
3102 | * If we are doing data journalling we don't support O_DIRECT | |
3103 | */ | |
3104 | if (ext4_should_journal_data(inode)) | |
3105 | return 0; | |
3106 | ||
46c7f254 TM |
3107 | /* Let buffer I/O handle the inline data case. */ |
3108 | if (ext4_has_inline_data(inode)) | |
3109 | return 0; | |
3110 | ||
a6cbcd4a | 3111 | trace_ext4_direct_IO_enter(inode, offset, count, rw); |
12e9b892 | 3112 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
16b1f05d | 3113 | ret = ext4_ext_direct_IO(rw, iocb, iter, offset); |
0562e0ba | 3114 | else |
16b1f05d | 3115 | ret = ext4_ind_direct_IO(rw, iocb, iter, offset); |
a6cbcd4a | 3116 | trace_ext4_direct_IO_exit(inode, offset, count, rw, ret); |
0562e0ba | 3117 | return ret; |
4c0425ff MC |
3118 | } |
3119 | ||
ac27a0ec | 3120 | /* |
617ba13b | 3121 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3122 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3123 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3124 | * not necessarily locked. | |
3125 | * | |
3126 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3127 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3128 | * or jbddirty because all the journalling code will explode. | |
3129 | * | |
3130 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3131 | * is called, propagate that into the buffers appropriately. | |
3132 | */ | |
617ba13b | 3133 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3134 | { |
3135 | SetPageChecked(page); | |
3136 | return __set_page_dirty_nobuffers(page); | |
3137 | } | |
3138 | ||
74d553aa | 3139 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3140 | .readpage = ext4_readpage, |
3141 | .readpages = ext4_readpages, | |
43ce1d23 | 3142 | .writepage = ext4_writepage, |
20970ba6 | 3143 | .writepages = ext4_writepages, |
8ab22b9a | 3144 | .write_begin = ext4_write_begin, |
74d553aa | 3145 | .write_end = ext4_write_end, |
8ab22b9a HH |
3146 | .bmap = ext4_bmap, |
3147 | .invalidatepage = ext4_invalidatepage, | |
3148 | .releasepage = ext4_releasepage, | |
3149 | .direct_IO = ext4_direct_IO, | |
3150 | .migratepage = buffer_migrate_page, | |
3151 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3152 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3153 | }; |
3154 | ||
617ba13b | 3155 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3156 | .readpage = ext4_readpage, |
3157 | .readpages = ext4_readpages, | |
43ce1d23 | 3158 | .writepage = ext4_writepage, |
20970ba6 | 3159 | .writepages = ext4_writepages, |
8ab22b9a HH |
3160 | .write_begin = ext4_write_begin, |
3161 | .write_end = ext4_journalled_write_end, | |
3162 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3163 | .bmap = ext4_bmap, | |
4520fb3c | 3164 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3165 | .releasepage = ext4_releasepage, |
84ebd795 | 3166 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3167 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3168 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3169 | }; |
3170 | ||
64769240 | 3171 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3172 | .readpage = ext4_readpage, |
3173 | .readpages = ext4_readpages, | |
43ce1d23 | 3174 | .writepage = ext4_writepage, |
20970ba6 | 3175 | .writepages = ext4_writepages, |
8ab22b9a HH |
3176 | .write_begin = ext4_da_write_begin, |
3177 | .write_end = ext4_da_write_end, | |
3178 | .bmap = ext4_bmap, | |
3179 | .invalidatepage = ext4_da_invalidatepage, | |
3180 | .releasepage = ext4_releasepage, | |
3181 | .direct_IO = ext4_direct_IO, | |
3182 | .migratepage = buffer_migrate_page, | |
3183 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3184 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3185 | }; |
3186 | ||
617ba13b | 3187 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3188 | { |
3d2b1582 LC |
3189 | switch (ext4_inode_journal_mode(inode)) { |
3190 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3191 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3192 | break; |
3193 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3194 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3195 | break; |
3196 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3197 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3198 | return; |
3d2b1582 LC |
3199 | default: |
3200 | BUG(); | |
3201 | } | |
74d553aa TT |
3202 | if (test_opt(inode->i_sb, DELALLOC)) |
3203 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3204 | else | |
3205 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3206 | } |
3207 | ||
923ae0ff | 3208 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3209 | struct address_space *mapping, loff_t from, loff_t length) |
3210 | { | |
3211 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3212 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
923ae0ff | 3213 | unsigned blocksize, pos; |
d863dc36 LC |
3214 | ext4_lblk_t iblock; |
3215 | struct inode *inode = mapping->host; | |
3216 | struct buffer_head *bh; | |
3217 | struct page *page; | |
3218 | int err = 0; | |
3219 | ||
3220 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3221 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3222 | if (!page) | |
3223 | return -ENOMEM; | |
3224 | ||
3225 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 LC |
3226 | |
3227 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3228 | ||
3229 | if (!page_has_buffers(page)) | |
3230 | create_empty_buffers(page, blocksize, 0); | |
3231 | ||
3232 | /* Find the buffer that contains "offset" */ | |
3233 | bh = page_buffers(page); | |
3234 | pos = blocksize; | |
3235 | while (offset >= pos) { | |
3236 | bh = bh->b_this_page; | |
3237 | iblock++; | |
3238 | pos += blocksize; | |
3239 | } | |
d863dc36 LC |
3240 | if (buffer_freed(bh)) { |
3241 | BUFFER_TRACE(bh, "freed: skip"); | |
3242 | goto unlock; | |
3243 | } | |
d863dc36 LC |
3244 | if (!buffer_mapped(bh)) { |
3245 | BUFFER_TRACE(bh, "unmapped"); | |
3246 | ext4_get_block(inode, iblock, bh, 0); | |
3247 | /* unmapped? It's a hole - nothing to do */ | |
3248 | if (!buffer_mapped(bh)) { | |
3249 | BUFFER_TRACE(bh, "still unmapped"); | |
3250 | goto unlock; | |
3251 | } | |
3252 | } | |
3253 | ||
3254 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3255 | if (PageUptodate(page)) | |
3256 | set_buffer_uptodate(bh); | |
3257 | ||
3258 | if (!buffer_uptodate(bh)) { | |
3259 | err = -EIO; | |
3260 | ll_rw_block(READ, 1, &bh); | |
3261 | wait_on_buffer(bh); | |
3262 | /* Uhhuh. Read error. Complain and punt. */ | |
3263 | if (!buffer_uptodate(bh)) | |
3264 | goto unlock; | |
3265 | } | |
d863dc36 LC |
3266 | if (ext4_should_journal_data(inode)) { |
3267 | BUFFER_TRACE(bh, "get write access"); | |
3268 | err = ext4_journal_get_write_access(handle, bh); | |
3269 | if (err) | |
3270 | goto unlock; | |
3271 | } | |
d863dc36 | 3272 | zero_user(page, offset, length); |
d863dc36 LC |
3273 | BUFFER_TRACE(bh, "zeroed end of block"); |
3274 | ||
d863dc36 LC |
3275 | if (ext4_should_journal_data(inode)) { |
3276 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3277 | } else { |
353eefd3 | 3278 | err = 0; |
d863dc36 | 3279 | mark_buffer_dirty(bh); |
0713ed0c LC |
3280 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) |
3281 | err = ext4_jbd2_file_inode(handle, inode); | |
3282 | } | |
d863dc36 LC |
3283 | |
3284 | unlock: | |
3285 | unlock_page(page); | |
3286 | page_cache_release(page); | |
3287 | return err; | |
3288 | } | |
3289 | ||
923ae0ff RZ |
3290 | /* |
3291 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3292 | * starting from file offset 'from'. The range to be zero'd must | |
3293 | * be contained with in one block. If the specified range exceeds | |
3294 | * the end of the block it will be shortened to end of the block | |
3295 | * that cooresponds to 'from' | |
3296 | */ | |
3297 | static int ext4_block_zero_page_range(handle_t *handle, | |
3298 | struct address_space *mapping, loff_t from, loff_t length) | |
3299 | { | |
3300 | struct inode *inode = mapping->host; | |
3301 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3302 | unsigned blocksize = inode->i_sb->s_blocksize; | |
3303 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3304 | ||
3305 | /* | |
3306 | * correct length if it does not fall between | |
3307 | * 'from' and the end of the block | |
3308 | */ | |
3309 | if (length > max || length < 0) | |
3310 | length = max; | |
3311 | ||
3312 | if (IS_DAX(inode)) | |
3313 | return dax_zero_page_range(inode, from, length, ext4_get_block); | |
3314 | return __ext4_block_zero_page_range(handle, mapping, from, length); | |
3315 | } | |
3316 | ||
94350ab5 MW |
3317 | /* |
3318 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3319 | * up to the end of the block which corresponds to `from'. | |
3320 | * This required during truncate. We need to physically zero the tail end | |
3321 | * of that block so it doesn't yield old data if the file is later grown. | |
3322 | */ | |
c197855e | 3323 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3324 | struct address_space *mapping, loff_t from) |
3325 | { | |
3326 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3327 | unsigned length; | |
3328 | unsigned blocksize; | |
3329 | struct inode *inode = mapping->host; | |
3330 | ||
3331 | blocksize = inode->i_sb->s_blocksize; | |
3332 | length = blocksize - (offset & (blocksize - 1)); | |
3333 | ||
3334 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3335 | } | |
3336 | ||
a87dd18c LC |
3337 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3338 | loff_t lstart, loff_t length) | |
3339 | { | |
3340 | struct super_block *sb = inode->i_sb; | |
3341 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3342 | unsigned partial_start, partial_end; |
a87dd18c LC |
3343 | ext4_fsblk_t start, end; |
3344 | loff_t byte_end = (lstart + length - 1); | |
3345 | int err = 0; | |
3346 | ||
e1be3a92 LC |
3347 | partial_start = lstart & (sb->s_blocksize - 1); |
3348 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3349 | ||
a87dd18c LC |
3350 | start = lstart >> sb->s_blocksize_bits; |
3351 | end = byte_end >> sb->s_blocksize_bits; | |
3352 | ||
3353 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3354 | if (start == end && |
3355 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3356 | err = ext4_block_zero_page_range(handle, mapping, |
3357 | lstart, length); | |
3358 | return err; | |
3359 | } | |
3360 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3361 | if (partial_start) { |
a87dd18c LC |
3362 | err = ext4_block_zero_page_range(handle, mapping, |
3363 | lstart, sb->s_blocksize); | |
3364 | if (err) | |
3365 | return err; | |
3366 | } | |
3367 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3368 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3369 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3370 | byte_end - partial_end, |
3371 | partial_end + 1); | |
a87dd18c LC |
3372 | return err; |
3373 | } | |
3374 | ||
91ef4caf DG |
3375 | int ext4_can_truncate(struct inode *inode) |
3376 | { | |
91ef4caf DG |
3377 | if (S_ISREG(inode->i_mode)) |
3378 | return 1; | |
3379 | if (S_ISDIR(inode->i_mode)) | |
3380 | return 1; | |
3381 | if (S_ISLNK(inode->i_mode)) | |
3382 | return !ext4_inode_is_fast_symlink(inode); | |
3383 | return 0; | |
3384 | } | |
3385 | ||
a4bb6b64 AH |
3386 | /* |
3387 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3388 | * associated with the given offset and length | |
3389 | * | |
3390 | * @inode: File inode | |
3391 | * @offset: The offset where the hole will begin | |
3392 | * @len: The length of the hole | |
3393 | * | |
4907cb7b | 3394 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3395 | */ |
3396 | ||
aeb2817a | 3397 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3398 | { |
26a4c0c6 TT |
3399 | struct super_block *sb = inode->i_sb; |
3400 | ext4_lblk_t first_block, stop_block; | |
3401 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3402 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3403 | handle_t *handle; |
3404 | unsigned int credits; | |
3405 | int ret = 0; | |
3406 | ||
a4bb6b64 | 3407 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3408 | return -EOPNOTSUPP; |
a4bb6b64 | 3409 | |
b8a86845 | 3410 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3411 | |
26a4c0c6 TT |
3412 | /* |
3413 | * Write out all dirty pages to avoid race conditions | |
3414 | * Then release them. | |
3415 | */ | |
3416 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3417 | ret = filemap_write_and_wait_range(mapping, offset, | |
3418 | offset + length - 1); | |
3419 | if (ret) | |
3420 | return ret; | |
3421 | } | |
3422 | ||
3423 | mutex_lock(&inode->i_mutex); | |
9ef06cec | 3424 | |
26a4c0c6 TT |
3425 | /* No need to punch hole beyond i_size */ |
3426 | if (offset >= inode->i_size) | |
3427 | goto out_mutex; | |
3428 | ||
3429 | /* | |
3430 | * If the hole extends beyond i_size, set the hole | |
3431 | * to end after the page that contains i_size | |
3432 | */ | |
3433 | if (offset + length > inode->i_size) { | |
3434 | length = inode->i_size + | |
3435 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3436 | offset; | |
3437 | } | |
3438 | ||
a361293f JK |
3439 | if (offset & (sb->s_blocksize - 1) || |
3440 | (offset + length) & (sb->s_blocksize - 1)) { | |
3441 | /* | |
3442 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3443 | * partial block | |
3444 | */ | |
3445 | ret = ext4_inode_attach_jinode(inode); | |
3446 | if (ret < 0) | |
3447 | goto out_mutex; | |
3448 | ||
3449 | } | |
3450 | ||
a87dd18c LC |
3451 | first_block_offset = round_up(offset, sb->s_blocksize); |
3452 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3453 | |
a87dd18c LC |
3454 | /* Now release the pages and zero block aligned part of pages*/ |
3455 | if (last_block_offset > first_block_offset) | |
3456 | truncate_pagecache_range(inode, first_block_offset, | |
3457 | last_block_offset); | |
26a4c0c6 TT |
3458 | |
3459 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3460 | ext4_inode_block_unlocked_dio(inode); | |
26a4c0c6 TT |
3461 | inode_dio_wait(inode); |
3462 | ||
3463 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3464 | credits = ext4_writepage_trans_blocks(inode); | |
3465 | else | |
3466 | credits = ext4_blocks_for_truncate(inode); | |
3467 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3468 | if (IS_ERR(handle)) { | |
3469 | ret = PTR_ERR(handle); | |
3470 | ext4_std_error(sb, ret); | |
3471 | goto out_dio; | |
3472 | } | |
3473 | ||
a87dd18c LC |
3474 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3475 | length); | |
3476 | if (ret) | |
3477 | goto out_stop; | |
26a4c0c6 TT |
3478 | |
3479 | first_block = (offset + sb->s_blocksize - 1) >> | |
3480 | EXT4_BLOCK_SIZE_BITS(sb); | |
3481 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3482 | ||
3483 | /* If there are no blocks to remove, return now */ | |
3484 | if (first_block >= stop_block) | |
3485 | goto out_stop; | |
3486 | ||
3487 | down_write(&EXT4_I(inode)->i_data_sem); | |
3488 | ext4_discard_preallocations(inode); | |
3489 | ||
3490 | ret = ext4_es_remove_extent(inode, first_block, | |
3491 | stop_block - first_block); | |
3492 | if (ret) { | |
3493 | up_write(&EXT4_I(inode)->i_data_sem); | |
3494 | goto out_stop; | |
3495 | } | |
3496 | ||
3497 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3498 | ret = ext4_ext_remove_space(inode, first_block, | |
3499 | stop_block - 1); | |
3500 | else | |
4f579ae7 | 3501 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
3502 | stop_block); |
3503 | ||
819c4920 | 3504 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3505 | if (IS_SYNC(inode)) |
3506 | ext4_handle_sync(handle); | |
e251f9bc MP |
3507 | |
3508 | /* Now release the pages again to reduce race window */ | |
3509 | if (last_block_offset > first_block_offset) | |
3510 | truncate_pagecache_range(inode, first_block_offset, | |
3511 | last_block_offset); | |
3512 | ||
26a4c0c6 TT |
3513 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3514 | ext4_mark_inode_dirty(handle, inode); | |
3515 | out_stop: | |
3516 | ext4_journal_stop(handle); | |
3517 | out_dio: | |
3518 | ext4_inode_resume_unlocked_dio(inode); | |
3519 | out_mutex: | |
3520 | mutex_unlock(&inode->i_mutex); | |
3521 | return ret; | |
a4bb6b64 AH |
3522 | } |
3523 | ||
a361293f JK |
3524 | int ext4_inode_attach_jinode(struct inode *inode) |
3525 | { | |
3526 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3527 | struct jbd2_inode *jinode; | |
3528 | ||
3529 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
3530 | return 0; | |
3531 | ||
3532 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
3533 | spin_lock(&inode->i_lock); | |
3534 | if (!ei->jinode) { | |
3535 | if (!jinode) { | |
3536 | spin_unlock(&inode->i_lock); | |
3537 | return -ENOMEM; | |
3538 | } | |
3539 | ei->jinode = jinode; | |
3540 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
3541 | jinode = NULL; | |
3542 | } | |
3543 | spin_unlock(&inode->i_lock); | |
3544 | if (unlikely(jinode != NULL)) | |
3545 | jbd2_free_inode(jinode); | |
3546 | return 0; | |
3547 | } | |
3548 | ||
ac27a0ec | 3549 | /* |
617ba13b | 3550 | * ext4_truncate() |
ac27a0ec | 3551 | * |
617ba13b MC |
3552 | * We block out ext4_get_block() block instantiations across the entire |
3553 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3554 | * simultaneously on behalf of the same inode. |
3555 | * | |
42b2aa86 | 3556 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3557 | * is one core, guiding principle: the file's tree must always be consistent on |
3558 | * disk. We must be able to restart the truncate after a crash. | |
3559 | * | |
3560 | * The file's tree may be transiently inconsistent in memory (although it | |
3561 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3562 | * the contents of (the filesystem + the journal) must be consistent and | |
3563 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3564 | * left-to-right works OK too). | |
3565 | * | |
3566 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3567 | * truncate against the orphan inode list. | |
3568 | * | |
3569 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3570 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3571 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3572 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3573 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3574 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3575 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3576 | */ |
617ba13b | 3577 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3578 | { |
819c4920 TT |
3579 | struct ext4_inode_info *ei = EXT4_I(inode); |
3580 | unsigned int credits; | |
3581 | handle_t *handle; | |
3582 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 3583 | |
19b5ef61 TT |
3584 | /* |
3585 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 3586 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
3587 | * have i_mutex locked because it's not necessary. |
3588 | */ | |
3589 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3590 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3591 | trace_ext4_truncate_enter(inode); |
3592 | ||
91ef4caf | 3593 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3594 | return; |
3595 | ||
12e9b892 | 3596 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3597 | |
5534fb5b | 3598 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3599 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3600 | |
aef1c851 TM |
3601 | if (ext4_has_inline_data(inode)) { |
3602 | int has_inline = 1; | |
3603 | ||
3604 | ext4_inline_data_truncate(inode, &has_inline); | |
3605 | if (has_inline) | |
3606 | return; | |
3607 | } | |
3608 | ||
a361293f JK |
3609 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
3610 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
3611 | if (ext4_inode_attach_jinode(inode) < 0) | |
3612 | return; | |
3613 | } | |
3614 | ||
819c4920 TT |
3615 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3616 | credits = ext4_writepage_trans_blocks(inode); | |
3617 | else | |
3618 | credits = ext4_blocks_for_truncate(inode); | |
3619 | ||
3620 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3621 | if (IS_ERR(handle)) { | |
3622 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3623 | return; | |
3624 | } | |
3625 | ||
eb3544c6 LC |
3626 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
3627 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
3628 | |
3629 | /* | |
3630 | * We add the inode to the orphan list, so that if this | |
3631 | * truncate spans multiple transactions, and we crash, we will | |
3632 | * resume the truncate when the filesystem recovers. It also | |
3633 | * marks the inode dirty, to catch the new size. | |
3634 | * | |
3635 | * Implication: the file must always be in a sane, consistent | |
3636 | * truncatable state while each transaction commits. | |
3637 | */ | |
3638 | if (ext4_orphan_add(handle, inode)) | |
3639 | goto out_stop; | |
3640 | ||
3641 | down_write(&EXT4_I(inode)->i_data_sem); | |
3642 | ||
3643 | ext4_discard_preallocations(inode); | |
3644 | ||
ff9893dc | 3645 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3646 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3647 | else |
819c4920 TT |
3648 | ext4_ind_truncate(handle, inode); |
3649 | ||
3650 | up_write(&ei->i_data_sem); | |
3651 | ||
3652 | if (IS_SYNC(inode)) | |
3653 | ext4_handle_sync(handle); | |
3654 | ||
3655 | out_stop: | |
3656 | /* | |
3657 | * If this was a simple ftruncate() and the file will remain alive, | |
3658 | * then we need to clear up the orphan record which we created above. | |
3659 | * However, if this was a real unlink then we were called by | |
58d86a50 | 3660 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
3661 | * orphan info for us. |
3662 | */ | |
3663 | if (inode->i_nlink) | |
3664 | ext4_orphan_del(handle, inode); | |
3665 | ||
3666 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3667 | ext4_mark_inode_dirty(handle, inode); | |
3668 | ext4_journal_stop(handle); | |
ac27a0ec | 3669 | |
0562e0ba | 3670 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3671 | } |
3672 | ||
ac27a0ec | 3673 | /* |
617ba13b | 3674 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3675 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3676 | * data in memory that is needed to recreate the on-disk version of this | |
3677 | * inode. | |
3678 | */ | |
617ba13b MC |
3679 | static int __ext4_get_inode_loc(struct inode *inode, |
3680 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3681 | { |
240799cd TT |
3682 | struct ext4_group_desc *gdp; |
3683 | struct buffer_head *bh; | |
3684 | struct super_block *sb = inode->i_sb; | |
3685 | ext4_fsblk_t block; | |
3686 | int inodes_per_block, inode_offset; | |
3687 | ||
3a06d778 | 3688 | iloc->bh = NULL; |
240799cd TT |
3689 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3690 | return -EIO; | |
ac27a0ec | 3691 | |
240799cd TT |
3692 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3693 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3694 | if (!gdp) | |
ac27a0ec DK |
3695 | return -EIO; |
3696 | ||
240799cd TT |
3697 | /* |
3698 | * Figure out the offset within the block group inode table | |
3699 | */ | |
00d09882 | 3700 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3701 | inode_offset = ((inode->i_ino - 1) % |
3702 | EXT4_INODES_PER_GROUP(sb)); | |
3703 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3704 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3705 | ||
3706 | bh = sb_getblk(sb, block); | |
aebf0243 | 3707 | if (unlikely(!bh)) |
860d21e2 | 3708 | return -ENOMEM; |
ac27a0ec DK |
3709 | if (!buffer_uptodate(bh)) { |
3710 | lock_buffer(bh); | |
9c83a923 HK |
3711 | |
3712 | /* | |
3713 | * If the buffer has the write error flag, we have failed | |
3714 | * to write out another inode in the same block. In this | |
3715 | * case, we don't have to read the block because we may | |
3716 | * read the old inode data successfully. | |
3717 | */ | |
3718 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3719 | set_buffer_uptodate(bh); | |
3720 | ||
ac27a0ec DK |
3721 | if (buffer_uptodate(bh)) { |
3722 | /* someone brought it uptodate while we waited */ | |
3723 | unlock_buffer(bh); | |
3724 | goto has_buffer; | |
3725 | } | |
3726 | ||
3727 | /* | |
3728 | * If we have all information of the inode in memory and this | |
3729 | * is the only valid inode in the block, we need not read the | |
3730 | * block. | |
3731 | */ | |
3732 | if (in_mem) { | |
3733 | struct buffer_head *bitmap_bh; | |
240799cd | 3734 | int i, start; |
ac27a0ec | 3735 | |
240799cd | 3736 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3737 | |
240799cd TT |
3738 | /* Is the inode bitmap in cache? */ |
3739 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3740 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3741 | goto make_io; |
3742 | ||
3743 | /* | |
3744 | * If the inode bitmap isn't in cache then the | |
3745 | * optimisation may end up performing two reads instead | |
3746 | * of one, so skip it. | |
3747 | */ | |
3748 | if (!buffer_uptodate(bitmap_bh)) { | |
3749 | brelse(bitmap_bh); | |
3750 | goto make_io; | |
3751 | } | |
240799cd | 3752 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3753 | if (i == inode_offset) |
3754 | continue; | |
617ba13b | 3755 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3756 | break; |
3757 | } | |
3758 | brelse(bitmap_bh); | |
240799cd | 3759 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3760 | /* all other inodes are free, so skip I/O */ |
3761 | memset(bh->b_data, 0, bh->b_size); | |
3762 | set_buffer_uptodate(bh); | |
3763 | unlock_buffer(bh); | |
3764 | goto has_buffer; | |
3765 | } | |
3766 | } | |
3767 | ||
3768 | make_io: | |
240799cd TT |
3769 | /* |
3770 | * If we need to do any I/O, try to pre-readahead extra | |
3771 | * blocks from the inode table. | |
3772 | */ | |
3773 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3774 | ext4_fsblk_t b, end, table; | |
3775 | unsigned num; | |
0d606e2c | 3776 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
3777 | |
3778 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3779 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 3780 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
3781 | if (table > b) |
3782 | b = table; | |
0d606e2c | 3783 | end = b + ra_blks; |
240799cd | 3784 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 3785 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 3786 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3787 | table += num / inodes_per_block; |
3788 | if (end > table) | |
3789 | end = table; | |
3790 | while (b <= end) | |
3791 | sb_breadahead(sb, b++); | |
3792 | } | |
3793 | ||
ac27a0ec DK |
3794 | /* |
3795 | * There are other valid inodes in the buffer, this inode | |
3796 | * has in-inode xattrs, or we don't have this inode in memory. | |
3797 | * Read the block from disk. | |
3798 | */ | |
0562e0ba | 3799 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3800 | get_bh(bh); |
3801 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 3802 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
3803 | wait_on_buffer(bh); |
3804 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3805 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3806 | "unable to read itable block"); | |
ac27a0ec DK |
3807 | brelse(bh); |
3808 | return -EIO; | |
3809 | } | |
3810 | } | |
3811 | has_buffer: | |
3812 | iloc->bh = bh; | |
3813 | return 0; | |
3814 | } | |
3815 | ||
617ba13b | 3816 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3817 | { |
3818 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3819 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3820 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3821 | } |
3822 | ||
617ba13b | 3823 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3824 | { |
617ba13b | 3825 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 3826 | unsigned int new_fl = 0; |
ac27a0ec | 3827 | |
617ba13b | 3828 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 3829 | new_fl |= S_SYNC; |
617ba13b | 3830 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 3831 | new_fl |= S_APPEND; |
617ba13b | 3832 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 3833 | new_fl |= S_IMMUTABLE; |
617ba13b | 3834 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 3835 | new_fl |= S_NOATIME; |
617ba13b | 3836 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 3837 | new_fl |= S_DIRSYNC; |
923ae0ff RZ |
3838 | if (test_opt(inode->i_sb, DAX)) |
3839 | new_fl |= S_DAX; | |
5f16f322 | 3840 | inode_set_flags(inode, new_fl, |
923ae0ff | 3841 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
3842 | } |
3843 | ||
ff9ddf7e JK |
3844 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3845 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3846 | { | |
84a8dce2 DM |
3847 | unsigned int vfs_fl; |
3848 | unsigned long old_fl, new_fl; | |
3849 | ||
3850 | do { | |
3851 | vfs_fl = ei->vfs_inode.i_flags; | |
3852 | old_fl = ei->i_flags; | |
3853 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3854 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3855 | EXT4_DIRSYNC_FL); | |
3856 | if (vfs_fl & S_SYNC) | |
3857 | new_fl |= EXT4_SYNC_FL; | |
3858 | if (vfs_fl & S_APPEND) | |
3859 | new_fl |= EXT4_APPEND_FL; | |
3860 | if (vfs_fl & S_IMMUTABLE) | |
3861 | new_fl |= EXT4_IMMUTABLE_FL; | |
3862 | if (vfs_fl & S_NOATIME) | |
3863 | new_fl |= EXT4_NOATIME_FL; | |
3864 | if (vfs_fl & S_DIRSYNC) | |
3865 | new_fl |= EXT4_DIRSYNC_FL; | |
3866 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3867 | } |
de9a55b8 | 3868 | |
0fc1b451 | 3869 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3870 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3871 | { |
3872 | blkcnt_t i_blocks ; | |
8180a562 AK |
3873 | struct inode *inode = &(ei->vfs_inode); |
3874 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3875 | |
3876 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3877 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3878 | /* we are using combined 48 bit field */ | |
3879 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3880 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 3881 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
3882 | /* i_blocks represent file system block size */ |
3883 | return i_blocks << (inode->i_blkbits - 9); | |
3884 | } else { | |
3885 | return i_blocks; | |
3886 | } | |
0fc1b451 AK |
3887 | } else { |
3888 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
3889 | } | |
3890 | } | |
ff9ddf7e | 3891 | |
152a7b0a TM |
3892 | static inline void ext4_iget_extra_inode(struct inode *inode, |
3893 | struct ext4_inode *raw_inode, | |
3894 | struct ext4_inode_info *ei) | |
3895 | { | |
3896 | __le32 *magic = (void *)raw_inode + | |
3897 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 3898 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 3899 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 3900 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
3901 | } else |
3902 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
3903 | } |
3904 | ||
1d1fe1ee | 3905 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 3906 | { |
617ba13b MC |
3907 | struct ext4_iloc iloc; |
3908 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 3909 | struct ext4_inode_info *ei; |
1d1fe1ee | 3910 | struct inode *inode; |
b436b9be | 3911 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 3912 | long ret; |
ac27a0ec | 3913 | int block; |
08cefc7a EB |
3914 | uid_t i_uid; |
3915 | gid_t i_gid; | |
ac27a0ec | 3916 | |
1d1fe1ee DH |
3917 | inode = iget_locked(sb, ino); |
3918 | if (!inode) | |
3919 | return ERR_PTR(-ENOMEM); | |
3920 | if (!(inode->i_state & I_NEW)) | |
3921 | return inode; | |
3922 | ||
3923 | ei = EXT4_I(inode); | |
7dc57615 | 3924 | iloc.bh = NULL; |
ac27a0ec | 3925 | |
1d1fe1ee DH |
3926 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
3927 | if (ret < 0) | |
ac27a0ec | 3928 | goto bad_inode; |
617ba13b | 3929 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
3930 | |
3931 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3932 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
3933 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
3934 | EXT4_INODE_SIZE(inode->i_sb)) { | |
3935 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
3936 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
3937 | EXT4_INODE_SIZE(inode->i_sb)); | |
3938 | ret = -EIO; | |
3939 | goto bad_inode; | |
3940 | } | |
3941 | } else | |
3942 | ei->i_extra_isize = 0; | |
3943 | ||
3944 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 3945 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
3946 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
3947 | __u32 csum; | |
3948 | __le32 inum = cpu_to_le32(inode->i_ino); | |
3949 | __le32 gen = raw_inode->i_generation; | |
3950 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
3951 | sizeof(inum)); | |
3952 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
3953 | sizeof(gen)); | |
3954 | } | |
3955 | ||
3956 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
3957 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
3958 | ret = -EIO; | |
3959 | goto bad_inode; | |
3960 | } | |
3961 | ||
ac27a0ec | 3962 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
3963 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
3964 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 3965 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
3966 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
3967 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 3968 | } |
08cefc7a EB |
3969 | i_uid_write(inode, i_uid); |
3970 | i_gid_write(inode, i_gid); | |
bfe86848 | 3971 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 3972 | |
353eb83c | 3973 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 3974 | ei->i_inline_off = 0; |
ac27a0ec DK |
3975 | ei->i_dir_start_lookup = 0; |
3976 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
3977 | /* We now have enough fields to check if the inode was active or not. | |
3978 | * This is needed because nfsd might try to access dead inodes | |
3979 | * the test is that same one that e2fsck uses | |
3980 | * NeilBrown 1999oct15 | |
3981 | */ | |
3982 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
3983 | if ((inode->i_mode == 0 || |
3984 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
3985 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 3986 | /* this inode is deleted */ |
1d1fe1ee | 3987 | ret = -ESTALE; |
ac27a0ec DK |
3988 | goto bad_inode; |
3989 | } | |
3990 | /* The only unlinked inodes we let through here have | |
3991 | * valid i_mode and are being read by the orphan | |
3992 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
3993 | * the process of deleting those. |
3994 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
3995 | * not initialized on a new filesystem. */ | |
ac27a0ec | 3996 | } |
ac27a0ec | 3997 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 3998 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 3999 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4000 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4001 | ei->i_file_acl |= |
4002 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4003 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4004 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4005 | #ifdef CONFIG_QUOTA |
4006 | ei->i_reserved_quota = 0; | |
4007 | #endif | |
ac27a0ec DK |
4008 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4009 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4010 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4011 | /* |
4012 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4013 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4014 | */ | |
617ba13b | 4015 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4016 | ei->i_data[block] = raw_inode->i_block[block]; |
4017 | INIT_LIST_HEAD(&ei->i_orphan); | |
4018 | ||
b436b9be JK |
4019 | /* |
4020 | * Set transaction id's of transactions that have to be committed | |
4021 | * to finish f[data]sync. We set them to currently running transaction | |
4022 | * as we cannot be sure that the inode or some of its metadata isn't | |
4023 | * part of the transaction - the inode could have been reclaimed and | |
4024 | * now it is reread from disk. | |
4025 | */ | |
4026 | if (journal) { | |
4027 | transaction_t *transaction; | |
4028 | tid_t tid; | |
4029 | ||
a931da6a | 4030 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4031 | if (journal->j_running_transaction) |
4032 | transaction = journal->j_running_transaction; | |
4033 | else | |
4034 | transaction = journal->j_committing_transaction; | |
4035 | if (transaction) | |
4036 | tid = transaction->t_tid; | |
4037 | else | |
4038 | tid = journal->j_commit_sequence; | |
a931da6a | 4039 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4040 | ei->i_sync_tid = tid; |
4041 | ei->i_datasync_tid = tid; | |
4042 | } | |
4043 | ||
0040d987 | 4044 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4045 | if (ei->i_extra_isize == 0) { |
4046 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4047 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4048 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4049 | } else { |
152a7b0a | 4050 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4051 | } |
814525f4 | 4052 | } |
ac27a0ec | 4053 | |
ef7f3835 KS |
4054 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4055 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4056 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4057 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4058 | ||
ed3654eb | 4059 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4060 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4061 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4062 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4063 | inode->i_version |= | |
4064 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4065 | } | |
25ec56b5 JNC |
4066 | } |
4067 | ||
c4b5a614 | 4068 | ret = 0; |
485c26ec | 4069 | if (ei->i_file_acl && |
1032988c | 4070 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4071 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4072 | ei->i_file_acl); | |
485c26ec TT |
4073 | ret = -EIO; |
4074 | goto bad_inode; | |
f19d5870 TM |
4075 | } else if (!ext4_has_inline_data(inode)) { |
4076 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4077 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4078 | (S_ISLNK(inode->i_mode) && | |
4079 | !ext4_inode_is_fast_symlink(inode)))) | |
4080 | /* Validate extent which is part of inode */ | |
4081 | ret = ext4_ext_check_inode(inode); | |
4082 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4083 | (S_ISLNK(inode->i_mode) && | |
4084 | !ext4_inode_is_fast_symlink(inode))) { | |
4085 | /* Validate block references which are part of inode */ | |
4086 | ret = ext4_ind_check_inode(inode); | |
4087 | } | |
fe2c8191 | 4088 | } |
567f3e9a | 4089 | if (ret) |
de9a55b8 | 4090 | goto bad_inode; |
7a262f7c | 4091 | |
ac27a0ec | 4092 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4093 | inode->i_op = &ext4_file_inode_operations; |
923ae0ff RZ |
4094 | if (test_opt(inode->i_sb, DAX)) |
4095 | inode->i_fop = &ext4_dax_file_operations; | |
4096 | else | |
4097 | inode->i_fop = &ext4_file_operations; | |
617ba13b | 4098 | ext4_set_aops(inode); |
ac27a0ec | 4099 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4100 | inode->i_op = &ext4_dir_inode_operations; |
4101 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4102 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4103 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4104 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4105 | nd_terminate_link(ei->i_data, inode->i_size, |
4106 | sizeof(ei->i_data) - 1); | |
4107 | } else { | |
617ba13b MC |
4108 | inode->i_op = &ext4_symlink_inode_operations; |
4109 | ext4_set_aops(inode); | |
ac27a0ec | 4110 | } |
563bdd61 TT |
4111 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4112 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4113 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4114 | if (raw_inode->i_block[0]) |
4115 | init_special_inode(inode, inode->i_mode, | |
4116 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4117 | else | |
4118 | init_special_inode(inode, inode->i_mode, | |
4119 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4120 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4121 | make_bad_inode(inode); | |
563bdd61 | 4122 | } else { |
563bdd61 | 4123 | ret = -EIO; |
24676da4 | 4124 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4125 | goto bad_inode; |
ac27a0ec | 4126 | } |
af5bc92d | 4127 | brelse(iloc.bh); |
617ba13b | 4128 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4129 | unlock_new_inode(inode); |
4130 | return inode; | |
ac27a0ec DK |
4131 | |
4132 | bad_inode: | |
567f3e9a | 4133 | brelse(iloc.bh); |
1d1fe1ee DH |
4134 | iget_failed(inode); |
4135 | return ERR_PTR(ret); | |
ac27a0ec DK |
4136 | } |
4137 | ||
f4bb2981 TT |
4138 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4139 | { | |
4140 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
4141 | return ERR_PTR(-EIO); | |
4142 | return ext4_iget(sb, ino); | |
4143 | } | |
4144 | ||
0fc1b451 AK |
4145 | static int ext4_inode_blocks_set(handle_t *handle, |
4146 | struct ext4_inode *raw_inode, | |
4147 | struct ext4_inode_info *ei) | |
4148 | { | |
4149 | struct inode *inode = &(ei->vfs_inode); | |
4150 | u64 i_blocks = inode->i_blocks; | |
4151 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4152 | |
4153 | if (i_blocks <= ~0U) { | |
4154 | /* | |
4907cb7b | 4155 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4156 | * as multiple of 512 bytes |
4157 | */ | |
8180a562 | 4158 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4159 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4160 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4161 | return 0; |
4162 | } | |
4163 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4164 | return -EFBIG; | |
4165 | ||
4166 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4167 | /* |
4168 | * i_blocks can be represented in a 48 bit variable | |
4169 | * as multiple of 512 bytes | |
4170 | */ | |
8180a562 | 4171 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4172 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4173 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4174 | } else { |
84a8dce2 | 4175 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4176 | /* i_block is stored in file system block size */ |
4177 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4178 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4179 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4180 | } |
f287a1a5 | 4181 | return 0; |
0fc1b451 AK |
4182 | } |
4183 | ||
a26f4992 TT |
4184 | struct other_inode { |
4185 | unsigned long orig_ino; | |
4186 | struct ext4_inode *raw_inode; | |
4187 | }; | |
4188 | ||
4189 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4190 | void *data) | |
4191 | { | |
4192 | struct other_inode *oi = (struct other_inode *) data; | |
4193 | ||
4194 | if ((inode->i_ino != ino) || | |
4195 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4196 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4197 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4198 | return 0; | |
4199 | spin_lock(&inode->i_lock); | |
4200 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4201 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4202 | (inode->i_state & I_DIRTY_TIME)) { | |
4203 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4204 | ||
4205 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4206 | spin_unlock(&inode->i_lock); | |
4207 | ||
4208 | spin_lock(&ei->i_raw_lock); | |
4209 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4210 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4211 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4212 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4213 | spin_unlock(&ei->i_raw_lock); | |
4214 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4215 | return -1; | |
4216 | } | |
4217 | spin_unlock(&inode->i_lock); | |
4218 | return -1; | |
4219 | } | |
4220 | ||
4221 | /* | |
4222 | * Opportunistically update the other time fields for other inodes in | |
4223 | * the same inode table block. | |
4224 | */ | |
4225 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4226 | unsigned long orig_ino, char *buf) | |
4227 | { | |
4228 | struct other_inode oi; | |
4229 | unsigned long ino; | |
4230 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4231 | int inode_size = EXT4_INODE_SIZE(sb); | |
4232 | ||
4233 | oi.orig_ino = orig_ino; | |
4234 | ino = orig_ino & ~(inodes_per_block - 1); | |
4235 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { | |
4236 | if (ino == orig_ino) | |
4237 | continue; | |
4238 | oi.raw_inode = (struct ext4_inode *) buf; | |
4239 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4240 | } | |
4241 | } | |
4242 | ||
ac27a0ec DK |
4243 | /* |
4244 | * Post the struct inode info into an on-disk inode location in the | |
4245 | * buffer-cache. This gobbles the caller's reference to the | |
4246 | * buffer_head in the inode location struct. | |
4247 | * | |
4248 | * The caller must have write access to iloc->bh. | |
4249 | */ | |
617ba13b | 4250 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4251 | struct inode *inode, |
830156c7 | 4252 | struct ext4_iloc *iloc) |
ac27a0ec | 4253 | { |
617ba13b MC |
4254 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4255 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4256 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4257 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4258 | int err = 0, rc, block; |
202ee5df | 4259 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4260 | uid_t i_uid; |
4261 | gid_t i_gid; | |
ac27a0ec | 4262 | |
202ee5df TT |
4263 | spin_lock(&ei->i_raw_lock); |
4264 | ||
4265 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4266 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4267 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4268 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4269 | |
ff9ddf7e | 4270 | ext4_get_inode_flags(ei); |
ac27a0ec | 4271 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4272 | i_uid = i_uid_read(inode); |
4273 | i_gid = i_gid_read(inode); | |
af5bc92d | 4274 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4275 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4276 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4277 | /* |
4278 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4279 | * re-used with the upper 16 bits of the uid/gid intact | |
4280 | */ | |
af5bc92d | 4281 | if (!ei->i_dtime) { |
ac27a0ec | 4282 | raw_inode->i_uid_high = |
08cefc7a | 4283 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4284 | raw_inode->i_gid_high = |
08cefc7a | 4285 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4286 | } else { |
4287 | raw_inode->i_uid_high = 0; | |
4288 | raw_inode->i_gid_high = 0; | |
4289 | } | |
4290 | } else { | |
08cefc7a EB |
4291 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4292 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4293 | raw_inode->i_uid_high = 0; |
4294 | raw_inode->i_gid_high = 0; | |
4295 | } | |
4296 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4297 | |
4298 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4299 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4300 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4301 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4302 | ||
bce92d56 LX |
4303 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4304 | if (err) { | |
202ee5df | 4305 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4306 | goto out_brelse; |
202ee5df | 4307 | } |
ac27a0ec | 4308 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4309 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4310 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4311 | raw_inode->i_file_acl_high = |
4312 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4313 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4314 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4315 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4316 | need_datasync = 1; | |
4317 | } | |
a48380f7 | 4318 | if (ei->i_disksize > 0x7fffffffULL) { |
a48380f7 AK |
4319 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
4320 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4321 | EXT4_SB(sb)->s_es->s_rev_level == | |
202ee5df TT |
4322 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4323 | set_large_file = 1; | |
ac27a0ec DK |
4324 | } |
4325 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4326 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4327 | if (old_valid_dev(inode->i_rdev)) { | |
4328 | raw_inode->i_block[0] = | |
4329 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4330 | raw_inode->i_block[1] = 0; | |
4331 | } else { | |
4332 | raw_inode->i_block[0] = 0; | |
4333 | raw_inode->i_block[1] = | |
4334 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4335 | raw_inode->i_block[2] = 0; | |
4336 | } | |
f19d5870 | 4337 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4338 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4339 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4340 | } |
ac27a0ec | 4341 | |
ed3654eb | 4342 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4343 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4344 | if (ei->i_extra_isize) { | |
4345 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4346 | raw_inode->i_version_hi = | |
4347 | cpu_to_le32(inode->i_version >> 32); | |
4348 | raw_inode->i_extra_isize = | |
4349 | cpu_to_le16(ei->i_extra_isize); | |
4350 | } | |
25ec56b5 | 4351 | } |
814525f4 | 4352 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 4353 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
4354 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
4355 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
4356 | bh->b_data); | |
202ee5df | 4357 | |
830156c7 | 4358 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4359 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4360 | if (!err) |
4361 | err = rc; | |
19f5fb7a | 4362 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 4363 | if (set_large_file) { |
5d601255 | 4364 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4365 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4366 | if (err) | |
4367 | goto out_brelse; | |
4368 | ext4_update_dynamic_rev(sb); | |
4369 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
4370 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); | |
4371 | ext4_handle_sync(handle); | |
4372 | err = ext4_handle_dirty_super(handle, sb); | |
4373 | } | |
b71fc079 | 4374 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4375 | out_brelse: |
af5bc92d | 4376 | brelse(bh); |
617ba13b | 4377 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4378 | return err; |
4379 | } | |
4380 | ||
4381 | /* | |
617ba13b | 4382 | * ext4_write_inode() |
ac27a0ec DK |
4383 | * |
4384 | * We are called from a few places: | |
4385 | * | |
87f7e416 | 4386 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4387 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4388 | * transaction to commit. |
ac27a0ec | 4389 | * |
87f7e416 TT |
4390 | * - Within flush work (sys_sync(), kupdate and such). |
4391 | * We wait on commit, if told to. | |
ac27a0ec | 4392 | * |
87f7e416 TT |
4393 | * - Within iput_final() -> write_inode_now() |
4394 | * We wait on commit, if told to. | |
ac27a0ec DK |
4395 | * |
4396 | * In all cases it is actually safe for us to return without doing anything, | |
4397 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4398 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4399 | * writeback. | |
ac27a0ec DK |
4400 | * |
4401 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4402 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4403 | * which we are interested. | |
4404 | * | |
4405 | * It would be a bug for them to not do this. The code: | |
4406 | * | |
4407 | * mark_inode_dirty(inode) | |
4408 | * stuff(); | |
4409 | * inode->i_size = expr; | |
4410 | * | |
87f7e416 TT |
4411 | * is in error because write_inode() could occur while `stuff()' is running, |
4412 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4413 | * superblock's dirty inode list. | |
ac27a0ec | 4414 | */ |
a9185b41 | 4415 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4416 | { |
91ac6f43 FM |
4417 | int err; |
4418 | ||
87f7e416 | 4419 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4420 | return 0; |
4421 | ||
91ac6f43 FM |
4422 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4423 | if (ext4_journal_current_handle()) { | |
4424 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4425 | dump_stack(); | |
4426 | return -EIO; | |
4427 | } | |
ac27a0ec | 4428 | |
10542c22 JK |
4429 | /* |
4430 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4431 | * ext4_sync_fs() will force the commit after everything is | |
4432 | * written. | |
4433 | */ | |
4434 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4435 | return 0; |
4436 | ||
4437 | err = ext4_force_commit(inode->i_sb); | |
4438 | } else { | |
4439 | struct ext4_iloc iloc; | |
ac27a0ec | 4440 | |
8b472d73 | 4441 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4442 | if (err) |
4443 | return err; | |
10542c22 JK |
4444 | /* |
4445 | * sync(2) will flush the whole buffer cache. No need to do | |
4446 | * it here separately for each inode. | |
4447 | */ | |
4448 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4449 | sync_dirty_buffer(iloc.bh); |
4450 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4451 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4452 | "IO error syncing inode"); | |
830156c7 FM |
4453 | err = -EIO; |
4454 | } | |
fd2dd9fb | 4455 | brelse(iloc.bh); |
91ac6f43 FM |
4456 | } |
4457 | return err; | |
ac27a0ec DK |
4458 | } |
4459 | ||
53e87268 JK |
4460 | /* |
4461 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4462 | * buffers that are attached to a page stradding i_size and are undergoing | |
4463 | * commit. In that case we have to wait for commit to finish and try again. | |
4464 | */ | |
4465 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4466 | { | |
4467 | struct page *page; | |
4468 | unsigned offset; | |
4469 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4470 | tid_t commit_tid = 0; | |
4471 | int ret; | |
4472 | ||
4473 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4474 | /* | |
4475 | * All buffers in the last page remain valid? Then there's nothing to | |
4476 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4477 | * blocksize case | |
4478 | */ | |
4479 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4480 | return; | |
4481 | while (1) { | |
4482 | page = find_lock_page(inode->i_mapping, | |
4483 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4484 | if (!page) | |
4485 | return; | |
ca99fdd2 LC |
4486 | ret = __ext4_journalled_invalidatepage(page, offset, |
4487 | PAGE_CACHE_SIZE - offset); | |
53e87268 JK |
4488 | unlock_page(page); |
4489 | page_cache_release(page); | |
4490 | if (ret != -EBUSY) | |
4491 | return; | |
4492 | commit_tid = 0; | |
4493 | read_lock(&journal->j_state_lock); | |
4494 | if (journal->j_committing_transaction) | |
4495 | commit_tid = journal->j_committing_transaction->t_tid; | |
4496 | read_unlock(&journal->j_state_lock); | |
4497 | if (commit_tid) | |
4498 | jbd2_log_wait_commit(journal, commit_tid); | |
4499 | } | |
4500 | } | |
4501 | ||
ac27a0ec | 4502 | /* |
617ba13b | 4503 | * ext4_setattr() |
ac27a0ec DK |
4504 | * |
4505 | * Called from notify_change. | |
4506 | * | |
4507 | * We want to trap VFS attempts to truncate the file as soon as | |
4508 | * possible. In particular, we want to make sure that when the VFS | |
4509 | * shrinks i_size, we put the inode on the orphan list and modify | |
4510 | * i_disksize immediately, so that during the subsequent flushing of | |
4511 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4512 | * commit will leave the blocks being flushed in an unused state on | |
4513 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4514 | * be freed, so we have a strong guarantee that no future commit will | |
4515 | * leave these blocks visible to the user.) | |
4516 | * | |
678aaf48 JK |
4517 | * Another thing we have to assure is that if we are in ordered mode |
4518 | * and inode is still attached to the committing transaction, we must | |
4519 | * we start writeout of all the dirty pages which are being truncated. | |
4520 | * This way we are sure that all the data written in the previous | |
4521 | * transaction are already on disk (truncate waits for pages under | |
4522 | * writeback). | |
4523 | * | |
4524 | * Called with inode->i_mutex down. | |
ac27a0ec | 4525 | */ |
617ba13b | 4526 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4527 | { |
4528 | struct inode *inode = dentry->d_inode; | |
4529 | int error, rc = 0; | |
3d287de3 | 4530 | int orphan = 0; |
ac27a0ec DK |
4531 | const unsigned int ia_valid = attr->ia_valid; |
4532 | ||
4533 | error = inode_change_ok(inode, attr); | |
4534 | if (error) | |
4535 | return error; | |
4536 | ||
12755627 | 4537 | if (is_quota_modification(inode, attr)) |
871a2931 | 4538 | dquot_initialize(inode); |
08cefc7a EB |
4539 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4540 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4541 | handle_t *handle; |
4542 | ||
4543 | /* (user+group)*(old+new) structure, inode write (sb, | |
4544 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4545 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4546 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4547 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4548 | if (IS_ERR(handle)) { |
4549 | error = PTR_ERR(handle); | |
4550 | goto err_out; | |
4551 | } | |
b43fa828 | 4552 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4553 | if (error) { |
617ba13b | 4554 | ext4_journal_stop(handle); |
ac27a0ec DK |
4555 | return error; |
4556 | } | |
4557 | /* Update corresponding info in inode so that everything is in | |
4558 | * one transaction */ | |
4559 | if (attr->ia_valid & ATTR_UID) | |
4560 | inode->i_uid = attr->ia_uid; | |
4561 | if (attr->ia_valid & ATTR_GID) | |
4562 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4563 | error = ext4_mark_inode_dirty(handle, inode); |
4564 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4565 | } |
4566 | ||
5208386c JK |
4567 | if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) { |
4568 | handle_t *handle; | |
562c72aa | 4569 | |
12e9b892 | 4570 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4571 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4572 | ||
0c095c7f TT |
4573 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4574 | return -EFBIG; | |
e2b46574 | 4575 | } |
dff6efc3 CH |
4576 | |
4577 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
4578 | inode_inc_iversion(inode); | |
4579 | ||
5208386c JK |
4580 | if (S_ISREG(inode->i_mode) && |
4581 | (attr->ia_size < inode->i_size)) { | |
4582 | if (ext4_should_order_data(inode)) { | |
4583 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 4584 | attr->ia_size); |
5208386c | 4585 | if (error) |
678aaf48 | 4586 | goto err_out; |
5208386c JK |
4587 | } |
4588 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); | |
4589 | if (IS_ERR(handle)) { | |
4590 | error = PTR_ERR(handle); | |
4591 | goto err_out; | |
4592 | } | |
4593 | if (ext4_handle_valid(handle)) { | |
4594 | error = ext4_orphan_add(handle, inode); | |
4595 | orphan = 1; | |
4596 | } | |
90e775b7 | 4597 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
4598 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4599 | rc = ext4_mark_inode_dirty(handle, inode); | |
4600 | if (!error) | |
4601 | error = rc; | |
90e775b7 JK |
4602 | /* |
4603 | * We have to update i_size under i_data_sem together | |
4604 | * with i_disksize to avoid races with writeback code | |
4605 | * running ext4_wb_update_i_disksize(). | |
4606 | */ | |
4607 | if (!error) | |
4608 | i_size_write(inode, attr->ia_size); | |
4609 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
4610 | ext4_journal_stop(handle); |
4611 | if (error) { | |
4612 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
4613 | goto err_out; |
4614 | } | |
d6320cbf JK |
4615 | } else { |
4616 | loff_t oldsize = inode->i_size; | |
4617 | ||
90e775b7 | 4618 | i_size_write(inode, attr->ia_size); |
d6320cbf JK |
4619 | pagecache_isize_extended(inode, oldsize, inode->i_size); |
4620 | } | |
53e87268 | 4621 | |
5208386c JK |
4622 | /* |
4623 | * Blocks are going to be removed from the inode. Wait | |
4624 | * for dio in flight. Temporarily disable | |
4625 | * dioread_nolock to prevent livelock. | |
4626 | */ | |
4627 | if (orphan) { | |
4628 | if (!ext4_should_journal_data(inode)) { | |
4629 | ext4_inode_block_unlocked_dio(inode); | |
4630 | inode_dio_wait(inode); | |
4631 | ext4_inode_resume_unlocked_dio(inode); | |
4632 | } else | |
4633 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 4634 | } |
5208386c JK |
4635 | /* |
4636 | * Truncate pagecache after we've waited for commit | |
4637 | * in data=journal mode to make pages freeable. | |
4638 | */ | |
923ae0ff | 4639 | truncate_pagecache(inode, inode->i_size); |
072bd7ea | 4640 | } |
5208386c JK |
4641 | /* |
4642 | * We want to call ext4_truncate() even if attr->ia_size == | |
4643 | * inode->i_size for cases like truncation of fallocated space | |
4644 | */ | |
4645 | if (attr->ia_valid & ATTR_SIZE) | |
4646 | ext4_truncate(inode); | |
ac27a0ec | 4647 | |
1025774c CH |
4648 | if (!rc) { |
4649 | setattr_copy(inode, attr); | |
4650 | mark_inode_dirty(inode); | |
4651 | } | |
4652 | ||
4653 | /* | |
4654 | * If the call to ext4_truncate failed to get a transaction handle at | |
4655 | * all, we need to clean up the in-core orphan list manually. | |
4656 | */ | |
3d287de3 | 4657 | if (orphan && inode->i_nlink) |
617ba13b | 4658 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4659 | |
4660 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 4661 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
4662 | |
4663 | err_out: | |
617ba13b | 4664 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4665 | if (!error) |
4666 | error = rc; | |
4667 | return error; | |
4668 | } | |
4669 | ||
3e3398a0 MC |
4670 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4671 | struct kstat *stat) | |
4672 | { | |
4673 | struct inode *inode; | |
8af8eecc | 4674 | unsigned long long delalloc_blocks; |
3e3398a0 MC |
4675 | |
4676 | inode = dentry->d_inode; | |
4677 | generic_fillattr(inode, stat); | |
4678 | ||
9206c561 AD |
4679 | /* |
4680 | * If there is inline data in the inode, the inode will normally not | |
4681 | * have data blocks allocated (it may have an external xattr block). | |
4682 | * Report at least one sector for such files, so tools like tar, rsync, | |
4683 | * others doen't incorrectly think the file is completely sparse. | |
4684 | */ | |
4685 | if (unlikely(ext4_has_inline_data(inode))) | |
4686 | stat->blocks += (stat->size + 511) >> 9; | |
4687 | ||
3e3398a0 MC |
4688 | /* |
4689 | * We can't update i_blocks if the block allocation is delayed | |
4690 | * otherwise in the case of system crash before the real block | |
4691 | * allocation is done, we will have i_blocks inconsistent with | |
4692 | * on-disk file blocks. | |
4693 | * We always keep i_blocks updated together with real | |
4694 | * allocation. But to not confuse with user, stat | |
4695 | * will return the blocks that include the delayed allocation | |
4696 | * blocks for this file. | |
4697 | */ | |
96607551 | 4698 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
4699 | EXT4_I(inode)->i_reserved_data_blocks); |
4700 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
4701 | return 0; |
4702 | } | |
ac27a0ec | 4703 | |
fffb2739 JK |
4704 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
4705 | int pextents) | |
a02908f1 | 4706 | { |
12e9b892 | 4707 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
4708 | return ext4_ind_trans_blocks(inode, lblocks); |
4709 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 4710 | } |
ac51d837 | 4711 | |
ac27a0ec | 4712 | /* |
a02908f1 MC |
4713 | * Account for index blocks, block groups bitmaps and block group |
4714 | * descriptor blocks if modify datablocks and index blocks | |
4715 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4716 | * |
a02908f1 | 4717 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4718 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4719 | * they could still across block group boundary. |
ac27a0ec | 4720 | * |
a02908f1 MC |
4721 | * Also account for superblock, inode, quota and xattr blocks |
4722 | */ | |
fffb2739 JK |
4723 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
4724 | int pextents) | |
a02908f1 | 4725 | { |
8df9675f TT |
4726 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4727 | int gdpblocks; | |
a02908f1 MC |
4728 | int idxblocks; |
4729 | int ret = 0; | |
4730 | ||
4731 | /* | |
fffb2739 JK |
4732 | * How many index blocks need to touch to map @lblocks logical blocks |
4733 | * to @pextents physical extents? | |
a02908f1 | 4734 | */ |
fffb2739 | 4735 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
4736 | |
4737 | ret = idxblocks; | |
4738 | ||
4739 | /* | |
4740 | * Now let's see how many group bitmaps and group descriptors need | |
4741 | * to account | |
4742 | */ | |
fffb2739 | 4743 | groups = idxblocks + pextents; |
a02908f1 | 4744 | gdpblocks = groups; |
8df9675f TT |
4745 | if (groups > ngroups) |
4746 | groups = ngroups; | |
a02908f1 MC |
4747 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4748 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4749 | ||
4750 | /* bitmaps and block group descriptor blocks */ | |
4751 | ret += groups + gdpblocks; | |
4752 | ||
4753 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4754 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4755 | ||
4756 | return ret; | |
4757 | } | |
4758 | ||
4759 | /* | |
25985edc | 4760 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4761 | * the modification of a single pages into a single transaction, |
4762 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4763 | * |
525f4ed8 | 4764 | * This could be called via ext4_write_begin() |
ac27a0ec | 4765 | * |
525f4ed8 | 4766 | * We need to consider the worse case, when |
a02908f1 | 4767 | * one new block per extent. |
ac27a0ec | 4768 | */ |
a86c6181 | 4769 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4770 | { |
617ba13b | 4771 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4772 | int ret; |
4773 | ||
fffb2739 | 4774 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 4775 | |
a02908f1 | 4776 | /* Account for data blocks for journalled mode */ |
617ba13b | 4777 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4778 | ret += bpp; |
ac27a0ec DK |
4779 | return ret; |
4780 | } | |
f3bd1f3f MC |
4781 | |
4782 | /* | |
4783 | * Calculate the journal credits for a chunk of data modification. | |
4784 | * | |
4785 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4786 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4787 | * |
4788 | * journal buffers for data blocks are not included here, as DIO | |
4789 | * and fallocate do no need to journal data buffers. | |
4790 | */ | |
4791 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4792 | { | |
4793 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4794 | } | |
4795 | ||
ac27a0ec | 4796 | /* |
617ba13b | 4797 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4798 | * Give this, we know that the caller already has write access to iloc->bh. |
4799 | */ | |
617ba13b | 4800 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4801 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4802 | { |
4803 | int err = 0; | |
4804 | ||
c64db50e | 4805 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4806 | inode_inc_iversion(inode); |
4807 | ||
ac27a0ec DK |
4808 | /* the do_update_inode consumes one bh->b_count */ |
4809 | get_bh(iloc->bh); | |
4810 | ||
dab291af | 4811 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4812 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4813 | put_bh(iloc->bh); |
4814 | return err; | |
4815 | } | |
4816 | ||
4817 | /* | |
4818 | * On success, We end up with an outstanding reference count against | |
4819 | * iloc->bh. This _must_ be cleaned up later. | |
4820 | */ | |
4821 | ||
4822 | int | |
617ba13b MC |
4823 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4824 | struct ext4_iloc *iloc) | |
ac27a0ec | 4825 | { |
0390131b FM |
4826 | int err; |
4827 | ||
4828 | err = ext4_get_inode_loc(inode, iloc); | |
4829 | if (!err) { | |
4830 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4831 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4832 | if (err) { | |
4833 | brelse(iloc->bh); | |
4834 | iloc->bh = NULL; | |
ac27a0ec DK |
4835 | } |
4836 | } | |
617ba13b | 4837 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4838 | return err; |
4839 | } | |
4840 | ||
6dd4ee7c KS |
4841 | /* |
4842 | * Expand an inode by new_extra_isize bytes. | |
4843 | * Returns 0 on success or negative error number on failure. | |
4844 | */ | |
1d03ec98 AK |
4845 | static int ext4_expand_extra_isize(struct inode *inode, |
4846 | unsigned int new_extra_isize, | |
4847 | struct ext4_iloc iloc, | |
4848 | handle_t *handle) | |
6dd4ee7c KS |
4849 | { |
4850 | struct ext4_inode *raw_inode; | |
4851 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4852 | |
4853 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4854 | return 0; | |
4855 | ||
4856 | raw_inode = ext4_raw_inode(&iloc); | |
4857 | ||
4858 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4859 | |
4860 | /* No extended attributes present */ | |
19f5fb7a TT |
4861 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4862 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4863 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4864 | new_extra_isize); | |
4865 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4866 | return 0; | |
4867 | } | |
4868 | ||
4869 | /* try to expand with EAs present */ | |
4870 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4871 | raw_inode, handle); | |
4872 | } | |
4873 | ||
ac27a0ec DK |
4874 | /* |
4875 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4876 | * dirtiness (it may still be data-dirty). | |
4877 | * This means that the in-core inode may be reaped by prune_icache | |
4878 | * without having to perform any I/O. This is a very good thing, | |
4879 | * because *any* task may call prune_icache - even ones which | |
4880 | * have a transaction open against a different journal. | |
4881 | * | |
4882 | * Is this cheating? Not really. Sure, we haven't written the | |
4883 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4884 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4885 | * we start and wait on commits. | |
ac27a0ec | 4886 | */ |
617ba13b | 4887 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4888 | { |
617ba13b | 4889 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4890 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4891 | static unsigned int mnt_count; | |
4892 | int err, ret; | |
ac27a0ec DK |
4893 | |
4894 | might_sleep(); | |
7ff9c073 | 4895 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4896 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4897 | if (ext4_handle_valid(handle) && |
4898 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4899 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4900 | /* |
4901 | * We need extra buffer credits since we may write into EA block | |
4902 | * with this same handle. If journal_extend fails, then it will | |
4903 | * only result in a minor loss of functionality for that inode. | |
4904 | * If this is felt to be critical, then e2fsck should be run to | |
4905 | * force a large enough s_min_extra_isize. | |
4906 | */ | |
4907 | if ((jbd2_journal_extend(handle, | |
4908 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4909 | ret = ext4_expand_extra_isize(inode, | |
4910 | sbi->s_want_extra_isize, | |
4911 | iloc, handle); | |
4912 | if (ret) { | |
19f5fb7a TT |
4913 | ext4_set_inode_state(inode, |
4914 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4915 | if (mnt_count != |
4916 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4917 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4918 | "Unable to expand inode %lu. Delete" |
4919 | " some EAs or run e2fsck.", | |
4920 | inode->i_ino); | |
c1bddad9 AK |
4921 | mnt_count = |
4922 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4923 | } |
4924 | } | |
4925 | } | |
4926 | } | |
ac27a0ec | 4927 | if (!err) |
617ba13b | 4928 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4929 | return err; |
4930 | } | |
4931 | ||
4932 | /* | |
617ba13b | 4933 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4934 | * |
4935 | * We're really interested in the case where a file is being extended. | |
4936 | * i_size has been changed by generic_commit_write() and we thus need | |
4937 | * to include the updated inode in the current transaction. | |
4938 | * | |
5dd4056d | 4939 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4940 | * are allocated to the file. |
4941 | * | |
4942 | * If the inode is marked synchronous, we don't honour that here - doing | |
4943 | * so would cause a commit on atime updates, which we don't bother doing. | |
4944 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
4945 | * |
4946 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
4947 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
4948 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 4949 | */ |
aa385729 | 4950 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4951 | { |
ac27a0ec DK |
4952 | handle_t *handle; |
4953 | ||
0ae45f63 TT |
4954 | if (flags == I_DIRTY_TIME) |
4955 | return; | |
9924a92a | 4956 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
4957 | if (IS_ERR(handle)) |
4958 | goto out; | |
f3dc272f | 4959 | |
f3dc272f CW |
4960 | ext4_mark_inode_dirty(handle, inode); |
4961 | ||
617ba13b | 4962 | ext4_journal_stop(handle); |
ac27a0ec DK |
4963 | out: |
4964 | return; | |
4965 | } | |
4966 | ||
4967 | #if 0 | |
4968 | /* | |
4969 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4970 | * it from being flushed to disk early. Unlike | |
617ba13b | 4971 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4972 | * returns no iloc structure, so the caller needs to repeat the iloc |
4973 | * lookup to mark the inode dirty later. | |
4974 | */ | |
617ba13b | 4975 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 4976 | { |
617ba13b | 4977 | struct ext4_iloc iloc; |
ac27a0ec DK |
4978 | |
4979 | int err = 0; | |
4980 | if (handle) { | |
617ba13b | 4981 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
4982 | if (!err) { |
4983 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 4984 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 4985 | if (!err) |
0390131b | 4986 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 4987 | NULL, |
0390131b | 4988 | iloc.bh); |
ac27a0ec DK |
4989 | brelse(iloc.bh); |
4990 | } | |
4991 | } | |
617ba13b | 4992 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4993 | return err; |
4994 | } | |
4995 | #endif | |
4996 | ||
617ba13b | 4997 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
4998 | { |
4999 | journal_t *journal; | |
5000 | handle_t *handle; | |
5001 | int err; | |
5002 | ||
5003 | /* | |
5004 | * We have to be very careful here: changing a data block's | |
5005 | * journaling status dynamically is dangerous. If we write a | |
5006 | * data block to the journal, change the status and then delete | |
5007 | * that block, we risk forgetting to revoke the old log record | |
5008 | * from the journal and so a subsequent replay can corrupt data. | |
5009 | * So, first we make sure that the journal is empty and that | |
5010 | * nobody is changing anything. | |
5011 | */ | |
5012 | ||
617ba13b | 5013 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5014 | if (!journal) |
5015 | return 0; | |
d699594d | 5016 | if (is_journal_aborted(journal)) |
ac27a0ec | 5017 | return -EROFS; |
2aff57b0 YY |
5018 | /* We have to allocate physical blocks for delalloc blocks |
5019 | * before flushing journal. otherwise delalloc blocks can not | |
5020 | * be allocated any more. even more truncate on delalloc blocks | |
5021 | * could trigger BUG by flushing delalloc blocks in journal. | |
5022 | * There is no delalloc block in non-journal data mode. | |
5023 | */ | |
5024 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5025 | err = ext4_alloc_da_blocks(inode); | |
5026 | if (err < 0) | |
5027 | return err; | |
5028 | } | |
ac27a0ec | 5029 | |
17335dcc DM |
5030 | /* Wait for all existing dio workers */ |
5031 | ext4_inode_block_unlocked_dio(inode); | |
5032 | inode_dio_wait(inode); | |
5033 | ||
dab291af | 5034 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5035 | |
5036 | /* | |
5037 | * OK, there are no updates running now, and all cached data is | |
5038 | * synced to disk. We are now in a completely consistent state | |
5039 | * which doesn't have anything in the journal, and we know that | |
5040 | * no filesystem updates are running, so it is safe to modify | |
5041 | * the inode's in-core data-journaling state flag now. | |
5042 | */ | |
5043 | ||
5044 | if (val) | |
12e9b892 | 5045 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5046 | else { |
4f879ca6 JK |
5047 | err = jbd2_journal_flush(journal); |
5048 | if (err < 0) { | |
5049 | jbd2_journal_unlock_updates(journal); | |
5050 | ext4_inode_resume_unlocked_dio(inode); | |
5051 | return err; | |
5052 | } | |
12e9b892 | 5053 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5054 | } |
617ba13b | 5055 | ext4_set_aops(inode); |
ac27a0ec | 5056 | |
dab291af | 5057 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5058 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5059 | |
5060 | /* Finally we can mark the inode as dirty. */ | |
5061 | ||
9924a92a | 5062 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5063 | if (IS_ERR(handle)) |
5064 | return PTR_ERR(handle); | |
5065 | ||
617ba13b | 5066 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5067 | ext4_handle_sync(handle); |
617ba13b MC |
5068 | ext4_journal_stop(handle); |
5069 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5070 | |
5071 | return err; | |
5072 | } | |
2e9ee850 AK |
5073 | |
5074 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5075 | { | |
5076 | return !buffer_mapped(bh); | |
5077 | } | |
5078 | ||
c2ec175c | 5079 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5080 | { |
c2ec175c | 5081 | struct page *page = vmf->page; |
2e9ee850 AK |
5082 | loff_t size; |
5083 | unsigned long len; | |
9ea7df53 | 5084 | int ret; |
2e9ee850 | 5085 | struct file *file = vma->vm_file; |
496ad9aa | 5086 | struct inode *inode = file_inode(file); |
2e9ee850 | 5087 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5088 | handle_t *handle; |
5089 | get_block_t *get_block; | |
5090 | int retries = 0; | |
2e9ee850 | 5091 | |
8e8ad8a5 | 5092 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5093 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5094 | /* Delalloc case is easy... */ |
5095 | if (test_opt(inode->i_sb, DELALLOC) && | |
5096 | !ext4_should_journal_data(inode) && | |
5097 | !ext4_nonda_switch(inode->i_sb)) { | |
5098 | do { | |
5099 | ret = __block_page_mkwrite(vma, vmf, | |
5100 | ext4_da_get_block_prep); | |
5101 | } while (ret == -ENOSPC && | |
5102 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5103 | goto out_ret; | |
2e9ee850 | 5104 | } |
0e499890 DW |
5105 | |
5106 | lock_page(page); | |
9ea7df53 JK |
5107 | size = i_size_read(inode); |
5108 | /* Page got truncated from under us? */ | |
5109 | if (page->mapping != mapping || page_offset(page) > size) { | |
5110 | unlock_page(page); | |
5111 | ret = VM_FAULT_NOPAGE; | |
5112 | goto out; | |
0e499890 | 5113 | } |
2e9ee850 AK |
5114 | |
5115 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5116 | len = size & ~PAGE_CACHE_MASK; | |
5117 | else | |
5118 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5119 | /* |
9ea7df53 JK |
5120 | * Return if we have all the buffers mapped. This avoids the need to do |
5121 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5122 | */ |
2e9ee850 | 5123 | if (page_has_buffers(page)) { |
f19d5870 TM |
5124 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5125 | 0, len, NULL, | |
5126 | ext4_bh_unmapped)) { | |
9ea7df53 | 5127 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5128 | wait_for_stable_page(page); |
9ea7df53 JK |
5129 | ret = VM_FAULT_LOCKED; |
5130 | goto out; | |
a827eaff | 5131 | } |
2e9ee850 | 5132 | } |
a827eaff | 5133 | unlock_page(page); |
9ea7df53 JK |
5134 | /* OK, we need to fill the hole... */ |
5135 | if (ext4_should_dioread_nolock(inode)) | |
5136 | get_block = ext4_get_block_write; | |
5137 | else | |
5138 | get_block = ext4_get_block; | |
5139 | retry_alloc: | |
9924a92a TT |
5140 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5141 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5142 | if (IS_ERR(handle)) { |
c2ec175c | 5143 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5144 | goto out; |
5145 | } | |
5146 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5147 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5148 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5149 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5150 | unlock_page(page); | |
5151 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5152 | ext4_journal_stop(handle); |
9ea7df53 JK |
5153 | goto out; |
5154 | } | |
5155 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5156 | } | |
5157 | ext4_journal_stop(handle); | |
5158 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5159 | goto retry_alloc; | |
5160 | out_ret: | |
5161 | ret = block_page_mkwrite_return(ret); | |
5162 | out: | |
8e8ad8a5 | 5163 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5164 | return ret; |
5165 | } |