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> | |
ac27a0ec DK |
23 | #include <linux/highuid.h> |
24 | #include <linux/pagemap.h> | |
c94c2acf | 25 | #include <linux/dax.h> |
ac27a0ec DK |
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> |
00a1a053 | 39 | #include <linux/bitops.h> |
9bffad1e | 40 | |
3dcf5451 | 41 | #include "ext4_jbd2.h" |
ac27a0ec DK |
42 | #include "xattr.h" |
43 | #include "acl.h" | |
9f125d64 | 44 | #include "truncate.h" |
ac27a0ec | 45 | |
9bffad1e TT |
46 | #include <trace/events/ext4.h> |
47 | ||
a1d6cc56 AK |
48 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
49 | ||
814525f4 DW |
50 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
51 | struct ext4_inode_info *ei) | |
52 | { | |
53 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
54 | __u16 csum_lo; | |
55 | __u16 csum_hi = 0; | |
56 | __u32 csum; | |
57 | ||
171a7f21 | 58 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
59 | raw->i_checksum_lo = 0; |
60 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
61 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 62 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
63 | raw->i_checksum_hi = 0; |
64 | } | |
65 | ||
66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
67 | EXT4_INODE_SIZE(inode->i_sb)); | |
68 | ||
171a7f21 | 69 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
70 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
71 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 72 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
73 | |
74 | return csum; | |
75 | } | |
76 | ||
77 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
78 | struct ext4_inode_info *ei) | |
79 | { | |
80 | __u32 provided, calculated; | |
81 | ||
82 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
83 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 84 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
85 | return 1; |
86 | ||
87 | provided = le16_to_cpu(raw->i_checksum_lo); | |
88 | calculated = ext4_inode_csum(inode, raw, ei); | |
89 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
90 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
91 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
92 | else | |
93 | calculated &= 0xFFFF; | |
94 | ||
95 | return provided == calculated; | |
96 | } | |
97 | ||
98 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
99 | struct ext4_inode_info *ei) | |
100 | { | |
101 | __u32 csum; | |
102 | ||
103 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
104 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 105 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
106 | return; |
107 | ||
108 | csum = ext4_inode_csum(inode, raw, ei); | |
109 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
110 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
111 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
112 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
113 | } | |
114 | ||
678aaf48 JK |
115 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
116 | loff_t new_size) | |
117 | { | |
7ff9c073 | 118 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
119 | /* |
120 | * If jinode is zero, then we never opened the file for | |
121 | * writing, so there's no need to call | |
122 | * jbd2_journal_begin_ordered_truncate() since there's no | |
123 | * outstanding writes we need to flush. | |
124 | */ | |
125 | if (!EXT4_I(inode)->jinode) | |
126 | return 0; | |
127 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
128 | EXT4_I(inode)->jinode, | |
129 | new_size); | |
678aaf48 JK |
130 | } |
131 | ||
d47992f8 LC |
132 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
133 | unsigned int length); | |
cb20d518 TT |
134 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
135 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
fffb2739 JK |
136 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
137 | int pextents); | |
64769240 | 138 | |
ac27a0ec DK |
139 | /* |
140 | * Test whether an inode is a fast symlink. | |
141 | */ | |
f348c252 | 142 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 143 | { |
65eddb56 YY |
144 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
145 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
ac27a0ec | 146 | |
bd9db175 ZL |
147 | if (ext4_has_inline_data(inode)) |
148 | return 0; | |
149 | ||
ac27a0ec DK |
150 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
151 | } | |
152 | ||
ac27a0ec DK |
153 | /* |
154 | * Restart the transaction associated with *handle. This does a commit, | |
155 | * so before we call here everything must be consistently dirtied against | |
156 | * this transaction. | |
157 | */ | |
fa5d1113 | 158 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 159 | int nblocks) |
ac27a0ec | 160 | { |
487caeef JK |
161 | int ret; |
162 | ||
163 | /* | |
e35fd660 | 164 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
165 | * moment, get_block can be called only for blocks inside i_size since |
166 | * page cache has been already dropped and writes are blocked by | |
167 | * i_mutex. So we can safely drop the i_data_sem here. | |
168 | */ | |
0390131b | 169 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 170 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 171 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 172 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 173 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 174 | ext4_discard_preallocations(inode); |
487caeef JK |
175 | |
176 | return ret; | |
ac27a0ec DK |
177 | } |
178 | ||
179 | /* | |
180 | * Called at the last iput() if i_nlink is zero. | |
181 | */ | |
0930fcc1 | 182 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
183 | { |
184 | handle_t *handle; | |
bc965ab3 | 185 | int err; |
ac27a0ec | 186 | |
7ff9c073 | 187 | trace_ext4_evict_inode(inode); |
2581fdc8 | 188 | |
0930fcc1 | 189 | if (inode->i_nlink) { |
2d859db3 JK |
190 | /* |
191 | * When journalling data dirty buffers are tracked only in the | |
192 | * journal. So although mm thinks everything is clean and | |
193 | * ready for reaping the inode might still have some pages to | |
194 | * write in the running transaction or waiting to be | |
195 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
196 | * (via truncate_inode_pages()) to discard these buffers can | |
197 | * cause data loss. Also even if we did not discard these | |
198 | * buffers, we would have no way to find them after the inode | |
199 | * is reaped and thus user could see stale data if he tries to | |
200 | * read them before the transaction is checkpointed. So be | |
201 | * careful and force everything to disk here... We use | |
202 | * ei->i_datasync_tid to store the newest transaction | |
203 | * containing inode's data. | |
204 | * | |
205 | * Note that directories do not have this problem because they | |
206 | * don't use page cache. | |
207 | */ | |
208 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
209 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
210 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
211 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
212 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
213 | ||
d76a3a77 | 214 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
215 | filemap_write_and_wait(&inode->i_data); |
216 | } | |
91b0abe3 | 217 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd JK |
218 | |
219 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); | |
0930fcc1 AV |
220 | goto no_delete; |
221 | } | |
222 | ||
e2bfb088 TT |
223 | if (is_bad_inode(inode)) |
224 | goto no_delete; | |
225 | dquot_initialize(inode); | |
907f4554 | 226 | |
678aaf48 JK |
227 | if (ext4_should_order_data(inode)) |
228 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 229 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 230 | |
5dc23bdd | 231 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); |
ac27a0ec | 232 | |
8e8ad8a5 JK |
233 | /* |
234 | * Protect us against freezing - iput() caller didn't have to have any | |
235 | * protection against it | |
236 | */ | |
237 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
238 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
239 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 240 | if (IS_ERR(handle)) { |
bc965ab3 | 241 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
242 | /* |
243 | * If we're going to skip the normal cleanup, we still need to | |
244 | * make sure that the in-core orphan linked list is properly | |
245 | * cleaned up. | |
246 | */ | |
617ba13b | 247 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 248 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
249 | goto no_delete; |
250 | } | |
251 | ||
252 | if (IS_SYNC(inode)) | |
0390131b | 253 | ext4_handle_sync(handle); |
ac27a0ec | 254 | inode->i_size = 0; |
bc965ab3 TT |
255 | err = ext4_mark_inode_dirty(handle, inode); |
256 | if (err) { | |
12062ddd | 257 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
258 | "couldn't mark inode dirty (err %d)", err); |
259 | goto stop_handle; | |
260 | } | |
ac27a0ec | 261 | if (inode->i_blocks) |
617ba13b | 262 | ext4_truncate(inode); |
bc965ab3 TT |
263 | |
264 | /* | |
265 | * ext4_ext_truncate() doesn't reserve any slop when it | |
266 | * restarts journal transactions; therefore there may not be | |
267 | * enough credits left in the handle to remove the inode from | |
268 | * the orphan list and set the dtime field. | |
269 | */ | |
0390131b | 270 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
271 | err = ext4_journal_extend(handle, 3); |
272 | if (err > 0) | |
273 | err = ext4_journal_restart(handle, 3); | |
274 | if (err != 0) { | |
12062ddd | 275 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
276 | "couldn't extend journal (err %d)", err); |
277 | stop_handle: | |
278 | ext4_journal_stop(handle); | |
45388219 | 279 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 280 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
281 | goto no_delete; |
282 | } | |
283 | } | |
284 | ||
ac27a0ec | 285 | /* |
617ba13b | 286 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 287 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 288 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 289 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 290 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
291 | * (Well, we could do this if we need to, but heck - it works) |
292 | */ | |
617ba13b MC |
293 | ext4_orphan_del(handle, inode); |
294 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
295 | |
296 | /* | |
297 | * One subtle ordering requirement: if anything has gone wrong | |
298 | * (transaction abort, IO errors, whatever), then we can still | |
299 | * do these next steps (the fs will already have been marked as | |
300 | * having errors), but we can't free the inode if the mark_dirty | |
301 | * fails. | |
302 | */ | |
617ba13b | 303 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 304 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 305 | ext4_clear_inode(inode); |
ac27a0ec | 306 | else |
617ba13b MC |
307 | ext4_free_inode(handle, inode); |
308 | ext4_journal_stop(handle); | |
8e8ad8a5 | 309 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
310 | return; |
311 | no_delete: | |
0930fcc1 | 312 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
313 | } |
314 | ||
a9e7f447 DM |
315 | #ifdef CONFIG_QUOTA |
316 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 317 | { |
a9e7f447 | 318 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 319 | } |
a9e7f447 | 320 | #endif |
9d0be502 | 321 | |
0637c6f4 TT |
322 | /* |
323 | * Called with i_data_sem down, which is important since we can call | |
324 | * ext4_discard_preallocations() from here. | |
325 | */ | |
5f634d06 AK |
326 | void ext4_da_update_reserve_space(struct inode *inode, |
327 | int used, int quota_claim) | |
12219aea AK |
328 | { |
329 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 330 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
331 | |
332 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 333 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 334 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 335 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 336 | "with only %d reserved data blocks", |
0637c6f4 TT |
337 | __func__, inode->i_ino, used, |
338 | ei->i_reserved_data_blocks); | |
339 | WARN_ON(1); | |
340 | used = ei->i_reserved_data_blocks; | |
341 | } | |
12219aea | 342 | |
0637c6f4 TT |
343 | /* Update per-inode reservations */ |
344 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 345 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 346 | |
12219aea | 347 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 348 | |
72b8ab9d ES |
349 | /* Update quota subsystem for data blocks */ |
350 | if (quota_claim) | |
7b415bf6 | 351 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 352 | else { |
5f634d06 AK |
353 | /* |
354 | * We did fallocate with an offset that is already delayed | |
355 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 356 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 357 | */ |
7b415bf6 | 358 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 359 | } |
d6014301 AK |
360 | |
361 | /* | |
362 | * If we have done all the pending block allocations and if | |
363 | * there aren't any writers on the inode, we can discard the | |
364 | * inode's preallocations. | |
365 | */ | |
0637c6f4 TT |
366 | if ((ei->i_reserved_data_blocks == 0) && |
367 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 368 | ext4_discard_preallocations(inode); |
12219aea AK |
369 | } |
370 | ||
e29136f8 | 371 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
372 | unsigned int line, |
373 | struct ext4_map_blocks *map) | |
6fd058f7 | 374 | { |
24676da4 TT |
375 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
376 | map->m_len)) { | |
c398eda0 TT |
377 | ext4_error_inode(inode, func, line, map->m_pblk, |
378 | "lblock %lu mapped to illegal pblock " | |
379 | "(length %d)", (unsigned long) map->m_lblk, | |
380 | map->m_len); | |
6a797d27 | 381 | return -EFSCORRUPTED; |
6fd058f7 TT |
382 | } |
383 | return 0; | |
384 | } | |
385 | ||
53085fac JK |
386 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
387 | ext4_lblk_t len) | |
388 | { | |
389 | int ret; | |
390 | ||
391 | if (ext4_encrypted_inode(inode)) | |
392 | return ext4_encrypted_zeroout(inode, lblk, pblk, len); | |
393 | ||
394 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
395 | if (ret > 0) | |
396 | ret = 0; | |
397 | ||
398 | return ret; | |
399 | } | |
400 | ||
e29136f8 | 401 | #define check_block_validity(inode, map) \ |
c398eda0 | 402 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 403 | |
921f266b DM |
404 | #ifdef ES_AGGRESSIVE_TEST |
405 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
406 | struct inode *inode, | |
407 | struct ext4_map_blocks *es_map, | |
408 | struct ext4_map_blocks *map, | |
409 | int flags) | |
410 | { | |
411 | int retval; | |
412 | ||
413 | map->m_flags = 0; | |
414 | /* | |
415 | * There is a race window that the result is not the same. | |
416 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
417 | * is that we lookup a block mapping in extent status tree with | |
418 | * out taking i_data_sem. So at the time the unwritten extent | |
419 | * could be converted. | |
420 | */ | |
2dcba478 | 421 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
422 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
423 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
424 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
425 | } else { | |
426 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
427 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
428 | } | |
2dcba478 | 429 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
430 | |
431 | /* | |
432 | * We don't check m_len because extent will be collpased in status | |
433 | * tree. So the m_len might not equal. | |
434 | */ | |
435 | if (es_map->m_lblk != map->m_lblk || | |
436 | es_map->m_flags != map->m_flags || | |
437 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 438 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
439 | "es_cached ex [%d/%d/%llu/%x] != " |
440 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
441 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
442 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
443 | map->m_len, map->m_pblk, map->m_flags, | |
444 | retval, flags); | |
445 | } | |
446 | } | |
447 | #endif /* ES_AGGRESSIVE_TEST */ | |
448 | ||
f5ab0d1f | 449 | /* |
e35fd660 | 450 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 451 | * and returns if the blocks are already mapped. |
f5ab0d1f | 452 | * |
f5ab0d1f MC |
453 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
454 | * and store the allocated blocks in the result buffer head and mark it | |
455 | * mapped. | |
456 | * | |
e35fd660 TT |
457 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
458 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
459 | * based files |
460 | * | |
556615dc LC |
461 | * On success, it returns the number of blocks being mapped or allocated. |
462 | * if create==0 and the blocks are pre-allocated and unwritten block, | |
f5ab0d1f MC |
463 | * the result buffer head is unmapped. If the create ==1, it will make sure |
464 | * the buffer head is mapped. | |
465 | * | |
466 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 467 | * that case, buffer head is unmapped |
f5ab0d1f MC |
468 | * |
469 | * It returns the error in case of allocation failure. | |
470 | */ | |
e35fd660 TT |
471 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
472 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 473 | { |
d100eef2 | 474 | struct extent_status es; |
0e855ac8 | 475 | int retval; |
b8a86845 | 476 | int ret = 0; |
921f266b DM |
477 | #ifdef ES_AGGRESSIVE_TEST |
478 | struct ext4_map_blocks orig_map; | |
479 | ||
480 | memcpy(&orig_map, map, sizeof(*map)); | |
481 | #endif | |
f5ab0d1f | 482 | |
e35fd660 TT |
483 | map->m_flags = 0; |
484 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
485 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
486 | (unsigned long) map->m_lblk); | |
d100eef2 | 487 | |
e861b5e9 TT |
488 | /* |
489 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
490 | */ | |
491 | if (unlikely(map->m_len > INT_MAX)) | |
492 | map->m_len = INT_MAX; | |
493 | ||
4adb6ab3 KM |
494 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
495 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 496 | return -EFSCORRUPTED; |
4adb6ab3 | 497 | |
d100eef2 ZL |
498 | /* Lookup extent status tree firstly */ |
499 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
500 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
501 | map->m_pblk = ext4_es_pblock(&es) + | |
502 | map->m_lblk - es.es_lblk; | |
503 | map->m_flags |= ext4_es_is_written(&es) ? | |
504 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
505 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
506 | if (retval > map->m_len) | |
507 | retval = map->m_len; | |
508 | map->m_len = retval; | |
509 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
510 | retval = 0; | |
511 | } else { | |
512 | BUG_ON(1); | |
513 | } | |
921f266b DM |
514 | #ifdef ES_AGGRESSIVE_TEST |
515 | ext4_map_blocks_es_recheck(handle, inode, map, | |
516 | &orig_map, flags); | |
517 | #endif | |
d100eef2 ZL |
518 | goto found; |
519 | } | |
520 | ||
4df3d265 | 521 | /* |
b920c755 TT |
522 | * Try to see if we can get the block without requesting a new |
523 | * file system block. | |
4df3d265 | 524 | */ |
2dcba478 | 525 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 526 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
527 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
528 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 529 | } else { |
a4e5d88b DM |
530 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
531 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 532 | } |
f7fec032 | 533 | if (retval > 0) { |
3be78c73 | 534 | unsigned int status; |
f7fec032 | 535 | |
44fb851d ZL |
536 | if (unlikely(retval != map->m_len)) { |
537 | ext4_warning(inode->i_sb, | |
538 | "ES len assertion failed for inode " | |
539 | "%lu: retval %d != map->m_len %d", | |
540 | inode->i_ino, retval, map->m_len); | |
541 | WARN_ON(1); | |
921f266b | 542 | } |
921f266b | 543 | |
f7fec032 ZL |
544 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
545 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
546 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 547 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
548 | ext4_find_delalloc_range(inode, map->m_lblk, |
549 | map->m_lblk + map->m_len - 1)) | |
550 | status |= EXTENT_STATUS_DELAYED; | |
551 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
552 | map->m_len, map->m_pblk, status); | |
553 | if (ret < 0) | |
554 | retval = ret; | |
555 | } | |
2dcba478 | 556 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 557 | |
d100eef2 | 558 | found: |
e35fd660 | 559 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 560 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
561 | if (ret != 0) |
562 | return ret; | |
563 | } | |
564 | ||
f5ab0d1f | 565 | /* If it is only a block(s) look up */ |
c2177057 | 566 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
567 | return retval; |
568 | ||
569 | /* | |
570 | * Returns if the blocks have already allocated | |
571 | * | |
572 | * Note that if blocks have been preallocated | |
df3ab170 | 573 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
574 | * with buffer head unmapped. |
575 | */ | |
e35fd660 | 576 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
577 | /* |
578 | * If we need to convert extent to unwritten | |
579 | * we continue and do the actual work in | |
580 | * ext4_ext_map_blocks() | |
581 | */ | |
582 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
583 | return retval; | |
4df3d265 | 584 | |
2a8964d6 | 585 | /* |
a25a4e1a ZL |
586 | * Here we clear m_flags because after allocating an new extent, |
587 | * it will be set again. | |
2a8964d6 | 588 | */ |
a25a4e1a | 589 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 590 | |
4df3d265 | 591 | /* |
556615dc | 592 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 593 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 594 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 595 | * with create == 1 flag. |
4df3d265 | 596 | */ |
c8b459f4 | 597 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 598 | |
4df3d265 AK |
599 | /* |
600 | * We need to check for EXT4 here because migrate | |
601 | * could have changed the inode type in between | |
602 | */ | |
12e9b892 | 603 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 604 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 605 | } else { |
e35fd660 | 606 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 607 | |
e35fd660 | 608 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
609 | /* |
610 | * We allocated new blocks which will result in | |
611 | * i_data's format changing. Force the migrate | |
612 | * to fail by clearing migrate flags | |
613 | */ | |
19f5fb7a | 614 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 615 | } |
d2a17637 | 616 | |
5f634d06 AK |
617 | /* |
618 | * Update reserved blocks/metadata blocks after successful | |
619 | * block allocation which had been deferred till now. We don't | |
620 | * support fallocate for non extent files. So we can update | |
621 | * reserve space here. | |
622 | */ | |
623 | if ((retval > 0) && | |
1296cc85 | 624 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
625 | ext4_da_update_reserve_space(inode, retval, 1); |
626 | } | |
2ac3b6e0 | 627 | |
f7fec032 | 628 | if (retval > 0) { |
3be78c73 | 629 | unsigned int status; |
f7fec032 | 630 | |
44fb851d ZL |
631 | if (unlikely(retval != map->m_len)) { |
632 | ext4_warning(inode->i_sb, | |
633 | "ES len assertion failed for inode " | |
634 | "%lu: retval %d != map->m_len %d", | |
635 | inode->i_ino, retval, map->m_len); | |
636 | WARN_ON(1); | |
921f266b | 637 | } |
921f266b | 638 | |
c86d8db3 JK |
639 | /* |
640 | * We have to zeroout blocks before inserting them into extent | |
641 | * status tree. Otherwise someone could look them up there and | |
642 | * use them before they are really zeroed. | |
643 | */ | |
644 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
645 | map->m_flags & EXT4_MAP_MAPPED && | |
646 | map->m_flags & EXT4_MAP_NEW) { | |
647 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
648 | map->m_pblk, map->m_len); | |
649 | if (ret) { | |
650 | retval = ret; | |
651 | goto out_sem; | |
652 | } | |
653 | } | |
654 | ||
adb23551 ZL |
655 | /* |
656 | * If the extent has been zeroed out, we don't need to update | |
657 | * extent status tree. | |
658 | */ | |
659 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
660 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
661 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 662 | goto out_sem; |
adb23551 | 663 | } |
f7fec032 ZL |
664 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
665 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
666 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 667 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
668 | ext4_find_delalloc_range(inode, map->m_lblk, |
669 | map->m_lblk + map->m_len - 1)) | |
670 | status |= EXTENT_STATUS_DELAYED; | |
671 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
672 | map->m_pblk, status); | |
c86d8db3 | 673 | if (ret < 0) { |
f7fec032 | 674 | retval = ret; |
c86d8db3 JK |
675 | goto out_sem; |
676 | } | |
5356f261 AK |
677 | } |
678 | ||
c86d8db3 | 679 | out_sem: |
4df3d265 | 680 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 681 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 682 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
683 | if (ret != 0) |
684 | return ret; | |
685 | } | |
0e855ac8 AK |
686 | return retval; |
687 | } | |
688 | ||
ed8ad838 JK |
689 | /* |
690 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
691 | * we have to be careful as someone else may be manipulating b_state as well. | |
692 | */ | |
693 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
694 | { | |
695 | unsigned long old_state; | |
696 | unsigned long new_state; | |
697 | ||
698 | flags &= EXT4_MAP_FLAGS; | |
699 | ||
700 | /* Dummy buffer_head? Set non-atomically. */ | |
701 | if (!bh->b_page) { | |
702 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
703 | return; | |
704 | } | |
705 | /* | |
706 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
707 | * once we get rid of using bh as a container for mapping information | |
708 | * to pass to / from get_block functions, this can go away. | |
709 | */ | |
710 | do { | |
711 | old_state = READ_ONCE(bh->b_state); | |
712 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
713 | } while (unlikely( | |
714 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
715 | } | |
716 | ||
f3bd1f3f MC |
717 | /* Maximum number of blocks we map for direct IO at once. */ |
718 | #define DIO_MAX_BLOCKS 4096 | |
719 | ||
2ed88685 TT |
720 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
721 | struct buffer_head *bh, int flags) | |
ac27a0ec | 722 | { |
3e4fdaf8 | 723 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 724 | struct ext4_map_blocks map; |
7fb5409d | 725 | int ret = 0, started = 0; |
f3bd1f3f | 726 | int dio_credits; |
ac27a0ec | 727 | |
46c7f254 TM |
728 | if (ext4_has_inline_data(inode)) |
729 | return -ERANGE; | |
730 | ||
2ed88685 TT |
731 | map.m_lblk = iblock; |
732 | map.m_len = bh->b_size >> inode->i_blkbits; | |
733 | ||
2dcba478 | 734 | if (flags && !handle) { |
7fb5409d | 735 | /* Direct IO write... */ |
2ed88685 TT |
736 | if (map.m_len > DIO_MAX_BLOCKS) |
737 | map.m_len = DIO_MAX_BLOCKS; | |
738 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
739 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
740 | dio_credits); | |
7fb5409d | 741 | if (IS_ERR(handle)) { |
ac27a0ec | 742 | ret = PTR_ERR(handle); |
2ed88685 | 743 | return ret; |
ac27a0ec | 744 | } |
7fb5409d | 745 | started = 1; |
ac27a0ec DK |
746 | } |
747 | ||
2ed88685 | 748 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 749 | if (ret > 0) { |
7b7a8665 CH |
750 | ext4_io_end_t *io_end = ext4_inode_aio(inode); |
751 | ||
2ed88685 | 752 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 753 | ext4_update_bh_state(bh, map.m_flags); |
7b7a8665 CH |
754 | if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) |
755 | set_buffer_defer_completion(bh); | |
2ed88685 | 756 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 757 | ret = 0; |
ac27a0ec | 758 | } |
7fb5409d JK |
759 | if (started) |
760 | ext4_journal_stop(handle); | |
ac27a0ec DK |
761 | return ret; |
762 | } | |
763 | ||
2ed88685 TT |
764 | int ext4_get_block(struct inode *inode, sector_t iblock, |
765 | struct buffer_head *bh, int create) | |
766 | { | |
767 | return _ext4_get_block(inode, iblock, bh, | |
768 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
769 | } | |
770 | ||
ac27a0ec DK |
771 | /* |
772 | * `handle' can be NULL if create is zero | |
773 | */ | |
617ba13b | 774 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 775 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 776 | { |
2ed88685 TT |
777 | struct ext4_map_blocks map; |
778 | struct buffer_head *bh; | |
c5e298ae | 779 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 780 | int err; |
ac27a0ec DK |
781 | |
782 | J_ASSERT(handle != NULL || create == 0); | |
783 | ||
2ed88685 TT |
784 | map.m_lblk = block; |
785 | map.m_len = 1; | |
c5e298ae | 786 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 787 | |
10560082 TT |
788 | if (err == 0) |
789 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 790 | if (err < 0) |
10560082 | 791 | return ERR_PTR(err); |
2ed88685 TT |
792 | |
793 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
794 | if (unlikely(!bh)) |
795 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
796 | if (map.m_flags & EXT4_MAP_NEW) { |
797 | J_ASSERT(create != 0); | |
798 | J_ASSERT(handle != NULL); | |
ac27a0ec | 799 | |
2ed88685 TT |
800 | /* |
801 | * Now that we do not always journal data, we should | |
802 | * keep in mind whether this should always journal the | |
803 | * new buffer as metadata. For now, regular file | |
804 | * writes use ext4_get_block instead, so it's not a | |
805 | * problem. | |
806 | */ | |
807 | lock_buffer(bh); | |
808 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
809 | err = ext4_journal_get_create_access(handle, bh); |
810 | if (unlikely(err)) { | |
811 | unlock_buffer(bh); | |
812 | goto errout; | |
813 | } | |
814 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
815 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
816 | set_buffer_uptodate(bh); | |
ac27a0ec | 817 | } |
2ed88685 TT |
818 | unlock_buffer(bh); |
819 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
820 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
821 | if (unlikely(err)) |
822 | goto errout; | |
823 | } else | |
2ed88685 | 824 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 825 | return bh; |
10560082 TT |
826 | errout: |
827 | brelse(bh); | |
828 | return ERR_PTR(err); | |
ac27a0ec DK |
829 | } |
830 | ||
617ba13b | 831 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 832 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 833 | { |
af5bc92d | 834 | struct buffer_head *bh; |
ac27a0ec | 835 | |
c5e298ae | 836 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 837 | if (IS_ERR(bh)) |
ac27a0ec | 838 | return bh; |
1c215028 | 839 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 840 | return bh; |
65299a3b | 841 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
842 | wait_on_buffer(bh); |
843 | if (buffer_uptodate(bh)) | |
844 | return bh; | |
845 | put_bh(bh); | |
1c215028 | 846 | return ERR_PTR(-EIO); |
ac27a0ec DK |
847 | } |
848 | ||
f19d5870 TM |
849 | int ext4_walk_page_buffers(handle_t *handle, |
850 | struct buffer_head *head, | |
851 | unsigned from, | |
852 | unsigned to, | |
853 | int *partial, | |
854 | int (*fn)(handle_t *handle, | |
855 | struct buffer_head *bh)) | |
ac27a0ec DK |
856 | { |
857 | struct buffer_head *bh; | |
858 | unsigned block_start, block_end; | |
859 | unsigned blocksize = head->b_size; | |
860 | int err, ret = 0; | |
861 | struct buffer_head *next; | |
862 | ||
af5bc92d TT |
863 | for (bh = head, block_start = 0; |
864 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 865 | block_start = block_end, bh = next) { |
ac27a0ec DK |
866 | next = bh->b_this_page; |
867 | block_end = block_start + blocksize; | |
868 | if (block_end <= from || block_start >= to) { | |
869 | if (partial && !buffer_uptodate(bh)) | |
870 | *partial = 1; | |
871 | continue; | |
872 | } | |
873 | err = (*fn)(handle, bh); | |
874 | if (!ret) | |
875 | ret = err; | |
876 | } | |
877 | return ret; | |
878 | } | |
879 | ||
880 | /* | |
881 | * To preserve ordering, it is essential that the hole instantiation and | |
882 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 883 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 884 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
885 | * prepare_write() is the right place. |
886 | * | |
36ade451 JK |
887 | * Also, this function can nest inside ext4_writepage(). In that case, we |
888 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
889 | * whole page. So we won't block on the journal in that case, which is good, | |
890 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 891 | * |
617ba13b | 892 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
893 | * quota file writes. If we were to commit the transaction while thus |
894 | * reentered, there can be a deadlock - we would be holding a quota | |
895 | * lock, and the commit would never complete if another thread had a | |
896 | * transaction open and was blocking on the quota lock - a ranking | |
897 | * violation. | |
898 | * | |
dab291af | 899 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
900 | * will _not_ run commit under these circumstances because handle->h_ref |
901 | * is elevated. We'll still have enough credits for the tiny quotafile | |
902 | * write. | |
903 | */ | |
f19d5870 TM |
904 | int do_journal_get_write_access(handle_t *handle, |
905 | struct buffer_head *bh) | |
ac27a0ec | 906 | { |
56d35a4c JK |
907 | int dirty = buffer_dirty(bh); |
908 | int ret; | |
909 | ||
ac27a0ec DK |
910 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
911 | return 0; | |
56d35a4c | 912 | /* |
ebdec241 | 913 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
914 | * the dirty bit as jbd2_journal_get_write_access() could complain |
915 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 916 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
917 | * the bit before releasing a page lock and thus writeback cannot |
918 | * ever write the buffer. | |
919 | */ | |
920 | if (dirty) | |
921 | clear_buffer_dirty(bh); | |
5d601255 | 922 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
923 | ret = ext4_journal_get_write_access(handle, bh); |
924 | if (!ret && dirty) | |
925 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
926 | return ret; | |
ac27a0ec DK |
927 | } |
928 | ||
2058f83a MH |
929 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
930 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
931 | get_block_t *get_block) | |
932 | { | |
933 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
934 | unsigned to = from + len; | |
935 | struct inode *inode = page->mapping->host; | |
936 | unsigned block_start, block_end; | |
937 | sector_t block; | |
938 | int err = 0; | |
939 | unsigned blocksize = inode->i_sb->s_blocksize; | |
940 | unsigned bbits; | |
941 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
942 | bool decrypt = false; | |
943 | ||
944 | BUG_ON(!PageLocked(page)); | |
945 | BUG_ON(from > PAGE_CACHE_SIZE); | |
946 | BUG_ON(to > PAGE_CACHE_SIZE); | |
947 | BUG_ON(from > to); | |
948 | ||
949 | if (!page_has_buffers(page)) | |
950 | create_empty_buffers(page, blocksize, 0); | |
951 | head = page_buffers(page); | |
952 | bbits = ilog2(blocksize); | |
953 | block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); | |
954 | ||
955 | for (bh = head, block_start = 0; bh != head || !block_start; | |
956 | block++, block_start = block_end, bh = bh->b_this_page) { | |
957 | block_end = block_start + blocksize; | |
958 | if (block_end <= from || block_start >= to) { | |
959 | if (PageUptodate(page)) { | |
960 | if (!buffer_uptodate(bh)) | |
961 | set_buffer_uptodate(bh); | |
962 | } | |
963 | continue; | |
964 | } | |
965 | if (buffer_new(bh)) | |
966 | clear_buffer_new(bh); | |
967 | if (!buffer_mapped(bh)) { | |
968 | WARN_ON(bh->b_size != blocksize); | |
969 | err = get_block(inode, block, bh, 1); | |
970 | if (err) | |
971 | break; | |
972 | if (buffer_new(bh)) { | |
973 | unmap_underlying_metadata(bh->b_bdev, | |
974 | bh->b_blocknr); | |
975 | if (PageUptodate(page)) { | |
976 | clear_buffer_new(bh); | |
977 | set_buffer_uptodate(bh); | |
978 | mark_buffer_dirty(bh); | |
979 | continue; | |
980 | } | |
981 | if (block_end > to || block_start < from) | |
982 | zero_user_segments(page, to, block_end, | |
983 | block_start, from); | |
984 | continue; | |
985 | } | |
986 | } | |
987 | if (PageUptodate(page)) { | |
988 | if (!buffer_uptodate(bh)) | |
989 | set_buffer_uptodate(bh); | |
990 | continue; | |
991 | } | |
992 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
993 | !buffer_unwritten(bh) && | |
994 | (block_start < from || block_end > to)) { | |
995 | ll_rw_block(READ, 1, &bh); | |
996 | *wait_bh++ = bh; | |
997 | decrypt = ext4_encrypted_inode(inode) && | |
998 | S_ISREG(inode->i_mode); | |
999 | } | |
1000 | } | |
1001 | /* | |
1002 | * If we issued read requests, let them complete. | |
1003 | */ | |
1004 | while (wait_bh > wait) { | |
1005 | wait_on_buffer(*--wait_bh); | |
1006 | if (!buffer_uptodate(*wait_bh)) | |
1007 | err = -EIO; | |
1008 | } | |
1009 | if (unlikely(err)) | |
1010 | page_zero_new_buffers(page, from, to); | |
1011 | else if (decrypt) | |
3684de8c | 1012 | err = ext4_decrypt(page); |
2058f83a MH |
1013 | return err; |
1014 | } | |
1015 | #endif | |
1016 | ||
bfc1af65 | 1017 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1018 | loff_t pos, unsigned len, unsigned flags, |
1019 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1020 | { |
af5bc92d | 1021 | struct inode *inode = mapping->host; |
1938a150 | 1022 | int ret, needed_blocks; |
ac27a0ec DK |
1023 | handle_t *handle; |
1024 | int retries = 0; | |
af5bc92d | 1025 | struct page *page; |
de9a55b8 | 1026 | pgoff_t index; |
af5bc92d | 1027 | unsigned from, to; |
bfc1af65 | 1028 | |
9bffad1e | 1029 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1030 | /* |
1031 | * Reserve one block more for addition to orphan list in case | |
1032 | * we allocate blocks but write fails for some reason | |
1033 | */ | |
1034 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1035 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1036 | from = pos & (PAGE_CACHE_SIZE - 1); |
1037 | to = from + len; | |
ac27a0ec | 1038 | |
f19d5870 TM |
1039 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1040 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1041 | flags, pagep); | |
1042 | if (ret < 0) | |
47564bfb TT |
1043 | return ret; |
1044 | if (ret == 1) | |
1045 | return 0; | |
f19d5870 TM |
1046 | } |
1047 | ||
47564bfb TT |
1048 | /* |
1049 | * grab_cache_page_write_begin() can take a long time if the | |
1050 | * system is thrashing due to memory pressure, or if the page | |
1051 | * is being written back. So grab it first before we start | |
1052 | * the transaction handle. This also allows us to allocate | |
1053 | * the page (if needed) without using GFP_NOFS. | |
1054 | */ | |
1055 | retry_grab: | |
1056 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1057 | if (!page) | |
1058 | return -ENOMEM; | |
1059 | unlock_page(page); | |
1060 | ||
1061 | retry_journal: | |
9924a92a | 1062 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1063 | if (IS_ERR(handle)) { |
47564bfb TT |
1064 | page_cache_release(page); |
1065 | return PTR_ERR(handle); | |
7479d2b9 | 1066 | } |
ac27a0ec | 1067 | |
47564bfb TT |
1068 | lock_page(page); |
1069 | if (page->mapping != mapping) { | |
1070 | /* The page got truncated from under us */ | |
1071 | unlock_page(page); | |
1072 | page_cache_release(page); | |
cf108bca | 1073 | ext4_journal_stop(handle); |
47564bfb | 1074 | goto retry_grab; |
cf108bca | 1075 | } |
7afe5aa5 DM |
1076 | /* In case writeback began while the page was unlocked */ |
1077 | wait_for_stable_page(page); | |
cf108bca | 1078 | |
2058f83a MH |
1079 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1080 | if (ext4_should_dioread_nolock(inode)) | |
1081 | ret = ext4_block_write_begin(page, pos, len, | |
1082 | ext4_get_block_write); | |
1083 | else | |
1084 | ret = ext4_block_write_begin(page, pos, len, | |
1085 | ext4_get_block); | |
1086 | #else | |
744692dc | 1087 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1088 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1089 | else |
6e1db88d | 1090 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1091 | #endif |
bfc1af65 | 1092 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1093 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1094 | from, to, NULL, | |
1095 | do_journal_get_write_access); | |
ac27a0ec | 1096 | } |
bfc1af65 NP |
1097 | |
1098 | if (ret) { | |
af5bc92d | 1099 | unlock_page(page); |
ae4d5372 | 1100 | /* |
6e1db88d | 1101 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1102 | * outside i_size. Trim these off again. Don't need |
1103 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1104 | * |
1105 | * Add inode to orphan list in case we crash before | |
1106 | * truncate finishes | |
ae4d5372 | 1107 | */ |
ffacfa7a | 1108 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1109 | ext4_orphan_add(handle, inode); |
1110 | ||
1111 | ext4_journal_stop(handle); | |
1112 | if (pos + len > inode->i_size) { | |
b9a4207d | 1113 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1114 | /* |
ffacfa7a | 1115 | * If truncate failed early the inode might |
1938a150 AK |
1116 | * still be on the orphan list; we need to |
1117 | * make sure the inode is removed from the | |
1118 | * orphan list in that case. | |
1119 | */ | |
1120 | if (inode->i_nlink) | |
1121 | ext4_orphan_del(NULL, inode); | |
1122 | } | |
bfc1af65 | 1123 | |
47564bfb TT |
1124 | if (ret == -ENOSPC && |
1125 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1126 | goto retry_journal; | |
1127 | page_cache_release(page); | |
1128 | return ret; | |
1129 | } | |
1130 | *pagep = page; | |
ac27a0ec DK |
1131 | return ret; |
1132 | } | |
1133 | ||
bfc1af65 NP |
1134 | /* For write_end() in data=journal mode */ |
1135 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1136 | { |
13fca323 | 1137 | int ret; |
ac27a0ec DK |
1138 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1139 | return 0; | |
1140 | set_buffer_uptodate(bh); | |
13fca323 TT |
1141 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1142 | clear_buffer_meta(bh); | |
1143 | clear_buffer_prio(bh); | |
1144 | return ret; | |
ac27a0ec DK |
1145 | } |
1146 | ||
eed4333f ZL |
1147 | /* |
1148 | * We need to pick up the new inode size which generic_commit_write gave us | |
1149 | * `file' can be NULL - eg, when called from page_symlink(). | |
1150 | * | |
1151 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1152 | * buffers are managed internally. | |
1153 | */ | |
1154 | static int ext4_write_end(struct file *file, | |
1155 | struct address_space *mapping, | |
1156 | loff_t pos, unsigned len, unsigned copied, | |
1157 | struct page *page, void *fsdata) | |
f8514083 | 1158 | { |
f8514083 | 1159 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1160 | struct inode *inode = mapping->host; |
0572639f | 1161 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1162 | int ret = 0, ret2; |
1163 | int i_size_changed = 0; | |
1164 | ||
1165 | trace_ext4_write_end(inode, pos, len, copied); | |
1166 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1167 | ret = ext4_jbd2_file_inode(handle, inode); | |
1168 | if (ret) { | |
1169 | unlock_page(page); | |
1170 | page_cache_release(page); | |
1171 | goto errout; | |
1172 | } | |
1173 | } | |
f8514083 | 1174 | |
42c832de TT |
1175 | if (ext4_has_inline_data(inode)) { |
1176 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1177 | copied, page); | |
1178 | if (ret < 0) | |
1179 | goto errout; | |
1180 | copied = ret; | |
1181 | } else | |
f19d5870 TM |
1182 | copied = block_write_end(file, mapping, pos, |
1183 | len, copied, page, fsdata); | |
f8514083 | 1184 | /* |
4631dbf6 | 1185 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1186 | * page writeout could otherwise come in and zero beyond i_size. |
1187 | */ | |
4631dbf6 | 1188 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 AK |
1189 | unlock_page(page); |
1190 | page_cache_release(page); | |
1191 | ||
0572639f XW |
1192 | if (old_size < pos) |
1193 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1194 | /* |
1195 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1196 | * makes the holding time of page lock longer. Second, it forces lock | |
1197 | * ordering of page lock and transaction start for journaling | |
1198 | * filesystems. | |
1199 | */ | |
1200 | if (i_size_changed) | |
1201 | ext4_mark_inode_dirty(handle, inode); | |
1202 | ||
ffacfa7a | 1203 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1204 | /* if we have allocated more blocks and copied |
1205 | * less. We will have blocks allocated outside | |
1206 | * inode->i_size. So truncate them | |
1207 | */ | |
1208 | ext4_orphan_add(handle, inode); | |
74d553aa | 1209 | errout: |
617ba13b | 1210 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1211 | if (!ret) |
1212 | ret = ret2; | |
bfc1af65 | 1213 | |
f8514083 | 1214 | if (pos + len > inode->i_size) { |
b9a4207d | 1215 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1216 | /* |
ffacfa7a | 1217 | * If truncate failed early the inode might still be |
f8514083 AK |
1218 | * on the orphan list; we need to make sure the inode |
1219 | * is removed from the orphan list in that case. | |
1220 | */ | |
1221 | if (inode->i_nlink) | |
1222 | ext4_orphan_del(NULL, inode); | |
1223 | } | |
1224 | ||
bfc1af65 | 1225 | return ret ? ret : copied; |
ac27a0ec DK |
1226 | } |
1227 | ||
b90197b6 TT |
1228 | /* |
1229 | * This is a private version of page_zero_new_buffers() which doesn't | |
1230 | * set the buffer to be dirty, since in data=journalled mode we need | |
1231 | * to call ext4_handle_dirty_metadata() instead. | |
1232 | */ | |
1233 | static void zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1234 | { | |
1235 | unsigned int block_start = 0, block_end; | |
1236 | struct buffer_head *head, *bh; | |
1237 | ||
1238 | bh = head = page_buffers(page); | |
1239 | do { | |
1240 | block_end = block_start + bh->b_size; | |
1241 | if (buffer_new(bh)) { | |
1242 | if (block_end > from && block_start < to) { | |
1243 | if (!PageUptodate(page)) { | |
1244 | unsigned start, size; | |
1245 | ||
1246 | start = max(from, block_start); | |
1247 | size = min(to, block_end) - start; | |
1248 | ||
1249 | zero_user(page, start, size); | |
1250 | set_buffer_uptodate(bh); | |
1251 | } | |
1252 | clear_buffer_new(bh); | |
1253 | } | |
1254 | } | |
1255 | block_start = block_end; | |
1256 | bh = bh->b_this_page; | |
1257 | } while (bh != head); | |
1258 | } | |
1259 | ||
bfc1af65 | 1260 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1261 | struct address_space *mapping, |
1262 | loff_t pos, unsigned len, unsigned copied, | |
1263 | struct page *page, void *fsdata) | |
ac27a0ec | 1264 | { |
617ba13b | 1265 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1266 | struct inode *inode = mapping->host; |
0572639f | 1267 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1268 | int ret = 0, ret2; |
1269 | int partial = 0; | |
bfc1af65 | 1270 | unsigned from, to; |
4631dbf6 | 1271 | int size_changed = 0; |
ac27a0ec | 1272 | |
9bffad1e | 1273 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1274 | from = pos & (PAGE_CACHE_SIZE - 1); |
1275 | to = from + len; | |
1276 | ||
441c8508 CW |
1277 | BUG_ON(!ext4_handle_valid(handle)); |
1278 | ||
3fdcfb66 TM |
1279 | if (ext4_has_inline_data(inode)) |
1280 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1281 | copied, page); | |
1282 | else { | |
1283 | if (copied < len) { | |
1284 | if (!PageUptodate(page)) | |
1285 | copied = 0; | |
b90197b6 | 1286 | zero_new_buffers(page, from+copied, to); |
3fdcfb66 | 1287 | } |
ac27a0ec | 1288 | |
3fdcfb66 TM |
1289 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1290 | to, &partial, write_end_fn); | |
1291 | if (!partial) | |
1292 | SetPageUptodate(page); | |
1293 | } | |
4631dbf6 | 1294 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1295 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1296 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 DM |
1297 | unlock_page(page); |
1298 | page_cache_release(page); | |
1299 | ||
0572639f XW |
1300 | if (old_size < pos) |
1301 | pagecache_isize_extended(inode, old_size, pos); | |
1302 | ||
4631dbf6 | 1303 | if (size_changed) { |
617ba13b | 1304 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1305 | if (!ret) |
1306 | ret = ret2; | |
1307 | } | |
bfc1af65 | 1308 | |
ffacfa7a | 1309 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1310 | /* if we have allocated more blocks and copied |
1311 | * less. We will have blocks allocated outside | |
1312 | * inode->i_size. So truncate them | |
1313 | */ | |
1314 | ext4_orphan_add(handle, inode); | |
1315 | ||
617ba13b | 1316 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1317 | if (!ret) |
1318 | ret = ret2; | |
f8514083 | 1319 | if (pos + len > inode->i_size) { |
b9a4207d | 1320 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1321 | /* |
ffacfa7a | 1322 | * If truncate failed early the inode might still be |
f8514083 AK |
1323 | * on the orphan list; we need to make sure the inode |
1324 | * is removed from the orphan list in that case. | |
1325 | */ | |
1326 | if (inode->i_nlink) | |
1327 | ext4_orphan_del(NULL, inode); | |
1328 | } | |
bfc1af65 NP |
1329 | |
1330 | return ret ? ret : copied; | |
ac27a0ec | 1331 | } |
d2a17637 | 1332 | |
9d0be502 | 1333 | /* |
c27e43a1 | 1334 | * Reserve space for a single cluster |
9d0be502 | 1335 | */ |
c27e43a1 | 1336 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1337 | { |
60e58e0f | 1338 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1339 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1340 | int ret; |
03179fe9 TT |
1341 | |
1342 | /* | |
1343 | * We will charge metadata quota at writeout time; this saves | |
1344 | * us from metadata over-estimation, though we may go over by | |
1345 | * a small amount in the end. Here we just reserve for data. | |
1346 | */ | |
1347 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1348 | if (ret) | |
1349 | return ret; | |
d2a17637 | 1350 | |
0637c6f4 | 1351 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1352 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1353 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1354 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1355 | return -ENOSPC; |
1356 | } | |
9d0be502 | 1357 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1358 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1359 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1360 | |
d2a17637 MC |
1361 | return 0; /* success */ |
1362 | } | |
1363 | ||
12219aea | 1364 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1365 | { |
1366 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1367 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1368 | |
cd213226 MC |
1369 | if (!to_free) |
1370 | return; /* Nothing to release, exit */ | |
1371 | ||
d2a17637 | 1372 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1373 | |
5a58ec87 | 1374 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1375 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1376 | /* |
0637c6f4 TT |
1377 | * if there aren't enough reserved blocks, then the |
1378 | * counter is messed up somewhere. Since this | |
1379 | * function is called from invalidate page, it's | |
1380 | * harmless to return without any action. | |
cd213226 | 1381 | */ |
8de5c325 | 1382 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1383 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1384 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1385 | ei->i_reserved_data_blocks); |
1386 | WARN_ON(1); | |
1387 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1388 | } |
0637c6f4 | 1389 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1390 | |
72b8ab9d | 1391 | /* update fs dirty data blocks counter */ |
57042651 | 1392 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1393 | |
d2a17637 | 1394 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1395 | |
7b415bf6 | 1396 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1397 | } |
1398 | ||
1399 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1400 | unsigned int offset, |
1401 | unsigned int length) | |
d2a17637 | 1402 | { |
9705acd6 | 1403 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1404 | struct buffer_head *head, *bh; |
1405 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1406 | struct inode *inode = page->mapping->host; |
1407 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1408 | unsigned int stop = offset + length; |
7b415bf6 | 1409 | int num_clusters; |
51865fda | 1410 | ext4_fsblk_t lblk; |
d2a17637 | 1411 | |
ca99fdd2 LC |
1412 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
1413 | ||
d2a17637 MC |
1414 | head = page_buffers(page); |
1415 | bh = head; | |
1416 | do { | |
1417 | unsigned int next_off = curr_off + bh->b_size; | |
1418 | ||
ca99fdd2 LC |
1419 | if (next_off > stop) |
1420 | break; | |
1421 | ||
d2a17637 MC |
1422 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1423 | to_release++; | |
9705acd6 | 1424 | contiguous_blks++; |
d2a17637 | 1425 | clear_buffer_delay(bh); |
9705acd6 LC |
1426 | } else if (contiguous_blks) { |
1427 | lblk = page->index << | |
1428 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1429 | lblk += (curr_off >> inode->i_blkbits) - | |
1430 | contiguous_blks; | |
1431 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1432 | contiguous_blks = 0; | |
d2a17637 MC |
1433 | } |
1434 | curr_off = next_off; | |
1435 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1436 | |
9705acd6 | 1437 | if (contiguous_blks) { |
51865fda | 1438 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1439 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1440 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1441 | } |
1442 | ||
7b415bf6 AK |
1443 | /* If we have released all the blocks belonging to a cluster, then we |
1444 | * need to release the reserved space for that cluster. */ | |
1445 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1446 | while (num_clusters > 0) { | |
7b415bf6 AK |
1447 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1448 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1449 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1450 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1451 | ext4_da_release_space(inode, 1); |
1452 | ||
1453 | num_clusters--; | |
1454 | } | |
d2a17637 | 1455 | } |
ac27a0ec | 1456 | |
64769240 AT |
1457 | /* |
1458 | * Delayed allocation stuff | |
1459 | */ | |
1460 | ||
4e7ea81d JK |
1461 | struct mpage_da_data { |
1462 | struct inode *inode; | |
1463 | struct writeback_control *wbc; | |
6b523df4 | 1464 | |
4e7ea81d JK |
1465 | pgoff_t first_page; /* The first page to write */ |
1466 | pgoff_t next_page; /* Current page to examine */ | |
1467 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1468 | /* |
4e7ea81d JK |
1469 | * Extent to map - this can be after first_page because that can be |
1470 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1471 | * is delalloc or unwritten. | |
791b7f08 | 1472 | */ |
4e7ea81d JK |
1473 | struct ext4_map_blocks map; |
1474 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1475 | }; | |
64769240 | 1476 | |
4e7ea81d JK |
1477 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1478 | bool invalidate) | |
c4a0c46e AK |
1479 | { |
1480 | int nr_pages, i; | |
1481 | pgoff_t index, end; | |
1482 | struct pagevec pvec; | |
1483 | struct inode *inode = mpd->inode; | |
1484 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1485 | |
1486 | /* This is necessary when next_page == 0. */ | |
1487 | if (mpd->first_page >= mpd->next_page) | |
1488 | return; | |
c4a0c46e | 1489 | |
c7f5938a CW |
1490 | index = mpd->first_page; |
1491 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1492 | if (invalidate) { |
1493 | ext4_lblk_t start, last; | |
1494 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1495 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1496 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1497 | } | |
51865fda | 1498 | |
66bea92c | 1499 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1500 | while (index <= end) { |
1501 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1502 | if (nr_pages == 0) | |
1503 | break; | |
1504 | for (i = 0; i < nr_pages; i++) { | |
1505 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1506 | if (page->index > end) |
c4a0c46e | 1507 | break; |
c4a0c46e AK |
1508 | BUG_ON(!PageLocked(page)); |
1509 | BUG_ON(PageWriteback(page)); | |
4e7ea81d JK |
1510 | if (invalidate) { |
1511 | block_invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
1512 | ClearPageUptodate(page); | |
1513 | } | |
c4a0c46e AK |
1514 | unlock_page(page); |
1515 | } | |
9b1d0998 JK |
1516 | index = pvec.pages[nr_pages - 1]->index + 1; |
1517 | pagevec_release(&pvec); | |
c4a0c46e | 1518 | } |
c4a0c46e AK |
1519 | } |
1520 | ||
df22291f AK |
1521 | static void ext4_print_free_blocks(struct inode *inode) |
1522 | { | |
1523 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1524 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1525 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1526 | |
1527 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1528 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1529 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1530 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1531 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1532 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1533 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1534 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1535 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1536 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1537 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1538 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1539 | ei->i_reserved_data_blocks); |
df22291f AK |
1540 | return; |
1541 | } | |
1542 | ||
c364b22c | 1543 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1544 | { |
c364b22c | 1545 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1546 | } |
1547 | ||
5356f261 AK |
1548 | /* |
1549 | * This function is grabs code from the very beginning of | |
1550 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1551 | * time. This function looks up the requested blocks and sets the | |
1552 | * buffer delay bit under the protection of i_data_sem. | |
1553 | */ | |
1554 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1555 | struct ext4_map_blocks *map, | |
1556 | struct buffer_head *bh) | |
1557 | { | |
d100eef2 | 1558 | struct extent_status es; |
5356f261 AK |
1559 | int retval; |
1560 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1561 | #ifdef ES_AGGRESSIVE_TEST |
1562 | struct ext4_map_blocks orig_map; | |
1563 | ||
1564 | memcpy(&orig_map, map, sizeof(*map)); | |
1565 | #endif | |
5356f261 AK |
1566 | |
1567 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1568 | invalid_block = ~0; | |
1569 | ||
1570 | map->m_flags = 0; | |
1571 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1572 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1573 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1574 | |
1575 | /* Lookup extent status tree firstly */ | |
1576 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1577 | if (ext4_es_is_hole(&es)) { |
1578 | retval = 0; | |
c8b459f4 | 1579 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1580 | goto add_delayed; |
1581 | } | |
1582 | ||
1583 | /* | |
1584 | * Delayed extent could be allocated by fallocate. | |
1585 | * So we need to check it. | |
1586 | */ | |
1587 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1588 | map_bh(bh, inode->i_sb, invalid_block); | |
1589 | set_buffer_new(bh); | |
1590 | set_buffer_delay(bh); | |
1591 | return 0; | |
1592 | } | |
1593 | ||
1594 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1595 | retval = es.es_len - (iblock - es.es_lblk); | |
1596 | if (retval > map->m_len) | |
1597 | retval = map->m_len; | |
1598 | map->m_len = retval; | |
1599 | if (ext4_es_is_written(&es)) | |
1600 | map->m_flags |= EXT4_MAP_MAPPED; | |
1601 | else if (ext4_es_is_unwritten(&es)) | |
1602 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1603 | else | |
1604 | BUG_ON(1); | |
1605 | ||
921f266b DM |
1606 | #ifdef ES_AGGRESSIVE_TEST |
1607 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1608 | #endif | |
d100eef2 ZL |
1609 | return retval; |
1610 | } | |
1611 | ||
5356f261 AK |
1612 | /* |
1613 | * Try to see if we can get the block without requesting a new | |
1614 | * file system block. | |
1615 | */ | |
c8b459f4 | 1616 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1617 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1618 | retval = 0; |
cbd7584e | 1619 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1620 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1621 | else |
2f8e0a7c | 1622 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1623 | |
d100eef2 | 1624 | add_delayed: |
5356f261 | 1625 | if (retval == 0) { |
f7fec032 | 1626 | int ret; |
5356f261 AK |
1627 | /* |
1628 | * XXX: __block_prepare_write() unmaps passed block, | |
1629 | * is it OK? | |
1630 | */ | |
386ad67c LC |
1631 | /* |
1632 | * If the block was allocated from previously allocated cluster, | |
1633 | * then we don't need to reserve it again. However we still need | |
1634 | * to reserve metadata for every block we're going to write. | |
1635 | */ | |
c27e43a1 | 1636 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1637 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1638 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1639 | if (ret) { |
5356f261 | 1640 | /* not enough space to reserve */ |
f7fec032 | 1641 | retval = ret; |
5356f261 | 1642 | goto out_unlock; |
f7fec032 | 1643 | } |
5356f261 AK |
1644 | } |
1645 | ||
f7fec032 ZL |
1646 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1647 | ~0, EXTENT_STATUS_DELAYED); | |
1648 | if (ret) { | |
1649 | retval = ret; | |
51865fda | 1650 | goto out_unlock; |
f7fec032 | 1651 | } |
51865fda | 1652 | |
5356f261 AK |
1653 | map_bh(bh, inode->i_sb, invalid_block); |
1654 | set_buffer_new(bh); | |
1655 | set_buffer_delay(bh); | |
f7fec032 ZL |
1656 | } else if (retval > 0) { |
1657 | int ret; | |
3be78c73 | 1658 | unsigned int status; |
f7fec032 | 1659 | |
44fb851d ZL |
1660 | if (unlikely(retval != map->m_len)) { |
1661 | ext4_warning(inode->i_sb, | |
1662 | "ES len assertion failed for inode " | |
1663 | "%lu: retval %d != map->m_len %d", | |
1664 | inode->i_ino, retval, map->m_len); | |
1665 | WARN_ON(1); | |
921f266b | 1666 | } |
921f266b | 1667 | |
f7fec032 ZL |
1668 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1669 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1670 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1671 | map->m_pblk, status); | |
1672 | if (ret != 0) | |
1673 | retval = ret; | |
5356f261 AK |
1674 | } |
1675 | ||
1676 | out_unlock: | |
1677 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1678 | ||
1679 | return retval; | |
1680 | } | |
1681 | ||
64769240 | 1682 | /* |
d91bd2c1 | 1683 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1684 | * ext4_da_write_begin(). It will either return mapped block or |
1685 | * reserve space for a single block. | |
29fa89d0 AK |
1686 | * |
1687 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1688 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1689 | * | |
1690 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1691 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1692 | * initialized properly. | |
64769240 | 1693 | */ |
9c3569b5 TM |
1694 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1695 | struct buffer_head *bh, int create) | |
64769240 | 1696 | { |
2ed88685 | 1697 | struct ext4_map_blocks map; |
64769240 AT |
1698 | int ret = 0; |
1699 | ||
1700 | BUG_ON(create == 0); | |
2ed88685 TT |
1701 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1702 | ||
1703 | map.m_lblk = iblock; | |
1704 | map.m_len = 1; | |
64769240 AT |
1705 | |
1706 | /* | |
1707 | * first, we need to know whether the block is allocated already | |
1708 | * preallocated blocks are unmapped but should treated | |
1709 | * the same as allocated blocks. | |
1710 | */ | |
5356f261 AK |
1711 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1712 | if (ret <= 0) | |
2ed88685 | 1713 | return ret; |
64769240 | 1714 | |
2ed88685 | 1715 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1716 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1717 | |
1718 | if (buffer_unwritten(bh)) { | |
1719 | /* A delayed write to unwritten bh should be marked | |
1720 | * new and mapped. Mapped ensures that we don't do | |
1721 | * get_block multiple times when we write to the same | |
1722 | * offset and new ensures that we do proper zero out | |
1723 | * for partial write. | |
1724 | */ | |
1725 | set_buffer_new(bh); | |
c8205636 | 1726 | set_buffer_mapped(bh); |
2ed88685 TT |
1727 | } |
1728 | return 0; | |
64769240 | 1729 | } |
61628a3f | 1730 | |
62e086be AK |
1731 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1732 | { | |
1733 | get_bh(bh); | |
1734 | return 0; | |
1735 | } | |
1736 | ||
1737 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1738 | { | |
1739 | put_bh(bh); | |
1740 | return 0; | |
1741 | } | |
1742 | ||
1743 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1744 | unsigned int len) |
1745 | { | |
1746 | struct address_space *mapping = page->mapping; | |
1747 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1748 | struct buffer_head *page_bufs = NULL; |
62e086be | 1749 | handle_t *handle = NULL; |
3fdcfb66 TM |
1750 | int ret = 0, err = 0; |
1751 | int inline_data = ext4_has_inline_data(inode); | |
1752 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1753 | |
cb20d518 | 1754 | ClearPageChecked(page); |
3fdcfb66 TM |
1755 | |
1756 | if (inline_data) { | |
1757 | BUG_ON(page->index != 0); | |
1758 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1759 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1760 | if (inode_bh == NULL) | |
1761 | goto out; | |
1762 | } else { | |
1763 | page_bufs = page_buffers(page); | |
1764 | if (!page_bufs) { | |
1765 | BUG(); | |
1766 | goto out; | |
1767 | } | |
1768 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1769 | NULL, bget_one); | |
1770 | } | |
bdf96838 TT |
1771 | /* |
1772 | * We need to release the page lock before we start the | |
1773 | * journal, so grab a reference so the page won't disappear | |
1774 | * out from under us. | |
1775 | */ | |
1776 | get_page(page); | |
62e086be AK |
1777 | unlock_page(page); |
1778 | ||
9924a92a TT |
1779 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1780 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1781 | if (IS_ERR(handle)) { |
1782 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1783 | put_page(page); |
1784 | goto out_no_pagelock; | |
62e086be | 1785 | } |
441c8508 CW |
1786 | BUG_ON(!ext4_handle_valid(handle)); |
1787 | ||
bdf96838 TT |
1788 | lock_page(page); |
1789 | put_page(page); | |
1790 | if (page->mapping != mapping) { | |
1791 | /* The page got truncated from under us */ | |
1792 | ext4_journal_stop(handle); | |
1793 | ret = 0; | |
1794 | goto out; | |
1795 | } | |
1796 | ||
3fdcfb66 | 1797 | if (inline_data) { |
5d601255 | 1798 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1799 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1800 | |
3fdcfb66 TM |
1801 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1802 | ||
1803 | } else { | |
1804 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1805 | do_journal_get_write_access); | |
1806 | ||
1807 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1808 | write_end_fn); | |
1809 | } | |
62e086be AK |
1810 | if (ret == 0) |
1811 | ret = err; | |
2d859db3 | 1812 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1813 | err = ext4_journal_stop(handle); |
1814 | if (!ret) | |
1815 | ret = err; | |
1816 | ||
3fdcfb66 | 1817 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1818 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1819 | NULL, bput_one); |
19f5fb7a | 1820 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1821 | out: |
bdf96838 TT |
1822 | unlock_page(page); |
1823 | out_no_pagelock: | |
3fdcfb66 | 1824 | brelse(inode_bh); |
62e086be AK |
1825 | return ret; |
1826 | } | |
1827 | ||
61628a3f | 1828 | /* |
43ce1d23 AK |
1829 | * Note that we don't need to start a transaction unless we're journaling data |
1830 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1831 | * need to file the inode to the transaction's list in ordered mode because if | |
1832 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1833 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1834 | * transaction the data will hit the disk. In case we are journaling data, we |
1835 | * cannot start transaction directly because transaction start ranks above page | |
1836 | * lock so we have to do some magic. | |
1837 | * | |
b920c755 | 1838 | * This function can get called via... |
20970ba6 | 1839 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1840 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1841 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1842 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1843 | * |
1844 | * We don't do any block allocation in this function. If we have page with | |
1845 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1846 | * is important for mmaped based write. So if we do with blocksize 1K | |
1847 | * truncate(f, 1024); | |
1848 | * a = mmap(f, 0, 4096); | |
1849 | * a[0] = 'a'; | |
1850 | * truncate(f, 4096); | |
1851 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1852 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1853 | * do_wp_page). So writepage should write the first block. If we modify |
1854 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1855 | * page_mkwrite callback will do the block allocation and mark the | |
1856 | * buffer_heads mapped. | |
1857 | * | |
1858 | * We redirty the page if we have any buffer_heads that is either delay or | |
1859 | * unwritten in the page. | |
1860 | * | |
1861 | * We can get recursively called as show below. | |
1862 | * | |
1863 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1864 | * ext4_writepage() | |
1865 | * | |
1866 | * But since we don't do any block allocation we should not deadlock. | |
1867 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1868 | */ |
43ce1d23 | 1869 | static int ext4_writepage(struct page *page, |
62e086be | 1870 | struct writeback_control *wbc) |
64769240 | 1871 | { |
f8bec370 | 1872 | int ret = 0; |
61628a3f | 1873 | loff_t size; |
498e5f24 | 1874 | unsigned int len; |
744692dc | 1875 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1876 | struct inode *inode = page->mapping->host; |
36ade451 | 1877 | struct ext4_io_submit io_submit; |
1c8349a1 | 1878 | bool keep_towrite = false; |
61628a3f | 1879 | |
a9c667f8 | 1880 | trace_ext4_writepage(page); |
f0e6c985 AK |
1881 | size = i_size_read(inode); |
1882 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1883 | len = size & ~PAGE_CACHE_MASK; | |
1884 | else | |
1885 | len = PAGE_CACHE_SIZE; | |
64769240 | 1886 | |
a42afc5f | 1887 | page_bufs = page_buffers(page); |
a42afc5f | 1888 | /* |
fe386132 JK |
1889 | * We cannot do block allocation or other extent handling in this |
1890 | * function. If there are buffers needing that, we have to redirty | |
1891 | * the page. But we may reach here when we do a journal commit via | |
1892 | * journal_submit_inode_data_buffers() and in that case we must write | |
1893 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
1894 | * |
1895 | * Also, if there is only one buffer per page (the fs block | |
1896 | * size == the page size), if one buffer needs block | |
1897 | * allocation or needs to modify the extent tree to clear the | |
1898 | * unwritten flag, we know that the page can't be written at | |
1899 | * all, so we might as well refuse the write immediately. | |
1900 | * Unfortunately if the block size != page size, we can't as | |
1901 | * easily detect this case using ext4_walk_page_buffers(), but | |
1902 | * for the extremely common case, this is an optimization that | |
1903 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 1904 | */ |
f19d5870 TM |
1905 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
1906 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 1907 | redirty_page_for_writepage(wbc, page); |
cccd147a TT |
1908 | if ((current->flags & PF_MEMALLOC) || |
1909 | (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) { | |
fe386132 JK |
1910 | /* |
1911 | * For memory cleaning there's no point in writing only | |
1912 | * some buffers. So just bail out. Warn if we came here | |
1913 | * from direct reclaim. | |
1914 | */ | |
1915 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
1916 | == PF_MEMALLOC); | |
f0e6c985 AK |
1917 | unlock_page(page); |
1918 | return 0; | |
1919 | } | |
1c8349a1 | 1920 | keep_towrite = true; |
a42afc5f | 1921 | } |
64769240 | 1922 | |
cb20d518 | 1923 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1924 | /* |
1925 | * It's mmapped pagecache. Add buffers and journal it. There | |
1926 | * doesn't seem much point in redirtying the page here. | |
1927 | */ | |
3f0ca309 | 1928 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1929 | |
97a851ed JK |
1930 | ext4_io_submit_init(&io_submit, wbc); |
1931 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
1932 | if (!io_submit.io_end) { | |
1933 | redirty_page_for_writepage(wbc, page); | |
1934 | unlock_page(page); | |
1935 | return -ENOMEM; | |
1936 | } | |
1c8349a1 | 1937 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 1938 | ext4_io_submit(&io_submit); |
97a851ed JK |
1939 | /* Drop io_end reference we got from init */ |
1940 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
1941 | return ret; |
1942 | } | |
1943 | ||
5f1132b2 JK |
1944 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
1945 | { | |
1946 | int len; | |
1947 | loff_t size = i_size_read(mpd->inode); | |
1948 | int err; | |
1949 | ||
1950 | BUG_ON(page->index != mpd->first_page); | |
1951 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1952 | len = size & ~PAGE_CACHE_MASK; | |
1953 | else | |
1954 | len = PAGE_CACHE_SIZE; | |
1955 | clear_page_dirty_for_io(page); | |
1c8349a1 | 1956 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
1957 | if (!err) |
1958 | mpd->wbc->nr_to_write--; | |
1959 | mpd->first_page++; | |
1960 | ||
1961 | return err; | |
1962 | } | |
1963 | ||
4e7ea81d JK |
1964 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
1965 | ||
61628a3f | 1966 | /* |
fffb2739 JK |
1967 | * mballoc gives us at most this number of blocks... |
1968 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 1969 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 1970 | */ |
fffb2739 | 1971 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 1972 | |
4e7ea81d JK |
1973 | /* |
1974 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
1975 | * | |
1976 | * @mpd - extent of blocks | |
1977 | * @lblk - logical number of the block in the file | |
09930042 | 1978 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 1979 | * |
09930042 JK |
1980 | * The function is used to collect contig. blocks in the same state. If the |
1981 | * buffer doesn't require mapping for writeback and we haven't started the | |
1982 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
1983 | * caller can write the buffer right away. Otherwise the function returns true | |
1984 | * if the block has been added to the extent, false if the block couldn't be | |
1985 | * added. | |
4e7ea81d | 1986 | */ |
09930042 JK |
1987 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
1988 | struct buffer_head *bh) | |
4e7ea81d JK |
1989 | { |
1990 | struct ext4_map_blocks *map = &mpd->map; | |
1991 | ||
09930042 JK |
1992 | /* Buffer that doesn't need mapping for writeback? */ |
1993 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
1994 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
1995 | /* So far no extent to map => we write the buffer right away */ | |
1996 | if (map->m_len == 0) | |
1997 | return true; | |
1998 | return false; | |
1999 | } | |
4e7ea81d JK |
2000 | |
2001 | /* First block in the extent? */ | |
2002 | if (map->m_len == 0) { | |
2003 | map->m_lblk = lblk; | |
2004 | map->m_len = 1; | |
09930042 JK |
2005 | map->m_flags = bh->b_state & BH_FLAGS; |
2006 | return true; | |
4e7ea81d JK |
2007 | } |
2008 | ||
09930042 JK |
2009 | /* Don't go larger than mballoc is willing to allocate */ |
2010 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2011 | return false; | |
2012 | ||
4e7ea81d JK |
2013 | /* Can we merge the block to our big extent? */ |
2014 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2015 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2016 | map->m_len++; |
09930042 | 2017 | return true; |
4e7ea81d | 2018 | } |
09930042 | 2019 | return false; |
4e7ea81d JK |
2020 | } |
2021 | ||
5f1132b2 JK |
2022 | /* |
2023 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2024 | * | |
2025 | * @mpd - extent of blocks for mapping | |
2026 | * @head - the first buffer in the page | |
2027 | * @bh - buffer we should start processing from | |
2028 | * @lblk - logical number of the block in the file corresponding to @bh | |
2029 | * | |
2030 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2031 | * the page for IO if all buffers in this page were mapped and there's no | |
2032 | * accumulated extent of buffers to map or add buffers in the page to the | |
2033 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2034 | * by processing the next page, 0 if it should stop adding buffers to the | |
2035 | * extent to map because we cannot extend it anymore. It can also return value | |
2036 | * < 0 in case of error during IO submission. | |
2037 | */ | |
2038 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2039 | struct buffer_head *head, | |
2040 | struct buffer_head *bh, | |
2041 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2042 | { |
2043 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2044 | int err; |
4e7ea81d JK |
2045 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
2046 | >> inode->i_blkbits; | |
2047 | ||
2048 | do { | |
2049 | BUG_ON(buffer_locked(bh)); | |
2050 | ||
09930042 | 2051 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2052 | /* Found extent to map? */ |
2053 | if (mpd->map.m_len) | |
5f1132b2 | 2054 | return 0; |
09930042 | 2055 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2056 | break; |
4e7ea81d | 2057 | } |
4e7ea81d | 2058 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2059 | /* So far everything mapped? Submit the page for IO. */ |
2060 | if (mpd->map.m_len == 0) { | |
2061 | err = mpage_submit_page(mpd, head->b_page); | |
2062 | if (err < 0) | |
2063 | return err; | |
2064 | } | |
2065 | return lblk < blocks; | |
4e7ea81d JK |
2066 | } |
2067 | ||
2068 | /* | |
2069 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2070 | * submit fully mapped pages for IO | |
2071 | * | |
2072 | * @mpd - description of extent to map, on return next extent to map | |
2073 | * | |
2074 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2075 | * to be already locked) and update buffer state according to new extent state. | |
2076 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2077 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2078 | * and do extent conversion after IO is finished. If the last page is not fully |
2079 | * mapped, we update @map to the next extent in the last page that needs | |
2080 | * mapping. Otherwise we submit the page for IO. | |
2081 | */ | |
2082 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2083 | { | |
2084 | struct pagevec pvec; | |
2085 | int nr_pages, i; | |
2086 | struct inode *inode = mpd->inode; | |
2087 | struct buffer_head *head, *bh; | |
2088 | int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; | |
4e7ea81d JK |
2089 | pgoff_t start, end; |
2090 | ext4_lblk_t lblk; | |
2091 | sector_t pblock; | |
2092 | int err; | |
2093 | ||
2094 | start = mpd->map.m_lblk >> bpp_bits; | |
2095 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2096 | lblk = start << bpp_bits; | |
2097 | pblock = mpd->map.m_pblk; | |
2098 | ||
2099 | pagevec_init(&pvec, 0); | |
2100 | while (start <= end) { | |
2101 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2102 | PAGEVEC_SIZE); | |
2103 | if (nr_pages == 0) | |
2104 | break; | |
2105 | for (i = 0; i < nr_pages; i++) { | |
2106 | struct page *page = pvec.pages[i]; | |
2107 | ||
2108 | if (page->index > end) | |
2109 | break; | |
70261f56 | 2110 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2111 | BUG_ON(page->index != start); |
2112 | bh = head = page_buffers(page); | |
2113 | do { | |
2114 | if (lblk < mpd->map.m_lblk) | |
2115 | continue; | |
2116 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2117 | /* | |
2118 | * Buffer after end of mapped extent. | |
2119 | * Find next buffer in the page to map. | |
2120 | */ | |
2121 | mpd->map.m_len = 0; | |
2122 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2123 | /* |
2124 | * FIXME: If dioread_nolock supports | |
2125 | * blocksize < pagesize, we need to make | |
2126 | * sure we add size mapped so far to | |
2127 | * io_end->size as the following call | |
2128 | * can submit the page for IO. | |
2129 | */ | |
2130 | err = mpage_process_page_bufs(mpd, head, | |
2131 | bh, lblk); | |
4e7ea81d | 2132 | pagevec_release(&pvec); |
5f1132b2 JK |
2133 | if (err > 0) |
2134 | err = 0; | |
2135 | return err; | |
4e7ea81d JK |
2136 | } |
2137 | if (buffer_delay(bh)) { | |
2138 | clear_buffer_delay(bh); | |
2139 | bh->b_blocknr = pblock++; | |
2140 | } | |
4e7ea81d | 2141 | clear_buffer_unwritten(bh); |
5f1132b2 | 2142 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2143 | |
2144 | /* | |
2145 | * FIXME: This is going to break if dioread_nolock | |
2146 | * supports blocksize < pagesize as we will try to | |
2147 | * convert potentially unmapped parts of inode. | |
2148 | */ | |
2149 | mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; | |
2150 | /* Page fully mapped - let IO run! */ | |
2151 | err = mpage_submit_page(mpd, page); | |
2152 | if (err < 0) { | |
2153 | pagevec_release(&pvec); | |
2154 | return err; | |
2155 | } | |
2156 | start++; | |
2157 | } | |
2158 | pagevec_release(&pvec); | |
2159 | } | |
2160 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2161 | mpd->map.m_len = 0; | |
2162 | mpd->map.m_flags = 0; | |
2163 | return 0; | |
2164 | } | |
2165 | ||
2166 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2167 | { | |
2168 | struct inode *inode = mpd->inode; | |
2169 | struct ext4_map_blocks *map = &mpd->map; | |
2170 | int get_blocks_flags; | |
090f32ee | 2171 | int err, dioread_nolock; |
4e7ea81d JK |
2172 | |
2173 | trace_ext4_da_write_pages_extent(inode, map); | |
2174 | /* | |
2175 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2176 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2177 | * where we have written into one or more preallocated blocks). It is |
2178 | * possible that we're going to need more metadata blocks than | |
2179 | * previously reserved. However we must not fail because we're in | |
2180 | * writeback and there is nothing we can do about it so it might result | |
2181 | * in data loss. So use reserved blocks to allocate metadata if | |
2182 | * possible. | |
2183 | * | |
754cfed6 TT |
2184 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2185 | * the blocks in question are delalloc blocks. This indicates | |
2186 | * that the blocks and quotas has already been checked when | |
2187 | * the data was copied into the page cache. | |
4e7ea81d JK |
2188 | */ |
2189 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
2190 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
090f32ee LC |
2191 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2192 | if (dioread_nolock) | |
4e7ea81d JK |
2193 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2194 | if (map->m_flags & (1 << BH_Delay)) | |
2195 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2196 | ||
2197 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2198 | if (err < 0) | |
2199 | return err; | |
090f32ee | 2200 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2201 | if (!mpd->io_submit.io_end->handle && |
2202 | ext4_handle_valid(handle)) { | |
2203 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2204 | handle->h_rsv_handle = NULL; | |
2205 | } | |
3613d228 | 2206 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2207 | } |
4e7ea81d JK |
2208 | |
2209 | BUG_ON(map->m_len == 0); | |
2210 | if (map->m_flags & EXT4_MAP_NEW) { | |
2211 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2212 | int i; | |
2213 | ||
2214 | for (i = 0; i < map->m_len; i++) | |
2215 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2216 | } | |
2217 | return 0; | |
2218 | } | |
2219 | ||
2220 | /* | |
2221 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2222 | * mpd->len and submit pages underlying it for IO | |
2223 | * | |
2224 | * @handle - handle for journal operations | |
2225 | * @mpd - extent to map | |
7534e854 JK |
2226 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2227 | * is no hope of writing the data. The caller should discard | |
2228 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2229 | * |
2230 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2231 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2232 | * them to initialized or split the described range from larger unwritten | |
2233 | * extent. Note that we need not map all the described range since allocation | |
2234 | * can return less blocks or the range is covered by more unwritten extents. We | |
2235 | * cannot map more because we are limited by reserved transaction credits. On | |
2236 | * the other hand we always make sure that the last touched page is fully | |
2237 | * mapped so that it can be written out (and thus forward progress is | |
2238 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2239 | */ | |
2240 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2241 | struct mpage_da_data *mpd, |
2242 | bool *give_up_on_write) | |
4e7ea81d JK |
2243 | { |
2244 | struct inode *inode = mpd->inode; | |
2245 | struct ext4_map_blocks *map = &mpd->map; | |
2246 | int err; | |
2247 | loff_t disksize; | |
6603120e | 2248 | int progress = 0; |
4e7ea81d JK |
2249 | |
2250 | mpd->io_submit.io_end->offset = | |
2251 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2252 | do { |
4e7ea81d JK |
2253 | err = mpage_map_one_extent(handle, mpd); |
2254 | if (err < 0) { | |
2255 | struct super_block *sb = inode->i_sb; | |
2256 | ||
cb530541 TT |
2257 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2258 | goto invalidate_dirty_pages; | |
4e7ea81d | 2259 | /* |
cb530541 TT |
2260 | * Let the uper layers retry transient errors. |
2261 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2262 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2263 | */ |
cb530541 | 2264 | if ((err == -ENOMEM) || |
6603120e DM |
2265 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2266 | if (progress) | |
2267 | goto update_disksize; | |
cb530541 | 2268 | return err; |
6603120e | 2269 | } |
cb530541 TT |
2270 | ext4_msg(sb, KERN_CRIT, |
2271 | "Delayed block allocation failed for " | |
2272 | "inode %lu at logical offset %llu with" | |
2273 | " max blocks %u with error %d", | |
2274 | inode->i_ino, | |
2275 | (unsigned long long)map->m_lblk, | |
2276 | (unsigned)map->m_len, -err); | |
2277 | ext4_msg(sb, KERN_CRIT, | |
2278 | "This should not happen!! Data will " | |
2279 | "be lost\n"); | |
2280 | if (err == -ENOSPC) | |
2281 | ext4_print_free_blocks(inode); | |
2282 | invalidate_dirty_pages: | |
2283 | *give_up_on_write = true; | |
4e7ea81d JK |
2284 | return err; |
2285 | } | |
6603120e | 2286 | progress = 1; |
4e7ea81d JK |
2287 | /* |
2288 | * Update buffer state, submit mapped pages, and get us new | |
2289 | * extent to map | |
2290 | */ | |
2291 | err = mpage_map_and_submit_buffers(mpd); | |
2292 | if (err < 0) | |
6603120e | 2293 | goto update_disksize; |
27d7c4ed | 2294 | } while (map->m_len); |
4e7ea81d | 2295 | |
6603120e | 2296 | update_disksize: |
622cad13 TT |
2297 | /* |
2298 | * Update on-disk size after IO is submitted. Races with | |
2299 | * truncate are avoided by checking i_size under i_data_sem. | |
2300 | */ | |
4e7ea81d | 2301 | disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; |
4e7ea81d JK |
2302 | if (disksize > EXT4_I(inode)->i_disksize) { |
2303 | int err2; | |
622cad13 TT |
2304 | loff_t i_size; |
2305 | ||
2306 | down_write(&EXT4_I(inode)->i_data_sem); | |
2307 | i_size = i_size_read(inode); | |
2308 | if (disksize > i_size) | |
2309 | disksize = i_size; | |
2310 | if (disksize > EXT4_I(inode)->i_disksize) | |
2311 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2312 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2313 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2314 | if (err2) |
2315 | ext4_error(inode->i_sb, | |
2316 | "Failed to mark inode %lu dirty", | |
2317 | inode->i_ino); | |
2318 | if (!err) | |
2319 | err = err2; | |
2320 | } | |
2321 | return err; | |
2322 | } | |
2323 | ||
fffb2739 JK |
2324 | /* |
2325 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2326 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2327 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2328 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2329 | * bpp - 1 blocks in bpp different extents. | |
2330 | */ | |
525f4ed8 MC |
2331 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2332 | { | |
fffb2739 | 2333 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2334 | |
fffb2739 JK |
2335 | return ext4_meta_trans_blocks(inode, |
2336 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2337 | } |
61628a3f | 2338 | |
8e48dcfb | 2339 | /* |
4e7ea81d JK |
2340 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2341 | * and underlying extent to map | |
2342 | * | |
2343 | * @mpd - where to look for pages | |
2344 | * | |
2345 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2346 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2347 | * extent of buffers underlying these pages that needs mapping (formed by | |
2348 | * either delayed or unwritten buffers). We also lock the pages containing | |
2349 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2350 | * mpd->lblk with length mpd->len blocks). | |
2351 | * | |
2352 | * Note that this function can attach bios to one io_end structure which are | |
2353 | * neither logically nor physically contiguous. Although it may seem as an | |
2354 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2355 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2356 | */ |
4e7ea81d | 2357 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2358 | { |
4e7ea81d JK |
2359 | struct address_space *mapping = mpd->inode->i_mapping; |
2360 | struct pagevec pvec; | |
2361 | unsigned int nr_pages; | |
aeac589a | 2362 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2363 | pgoff_t index = mpd->first_page; |
2364 | pgoff_t end = mpd->last_page; | |
2365 | int tag; | |
2366 | int i, err = 0; | |
2367 | int blkbits = mpd->inode->i_blkbits; | |
2368 | ext4_lblk_t lblk; | |
2369 | struct buffer_head *head; | |
8e48dcfb | 2370 | |
4e7ea81d | 2371 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2372 | tag = PAGECACHE_TAG_TOWRITE; |
2373 | else | |
2374 | tag = PAGECACHE_TAG_DIRTY; | |
2375 | ||
4e7ea81d JK |
2376 | pagevec_init(&pvec, 0); |
2377 | mpd->map.m_len = 0; | |
2378 | mpd->next_page = index; | |
4f01b02c | 2379 | while (index <= end) { |
5b41d924 | 2380 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2381 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2382 | if (nr_pages == 0) | |
4e7ea81d | 2383 | goto out; |
8e48dcfb TT |
2384 | |
2385 | for (i = 0; i < nr_pages; i++) { | |
2386 | struct page *page = pvec.pages[i]; | |
2387 | ||
2388 | /* | |
2389 | * At this point, the page may be truncated or | |
2390 | * invalidated (changing page->mapping to NULL), or | |
2391 | * even swizzled back from swapper_space to tmpfs file | |
2392 | * mapping. However, page->index will not change | |
2393 | * because we have a reference on the page. | |
2394 | */ | |
4f01b02c TT |
2395 | if (page->index > end) |
2396 | goto out; | |
8e48dcfb | 2397 | |
aeac589a ML |
2398 | /* |
2399 | * Accumulated enough dirty pages? This doesn't apply | |
2400 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2401 | * keep going because someone may be concurrently | |
2402 | * dirtying pages, and we might have synced a lot of | |
2403 | * newly appeared dirty pages, but have not synced all | |
2404 | * of the old dirty pages. | |
2405 | */ | |
2406 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2407 | goto out; | |
2408 | ||
4e7ea81d JK |
2409 | /* If we can't merge this page, we are done. */ |
2410 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2411 | goto out; | |
78aaced3 | 2412 | |
8e48dcfb | 2413 | lock_page(page); |
8e48dcfb | 2414 | /* |
4e7ea81d JK |
2415 | * If the page is no longer dirty, or its mapping no |
2416 | * longer corresponds to inode we are writing (which | |
2417 | * means it has been truncated or invalidated), or the | |
2418 | * page is already under writeback and we are not doing | |
2419 | * a data integrity writeback, skip the page | |
8e48dcfb | 2420 | */ |
4f01b02c TT |
2421 | if (!PageDirty(page) || |
2422 | (PageWriteback(page) && | |
4e7ea81d | 2423 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2424 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2425 | unlock_page(page); |
2426 | continue; | |
2427 | } | |
2428 | ||
7cb1a535 | 2429 | wait_on_page_writeback(page); |
8e48dcfb | 2430 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2431 | |
4e7ea81d | 2432 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2433 | mpd->first_page = page->index; |
8eb9e5ce | 2434 | mpd->next_page = page->index + 1; |
f8bec370 | 2435 | /* Add all dirty buffers to mpd */ |
4e7ea81d JK |
2436 | lblk = ((ext4_lblk_t)page->index) << |
2437 | (PAGE_CACHE_SHIFT - blkbits); | |
f8bec370 | 2438 | head = page_buffers(page); |
5f1132b2 JK |
2439 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2440 | if (err <= 0) | |
4e7ea81d | 2441 | goto out; |
5f1132b2 | 2442 | err = 0; |
aeac589a | 2443 | left--; |
8e48dcfb TT |
2444 | } |
2445 | pagevec_release(&pvec); | |
2446 | cond_resched(); | |
2447 | } | |
4f01b02c | 2448 | return 0; |
8eb9e5ce TT |
2449 | out: |
2450 | pagevec_release(&pvec); | |
4e7ea81d | 2451 | return err; |
8e48dcfb TT |
2452 | } |
2453 | ||
20970ba6 TT |
2454 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2455 | void *data) | |
2456 | { | |
2457 | struct address_space *mapping = data; | |
2458 | int ret = ext4_writepage(page, wbc); | |
2459 | mapping_set_error(mapping, ret); | |
2460 | return ret; | |
2461 | } | |
2462 | ||
2463 | static int ext4_writepages(struct address_space *mapping, | |
2464 | struct writeback_control *wbc) | |
64769240 | 2465 | { |
4e7ea81d JK |
2466 | pgoff_t writeback_index = 0; |
2467 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2468 | int range_whole = 0; |
4e7ea81d | 2469 | int cycled = 1; |
61628a3f | 2470 | handle_t *handle = NULL; |
df22291f | 2471 | struct mpage_da_data mpd; |
5e745b04 | 2472 | struct inode *inode = mapping->host; |
6b523df4 | 2473 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2474 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2475 | bool done; |
1bce63d1 | 2476 | struct blk_plug plug; |
cb530541 | 2477 | bool give_up_on_write = false; |
61628a3f | 2478 | |
20970ba6 | 2479 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2480 | |
61628a3f MC |
2481 | /* |
2482 | * No pages to write? This is mainly a kludge to avoid starting | |
2483 | * a transaction for special inodes like journal inode on last iput() | |
2484 | * because that could violate lock ordering on umount | |
2485 | */ | |
a1d6cc56 | 2486 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2487 | goto out_writepages; |
2a21e37e | 2488 | |
20970ba6 TT |
2489 | if (ext4_should_journal_data(inode)) { |
2490 | struct blk_plug plug; | |
20970ba6 TT |
2491 | |
2492 | blk_start_plug(&plug); | |
2493 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2494 | blk_finish_plug(&plug); | |
bbf023c7 | 2495 | goto out_writepages; |
20970ba6 TT |
2496 | } |
2497 | ||
2a21e37e TT |
2498 | /* |
2499 | * If the filesystem has aborted, it is read-only, so return | |
2500 | * right away instead of dumping stack traces later on that | |
2501 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2502 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2503 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2504 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2505 | * *never* be called, so if that ever happens, we would want |
2506 | * the stack trace. | |
2507 | */ | |
bbf023c7 ML |
2508 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2509 | ret = -EROFS; | |
2510 | goto out_writepages; | |
2511 | } | |
2a21e37e | 2512 | |
6b523df4 JK |
2513 | if (ext4_should_dioread_nolock(inode)) { |
2514 | /* | |
70261f56 | 2515 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2516 | * the page and we may dirty the inode. |
2517 | */ | |
2518 | rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); | |
2519 | } | |
2520 | ||
4e7ea81d JK |
2521 | /* |
2522 | * If we have inline data and arrive here, it means that | |
2523 | * we will soon create the block for the 1st page, so | |
2524 | * we'd better clear the inline data here. | |
2525 | */ | |
2526 | if (ext4_has_inline_data(inode)) { | |
2527 | /* Just inode will be modified... */ | |
2528 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2529 | if (IS_ERR(handle)) { | |
2530 | ret = PTR_ERR(handle); | |
2531 | goto out_writepages; | |
2532 | } | |
2533 | BUG_ON(ext4_test_inode_state(inode, | |
2534 | EXT4_STATE_MAY_INLINE_DATA)); | |
2535 | ext4_destroy_inline_data(handle, inode); | |
2536 | ext4_journal_stop(handle); | |
2537 | } | |
2538 | ||
22208ded AK |
2539 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2540 | range_whole = 1; | |
61628a3f | 2541 | |
2acf2c26 | 2542 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2543 | writeback_index = mapping->writeback_index; |
2544 | if (writeback_index) | |
2acf2c26 | 2545 | cycled = 0; |
4e7ea81d JK |
2546 | mpd.first_page = writeback_index; |
2547 | mpd.last_page = -1; | |
5b41d924 | 2548 | } else { |
4e7ea81d JK |
2549 | mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; |
2550 | mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; | |
5b41d924 | 2551 | } |
a1d6cc56 | 2552 | |
4e7ea81d JK |
2553 | mpd.inode = inode; |
2554 | mpd.wbc = wbc; | |
2555 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2556 | retry: |
6e6938b6 | 2557 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2558 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2559 | done = false; | |
1bce63d1 | 2560 | blk_start_plug(&plug); |
4e7ea81d JK |
2561 | while (!done && mpd.first_page <= mpd.last_page) { |
2562 | /* For each extent of pages we use new io_end */ | |
2563 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2564 | if (!mpd.io_submit.io_end) { | |
2565 | ret = -ENOMEM; | |
2566 | break; | |
2567 | } | |
a1d6cc56 AK |
2568 | |
2569 | /* | |
4e7ea81d JK |
2570 | * We have two constraints: We find one extent to map and we |
2571 | * must always write out whole page (makes a difference when | |
2572 | * blocksize < pagesize) so that we don't block on IO when we | |
2573 | * try to write out the rest of the page. Journalled mode is | |
2574 | * not supported by delalloc. | |
a1d6cc56 AK |
2575 | */ |
2576 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2577 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2578 | |
4e7ea81d | 2579 | /* start a new transaction */ |
6b523df4 JK |
2580 | handle = ext4_journal_start_with_reserve(inode, |
2581 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2582 | if (IS_ERR(handle)) { |
2583 | ret = PTR_ERR(handle); | |
1693918e | 2584 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2585 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2586 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2587 | /* Release allocated io_end */ |
2588 | ext4_put_io_end(mpd.io_submit.io_end); | |
2589 | break; | |
61628a3f | 2590 | } |
f63e6005 | 2591 | |
4e7ea81d JK |
2592 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2593 | ret = mpage_prepare_extent_to_map(&mpd); | |
2594 | if (!ret) { | |
2595 | if (mpd.map.m_len) | |
cb530541 TT |
2596 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2597 | &give_up_on_write); | |
4e7ea81d JK |
2598 | else { |
2599 | /* | |
2600 | * We scanned the whole range (or exhausted | |
2601 | * nr_to_write), submitted what was mapped and | |
2602 | * didn't find anything needing mapping. We are | |
2603 | * done. | |
2604 | */ | |
2605 | done = true; | |
2606 | } | |
f63e6005 | 2607 | } |
61628a3f | 2608 | ext4_journal_stop(handle); |
4e7ea81d JK |
2609 | /* Submit prepared bio */ |
2610 | ext4_io_submit(&mpd.io_submit); | |
2611 | /* Unlock pages we didn't use */ | |
cb530541 | 2612 | mpage_release_unused_pages(&mpd, give_up_on_write); |
4e7ea81d JK |
2613 | /* Drop our io_end reference we got from init */ |
2614 | ext4_put_io_end(mpd.io_submit.io_end); | |
2615 | ||
2616 | if (ret == -ENOSPC && sbi->s_journal) { | |
2617 | /* | |
2618 | * Commit the transaction which would | |
22208ded AK |
2619 | * free blocks released in the transaction |
2620 | * and try again | |
2621 | */ | |
df22291f | 2622 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2623 | ret = 0; |
4e7ea81d JK |
2624 | continue; |
2625 | } | |
2626 | /* Fatal error - ENOMEM, EIO... */ | |
2627 | if (ret) | |
61628a3f | 2628 | break; |
a1d6cc56 | 2629 | } |
1bce63d1 | 2630 | blk_finish_plug(&plug); |
9c12a831 | 2631 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2632 | cycled = 1; |
4e7ea81d JK |
2633 | mpd.last_page = writeback_index - 1; |
2634 | mpd.first_page = 0; | |
2acf2c26 AK |
2635 | goto retry; |
2636 | } | |
22208ded AK |
2637 | |
2638 | /* Update index */ | |
22208ded AK |
2639 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2640 | /* | |
4e7ea81d | 2641 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2642 | * mode will write it back later |
2643 | */ | |
4e7ea81d | 2644 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2645 | |
61628a3f | 2646 | out_writepages: |
20970ba6 TT |
2647 | trace_ext4_writepages_result(inode, wbc, ret, |
2648 | nr_to_write - wbc->nr_to_write); | |
61628a3f | 2649 | return ret; |
64769240 AT |
2650 | } |
2651 | ||
79f0be8d AK |
2652 | static int ext4_nonda_switch(struct super_block *sb) |
2653 | { | |
5c1ff336 | 2654 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2655 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2656 | ||
2657 | /* | |
2658 | * switch to non delalloc mode if we are running low | |
2659 | * on free block. The free block accounting via percpu | |
179f7ebf | 2660 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2661 | * accumulated on each CPU without updating global counters |
2662 | * Delalloc need an accurate free block accounting. So switch | |
2663 | * to non delalloc when we are near to error range. | |
2664 | */ | |
5c1ff336 EW |
2665 | free_clusters = |
2666 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2667 | dirty_clusters = | |
2668 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2669 | /* |
2670 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2671 | */ | |
5c1ff336 | 2672 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2673 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2674 | |
5c1ff336 EW |
2675 | if (2 * free_clusters < 3 * dirty_clusters || |
2676 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2677 | /* |
c8afb446 ES |
2678 | * free block count is less than 150% of dirty blocks |
2679 | * or free blocks is less than watermark | |
79f0be8d AK |
2680 | */ |
2681 | return 1; | |
2682 | } | |
2683 | return 0; | |
2684 | } | |
2685 | ||
0ff8947f ES |
2686 | /* We always reserve for an inode update; the superblock could be there too */ |
2687 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2688 | { | |
e2b911c5 | 2689 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2690 | return 1; |
2691 | ||
2692 | if (pos + len <= 0x7fffffffULL) | |
2693 | return 1; | |
2694 | ||
2695 | /* We might need to update the superblock to set LARGE_FILE */ | |
2696 | return 2; | |
2697 | } | |
2698 | ||
64769240 | 2699 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2700 | loff_t pos, unsigned len, unsigned flags, |
2701 | struct page **pagep, void **fsdata) | |
64769240 | 2702 | { |
72b8ab9d | 2703 | int ret, retries = 0; |
64769240 AT |
2704 | struct page *page; |
2705 | pgoff_t index; | |
64769240 AT |
2706 | struct inode *inode = mapping->host; |
2707 | handle_t *handle; | |
2708 | ||
2709 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2710 | |
2711 | if (ext4_nonda_switch(inode->i_sb)) { | |
2712 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2713 | return ext4_write_begin(file, mapping, pos, | |
2714 | len, flags, pagep, fsdata); | |
2715 | } | |
2716 | *fsdata = (void *)0; | |
9bffad1e | 2717 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2718 | |
2719 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2720 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2721 | pos, len, flags, | |
2722 | pagep, fsdata); | |
2723 | if (ret < 0) | |
47564bfb TT |
2724 | return ret; |
2725 | if (ret == 1) | |
2726 | return 0; | |
9c3569b5 TM |
2727 | } |
2728 | ||
47564bfb TT |
2729 | /* |
2730 | * grab_cache_page_write_begin() can take a long time if the | |
2731 | * system is thrashing due to memory pressure, or if the page | |
2732 | * is being written back. So grab it first before we start | |
2733 | * the transaction handle. This also allows us to allocate | |
2734 | * the page (if needed) without using GFP_NOFS. | |
2735 | */ | |
2736 | retry_grab: | |
2737 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2738 | if (!page) | |
2739 | return -ENOMEM; | |
2740 | unlock_page(page); | |
2741 | ||
64769240 AT |
2742 | /* |
2743 | * With delayed allocation, we don't log the i_disksize update | |
2744 | * if there is delayed block allocation. But we still need | |
2745 | * to journalling the i_disksize update if writes to the end | |
2746 | * of file which has an already mapped buffer. | |
2747 | */ | |
47564bfb | 2748 | retry_journal: |
0ff8947f ES |
2749 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2750 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2751 | if (IS_ERR(handle)) { |
47564bfb TT |
2752 | page_cache_release(page); |
2753 | return PTR_ERR(handle); | |
64769240 AT |
2754 | } |
2755 | ||
47564bfb TT |
2756 | lock_page(page); |
2757 | if (page->mapping != mapping) { | |
2758 | /* The page got truncated from under us */ | |
2759 | unlock_page(page); | |
2760 | page_cache_release(page); | |
d5a0d4f7 | 2761 | ext4_journal_stop(handle); |
47564bfb | 2762 | goto retry_grab; |
d5a0d4f7 | 2763 | } |
47564bfb | 2764 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2765 | wait_for_stable_page(page); |
64769240 | 2766 | |
2058f83a MH |
2767 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
2768 | ret = ext4_block_write_begin(page, pos, len, | |
2769 | ext4_da_get_block_prep); | |
2770 | #else | |
6e1db88d | 2771 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2772 | #endif |
64769240 AT |
2773 | if (ret < 0) { |
2774 | unlock_page(page); | |
2775 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2776 | /* |
2777 | * block_write_begin may have instantiated a few blocks | |
2778 | * outside i_size. Trim these off again. Don't need | |
2779 | * i_size_read because we hold i_mutex. | |
2780 | */ | |
2781 | if (pos + len > inode->i_size) | |
b9a4207d | 2782 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2783 | |
2784 | if (ret == -ENOSPC && | |
2785 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2786 | goto retry_journal; | |
2787 | ||
2788 | page_cache_release(page); | |
2789 | return ret; | |
64769240 AT |
2790 | } |
2791 | ||
47564bfb | 2792 | *pagep = page; |
64769240 AT |
2793 | return ret; |
2794 | } | |
2795 | ||
632eaeab MC |
2796 | /* |
2797 | * Check if we should update i_disksize | |
2798 | * when write to the end of file but not require block allocation | |
2799 | */ | |
2800 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2801 | unsigned long offset) |
632eaeab MC |
2802 | { |
2803 | struct buffer_head *bh; | |
2804 | struct inode *inode = page->mapping->host; | |
2805 | unsigned int idx; | |
2806 | int i; | |
2807 | ||
2808 | bh = page_buffers(page); | |
2809 | idx = offset >> inode->i_blkbits; | |
2810 | ||
af5bc92d | 2811 | for (i = 0; i < idx; i++) |
632eaeab MC |
2812 | bh = bh->b_this_page; |
2813 | ||
29fa89d0 | 2814 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2815 | return 0; |
2816 | return 1; | |
2817 | } | |
2818 | ||
64769240 | 2819 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2820 | struct address_space *mapping, |
2821 | loff_t pos, unsigned len, unsigned copied, | |
2822 | struct page *page, void *fsdata) | |
64769240 AT |
2823 | { |
2824 | struct inode *inode = mapping->host; | |
2825 | int ret = 0, ret2; | |
2826 | handle_t *handle = ext4_journal_current_handle(); | |
2827 | loff_t new_i_size; | |
632eaeab | 2828 | unsigned long start, end; |
79f0be8d AK |
2829 | int write_mode = (int)(unsigned long)fsdata; |
2830 | ||
74d553aa TT |
2831 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2832 | return ext4_write_end(file, mapping, pos, | |
2833 | len, copied, page, fsdata); | |
632eaeab | 2834 | |
9bffad1e | 2835 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2836 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2837 | end = start + copied - 1; |
64769240 AT |
2838 | |
2839 | /* | |
2840 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2841 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2842 | * into that. | |
2843 | */ | |
64769240 | 2844 | new_i_size = pos + copied; |
ea51d132 | 2845 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2846 | if (ext4_has_inline_data(inode) || |
2847 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 2848 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
2849 | /* We need to mark inode dirty even if |
2850 | * new_i_size is less that inode->i_size | |
2851 | * bu greater than i_disksize.(hint delalloc) | |
2852 | */ | |
2853 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2854 | } |
632eaeab | 2855 | } |
9c3569b5 TM |
2856 | |
2857 | if (write_mode != CONVERT_INLINE_DATA && | |
2858 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2859 | ext4_has_inline_data(inode)) | |
2860 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2861 | page); | |
2862 | else | |
2863 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2864 | page, fsdata); |
9c3569b5 | 2865 | |
64769240 AT |
2866 | copied = ret2; |
2867 | if (ret2 < 0) | |
2868 | ret = ret2; | |
2869 | ret2 = ext4_journal_stop(handle); | |
2870 | if (!ret) | |
2871 | ret = ret2; | |
2872 | ||
2873 | return ret ? ret : copied; | |
2874 | } | |
2875 | ||
d47992f8 LC |
2876 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
2877 | unsigned int length) | |
64769240 | 2878 | { |
64769240 AT |
2879 | /* |
2880 | * Drop reserved blocks | |
2881 | */ | |
2882 | BUG_ON(!PageLocked(page)); | |
2883 | if (!page_has_buffers(page)) | |
2884 | goto out; | |
2885 | ||
ca99fdd2 | 2886 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
2887 | |
2888 | out: | |
d47992f8 | 2889 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
2890 | |
2891 | return; | |
2892 | } | |
2893 | ||
ccd2506b TT |
2894 | /* |
2895 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2896 | */ | |
2897 | int ext4_alloc_da_blocks(struct inode *inode) | |
2898 | { | |
fb40ba0d TT |
2899 | trace_ext4_alloc_da_blocks(inode); |
2900 | ||
71d4f7d0 | 2901 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
2902 | return 0; |
2903 | ||
2904 | /* | |
2905 | * We do something simple for now. The filemap_flush() will | |
2906 | * also start triggering a write of the data blocks, which is | |
2907 | * not strictly speaking necessary (and for users of | |
2908 | * laptop_mode, not even desirable). However, to do otherwise | |
2909 | * would require replicating code paths in: | |
de9a55b8 | 2910 | * |
20970ba6 | 2911 | * ext4_writepages() -> |
ccd2506b TT |
2912 | * write_cache_pages() ---> (via passed in callback function) |
2913 | * __mpage_da_writepage() --> | |
2914 | * mpage_add_bh_to_extent() | |
2915 | * mpage_da_map_blocks() | |
2916 | * | |
2917 | * The problem is that write_cache_pages(), located in | |
2918 | * mm/page-writeback.c, marks pages clean in preparation for | |
2919 | * doing I/O, which is not desirable if we're not planning on | |
2920 | * doing I/O at all. | |
2921 | * | |
2922 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2923 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2924 | * would be ugly in the extreme. So instead we would need to |
2925 | * replicate parts of the code in the above functions, | |
25985edc | 2926 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2927 | * write out the pages, but rather only collect contiguous |
2928 | * logical block extents, call the multi-block allocator, and | |
2929 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2930 | * |
ccd2506b TT |
2931 | * For now, though, we'll cheat by calling filemap_flush(), |
2932 | * which will map the blocks, and start the I/O, but not | |
2933 | * actually wait for the I/O to complete. | |
2934 | */ | |
2935 | return filemap_flush(inode->i_mapping); | |
2936 | } | |
64769240 | 2937 | |
ac27a0ec DK |
2938 | /* |
2939 | * bmap() is special. It gets used by applications such as lilo and by | |
2940 | * the swapper to find the on-disk block of a specific piece of data. | |
2941 | * | |
2942 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2943 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2944 | * filesystem and enables swap, then they may get a nasty shock when the |
2945 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2946 | * the original zero's written out previously to the journal and | |
2947 | * awaiting writeback in the kernel's buffer cache. | |
2948 | * | |
2949 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2950 | * take extra steps to flush any blocks which might be in the cache. | |
2951 | */ | |
617ba13b | 2952 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2953 | { |
2954 | struct inode *inode = mapping->host; | |
2955 | journal_t *journal; | |
2956 | int err; | |
2957 | ||
46c7f254 TM |
2958 | /* |
2959 | * We can get here for an inline file via the FIBMAP ioctl | |
2960 | */ | |
2961 | if (ext4_has_inline_data(inode)) | |
2962 | return 0; | |
2963 | ||
64769240 AT |
2964 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2965 | test_opt(inode->i_sb, DELALLOC)) { | |
2966 | /* | |
2967 | * With delalloc we want to sync the file | |
2968 | * so that we can make sure we allocate | |
2969 | * blocks for file | |
2970 | */ | |
2971 | filemap_write_and_wait(mapping); | |
2972 | } | |
2973 | ||
19f5fb7a TT |
2974 | if (EXT4_JOURNAL(inode) && |
2975 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2976 | /* |
2977 | * This is a REALLY heavyweight approach, but the use of | |
2978 | * bmap on dirty files is expected to be extremely rare: | |
2979 | * only if we run lilo or swapon on a freshly made file | |
2980 | * do we expect this to happen. | |
2981 | * | |
2982 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2983 | * represent an unprivileged user DOS attack --- we'd be | |
2984 | * in trouble if mortal users could trigger this path at | |
2985 | * will.) | |
2986 | * | |
617ba13b | 2987 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2988 | * regular files. If somebody wants to bmap a directory |
2989 | * or symlink and gets confused because the buffer | |
2990 | * hasn't yet been flushed to disk, they deserve | |
2991 | * everything they get. | |
2992 | */ | |
2993 | ||
19f5fb7a | 2994 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2995 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2996 | jbd2_journal_lock_updates(journal); |
2997 | err = jbd2_journal_flush(journal); | |
2998 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2999 | |
3000 | if (err) | |
3001 | return 0; | |
3002 | } | |
3003 | ||
af5bc92d | 3004 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3005 | } |
3006 | ||
617ba13b | 3007 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3008 | { |
46c7f254 TM |
3009 | int ret = -EAGAIN; |
3010 | struct inode *inode = page->mapping->host; | |
3011 | ||
0562e0ba | 3012 | trace_ext4_readpage(page); |
46c7f254 TM |
3013 | |
3014 | if (ext4_has_inline_data(inode)) | |
3015 | ret = ext4_readpage_inline(inode, page); | |
3016 | ||
3017 | if (ret == -EAGAIN) | |
f64e02fe | 3018 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3019 | |
3020 | return ret; | |
ac27a0ec DK |
3021 | } |
3022 | ||
3023 | static int | |
617ba13b | 3024 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3025 | struct list_head *pages, unsigned nr_pages) |
3026 | { | |
46c7f254 TM |
3027 | struct inode *inode = mapping->host; |
3028 | ||
3029 | /* If the file has inline data, no need to do readpages. */ | |
3030 | if (ext4_has_inline_data(inode)) | |
3031 | return 0; | |
3032 | ||
f64e02fe | 3033 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3034 | } |
3035 | ||
d47992f8 LC |
3036 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3037 | unsigned int length) | |
ac27a0ec | 3038 | { |
ca99fdd2 | 3039 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3040 | |
4520fb3c JK |
3041 | /* No journalling happens on data buffers when this function is used */ |
3042 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3043 | ||
ca99fdd2 | 3044 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3045 | } |
3046 | ||
53e87268 | 3047 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3048 | unsigned int offset, |
3049 | unsigned int length) | |
4520fb3c JK |
3050 | { |
3051 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3052 | ||
ca99fdd2 | 3053 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3054 | |
ac27a0ec DK |
3055 | /* |
3056 | * If it's a full truncate we just forget about the pending dirtying | |
3057 | */ | |
ca99fdd2 | 3058 | if (offset == 0 && length == PAGE_CACHE_SIZE) |
ac27a0ec DK |
3059 | ClearPageChecked(page); |
3060 | ||
ca99fdd2 | 3061 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3062 | } |
3063 | ||
3064 | /* Wrapper for aops... */ | |
3065 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3066 | unsigned int offset, |
3067 | unsigned int length) | |
53e87268 | 3068 | { |
ca99fdd2 | 3069 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3070 | } |
3071 | ||
617ba13b | 3072 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3073 | { |
617ba13b | 3074 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3075 | |
0562e0ba JZ |
3076 | trace_ext4_releasepage(page); |
3077 | ||
e1c36595 JK |
3078 | /* Page has dirty journalled data -> cannot release */ |
3079 | if (PageChecked(page)) | |
ac27a0ec | 3080 | return 0; |
0390131b FM |
3081 | if (journal) |
3082 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3083 | else | |
3084 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3085 | } |
3086 | ||
2ed88685 TT |
3087 | /* |
3088 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3089 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3090 | * The extent will be converted to initialized after the IO is complete. | |
3091 | */ | |
f19d5870 | 3092 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3093 | struct buffer_head *bh_result, int create) |
3094 | { | |
c7064ef1 | 3095 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3096 | inode->i_ino, create); |
2ed88685 TT |
3097 | return _ext4_get_block(inode, iblock, bh_result, |
3098 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3099 | } |
3100 | ||
2dcba478 | 3101 | static int ext4_get_block_overwrite(struct inode *inode, sector_t iblock, |
8b0f165f | 3102 | struct buffer_head *bh_result, int create) |
729f52c6 | 3103 | { |
2dcba478 JK |
3104 | int ret; |
3105 | ||
3106 | ext4_debug("ext4_get_block_overwrite: inode %lu, create flag %d\n", | |
8b0f165f | 3107 | inode->i_ino, create); |
2dcba478 JK |
3108 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
3109 | /* | |
3110 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
3111 | * that. | |
3112 | */ | |
3113 | WARN_ON_ONCE(!buffer_mapped(bh_result)); | |
3114 | ||
3115 | return ret; | |
729f52c6 ZL |
3116 | } |
3117 | ||
ba5843f5 JK |
3118 | #ifdef CONFIG_FS_DAX |
3119 | int ext4_dax_mmap_get_block(struct inode *inode, sector_t iblock, | |
3120 | struct buffer_head *bh_result, int create) | |
ed923b57 | 3121 | { |
ba5843f5 JK |
3122 | int ret, err; |
3123 | int credits; | |
3124 | struct ext4_map_blocks map; | |
3125 | handle_t *handle = NULL; | |
3126 | int flags = 0; | |
c86d8db3 | 3127 | |
ba5843f5 | 3128 | ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n", |
ed923b57 | 3129 | inode->i_ino, create); |
ba5843f5 JK |
3130 | map.m_lblk = iblock; |
3131 | map.m_len = bh_result->b_size >> inode->i_blkbits; | |
3132 | credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
3133 | if (create) { | |
3134 | flags |= EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_CREATE_ZERO; | |
3135 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits); | |
3136 | if (IS_ERR(handle)) { | |
3137 | ret = PTR_ERR(handle); | |
3138 | return ret; | |
3139 | } | |
3140 | } | |
3141 | ||
3142 | ret = ext4_map_blocks(handle, inode, &map, flags); | |
3143 | if (create) { | |
3144 | err = ext4_journal_stop(handle); | |
3145 | if (ret >= 0 && err < 0) | |
3146 | ret = err; | |
3147 | } | |
3148 | if (ret <= 0) | |
3149 | goto out; | |
3150 | if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
3151 | int err2; | |
3152 | ||
3153 | /* | |
3154 | * We are protected by i_mmap_sem so we know block cannot go | |
3155 | * away from under us even though we dropped i_data_sem. | |
3156 | * Convert extent to written and write zeros there. | |
3157 | * | |
3158 | * Note: We may get here even when create == 0. | |
3159 | */ | |
3160 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits); | |
3161 | if (IS_ERR(handle)) { | |
3162 | ret = PTR_ERR(handle); | |
3163 | goto out; | |
3164 | } | |
3165 | ||
3166 | err = ext4_map_blocks(handle, inode, &map, | |
3167 | EXT4_GET_BLOCKS_CONVERT | EXT4_GET_BLOCKS_CREATE_ZERO); | |
3168 | if (err < 0) | |
3169 | ret = err; | |
3170 | err2 = ext4_journal_stop(handle); | |
3171 | if (err2 < 0 && ret > 0) | |
3172 | ret = err2; | |
3173 | } | |
3174 | out: | |
3175 | WARN_ON_ONCE(ret == 0 && create); | |
3176 | if (ret > 0) { | |
3177 | map_bh(bh_result, inode->i_sb, map.m_pblk); | |
3178 | bh_result->b_state = (bh_result->b_state & ~EXT4_MAP_FLAGS) | | |
3179 | map.m_flags; | |
3180 | /* | |
3181 | * At least for now we have to clear BH_New so that DAX code | |
3182 | * doesn't attempt to zero blocks again in a racy way. | |
3183 | */ | |
3184 | bh_result->b_state &= ~(1 << BH_New); | |
3185 | bh_result->b_size = map.m_len << inode->i_blkbits; | |
3186 | ret = 0; | |
3187 | } | |
3188 | return ret; | |
ed923b57 | 3189 | } |
ba5843f5 | 3190 | #endif |
ed923b57 | 3191 | |
4c0425ff | 3192 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3193 | ssize_t size, void *private) |
4c0425ff MC |
3194 | { |
3195 | ext4_io_end_t *io_end = iocb->private; | |
4c0425ff | 3196 | |
97a851ed | 3197 | /* if not async direct IO just return */ |
7b7a8665 | 3198 | if (!io_end) |
97a851ed | 3199 | return; |
4b70df18 | 3200 | |
88635ca2 | 3201 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3202 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3203 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3204 | size); | |
8d5d02e6 | 3205 | |
b5a7e970 | 3206 | iocb->private = NULL; |
4c0425ff MC |
3207 | io_end->offset = offset; |
3208 | io_end->size = size; | |
7b7a8665 | 3209 | ext4_put_io_end(io_end); |
4c0425ff | 3210 | } |
c7064ef1 | 3211 | |
4c0425ff MC |
3212 | /* |
3213 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3214 | * preallocated extents, and those write extend the file, no need to | |
3215 | * fall back to buffered IO. | |
3216 | * | |
556615dc | 3217 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3218 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3219 | * still keep the range to write as unwritten. |
4c0425ff | 3220 | * |
69c499d1 | 3221 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3222 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3223 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3224 | * when async direct IO completed. |
4c0425ff MC |
3225 | * |
3226 | * If the O_DIRECT write will extend the file then add this inode to the | |
3227 | * orphan list. So recovery will truncate it back to the original size | |
3228 | * if the machine crashes during the write. | |
3229 | * | |
3230 | */ | |
6f673763 OS |
3231 | static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
3232 | loff_t offset) | |
4c0425ff MC |
3233 | { |
3234 | struct file *file = iocb->ki_filp; | |
3235 | struct inode *inode = file->f_mapping->host; | |
3236 | ssize_t ret; | |
a6cbcd4a | 3237 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3238 | int overwrite = 0; |
3239 | get_block_t *get_block_func = NULL; | |
3240 | int dio_flags = 0; | |
4c0425ff | 3241 | loff_t final_size = offset + count; |
97a851ed | 3242 | ext4_io_end_t *io_end = NULL; |
729f52c6 | 3243 | |
69c499d1 | 3244 | /* Use the old path for reads and writes beyond i_size. */ |
6f673763 OS |
3245 | if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size) |
3246 | return ext4_ind_direct_IO(iocb, iter, offset); | |
4bd809db | 3247 | |
69c499d1 | 3248 | BUG_ON(iocb->private == NULL); |
4bd809db | 3249 | |
e8340395 JK |
3250 | /* |
3251 | * Make all waiters for direct IO properly wait also for extent | |
3252 | * conversion. This also disallows race between truncate() and | |
3253 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3254 | */ | |
6f673763 | 3255 | if (iov_iter_rw(iter) == WRITE) |
fe0f07d0 | 3256 | inode_dio_begin(inode); |
e8340395 | 3257 | |
69c499d1 TT |
3258 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3259 | overwrite = *((int *)iocb->private); | |
4bd809db | 3260 | |
2dcba478 | 3261 | if (overwrite) |
5955102c | 3262 | inode_unlock(inode); |
8d5d02e6 | 3263 | |
69c499d1 TT |
3264 | /* |
3265 | * We could direct write to holes and fallocate. | |
3266 | * | |
3267 | * Allocated blocks to fill the hole are marked as | |
556615dc | 3268 | * unwritten to prevent parallel buffered read to expose |
69c499d1 TT |
3269 | * the stale data before DIO complete the data IO. |
3270 | * | |
3271 | * As to previously fallocated extents, ext4 get_block will | |
3272 | * just simply mark the buffer mapped but still keep the | |
556615dc | 3273 | * extents unwritten. |
69c499d1 TT |
3274 | * |
3275 | * For non AIO case, we will convert those unwritten extents | |
3276 | * to written after return back from blockdev_direct_IO. | |
3277 | * | |
3278 | * For async DIO, the conversion needs to be deferred when the | |
3279 | * IO is completed. The ext4 end_io callback function will be | |
3280 | * called to take care of the conversion work. Here for async | |
3281 | * case, we allocate an io_end structure to hook to the iocb. | |
3282 | */ | |
3283 | iocb->private = NULL; | |
69c499d1 | 3284 | if (overwrite) { |
2dcba478 | 3285 | get_block_func = ext4_get_block_overwrite; |
69c499d1 | 3286 | } else { |
74dae427 JK |
3287 | ext4_inode_aio_set(inode, NULL); |
3288 | if (!is_sync_kiocb(iocb)) { | |
3289 | io_end = ext4_init_io_end(inode, GFP_NOFS); | |
3290 | if (!io_end) { | |
3291 | ret = -ENOMEM; | |
3292 | goto retake_lock; | |
3293 | } | |
3294 | /* | |
3295 | * Grab reference for DIO. Will be dropped in | |
3296 | * ext4_end_io_dio() | |
3297 | */ | |
3298 | iocb->private = ext4_get_io_end(io_end); | |
3299 | /* | |
3300 | * we save the io structure for current async direct | |
3301 | * IO, so that later ext4_map_blocks() could flag the | |
3302 | * io structure whether there is a unwritten extents | |
3303 | * needs to be converted when IO is completed. | |
3304 | */ | |
3305 | ext4_inode_aio_set(inode, io_end); | |
3306 | } | |
69c499d1 TT |
3307 | get_block_func = ext4_get_block_write; |
3308 | dio_flags = DIO_LOCKING; | |
3309 | } | |
2058f83a MH |
3310 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3311 | BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); | |
3312 | #endif | |
923ae0ff | 3313 | if (IS_DAX(inode)) |
a95cd631 | 3314 | ret = dax_do_io(iocb, inode, iter, offset, get_block_func, |
923ae0ff RZ |
3315 | ext4_end_io_dio, dio_flags); |
3316 | else | |
17f8c842 | 3317 | ret = __blockdev_direct_IO(iocb, inode, |
923ae0ff RZ |
3318 | inode->i_sb->s_bdev, iter, offset, |
3319 | get_block_func, | |
3320 | ext4_end_io_dio, NULL, dio_flags); | |
69c499d1 | 3321 | |
69c499d1 | 3322 | /* |
97a851ed JK |
3323 | * Put our reference to io_end. This can free the io_end structure e.g. |
3324 | * in sync IO case or in case of error. It can even perform extent | |
3325 | * conversion if all bios we submitted finished before we got here. | |
3326 | * Note that in that case iocb->private can be already set to NULL | |
3327 | * here. | |
69c499d1 | 3328 | */ |
97a851ed JK |
3329 | if (io_end) { |
3330 | ext4_inode_aio_set(inode, NULL); | |
3331 | ext4_put_io_end(io_end); | |
3332 | /* | |
3333 | * When no IO was submitted ext4_end_io_dio() was not | |
3334 | * called so we have to put iocb's reference. | |
3335 | */ | |
3336 | if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { | |
3337 | WARN_ON(iocb->private != io_end); | |
3338 | WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
97a851ed JK |
3339 | ext4_put_io_end(io_end); |
3340 | iocb->private = NULL; | |
3341 | } | |
3342 | } | |
3343 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3344 | EXT4_STATE_DIO_UNWRITTEN)) { |
3345 | int err; | |
3346 | /* | |
3347 | * for non AIO case, since the IO is already | |
3348 | * completed, we could do the conversion right here | |
3349 | */ | |
6b523df4 | 3350 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3351 | offset, ret); |
3352 | if (err < 0) | |
3353 | ret = err; | |
3354 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3355 | } | |
4bd809db | 3356 | |
69c499d1 | 3357 | retake_lock: |
6f673763 | 3358 | if (iov_iter_rw(iter) == WRITE) |
fe0f07d0 | 3359 | inode_dio_end(inode); |
69c499d1 | 3360 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3361 | if (overwrite) |
5955102c | 3362 | inode_lock(inode); |
8d5d02e6 | 3363 | |
69c499d1 | 3364 | return ret; |
4c0425ff MC |
3365 | } |
3366 | ||
22c6186e OS |
3367 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
3368 | loff_t offset) | |
4c0425ff MC |
3369 | { |
3370 | struct file *file = iocb->ki_filp; | |
3371 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3372 | size_t count = iov_iter_count(iter); |
0562e0ba | 3373 | ssize_t ret; |
4c0425ff | 3374 | |
2058f83a MH |
3375 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3376 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3377 | return 0; | |
3378 | #endif | |
3379 | ||
84ebd795 TT |
3380 | /* |
3381 | * If we are doing data journalling we don't support O_DIRECT | |
3382 | */ | |
3383 | if (ext4_should_journal_data(inode)) | |
3384 | return 0; | |
3385 | ||
46c7f254 TM |
3386 | /* Let buffer I/O handle the inline data case. */ |
3387 | if (ext4_has_inline_data(inode)) | |
3388 | return 0; | |
3389 | ||
6f673763 | 3390 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
12e9b892 | 3391 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
6f673763 | 3392 | ret = ext4_ext_direct_IO(iocb, iter, offset); |
0562e0ba | 3393 | else |
6f673763 OS |
3394 | ret = ext4_ind_direct_IO(iocb, iter, offset); |
3395 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); | |
0562e0ba | 3396 | return ret; |
4c0425ff MC |
3397 | } |
3398 | ||
ac27a0ec | 3399 | /* |
617ba13b | 3400 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3401 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3402 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3403 | * not necessarily locked. | |
3404 | * | |
3405 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3406 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3407 | * or jbddirty because all the journalling code will explode. | |
3408 | * | |
3409 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3410 | * is called, propagate that into the buffers appropriately. | |
3411 | */ | |
617ba13b | 3412 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3413 | { |
3414 | SetPageChecked(page); | |
3415 | return __set_page_dirty_nobuffers(page); | |
3416 | } | |
3417 | ||
74d553aa | 3418 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3419 | .readpage = ext4_readpage, |
3420 | .readpages = ext4_readpages, | |
43ce1d23 | 3421 | .writepage = ext4_writepage, |
20970ba6 | 3422 | .writepages = ext4_writepages, |
8ab22b9a | 3423 | .write_begin = ext4_write_begin, |
74d553aa | 3424 | .write_end = ext4_write_end, |
8ab22b9a HH |
3425 | .bmap = ext4_bmap, |
3426 | .invalidatepage = ext4_invalidatepage, | |
3427 | .releasepage = ext4_releasepage, | |
3428 | .direct_IO = ext4_direct_IO, | |
3429 | .migratepage = buffer_migrate_page, | |
3430 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3431 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3432 | }; |
3433 | ||
617ba13b | 3434 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3435 | .readpage = ext4_readpage, |
3436 | .readpages = ext4_readpages, | |
43ce1d23 | 3437 | .writepage = ext4_writepage, |
20970ba6 | 3438 | .writepages = ext4_writepages, |
8ab22b9a HH |
3439 | .write_begin = ext4_write_begin, |
3440 | .write_end = ext4_journalled_write_end, | |
3441 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3442 | .bmap = ext4_bmap, | |
4520fb3c | 3443 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3444 | .releasepage = ext4_releasepage, |
84ebd795 | 3445 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3446 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3447 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3448 | }; |
3449 | ||
64769240 | 3450 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3451 | .readpage = ext4_readpage, |
3452 | .readpages = ext4_readpages, | |
43ce1d23 | 3453 | .writepage = ext4_writepage, |
20970ba6 | 3454 | .writepages = ext4_writepages, |
8ab22b9a HH |
3455 | .write_begin = ext4_da_write_begin, |
3456 | .write_end = ext4_da_write_end, | |
3457 | .bmap = ext4_bmap, | |
3458 | .invalidatepage = ext4_da_invalidatepage, | |
3459 | .releasepage = ext4_releasepage, | |
3460 | .direct_IO = ext4_direct_IO, | |
3461 | .migratepage = buffer_migrate_page, | |
3462 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3463 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3464 | }; |
3465 | ||
617ba13b | 3466 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3467 | { |
3d2b1582 LC |
3468 | switch (ext4_inode_journal_mode(inode)) { |
3469 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3470 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3471 | break; |
3472 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3473 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3474 | break; |
3475 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3476 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3477 | return; |
3d2b1582 LC |
3478 | default: |
3479 | BUG(); | |
3480 | } | |
74d553aa TT |
3481 | if (test_opt(inode->i_sb, DELALLOC)) |
3482 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3483 | else | |
3484 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3485 | } |
3486 | ||
923ae0ff | 3487 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3488 | struct address_space *mapping, loff_t from, loff_t length) |
3489 | { | |
3490 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3491 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
923ae0ff | 3492 | unsigned blocksize, pos; |
d863dc36 LC |
3493 | ext4_lblk_t iblock; |
3494 | struct inode *inode = mapping->host; | |
3495 | struct buffer_head *bh; | |
3496 | struct page *page; | |
3497 | int err = 0; | |
3498 | ||
3499 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
c62d2555 | 3500 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3501 | if (!page) |
3502 | return -ENOMEM; | |
3503 | ||
3504 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 LC |
3505 | |
3506 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3507 | ||
3508 | if (!page_has_buffers(page)) | |
3509 | create_empty_buffers(page, blocksize, 0); | |
3510 | ||
3511 | /* Find the buffer that contains "offset" */ | |
3512 | bh = page_buffers(page); | |
3513 | pos = blocksize; | |
3514 | while (offset >= pos) { | |
3515 | bh = bh->b_this_page; | |
3516 | iblock++; | |
3517 | pos += blocksize; | |
3518 | } | |
d863dc36 LC |
3519 | if (buffer_freed(bh)) { |
3520 | BUFFER_TRACE(bh, "freed: skip"); | |
3521 | goto unlock; | |
3522 | } | |
d863dc36 LC |
3523 | if (!buffer_mapped(bh)) { |
3524 | BUFFER_TRACE(bh, "unmapped"); | |
3525 | ext4_get_block(inode, iblock, bh, 0); | |
3526 | /* unmapped? It's a hole - nothing to do */ | |
3527 | if (!buffer_mapped(bh)) { | |
3528 | BUFFER_TRACE(bh, "still unmapped"); | |
3529 | goto unlock; | |
3530 | } | |
3531 | } | |
3532 | ||
3533 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3534 | if (PageUptodate(page)) | |
3535 | set_buffer_uptodate(bh); | |
3536 | ||
3537 | if (!buffer_uptodate(bh)) { | |
3538 | err = -EIO; | |
3539 | ll_rw_block(READ, 1, &bh); | |
3540 | wait_on_buffer(bh); | |
3541 | /* Uhhuh. Read error. Complain and punt. */ | |
3542 | if (!buffer_uptodate(bh)) | |
3543 | goto unlock; | |
c9c7429c MH |
3544 | if (S_ISREG(inode->i_mode) && |
3545 | ext4_encrypted_inode(inode)) { | |
3546 | /* We expect the key to be set. */ | |
3547 | BUG_ON(!ext4_has_encryption_key(inode)); | |
3548 | BUG_ON(blocksize != PAGE_CACHE_SIZE); | |
3684de8c | 3549 | WARN_ON_ONCE(ext4_decrypt(page)); |
c9c7429c | 3550 | } |
d863dc36 | 3551 | } |
d863dc36 LC |
3552 | if (ext4_should_journal_data(inode)) { |
3553 | BUFFER_TRACE(bh, "get write access"); | |
3554 | err = ext4_journal_get_write_access(handle, bh); | |
3555 | if (err) | |
3556 | goto unlock; | |
3557 | } | |
d863dc36 | 3558 | zero_user(page, offset, length); |
d863dc36 LC |
3559 | BUFFER_TRACE(bh, "zeroed end of block"); |
3560 | ||
d863dc36 LC |
3561 | if (ext4_should_journal_data(inode)) { |
3562 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3563 | } else { |
353eefd3 | 3564 | err = 0; |
d863dc36 | 3565 | mark_buffer_dirty(bh); |
0713ed0c LC |
3566 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) |
3567 | err = ext4_jbd2_file_inode(handle, inode); | |
3568 | } | |
d863dc36 LC |
3569 | |
3570 | unlock: | |
3571 | unlock_page(page); | |
3572 | page_cache_release(page); | |
3573 | return err; | |
3574 | } | |
3575 | ||
923ae0ff RZ |
3576 | /* |
3577 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3578 | * starting from file offset 'from'. The range to be zero'd must | |
3579 | * be contained with in one block. If the specified range exceeds | |
3580 | * the end of the block it will be shortened to end of the block | |
3581 | * that cooresponds to 'from' | |
3582 | */ | |
3583 | static int ext4_block_zero_page_range(handle_t *handle, | |
3584 | struct address_space *mapping, loff_t from, loff_t length) | |
3585 | { | |
3586 | struct inode *inode = mapping->host; | |
3587 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3588 | unsigned blocksize = inode->i_sb->s_blocksize; | |
3589 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3590 | ||
3591 | /* | |
3592 | * correct length if it does not fall between | |
3593 | * 'from' and the end of the block | |
3594 | */ | |
3595 | if (length > max || length < 0) | |
3596 | length = max; | |
3597 | ||
3598 | if (IS_DAX(inode)) | |
3599 | return dax_zero_page_range(inode, from, length, ext4_get_block); | |
3600 | return __ext4_block_zero_page_range(handle, mapping, from, length); | |
3601 | } | |
3602 | ||
94350ab5 MW |
3603 | /* |
3604 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3605 | * up to the end of the block which corresponds to `from'. | |
3606 | * This required during truncate. We need to physically zero the tail end | |
3607 | * of that block so it doesn't yield old data if the file is later grown. | |
3608 | */ | |
c197855e | 3609 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3610 | struct address_space *mapping, loff_t from) |
3611 | { | |
3612 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3613 | unsigned length; | |
3614 | unsigned blocksize; | |
3615 | struct inode *inode = mapping->host; | |
3616 | ||
3617 | blocksize = inode->i_sb->s_blocksize; | |
3618 | length = blocksize - (offset & (blocksize - 1)); | |
3619 | ||
3620 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3621 | } | |
3622 | ||
a87dd18c LC |
3623 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3624 | loff_t lstart, loff_t length) | |
3625 | { | |
3626 | struct super_block *sb = inode->i_sb; | |
3627 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3628 | unsigned partial_start, partial_end; |
a87dd18c LC |
3629 | ext4_fsblk_t start, end; |
3630 | loff_t byte_end = (lstart + length - 1); | |
3631 | int err = 0; | |
3632 | ||
e1be3a92 LC |
3633 | partial_start = lstart & (sb->s_blocksize - 1); |
3634 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3635 | ||
a87dd18c LC |
3636 | start = lstart >> sb->s_blocksize_bits; |
3637 | end = byte_end >> sb->s_blocksize_bits; | |
3638 | ||
3639 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3640 | if (start == end && |
3641 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3642 | err = ext4_block_zero_page_range(handle, mapping, |
3643 | lstart, length); | |
3644 | return err; | |
3645 | } | |
3646 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3647 | if (partial_start) { |
a87dd18c LC |
3648 | err = ext4_block_zero_page_range(handle, mapping, |
3649 | lstart, sb->s_blocksize); | |
3650 | if (err) | |
3651 | return err; | |
3652 | } | |
3653 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3654 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3655 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3656 | byte_end - partial_end, |
3657 | partial_end + 1); | |
a87dd18c LC |
3658 | return err; |
3659 | } | |
3660 | ||
91ef4caf DG |
3661 | int ext4_can_truncate(struct inode *inode) |
3662 | { | |
91ef4caf DG |
3663 | if (S_ISREG(inode->i_mode)) |
3664 | return 1; | |
3665 | if (S_ISDIR(inode->i_mode)) | |
3666 | return 1; | |
3667 | if (S_ISLNK(inode->i_mode)) | |
3668 | return !ext4_inode_is_fast_symlink(inode); | |
3669 | return 0; | |
3670 | } | |
3671 | ||
01127848 JK |
3672 | /* |
3673 | * We have to make sure i_disksize gets properly updated before we truncate | |
3674 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3675 | * can get lost as it may have been postponed to submission of writeback but | |
3676 | * that will never happen after we truncate page cache. | |
3677 | */ | |
3678 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3679 | loff_t len) | |
3680 | { | |
3681 | handle_t *handle; | |
3682 | loff_t size = i_size_read(inode); | |
3683 | ||
5955102c | 3684 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3685 | if (offset > size || offset + len < size) |
3686 | return 0; | |
3687 | ||
3688 | if (EXT4_I(inode)->i_disksize >= size) | |
3689 | return 0; | |
3690 | ||
3691 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3692 | if (IS_ERR(handle)) | |
3693 | return PTR_ERR(handle); | |
3694 | ext4_update_i_disksize(inode, size); | |
3695 | ext4_mark_inode_dirty(handle, inode); | |
3696 | ext4_journal_stop(handle); | |
3697 | ||
3698 | return 0; | |
3699 | } | |
3700 | ||
a4bb6b64 AH |
3701 | /* |
3702 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3703 | * associated with the given offset and length | |
3704 | * | |
3705 | * @inode: File inode | |
3706 | * @offset: The offset where the hole will begin | |
3707 | * @len: The length of the hole | |
3708 | * | |
4907cb7b | 3709 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3710 | */ |
3711 | ||
aeb2817a | 3712 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3713 | { |
26a4c0c6 TT |
3714 | struct super_block *sb = inode->i_sb; |
3715 | ext4_lblk_t first_block, stop_block; | |
3716 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3717 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3718 | handle_t *handle; |
3719 | unsigned int credits; | |
3720 | int ret = 0; | |
3721 | ||
a4bb6b64 | 3722 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3723 | return -EOPNOTSUPP; |
a4bb6b64 | 3724 | |
b8a86845 | 3725 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3726 | |
26a4c0c6 TT |
3727 | /* |
3728 | * Write out all dirty pages to avoid race conditions | |
3729 | * Then release them. | |
3730 | */ | |
3731 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3732 | ret = filemap_write_and_wait_range(mapping, offset, | |
3733 | offset + length - 1); | |
3734 | if (ret) | |
3735 | return ret; | |
3736 | } | |
3737 | ||
5955102c | 3738 | inode_lock(inode); |
9ef06cec | 3739 | |
26a4c0c6 TT |
3740 | /* No need to punch hole beyond i_size */ |
3741 | if (offset >= inode->i_size) | |
3742 | goto out_mutex; | |
3743 | ||
3744 | /* | |
3745 | * If the hole extends beyond i_size, set the hole | |
3746 | * to end after the page that contains i_size | |
3747 | */ | |
3748 | if (offset + length > inode->i_size) { | |
3749 | length = inode->i_size + | |
3750 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3751 | offset; | |
3752 | } | |
3753 | ||
a361293f JK |
3754 | if (offset & (sb->s_blocksize - 1) || |
3755 | (offset + length) & (sb->s_blocksize - 1)) { | |
3756 | /* | |
3757 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3758 | * partial block | |
3759 | */ | |
3760 | ret = ext4_inode_attach_jinode(inode); | |
3761 | if (ret < 0) | |
3762 | goto out_mutex; | |
3763 | ||
3764 | } | |
3765 | ||
ea3d7209 JK |
3766 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
3767 | ext4_inode_block_unlocked_dio(inode); | |
3768 | inode_dio_wait(inode); | |
3769 | ||
3770 | /* | |
3771 | * Prevent page faults from reinstantiating pages we have released from | |
3772 | * page cache. | |
3773 | */ | |
3774 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
3775 | first_block_offset = round_up(offset, sb->s_blocksize); |
3776 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3777 | |
a87dd18c | 3778 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
3779 | if (last_block_offset > first_block_offset) { |
3780 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
3781 | if (ret) | |
3782 | goto out_dio; | |
a87dd18c LC |
3783 | truncate_pagecache_range(inode, first_block_offset, |
3784 | last_block_offset); | |
01127848 | 3785 | } |
26a4c0c6 TT |
3786 | |
3787 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3788 | credits = ext4_writepage_trans_blocks(inode); | |
3789 | else | |
3790 | credits = ext4_blocks_for_truncate(inode); | |
3791 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3792 | if (IS_ERR(handle)) { | |
3793 | ret = PTR_ERR(handle); | |
3794 | ext4_std_error(sb, ret); | |
3795 | goto out_dio; | |
3796 | } | |
3797 | ||
a87dd18c LC |
3798 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3799 | length); | |
3800 | if (ret) | |
3801 | goto out_stop; | |
26a4c0c6 TT |
3802 | |
3803 | first_block = (offset + sb->s_blocksize - 1) >> | |
3804 | EXT4_BLOCK_SIZE_BITS(sb); | |
3805 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3806 | ||
3807 | /* If there are no blocks to remove, return now */ | |
3808 | if (first_block >= stop_block) | |
3809 | goto out_stop; | |
3810 | ||
3811 | down_write(&EXT4_I(inode)->i_data_sem); | |
3812 | ext4_discard_preallocations(inode); | |
3813 | ||
3814 | ret = ext4_es_remove_extent(inode, first_block, | |
3815 | stop_block - first_block); | |
3816 | if (ret) { | |
3817 | up_write(&EXT4_I(inode)->i_data_sem); | |
3818 | goto out_stop; | |
3819 | } | |
3820 | ||
3821 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3822 | ret = ext4_ext_remove_space(inode, first_block, | |
3823 | stop_block - 1); | |
3824 | else | |
4f579ae7 | 3825 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
3826 | stop_block); |
3827 | ||
819c4920 | 3828 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3829 | if (IS_SYNC(inode)) |
3830 | ext4_handle_sync(handle); | |
e251f9bc | 3831 | |
26a4c0c6 TT |
3832 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3833 | ext4_mark_inode_dirty(handle, inode); | |
3834 | out_stop: | |
3835 | ext4_journal_stop(handle); | |
3836 | out_dio: | |
ea3d7209 | 3837 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 TT |
3838 | ext4_inode_resume_unlocked_dio(inode); |
3839 | out_mutex: | |
5955102c | 3840 | inode_unlock(inode); |
26a4c0c6 | 3841 | return ret; |
a4bb6b64 AH |
3842 | } |
3843 | ||
a361293f JK |
3844 | int ext4_inode_attach_jinode(struct inode *inode) |
3845 | { | |
3846 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3847 | struct jbd2_inode *jinode; | |
3848 | ||
3849 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
3850 | return 0; | |
3851 | ||
3852 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
3853 | spin_lock(&inode->i_lock); | |
3854 | if (!ei->jinode) { | |
3855 | if (!jinode) { | |
3856 | spin_unlock(&inode->i_lock); | |
3857 | return -ENOMEM; | |
3858 | } | |
3859 | ei->jinode = jinode; | |
3860 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
3861 | jinode = NULL; | |
3862 | } | |
3863 | spin_unlock(&inode->i_lock); | |
3864 | if (unlikely(jinode != NULL)) | |
3865 | jbd2_free_inode(jinode); | |
3866 | return 0; | |
3867 | } | |
3868 | ||
ac27a0ec | 3869 | /* |
617ba13b | 3870 | * ext4_truncate() |
ac27a0ec | 3871 | * |
617ba13b MC |
3872 | * We block out ext4_get_block() block instantiations across the entire |
3873 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3874 | * simultaneously on behalf of the same inode. |
3875 | * | |
42b2aa86 | 3876 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3877 | * is one core, guiding principle: the file's tree must always be consistent on |
3878 | * disk. We must be able to restart the truncate after a crash. | |
3879 | * | |
3880 | * The file's tree may be transiently inconsistent in memory (although it | |
3881 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3882 | * the contents of (the filesystem + the journal) must be consistent and | |
3883 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3884 | * left-to-right works OK too). | |
3885 | * | |
3886 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3887 | * truncate against the orphan inode list. | |
3888 | * | |
3889 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3890 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3891 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3892 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3893 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3894 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3895 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3896 | */ |
617ba13b | 3897 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3898 | { |
819c4920 TT |
3899 | struct ext4_inode_info *ei = EXT4_I(inode); |
3900 | unsigned int credits; | |
3901 | handle_t *handle; | |
3902 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 3903 | |
19b5ef61 TT |
3904 | /* |
3905 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 3906 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
3907 | * have i_mutex locked because it's not necessary. |
3908 | */ | |
3909 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 3910 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
3911 | trace_ext4_truncate_enter(inode); |
3912 | ||
91ef4caf | 3913 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3914 | return; |
3915 | ||
12e9b892 | 3916 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3917 | |
5534fb5b | 3918 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3919 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3920 | |
aef1c851 TM |
3921 | if (ext4_has_inline_data(inode)) { |
3922 | int has_inline = 1; | |
3923 | ||
3924 | ext4_inline_data_truncate(inode, &has_inline); | |
3925 | if (has_inline) | |
3926 | return; | |
3927 | } | |
3928 | ||
a361293f JK |
3929 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
3930 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
3931 | if (ext4_inode_attach_jinode(inode) < 0) | |
3932 | return; | |
3933 | } | |
3934 | ||
819c4920 TT |
3935 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3936 | credits = ext4_writepage_trans_blocks(inode); | |
3937 | else | |
3938 | credits = ext4_blocks_for_truncate(inode); | |
3939 | ||
3940 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3941 | if (IS_ERR(handle)) { | |
3942 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3943 | return; | |
3944 | } | |
3945 | ||
eb3544c6 LC |
3946 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
3947 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
3948 | |
3949 | /* | |
3950 | * We add the inode to the orphan list, so that if this | |
3951 | * truncate spans multiple transactions, and we crash, we will | |
3952 | * resume the truncate when the filesystem recovers. It also | |
3953 | * marks the inode dirty, to catch the new size. | |
3954 | * | |
3955 | * Implication: the file must always be in a sane, consistent | |
3956 | * truncatable state while each transaction commits. | |
3957 | */ | |
3958 | if (ext4_orphan_add(handle, inode)) | |
3959 | goto out_stop; | |
3960 | ||
3961 | down_write(&EXT4_I(inode)->i_data_sem); | |
3962 | ||
3963 | ext4_discard_preallocations(inode); | |
3964 | ||
ff9893dc | 3965 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3966 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3967 | else |
819c4920 TT |
3968 | ext4_ind_truncate(handle, inode); |
3969 | ||
3970 | up_write(&ei->i_data_sem); | |
3971 | ||
3972 | if (IS_SYNC(inode)) | |
3973 | ext4_handle_sync(handle); | |
3974 | ||
3975 | out_stop: | |
3976 | /* | |
3977 | * If this was a simple ftruncate() and the file will remain alive, | |
3978 | * then we need to clear up the orphan record which we created above. | |
3979 | * However, if this was a real unlink then we were called by | |
58d86a50 | 3980 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
3981 | * orphan info for us. |
3982 | */ | |
3983 | if (inode->i_nlink) | |
3984 | ext4_orphan_del(handle, inode); | |
3985 | ||
3986 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3987 | ext4_mark_inode_dirty(handle, inode); | |
3988 | ext4_journal_stop(handle); | |
ac27a0ec | 3989 | |
0562e0ba | 3990 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3991 | } |
3992 | ||
ac27a0ec | 3993 | /* |
617ba13b | 3994 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3995 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3996 | * data in memory that is needed to recreate the on-disk version of this | |
3997 | * inode. | |
3998 | */ | |
617ba13b MC |
3999 | static int __ext4_get_inode_loc(struct inode *inode, |
4000 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4001 | { |
240799cd TT |
4002 | struct ext4_group_desc *gdp; |
4003 | struct buffer_head *bh; | |
4004 | struct super_block *sb = inode->i_sb; | |
4005 | ext4_fsblk_t block; | |
4006 | int inodes_per_block, inode_offset; | |
4007 | ||
3a06d778 | 4008 | iloc->bh = NULL; |
240799cd | 4009 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 4010 | return -EFSCORRUPTED; |
ac27a0ec | 4011 | |
240799cd TT |
4012 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4013 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4014 | if (!gdp) | |
ac27a0ec DK |
4015 | return -EIO; |
4016 | ||
240799cd TT |
4017 | /* |
4018 | * Figure out the offset within the block group inode table | |
4019 | */ | |
00d09882 | 4020 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4021 | inode_offset = ((inode->i_ino - 1) % |
4022 | EXT4_INODES_PER_GROUP(sb)); | |
4023 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4024 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4025 | ||
4026 | bh = sb_getblk(sb, block); | |
aebf0243 | 4027 | if (unlikely(!bh)) |
860d21e2 | 4028 | return -ENOMEM; |
ac27a0ec DK |
4029 | if (!buffer_uptodate(bh)) { |
4030 | lock_buffer(bh); | |
9c83a923 HK |
4031 | |
4032 | /* | |
4033 | * If the buffer has the write error flag, we have failed | |
4034 | * to write out another inode in the same block. In this | |
4035 | * case, we don't have to read the block because we may | |
4036 | * read the old inode data successfully. | |
4037 | */ | |
4038 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4039 | set_buffer_uptodate(bh); | |
4040 | ||
ac27a0ec DK |
4041 | if (buffer_uptodate(bh)) { |
4042 | /* someone brought it uptodate while we waited */ | |
4043 | unlock_buffer(bh); | |
4044 | goto has_buffer; | |
4045 | } | |
4046 | ||
4047 | /* | |
4048 | * If we have all information of the inode in memory and this | |
4049 | * is the only valid inode in the block, we need not read the | |
4050 | * block. | |
4051 | */ | |
4052 | if (in_mem) { | |
4053 | struct buffer_head *bitmap_bh; | |
240799cd | 4054 | int i, start; |
ac27a0ec | 4055 | |
240799cd | 4056 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4057 | |
240799cd TT |
4058 | /* Is the inode bitmap in cache? */ |
4059 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4060 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4061 | goto make_io; |
4062 | ||
4063 | /* | |
4064 | * If the inode bitmap isn't in cache then the | |
4065 | * optimisation may end up performing two reads instead | |
4066 | * of one, so skip it. | |
4067 | */ | |
4068 | if (!buffer_uptodate(bitmap_bh)) { | |
4069 | brelse(bitmap_bh); | |
4070 | goto make_io; | |
4071 | } | |
240799cd | 4072 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4073 | if (i == inode_offset) |
4074 | continue; | |
617ba13b | 4075 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4076 | break; |
4077 | } | |
4078 | brelse(bitmap_bh); | |
240799cd | 4079 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4080 | /* all other inodes are free, so skip I/O */ |
4081 | memset(bh->b_data, 0, bh->b_size); | |
4082 | set_buffer_uptodate(bh); | |
4083 | unlock_buffer(bh); | |
4084 | goto has_buffer; | |
4085 | } | |
4086 | } | |
4087 | ||
4088 | make_io: | |
240799cd TT |
4089 | /* |
4090 | * If we need to do any I/O, try to pre-readahead extra | |
4091 | * blocks from the inode table. | |
4092 | */ | |
4093 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4094 | ext4_fsblk_t b, end, table; | |
4095 | unsigned num; | |
0d606e2c | 4096 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4097 | |
4098 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4099 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4100 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4101 | if (table > b) |
4102 | b = table; | |
0d606e2c | 4103 | end = b + ra_blks; |
240799cd | 4104 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4105 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4106 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4107 | table += num / inodes_per_block; |
4108 | if (end > table) | |
4109 | end = table; | |
4110 | while (b <= end) | |
4111 | sb_breadahead(sb, b++); | |
4112 | } | |
4113 | ||
ac27a0ec DK |
4114 | /* |
4115 | * There are other valid inodes in the buffer, this inode | |
4116 | * has in-inode xattrs, or we don't have this inode in memory. | |
4117 | * Read the block from disk. | |
4118 | */ | |
0562e0ba | 4119 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4120 | get_bh(bh); |
4121 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 4122 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4123 | wait_on_buffer(bh); |
4124 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4125 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4126 | "unable to read itable block"); | |
ac27a0ec DK |
4127 | brelse(bh); |
4128 | return -EIO; | |
4129 | } | |
4130 | } | |
4131 | has_buffer: | |
4132 | iloc->bh = bh; | |
4133 | return 0; | |
4134 | } | |
4135 | ||
617ba13b | 4136 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4137 | { |
4138 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4139 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4140 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4141 | } |
4142 | ||
617ba13b | 4143 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4144 | { |
617ba13b | 4145 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4146 | unsigned int new_fl = 0; |
ac27a0ec | 4147 | |
617ba13b | 4148 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4149 | new_fl |= S_SYNC; |
617ba13b | 4150 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4151 | new_fl |= S_APPEND; |
617ba13b | 4152 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4153 | new_fl |= S_IMMUTABLE; |
617ba13b | 4154 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4155 | new_fl |= S_NOATIME; |
617ba13b | 4156 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4157 | new_fl |= S_DIRSYNC; |
923ae0ff RZ |
4158 | if (test_opt(inode->i_sb, DAX)) |
4159 | new_fl |= S_DAX; | |
5f16f322 | 4160 | inode_set_flags(inode, new_fl, |
923ae0ff | 4161 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
4162 | } |
4163 | ||
ff9ddf7e JK |
4164 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4165 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4166 | { | |
84a8dce2 DM |
4167 | unsigned int vfs_fl; |
4168 | unsigned long old_fl, new_fl; | |
4169 | ||
4170 | do { | |
4171 | vfs_fl = ei->vfs_inode.i_flags; | |
4172 | old_fl = ei->i_flags; | |
4173 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4174 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4175 | EXT4_DIRSYNC_FL); | |
4176 | if (vfs_fl & S_SYNC) | |
4177 | new_fl |= EXT4_SYNC_FL; | |
4178 | if (vfs_fl & S_APPEND) | |
4179 | new_fl |= EXT4_APPEND_FL; | |
4180 | if (vfs_fl & S_IMMUTABLE) | |
4181 | new_fl |= EXT4_IMMUTABLE_FL; | |
4182 | if (vfs_fl & S_NOATIME) | |
4183 | new_fl |= EXT4_NOATIME_FL; | |
4184 | if (vfs_fl & S_DIRSYNC) | |
4185 | new_fl |= EXT4_DIRSYNC_FL; | |
4186 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4187 | } |
de9a55b8 | 4188 | |
0fc1b451 | 4189 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4190 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4191 | { |
4192 | blkcnt_t i_blocks ; | |
8180a562 AK |
4193 | struct inode *inode = &(ei->vfs_inode); |
4194 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4195 | |
e2b911c5 | 4196 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4197 | /* we are using combined 48 bit field */ |
4198 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4199 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4200 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4201 | /* i_blocks represent file system block size */ |
4202 | return i_blocks << (inode->i_blkbits - 9); | |
4203 | } else { | |
4204 | return i_blocks; | |
4205 | } | |
0fc1b451 AK |
4206 | } else { |
4207 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4208 | } | |
4209 | } | |
ff9ddf7e | 4210 | |
152a7b0a TM |
4211 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4212 | struct ext4_inode *raw_inode, | |
4213 | struct ext4_inode_info *ei) | |
4214 | { | |
4215 | __le32 *magic = (void *)raw_inode + | |
4216 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4217 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4218 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4219 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4220 | } else |
4221 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4222 | } |
4223 | ||
040cb378 LX |
4224 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4225 | { | |
4226 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, EXT4_FEATURE_RO_COMPAT_PROJECT)) | |
4227 | return -EOPNOTSUPP; | |
4228 | *projid = EXT4_I(inode)->i_projid; | |
4229 | return 0; | |
4230 | } | |
4231 | ||
1d1fe1ee | 4232 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4233 | { |
617ba13b MC |
4234 | struct ext4_iloc iloc; |
4235 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4236 | struct ext4_inode_info *ei; |
1d1fe1ee | 4237 | struct inode *inode; |
b436b9be | 4238 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4239 | long ret; |
ac27a0ec | 4240 | int block; |
08cefc7a EB |
4241 | uid_t i_uid; |
4242 | gid_t i_gid; | |
040cb378 | 4243 | projid_t i_projid; |
ac27a0ec | 4244 | |
1d1fe1ee DH |
4245 | inode = iget_locked(sb, ino); |
4246 | if (!inode) | |
4247 | return ERR_PTR(-ENOMEM); | |
4248 | if (!(inode->i_state & I_NEW)) | |
4249 | return inode; | |
4250 | ||
4251 | ei = EXT4_I(inode); | |
7dc57615 | 4252 | iloc.bh = NULL; |
ac27a0ec | 4253 | |
1d1fe1ee DH |
4254 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4255 | if (ret < 0) | |
ac27a0ec | 4256 | goto bad_inode; |
617ba13b | 4257 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4258 | |
4259 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4260 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4261 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4262 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4263 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4264 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4265 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4266 | ret = -EFSCORRUPTED; |
814525f4 DW |
4267 | goto bad_inode; |
4268 | } | |
4269 | } else | |
4270 | ei->i_extra_isize = 0; | |
4271 | ||
4272 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4273 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4274 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4275 | __u32 csum; | |
4276 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4277 | __le32 gen = raw_inode->i_generation; | |
4278 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4279 | sizeof(inum)); | |
4280 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4281 | sizeof(gen)); | |
4282 | } | |
4283 | ||
4284 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4285 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4286 | ret = -EFSBADCRC; |
814525f4 DW |
4287 | goto bad_inode; |
4288 | } | |
4289 | ||
ac27a0ec | 4290 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4291 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4292 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
040cb378 LX |
4293 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) && |
4294 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
4295 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4296 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4297 | else | |
4298 | i_projid = EXT4_DEF_PROJID; | |
4299 | ||
af5bc92d | 4300 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4301 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4302 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4303 | } |
08cefc7a EB |
4304 | i_uid_write(inode, i_uid); |
4305 | i_gid_write(inode, i_gid); | |
040cb378 | 4306 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4307 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4308 | |
353eb83c | 4309 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4310 | ei->i_inline_off = 0; |
ac27a0ec DK |
4311 | ei->i_dir_start_lookup = 0; |
4312 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4313 | /* We now have enough fields to check if the inode was active or not. | |
4314 | * This is needed because nfsd might try to access dead inodes | |
4315 | * the test is that same one that e2fsck uses | |
4316 | * NeilBrown 1999oct15 | |
4317 | */ | |
4318 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4319 | if ((inode->i_mode == 0 || |
4320 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4321 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4322 | /* this inode is deleted */ |
1d1fe1ee | 4323 | ret = -ESTALE; |
ac27a0ec DK |
4324 | goto bad_inode; |
4325 | } | |
4326 | /* The only unlinked inodes we let through here have | |
4327 | * valid i_mode and are being read by the orphan | |
4328 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4329 | * the process of deleting those. |
4330 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4331 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4332 | } |
ac27a0ec | 4333 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4334 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4335 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4336 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4337 | ei->i_file_acl |= |
4338 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4339 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4340 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4341 | #ifdef CONFIG_QUOTA |
4342 | ei->i_reserved_quota = 0; | |
4343 | #endif | |
ac27a0ec DK |
4344 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4345 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4346 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4347 | /* |
4348 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4349 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4350 | */ | |
617ba13b | 4351 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4352 | ei->i_data[block] = raw_inode->i_block[block]; |
4353 | INIT_LIST_HEAD(&ei->i_orphan); | |
4354 | ||
b436b9be JK |
4355 | /* |
4356 | * Set transaction id's of transactions that have to be committed | |
4357 | * to finish f[data]sync. We set them to currently running transaction | |
4358 | * as we cannot be sure that the inode or some of its metadata isn't | |
4359 | * part of the transaction - the inode could have been reclaimed and | |
4360 | * now it is reread from disk. | |
4361 | */ | |
4362 | if (journal) { | |
4363 | transaction_t *transaction; | |
4364 | tid_t tid; | |
4365 | ||
a931da6a | 4366 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4367 | if (journal->j_running_transaction) |
4368 | transaction = journal->j_running_transaction; | |
4369 | else | |
4370 | transaction = journal->j_committing_transaction; | |
4371 | if (transaction) | |
4372 | tid = transaction->t_tid; | |
4373 | else | |
4374 | tid = journal->j_commit_sequence; | |
a931da6a | 4375 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4376 | ei->i_sync_tid = tid; |
4377 | ei->i_datasync_tid = tid; | |
4378 | } | |
4379 | ||
0040d987 | 4380 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4381 | if (ei->i_extra_isize == 0) { |
4382 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4383 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4384 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4385 | } else { |
152a7b0a | 4386 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4387 | } |
814525f4 | 4388 | } |
ac27a0ec | 4389 | |
ef7f3835 KS |
4390 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4391 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4392 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4393 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4394 | ||
ed3654eb | 4395 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4396 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4397 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4398 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4399 | inode->i_version |= | |
4400 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4401 | } | |
25ec56b5 JNC |
4402 | } |
4403 | ||
c4b5a614 | 4404 | ret = 0; |
485c26ec | 4405 | if (ei->i_file_acl && |
1032988c | 4406 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4407 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4408 | ei->i_file_acl); | |
6a797d27 | 4409 | ret = -EFSCORRUPTED; |
485c26ec | 4410 | goto bad_inode; |
f19d5870 TM |
4411 | } else if (!ext4_has_inline_data(inode)) { |
4412 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4413 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4414 | (S_ISLNK(inode->i_mode) && | |
4415 | !ext4_inode_is_fast_symlink(inode)))) | |
4416 | /* Validate extent which is part of inode */ | |
4417 | ret = ext4_ext_check_inode(inode); | |
4418 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4419 | (S_ISLNK(inode->i_mode) && | |
4420 | !ext4_inode_is_fast_symlink(inode))) { | |
4421 | /* Validate block references which are part of inode */ | |
4422 | ret = ext4_ind_check_inode(inode); | |
4423 | } | |
fe2c8191 | 4424 | } |
567f3e9a | 4425 | if (ret) |
de9a55b8 | 4426 | goto bad_inode; |
7a262f7c | 4427 | |
ac27a0ec | 4428 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4429 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4430 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4431 | ext4_set_aops(inode); |
ac27a0ec | 4432 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4433 | inode->i_op = &ext4_dir_inode_operations; |
4434 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4435 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4436 | if (ext4_encrypted_inode(inode)) { |
4437 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4438 | ext4_set_aops(inode); | |
4439 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4440 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4441 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4442 | nd_terminate_link(ei->i_data, inode->i_size, |
4443 | sizeof(ei->i_data) - 1); | |
4444 | } else { | |
617ba13b MC |
4445 | inode->i_op = &ext4_symlink_inode_operations; |
4446 | ext4_set_aops(inode); | |
ac27a0ec | 4447 | } |
21fc61c7 | 4448 | inode_nohighmem(inode); |
563bdd61 TT |
4449 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4450 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4451 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4452 | if (raw_inode->i_block[0]) |
4453 | init_special_inode(inode, inode->i_mode, | |
4454 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4455 | else | |
4456 | init_special_inode(inode, inode->i_mode, | |
4457 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4458 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4459 | make_bad_inode(inode); | |
563bdd61 | 4460 | } else { |
6a797d27 | 4461 | ret = -EFSCORRUPTED; |
24676da4 | 4462 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4463 | goto bad_inode; |
ac27a0ec | 4464 | } |
af5bc92d | 4465 | brelse(iloc.bh); |
617ba13b | 4466 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4467 | unlock_new_inode(inode); |
4468 | return inode; | |
ac27a0ec DK |
4469 | |
4470 | bad_inode: | |
567f3e9a | 4471 | brelse(iloc.bh); |
1d1fe1ee DH |
4472 | iget_failed(inode); |
4473 | return ERR_PTR(ret); | |
ac27a0ec DK |
4474 | } |
4475 | ||
f4bb2981 TT |
4476 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4477 | { | |
4478 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4479 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4480 | return ext4_iget(sb, ino); |
4481 | } | |
4482 | ||
0fc1b451 AK |
4483 | static int ext4_inode_blocks_set(handle_t *handle, |
4484 | struct ext4_inode *raw_inode, | |
4485 | struct ext4_inode_info *ei) | |
4486 | { | |
4487 | struct inode *inode = &(ei->vfs_inode); | |
4488 | u64 i_blocks = inode->i_blocks; | |
4489 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4490 | |
4491 | if (i_blocks <= ~0U) { | |
4492 | /* | |
4907cb7b | 4493 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4494 | * as multiple of 512 bytes |
4495 | */ | |
8180a562 | 4496 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4497 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4498 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4499 | return 0; |
4500 | } | |
e2b911c5 | 4501 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4502 | return -EFBIG; |
4503 | ||
4504 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4505 | /* |
4506 | * i_blocks can be represented in a 48 bit variable | |
4507 | * as multiple of 512 bytes | |
4508 | */ | |
8180a562 | 4509 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4510 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4511 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4512 | } else { |
84a8dce2 | 4513 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4514 | /* i_block is stored in file system block size */ |
4515 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4516 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4517 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4518 | } |
f287a1a5 | 4519 | return 0; |
0fc1b451 AK |
4520 | } |
4521 | ||
a26f4992 TT |
4522 | struct other_inode { |
4523 | unsigned long orig_ino; | |
4524 | struct ext4_inode *raw_inode; | |
4525 | }; | |
4526 | ||
4527 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4528 | void *data) | |
4529 | { | |
4530 | struct other_inode *oi = (struct other_inode *) data; | |
4531 | ||
4532 | if ((inode->i_ino != ino) || | |
4533 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4534 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4535 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4536 | return 0; | |
4537 | spin_lock(&inode->i_lock); | |
4538 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4539 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4540 | (inode->i_state & I_DIRTY_TIME)) { | |
4541 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4542 | ||
4543 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4544 | spin_unlock(&inode->i_lock); | |
4545 | ||
4546 | spin_lock(&ei->i_raw_lock); | |
4547 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4548 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4549 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4550 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4551 | spin_unlock(&ei->i_raw_lock); | |
4552 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4553 | return -1; | |
4554 | } | |
4555 | spin_unlock(&inode->i_lock); | |
4556 | return -1; | |
4557 | } | |
4558 | ||
4559 | /* | |
4560 | * Opportunistically update the other time fields for other inodes in | |
4561 | * the same inode table block. | |
4562 | */ | |
4563 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4564 | unsigned long orig_ino, char *buf) | |
4565 | { | |
4566 | struct other_inode oi; | |
4567 | unsigned long ino; | |
4568 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4569 | int inode_size = EXT4_INODE_SIZE(sb); | |
4570 | ||
4571 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
4572 | /* |
4573 | * Calculate the first inode in the inode table block. Inode | |
4574 | * numbers are one-based. That is, the first inode in a block | |
4575 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4576 | */ | |
4577 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
4578 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4579 | if (ino == orig_ino) | |
4580 | continue; | |
4581 | oi.raw_inode = (struct ext4_inode *) buf; | |
4582 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4583 | } | |
4584 | } | |
4585 | ||
ac27a0ec DK |
4586 | /* |
4587 | * Post the struct inode info into an on-disk inode location in the | |
4588 | * buffer-cache. This gobbles the caller's reference to the | |
4589 | * buffer_head in the inode location struct. | |
4590 | * | |
4591 | * The caller must have write access to iloc->bh. | |
4592 | */ | |
617ba13b | 4593 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4594 | struct inode *inode, |
830156c7 | 4595 | struct ext4_iloc *iloc) |
ac27a0ec | 4596 | { |
617ba13b MC |
4597 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4598 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4599 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4600 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4601 | int err = 0, rc, block; |
202ee5df | 4602 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4603 | uid_t i_uid; |
4604 | gid_t i_gid; | |
040cb378 | 4605 | projid_t i_projid; |
ac27a0ec | 4606 | |
202ee5df TT |
4607 | spin_lock(&ei->i_raw_lock); |
4608 | ||
4609 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4610 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4611 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4612 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4613 | |
ff9ddf7e | 4614 | ext4_get_inode_flags(ei); |
ac27a0ec | 4615 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4616 | i_uid = i_uid_read(inode); |
4617 | i_gid = i_gid_read(inode); | |
040cb378 | 4618 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 4619 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4620 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4621 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4622 | /* |
4623 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4624 | * re-used with the upper 16 bits of the uid/gid intact | |
4625 | */ | |
af5bc92d | 4626 | if (!ei->i_dtime) { |
ac27a0ec | 4627 | raw_inode->i_uid_high = |
08cefc7a | 4628 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4629 | raw_inode->i_gid_high = |
08cefc7a | 4630 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4631 | } else { |
4632 | raw_inode->i_uid_high = 0; | |
4633 | raw_inode->i_gid_high = 0; | |
4634 | } | |
4635 | } else { | |
08cefc7a EB |
4636 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4637 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4638 | raw_inode->i_uid_high = 0; |
4639 | raw_inode->i_gid_high = 0; | |
4640 | } | |
4641 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4642 | |
4643 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4644 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4645 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4646 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4647 | ||
bce92d56 LX |
4648 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4649 | if (err) { | |
202ee5df | 4650 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4651 | goto out_brelse; |
202ee5df | 4652 | } |
ac27a0ec | 4653 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4654 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4655 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4656 | raw_inode->i_file_acl_high = |
4657 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4658 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4659 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4660 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4661 | need_datasync = 1; | |
4662 | } | |
a48380f7 | 4663 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 4664 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 4665 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
4666 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4667 | set_large_file = 1; | |
ac27a0ec DK |
4668 | } |
4669 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4670 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4671 | if (old_valid_dev(inode->i_rdev)) { | |
4672 | raw_inode->i_block[0] = | |
4673 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4674 | raw_inode->i_block[1] = 0; | |
4675 | } else { | |
4676 | raw_inode->i_block[0] = 0; | |
4677 | raw_inode->i_block[1] = | |
4678 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4679 | raw_inode->i_block[2] = 0; | |
4680 | } | |
f19d5870 | 4681 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4682 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4683 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4684 | } |
ac27a0ec | 4685 | |
ed3654eb | 4686 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4687 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4688 | if (ei->i_extra_isize) { | |
4689 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4690 | raw_inode->i_version_hi = | |
4691 | cpu_to_le32(inode->i_version >> 32); | |
4692 | raw_inode->i_extra_isize = | |
4693 | cpu_to_le16(ei->i_extra_isize); | |
4694 | } | |
25ec56b5 | 4695 | } |
040cb378 LX |
4696 | |
4697 | BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
4698 | EXT4_FEATURE_RO_COMPAT_PROJECT) && | |
4699 | i_projid != EXT4_DEF_PROJID); | |
4700 | ||
4701 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
4702 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4703 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
4704 | ||
814525f4 | 4705 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 4706 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
4707 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
4708 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
4709 | bh->b_data); | |
202ee5df | 4710 | |
830156c7 | 4711 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4712 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4713 | if (!err) |
4714 | err = rc; | |
19f5fb7a | 4715 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 4716 | if (set_large_file) { |
5d601255 | 4717 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4718 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4719 | if (err) | |
4720 | goto out_brelse; | |
4721 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 4722 | ext4_set_feature_large_file(sb); |
202ee5df TT |
4723 | ext4_handle_sync(handle); |
4724 | err = ext4_handle_dirty_super(handle, sb); | |
4725 | } | |
b71fc079 | 4726 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4727 | out_brelse: |
af5bc92d | 4728 | brelse(bh); |
617ba13b | 4729 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4730 | return err; |
4731 | } | |
4732 | ||
4733 | /* | |
617ba13b | 4734 | * ext4_write_inode() |
ac27a0ec DK |
4735 | * |
4736 | * We are called from a few places: | |
4737 | * | |
87f7e416 | 4738 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4739 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4740 | * transaction to commit. |
ac27a0ec | 4741 | * |
87f7e416 TT |
4742 | * - Within flush work (sys_sync(), kupdate and such). |
4743 | * We wait on commit, if told to. | |
ac27a0ec | 4744 | * |
87f7e416 TT |
4745 | * - Within iput_final() -> write_inode_now() |
4746 | * We wait on commit, if told to. | |
ac27a0ec DK |
4747 | * |
4748 | * In all cases it is actually safe for us to return without doing anything, | |
4749 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4750 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4751 | * writeback. | |
ac27a0ec DK |
4752 | * |
4753 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4754 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4755 | * which we are interested. | |
4756 | * | |
4757 | * It would be a bug for them to not do this. The code: | |
4758 | * | |
4759 | * mark_inode_dirty(inode) | |
4760 | * stuff(); | |
4761 | * inode->i_size = expr; | |
4762 | * | |
87f7e416 TT |
4763 | * is in error because write_inode() could occur while `stuff()' is running, |
4764 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4765 | * superblock's dirty inode list. | |
ac27a0ec | 4766 | */ |
a9185b41 | 4767 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4768 | { |
91ac6f43 FM |
4769 | int err; |
4770 | ||
87f7e416 | 4771 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4772 | return 0; |
4773 | ||
91ac6f43 FM |
4774 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4775 | if (ext4_journal_current_handle()) { | |
4776 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4777 | dump_stack(); | |
4778 | return -EIO; | |
4779 | } | |
ac27a0ec | 4780 | |
10542c22 JK |
4781 | /* |
4782 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4783 | * ext4_sync_fs() will force the commit after everything is | |
4784 | * written. | |
4785 | */ | |
4786 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4787 | return 0; |
4788 | ||
4789 | err = ext4_force_commit(inode->i_sb); | |
4790 | } else { | |
4791 | struct ext4_iloc iloc; | |
ac27a0ec | 4792 | |
8b472d73 | 4793 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4794 | if (err) |
4795 | return err; | |
10542c22 JK |
4796 | /* |
4797 | * sync(2) will flush the whole buffer cache. No need to do | |
4798 | * it here separately for each inode. | |
4799 | */ | |
4800 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4801 | sync_dirty_buffer(iloc.bh); |
4802 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4803 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4804 | "IO error syncing inode"); | |
830156c7 FM |
4805 | err = -EIO; |
4806 | } | |
fd2dd9fb | 4807 | brelse(iloc.bh); |
91ac6f43 FM |
4808 | } |
4809 | return err; | |
ac27a0ec DK |
4810 | } |
4811 | ||
53e87268 JK |
4812 | /* |
4813 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4814 | * buffers that are attached to a page stradding i_size and are undergoing | |
4815 | * commit. In that case we have to wait for commit to finish and try again. | |
4816 | */ | |
4817 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4818 | { | |
4819 | struct page *page; | |
4820 | unsigned offset; | |
4821 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4822 | tid_t commit_tid = 0; | |
4823 | int ret; | |
4824 | ||
4825 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4826 | /* | |
4827 | * All buffers in the last page remain valid? Then there's nothing to | |
4828 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4829 | * blocksize case | |
4830 | */ | |
4831 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4832 | return; | |
4833 | while (1) { | |
4834 | page = find_lock_page(inode->i_mapping, | |
4835 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4836 | if (!page) | |
4837 | return; | |
ca99fdd2 LC |
4838 | ret = __ext4_journalled_invalidatepage(page, offset, |
4839 | PAGE_CACHE_SIZE - offset); | |
53e87268 JK |
4840 | unlock_page(page); |
4841 | page_cache_release(page); | |
4842 | if (ret != -EBUSY) | |
4843 | return; | |
4844 | commit_tid = 0; | |
4845 | read_lock(&journal->j_state_lock); | |
4846 | if (journal->j_committing_transaction) | |
4847 | commit_tid = journal->j_committing_transaction->t_tid; | |
4848 | read_unlock(&journal->j_state_lock); | |
4849 | if (commit_tid) | |
4850 | jbd2_log_wait_commit(journal, commit_tid); | |
4851 | } | |
4852 | } | |
4853 | ||
ac27a0ec | 4854 | /* |
617ba13b | 4855 | * ext4_setattr() |
ac27a0ec DK |
4856 | * |
4857 | * Called from notify_change. | |
4858 | * | |
4859 | * We want to trap VFS attempts to truncate the file as soon as | |
4860 | * possible. In particular, we want to make sure that when the VFS | |
4861 | * shrinks i_size, we put the inode on the orphan list and modify | |
4862 | * i_disksize immediately, so that during the subsequent flushing of | |
4863 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4864 | * commit will leave the blocks being flushed in an unused state on | |
4865 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4866 | * be freed, so we have a strong guarantee that no future commit will | |
4867 | * leave these blocks visible to the user.) | |
4868 | * | |
678aaf48 JK |
4869 | * Another thing we have to assure is that if we are in ordered mode |
4870 | * and inode is still attached to the committing transaction, we must | |
4871 | * we start writeout of all the dirty pages which are being truncated. | |
4872 | * This way we are sure that all the data written in the previous | |
4873 | * transaction are already on disk (truncate waits for pages under | |
4874 | * writeback). | |
4875 | * | |
4876 | * Called with inode->i_mutex down. | |
ac27a0ec | 4877 | */ |
617ba13b | 4878 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 4879 | { |
2b0143b5 | 4880 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 4881 | int error, rc = 0; |
3d287de3 | 4882 | int orphan = 0; |
ac27a0ec DK |
4883 | const unsigned int ia_valid = attr->ia_valid; |
4884 | ||
4885 | error = inode_change_ok(inode, attr); | |
4886 | if (error) | |
4887 | return error; | |
4888 | ||
a7cdadee JK |
4889 | if (is_quota_modification(inode, attr)) { |
4890 | error = dquot_initialize(inode); | |
4891 | if (error) | |
4892 | return error; | |
4893 | } | |
08cefc7a EB |
4894 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4895 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4896 | handle_t *handle; |
4897 | ||
4898 | /* (user+group)*(old+new) structure, inode write (sb, | |
4899 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4900 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4901 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4902 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4903 | if (IS_ERR(handle)) { |
4904 | error = PTR_ERR(handle); | |
4905 | goto err_out; | |
4906 | } | |
b43fa828 | 4907 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4908 | if (error) { |
617ba13b | 4909 | ext4_journal_stop(handle); |
ac27a0ec DK |
4910 | return error; |
4911 | } | |
4912 | /* Update corresponding info in inode so that everything is in | |
4913 | * one transaction */ | |
4914 | if (attr->ia_valid & ATTR_UID) | |
4915 | inode->i_uid = attr->ia_uid; | |
4916 | if (attr->ia_valid & ATTR_GID) | |
4917 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4918 | error = ext4_mark_inode_dirty(handle, inode); |
4919 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4920 | } |
4921 | ||
3da40c7b | 4922 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 4923 | handle_t *handle; |
3da40c7b JB |
4924 | loff_t oldsize = inode->i_size; |
4925 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 4926 | |
12e9b892 | 4927 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4928 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4929 | ||
0c095c7f TT |
4930 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4931 | return -EFBIG; | |
e2b46574 | 4932 | } |
3da40c7b JB |
4933 | if (!S_ISREG(inode->i_mode)) |
4934 | return -EINVAL; | |
dff6efc3 CH |
4935 | |
4936 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
4937 | inode_inc_iversion(inode); | |
4938 | ||
3da40c7b | 4939 | if (ext4_should_order_data(inode) && |
5208386c | 4940 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 4941 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 4942 | attr->ia_size); |
3da40c7b JB |
4943 | if (error) |
4944 | goto err_out; | |
4945 | } | |
4946 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
4947 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
4948 | if (IS_ERR(handle)) { | |
4949 | error = PTR_ERR(handle); | |
4950 | goto err_out; | |
4951 | } | |
3da40c7b | 4952 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
4953 | error = ext4_orphan_add(handle, inode); |
4954 | orphan = 1; | |
4955 | } | |
911af577 EG |
4956 | /* |
4957 | * Update c/mtime on truncate up, ext4_truncate() will | |
4958 | * update c/mtime in shrink case below | |
4959 | */ | |
4960 | if (!shrink) { | |
4961 | inode->i_mtime = ext4_current_time(inode); | |
4962 | inode->i_ctime = inode->i_mtime; | |
4963 | } | |
90e775b7 | 4964 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
4965 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4966 | rc = ext4_mark_inode_dirty(handle, inode); | |
4967 | if (!error) | |
4968 | error = rc; | |
90e775b7 JK |
4969 | /* |
4970 | * We have to update i_size under i_data_sem together | |
4971 | * with i_disksize to avoid races with writeback code | |
4972 | * running ext4_wb_update_i_disksize(). | |
4973 | */ | |
4974 | if (!error) | |
4975 | i_size_write(inode, attr->ia_size); | |
4976 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
4977 | ext4_journal_stop(handle); |
4978 | if (error) { | |
3da40c7b JB |
4979 | if (orphan) |
4980 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
4981 | goto err_out; |
4982 | } | |
d6320cbf | 4983 | } |
3da40c7b JB |
4984 | if (!shrink) |
4985 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 4986 | |
5208386c JK |
4987 | /* |
4988 | * Blocks are going to be removed from the inode. Wait | |
4989 | * for dio in flight. Temporarily disable | |
4990 | * dioread_nolock to prevent livelock. | |
4991 | */ | |
4992 | if (orphan) { | |
4993 | if (!ext4_should_journal_data(inode)) { | |
4994 | ext4_inode_block_unlocked_dio(inode); | |
4995 | inode_dio_wait(inode); | |
4996 | ext4_inode_resume_unlocked_dio(inode); | |
4997 | } else | |
4998 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 4999 | } |
ea3d7209 | 5000 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
5001 | /* |
5002 | * Truncate pagecache after we've waited for commit | |
5003 | * in data=journal mode to make pages freeable. | |
5004 | */ | |
923ae0ff | 5005 | truncate_pagecache(inode, inode->i_size); |
3da40c7b JB |
5006 | if (shrink) |
5007 | ext4_truncate(inode); | |
ea3d7209 | 5008 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5009 | } |
ac27a0ec | 5010 | |
1025774c CH |
5011 | if (!rc) { |
5012 | setattr_copy(inode, attr); | |
5013 | mark_inode_dirty(inode); | |
5014 | } | |
5015 | ||
5016 | /* | |
5017 | * If the call to ext4_truncate failed to get a transaction handle at | |
5018 | * all, we need to clean up the in-core orphan list manually. | |
5019 | */ | |
3d287de3 | 5020 | if (orphan && inode->i_nlink) |
617ba13b | 5021 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5022 | |
5023 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 5024 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5025 | |
5026 | err_out: | |
617ba13b | 5027 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5028 | if (!error) |
5029 | error = rc; | |
5030 | return error; | |
5031 | } | |
5032 | ||
3e3398a0 MC |
5033 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5034 | struct kstat *stat) | |
5035 | { | |
5036 | struct inode *inode; | |
8af8eecc | 5037 | unsigned long long delalloc_blocks; |
3e3398a0 | 5038 | |
2b0143b5 | 5039 | inode = d_inode(dentry); |
3e3398a0 MC |
5040 | generic_fillattr(inode, stat); |
5041 | ||
9206c561 AD |
5042 | /* |
5043 | * If there is inline data in the inode, the inode will normally not | |
5044 | * have data blocks allocated (it may have an external xattr block). | |
5045 | * Report at least one sector for such files, so tools like tar, rsync, | |
5046 | * others doen't incorrectly think the file is completely sparse. | |
5047 | */ | |
5048 | if (unlikely(ext4_has_inline_data(inode))) | |
5049 | stat->blocks += (stat->size + 511) >> 9; | |
5050 | ||
3e3398a0 MC |
5051 | /* |
5052 | * We can't update i_blocks if the block allocation is delayed | |
5053 | * otherwise in the case of system crash before the real block | |
5054 | * allocation is done, we will have i_blocks inconsistent with | |
5055 | * on-disk file blocks. | |
5056 | * We always keep i_blocks updated together with real | |
5057 | * allocation. But to not confuse with user, stat | |
5058 | * will return the blocks that include the delayed allocation | |
5059 | * blocks for this file. | |
5060 | */ | |
96607551 | 5061 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5062 | EXT4_I(inode)->i_reserved_data_blocks); |
5063 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5064 | return 0; |
5065 | } | |
ac27a0ec | 5066 | |
fffb2739 JK |
5067 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5068 | int pextents) | |
a02908f1 | 5069 | { |
12e9b892 | 5070 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5071 | return ext4_ind_trans_blocks(inode, lblocks); |
5072 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5073 | } |
ac51d837 | 5074 | |
ac27a0ec | 5075 | /* |
a02908f1 MC |
5076 | * Account for index blocks, block groups bitmaps and block group |
5077 | * descriptor blocks if modify datablocks and index blocks | |
5078 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5079 | * |
a02908f1 | 5080 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5081 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5082 | * they could still across block group boundary. |
ac27a0ec | 5083 | * |
a02908f1 MC |
5084 | * Also account for superblock, inode, quota and xattr blocks |
5085 | */ | |
fffb2739 JK |
5086 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
5087 | int pextents) | |
a02908f1 | 5088 | { |
8df9675f TT |
5089 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5090 | int gdpblocks; | |
a02908f1 MC |
5091 | int idxblocks; |
5092 | int ret = 0; | |
5093 | ||
5094 | /* | |
fffb2739 JK |
5095 | * How many index blocks need to touch to map @lblocks logical blocks |
5096 | * to @pextents physical extents? | |
a02908f1 | 5097 | */ |
fffb2739 | 5098 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5099 | |
5100 | ret = idxblocks; | |
5101 | ||
5102 | /* | |
5103 | * Now let's see how many group bitmaps and group descriptors need | |
5104 | * to account | |
5105 | */ | |
fffb2739 | 5106 | groups = idxblocks + pextents; |
a02908f1 | 5107 | gdpblocks = groups; |
8df9675f TT |
5108 | if (groups > ngroups) |
5109 | groups = ngroups; | |
a02908f1 MC |
5110 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5111 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5112 | ||
5113 | /* bitmaps and block group descriptor blocks */ | |
5114 | ret += groups + gdpblocks; | |
5115 | ||
5116 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5117 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5118 | ||
5119 | return ret; | |
5120 | } | |
5121 | ||
5122 | /* | |
25985edc | 5123 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5124 | * the modification of a single pages into a single transaction, |
5125 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5126 | * |
525f4ed8 | 5127 | * This could be called via ext4_write_begin() |
ac27a0ec | 5128 | * |
525f4ed8 | 5129 | * We need to consider the worse case, when |
a02908f1 | 5130 | * one new block per extent. |
ac27a0ec | 5131 | */ |
a86c6181 | 5132 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5133 | { |
617ba13b | 5134 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5135 | int ret; |
5136 | ||
fffb2739 | 5137 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5138 | |
a02908f1 | 5139 | /* Account for data blocks for journalled mode */ |
617ba13b | 5140 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5141 | ret += bpp; |
ac27a0ec DK |
5142 | return ret; |
5143 | } | |
f3bd1f3f MC |
5144 | |
5145 | /* | |
5146 | * Calculate the journal credits for a chunk of data modification. | |
5147 | * | |
5148 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5149 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5150 | * |
5151 | * journal buffers for data blocks are not included here, as DIO | |
5152 | * and fallocate do no need to journal data buffers. | |
5153 | */ | |
5154 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5155 | { | |
5156 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5157 | } | |
5158 | ||
ac27a0ec | 5159 | /* |
617ba13b | 5160 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5161 | * Give this, we know that the caller already has write access to iloc->bh. |
5162 | */ | |
617ba13b | 5163 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5164 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5165 | { |
5166 | int err = 0; | |
5167 | ||
c64db50e | 5168 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5169 | inode_inc_iversion(inode); |
5170 | ||
ac27a0ec DK |
5171 | /* the do_update_inode consumes one bh->b_count */ |
5172 | get_bh(iloc->bh); | |
5173 | ||
dab291af | 5174 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5175 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5176 | put_bh(iloc->bh); |
5177 | return err; | |
5178 | } | |
5179 | ||
5180 | /* | |
5181 | * On success, We end up with an outstanding reference count against | |
5182 | * iloc->bh. This _must_ be cleaned up later. | |
5183 | */ | |
5184 | ||
5185 | int | |
617ba13b MC |
5186 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5187 | struct ext4_iloc *iloc) | |
ac27a0ec | 5188 | { |
0390131b FM |
5189 | int err; |
5190 | ||
5191 | err = ext4_get_inode_loc(inode, iloc); | |
5192 | if (!err) { | |
5193 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5194 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5195 | if (err) { | |
5196 | brelse(iloc->bh); | |
5197 | iloc->bh = NULL; | |
ac27a0ec DK |
5198 | } |
5199 | } | |
617ba13b | 5200 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5201 | return err; |
5202 | } | |
5203 | ||
6dd4ee7c KS |
5204 | /* |
5205 | * Expand an inode by new_extra_isize bytes. | |
5206 | * Returns 0 on success or negative error number on failure. | |
5207 | */ | |
1d03ec98 AK |
5208 | static int ext4_expand_extra_isize(struct inode *inode, |
5209 | unsigned int new_extra_isize, | |
5210 | struct ext4_iloc iloc, | |
5211 | handle_t *handle) | |
6dd4ee7c KS |
5212 | { |
5213 | struct ext4_inode *raw_inode; | |
5214 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5215 | |
5216 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5217 | return 0; | |
5218 | ||
5219 | raw_inode = ext4_raw_inode(&iloc); | |
5220 | ||
5221 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5222 | |
5223 | /* No extended attributes present */ | |
19f5fb7a TT |
5224 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5225 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5226 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5227 | new_extra_isize); | |
5228 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5229 | return 0; | |
5230 | } | |
5231 | ||
5232 | /* try to expand with EAs present */ | |
5233 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5234 | raw_inode, handle); | |
5235 | } | |
5236 | ||
ac27a0ec DK |
5237 | /* |
5238 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5239 | * dirtiness (it may still be data-dirty). | |
5240 | * This means that the in-core inode may be reaped by prune_icache | |
5241 | * without having to perform any I/O. This is a very good thing, | |
5242 | * because *any* task may call prune_icache - even ones which | |
5243 | * have a transaction open against a different journal. | |
5244 | * | |
5245 | * Is this cheating? Not really. Sure, we haven't written the | |
5246 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5247 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5248 | * we start and wait on commits. | |
ac27a0ec | 5249 | */ |
617ba13b | 5250 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5251 | { |
617ba13b | 5252 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5253 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5254 | static unsigned int mnt_count; | |
5255 | int err, ret; | |
ac27a0ec DK |
5256 | |
5257 | might_sleep(); | |
7ff9c073 | 5258 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5259 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5260 | if (ext4_handle_valid(handle) && |
5261 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5262 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5263 | /* |
5264 | * We need extra buffer credits since we may write into EA block | |
5265 | * with this same handle. If journal_extend fails, then it will | |
5266 | * only result in a minor loss of functionality for that inode. | |
5267 | * If this is felt to be critical, then e2fsck should be run to | |
5268 | * force a large enough s_min_extra_isize. | |
5269 | */ | |
5270 | if ((jbd2_journal_extend(handle, | |
5271 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5272 | ret = ext4_expand_extra_isize(inode, | |
5273 | sbi->s_want_extra_isize, | |
5274 | iloc, handle); | |
5275 | if (ret) { | |
19f5fb7a TT |
5276 | ext4_set_inode_state(inode, |
5277 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5278 | if (mnt_count != |
5279 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5280 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5281 | "Unable to expand inode %lu. Delete" |
5282 | " some EAs or run e2fsck.", | |
5283 | inode->i_ino); | |
c1bddad9 AK |
5284 | mnt_count = |
5285 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5286 | } |
5287 | } | |
5288 | } | |
5289 | } | |
ac27a0ec | 5290 | if (!err) |
617ba13b | 5291 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5292 | return err; |
5293 | } | |
5294 | ||
5295 | /* | |
617ba13b | 5296 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5297 | * |
5298 | * We're really interested in the case where a file is being extended. | |
5299 | * i_size has been changed by generic_commit_write() and we thus need | |
5300 | * to include the updated inode in the current transaction. | |
5301 | * | |
5dd4056d | 5302 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5303 | * are allocated to the file. |
5304 | * | |
5305 | * If the inode is marked synchronous, we don't honour that here - doing | |
5306 | * so would cause a commit on atime updates, which we don't bother doing. | |
5307 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5308 | * |
5309 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5310 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5311 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5312 | */ |
aa385729 | 5313 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5314 | { |
ac27a0ec DK |
5315 | handle_t *handle; |
5316 | ||
0ae45f63 TT |
5317 | if (flags == I_DIRTY_TIME) |
5318 | return; | |
9924a92a | 5319 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5320 | if (IS_ERR(handle)) |
5321 | goto out; | |
f3dc272f | 5322 | |
f3dc272f CW |
5323 | ext4_mark_inode_dirty(handle, inode); |
5324 | ||
617ba13b | 5325 | ext4_journal_stop(handle); |
ac27a0ec DK |
5326 | out: |
5327 | return; | |
5328 | } | |
5329 | ||
5330 | #if 0 | |
5331 | /* | |
5332 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5333 | * it from being flushed to disk early. Unlike | |
617ba13b | 5334 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5335 | * returns no iloc structure, so the caller needs to repeat the iloc |
5336 | * lookup to mark the inode dirty later. | |
5337 | */ | |
617ba13b | 5338 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5339 | { |
617ba13b | 5340 | struct ext4_iloc iloc; |
ac27a0ec DK |
5341 | |
5342 | int err = 0; | |
5343 | if (handle) { | |
617ba13b | 5344 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5345 | if (!err) { |
5346 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5347 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5348 | if (!err) |
0390131b | 5349 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5350 | NULL, |
0390131b | 5351 | iloc.bh); |
ac27a0ec DK |
5352 | brelse(iloc.bh); |
5353 | } | |
5354 | } | |
617ba13b | 5355 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5356 | return err; |
5357 | } | |
5358 | #endif | |
5359 | ||
617ba13b | 5360 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5361 | { |
5362 | journal_t *journal; | |
5363 | handle_t *handle; | |
5364 | int err; | |
5365 | ||
5366 | /* | |
5367 | * We have to be very careful here: changing a data block's | |
5368 | * journaling status dynamically is dangerous. If we write a | |
5369 | * data block to the journal, change the status and then delete | |
5370 | * that block, we risk forgetting to revoke the old log record | |
5371 | * from the journal and so a subsequent replay can corrupt data. | |
5372 | * So, first we make sure that the journal is empty and that | |
5373 | * nobody is changing anything. | |
5374 | */ | |
5375 | ||
617ba13b | 5376 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5377 | if (!journal) |
5378 | return 0; | |
d699594d | 5379 | if (is_journal_aborted(journal)) |
ac27a0ec | 5380 | return -EROFS; |
2aff57b0 YY |
5381 | /* We have to allocate physical blocks for delalloc blocks |
5382 | * before flushing journal. otherwise delalloc blocks can not | |
5383 | * be allocated any more. even more truncate on delalloc blocks | |
5384 | * could trigger BUG by flushing delalloc blocks in journal. | |
5385 | * There is no delalloc block in non-journal data mode. | |
5386 | */ | |
5387 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5388 | err = ext4_alloc_da_blocks(inode); | |
5389 | if (err < 0) | |
5390 | return err; | |
5391 | } | |
ac27a0ec | 5392 | |
17335dcc DM |
5393 | /* Wait for all existing dio workers */ |
5394 | ext4_inode_block_unlocked_dio(inode); | |
5395 | inode_dio_wait(inode); | |
5396 | ||
dab291af | 5397 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5398 | |
5399 | /* | |
5400 | * OK, there are no updates running now, and all cached data is | |
5401 | * synced to disk. We are now in a completely consistent state | |
5402 | * which doesn't have anything in the journal, and we know that | |
5403 | * no filesystem updates are running, so it is safe to modify | |
5404 | * the inode's in-core data-journaling state flag now. | |
5405 | */ | |
5406 | ||
5407 | if (val) | |
12e9b892 | 5408 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5409 | else { |
4f879ca6 JK |
5410 | err = jbd2_journal_flush(journal); |
5411 | if (err < 0) { | |
5412 | jbd2_journal_unlock_updates(journal); | |
5413 | ext4_inode_resume_unlocked_dio(inode); | |
5414 | return err; | |
5415 | } | |
12e9b892 | 5416 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5417 | } |
617ba13b | 5418 | ext4_set_aops(inode); |
ac27a0ec | 5419 | |
dab291af | 5420 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5421 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5422 | |
5423 | /* Finally we can mark the inode as dirty. */ | |
5424 | ||
9924a92a | 5425 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5426 | if (IS_ERR(handle)) |
5427 | return PTR_ERR(handle); | |
5428 | ||
617ba13b | 5429 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5430 | ext4_handle_sync(handle); |
617ba13b MC |
5431 | ext4_journal_stop(handle); |
5432 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5433 | |
5434 | return err; | |
5435 | } | |
2e9ee850 AK |
5436 | |
5437 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5438 | { | |
5439 | return !buffer_mapped(bh); | |
5440 | } | |
5441 | ||
c2ec175c | 5442 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5443 | { |
c2ec175c | 5444 | struct page *page = vmf->page; |
2e9ee850 AK |
5445 | loff_t size; |
5446 | unsigned long len; | |
9ea7df53 | 5447 | int ret; |
2e9ee850 | 5448 | struct file *file = vma->vm_file; |
496ad9aa | 5449 | struct inode *inode = file_inode(file); |
2e9ee850 | 5450 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5451 | handle_t *handle; |
5452 | get_block_t *get_block; | |
5453 | int retries = 0; | |
2e9ee850 | 5454 | |
8e8ad8a5 | 5455 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5456 | file_update_time(vma->vm_file); |
ea3d7209 JK |
5457 | |
5458 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
9ea7df53 JK |
5459 | /* Delalloc case is easy... */ |
5460 | if (test_opt(inode->i_sb, DELALLOC) && | |
5461 | !ext4_should_journal_data(inode) && | |
5462 | !ext4_nonda_switch(inode->i_sb)) { | |
5463 | do { | |
5c500029 | 5464 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
5465 | ext4_da_get_block_prep); |
5466 | } while (ret == -ENOSPC && | |
5467 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5468 | goto out_ret; | |
2e9ee850 | 5469 | } |
0e499890 DW |
5470 | |
5471 | lock_page(page); | |
9ea7df53 JK |
5472 | size = i_size_read(inode); |
5473 | /* Page got truncated from under us? */ | |
5474 | if (page->mapping != mapping || page_offset(page) > size) { | |
5475 | unlock_page(page); | |
5476 | ret = VM_FAULT_NOPAGE; | |
5477 | goto out; | |
0e499890 | 5478 | } |
2e9ee850 AK |
5479 | |
5480 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5481 | len = size & ~PAGE_CACHE_MASK; | |
5482 | else | |
5483 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5484 | /* |
9ea7df53 JK |
5485 | * Return if we have all the buffers mapped. This avoids the need to do |
5486 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5487 | */ |
2e9ee850 | 5488 | if (page_has_buffers(page)) { |
f19d5870 TM |
5489 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5490 | 0, len, NULL, | |
5491 | ext4_bh_unmapped)) { | |
9ea7df53 | 5492 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5493 | wait_for_stable_page(page); |
9ea7df53 JK |
5494 | ret = VM_FAULT_LOCKED; |
5495 | goto out; | |
a827eaff | 5496 | } |
2e9ee850 | 5497 | } |
a827eaff | 5498 | unlock_page(page); |
9ea7df53 JK |
5499 | /* OK, we need to fill the hole... */ |
5500 | if (ext4_should_dioread_nolock(inode)) | |
5501 | get_block = ext4_get_block_write; | |
5502 | else | |
5503 | get_block = ext4_get_block; | |
5504 | retry_alloc: | |
9924a92a TT |
5505 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5506 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5507 | if (IS_ERR(handle)) { |
c2ec175c | 5508 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5509 | goto out; |
5510 | } | |
5c500029 | 5511 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 5512 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 5513 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5514 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5515 | unlock_page(page); | |
5516 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5517 | ext4_journal_stop(handle); |
9ea7df53 JK |
5518 | goto out; |
5519 | } | |
5520 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5521 | } | |
5522 | ext4_journal_stop(handle); | |
5523 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5524 | goto retry_alloc; | |
5525 | out_ret: | |
5526 | ret = block_page_mkwrite_return(ret); | |
5527 | out: | |
ea3d7209 | 5528 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 5529 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5530 | return ret; |
5531 | } | |
ea3d7209 JK |
5532 | |
5533 | int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
5534 | { | |
5535 | struct inode *inode = file_inode(vma->vm_file); | |
5536 | int err; | |
5537 | ||
5538 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
5539 | err = filemap_fault(vma, vmf); | |
5540 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
5541 | ||
5542 | return err; | |
5543 | } |