]>
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 | ||
f3bd1f3f MC |
689 | /* Maximum number of blocks we map for direct IO at once. */ |
690 | #define DIO_MAX_BLOCKS 4096 | |
691 | ||
2ed88685 TT |
692 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
693 | struct buffer_head *bh, int flags) | |
ac27a0ec | 694 | { |
3e4fdaf8 | 695 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 696 | struct ext4_map_blocks map; |
7fb5409d | 697 | int ret = 0, started = 0; |
f3bd1f3f | 698 | int dio_credits; |
ac27a0ec | 699 | |
46c7f254 TM |
700 | if (ext4_has_inline_data(inode)) |
701 | return -ERANGE; | |
702 | ||
2ed88685 TT |
703 | map.m_lblk = iblock; |
704 | map.m_len = bh->b_size >> inode->i_blkbits; | |
705 | ||
2dcba478 | 706 | if (flags && !handle) { |
7fb5409d | 707 | /* Direct IO write... */ |
2ed88685 TT |
708 | if (map.m_len > DIO_MAX_BLOCKS) |
709 | map.m_len = DIO_MAX_BLOCKS; | |
710 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
711 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
712 | dio_credits); | |
7fb5409d | 713 | if (IS_ERR(handle)) { |
ac27a0ec | 714 | ret = PTR_ERR(handle); |
2ed88685 | 715 | return ret; |
ac27a0ec | 716 | } |
7fb5409d | 717 | started = 1; |
ac27a0ec DK |
718 | } |
719 | ||
2ed88685 | 720 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 721 | if (ret > 0) { |
7b7a8665 CH |
722 | ext4_io_end_t *io_end = ext4_inode_aio(inode); |
723 | ||
2ed88685 TT |
724 | map_bh(bh, inode->i_sb, map.m_pblk); |
725 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
e842f290 DC |
726 | if (IS_DAX(inode) && buffer_unwritten(bh)) { |
727 | /* | |
728 | * dgc: I suspect unwritten conversion on ext4+DAX is | |
729 | * fundamentally broken here when there are concurrent | |
730 | * read/write in progress on this inode. | |
731 | */ | |
732 | WARN_ON_ONCE(io_end); | |
923ae0ff RZ |
733 | bh->b_assoc_map = inode->i_mapping; |
734 | bh->b_private = (void *)(unsigned long)iblock; | |
923ae0ff | 735 | } |
7b7a8665 CH |
736 | if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) |
737 | set_buffer_defer_completion(bh); | |
2ed88685 | 738 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 739 | ret = 0; |
ac27a0ec | 740 | } |
7fb5409d JK |
741 | if (started) |
742 | ext4_journal_stop(handle); | |
ac27a0ec DK |
743 | return ret; |
744 | } | |
745 | ||
2ed88685 TT |
746 | int ext4_get_block(struct inode *inode, sector_t iblock, |
747 | struct buffer_head *bh, int create) | |
748 | { | |
749 | return _ext4_get_block(inode, iblock, bh, | |
750 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
751 | } | |
752 | ||
ac27a0ec DK |
753 | /* |
754 | * `handle' can be NULL if create is zero | |
755 | */ | |
617ba13b | 756 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 757 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 758 | { |
2ed88685 TT |
759 | struct ext4_map_blocks map; |
760 | struct buffer_head *bh; | |
c5e298ae | 761 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 762 | int err; |
ac27a0ec DK |
763 | |
764 | J_ASSERT(handle != NULL || create == 0); | |
765 | ||
2ed88685 TT |
766 | map.m_lblk = block; |
767 | map.m_len = 1; | |
c5e298ae | 768 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 769 | |
10560082 TT |
770 | if (err == 0) |
771 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 772 | if (err < 0) |
10560082 | 773 | return ERR_PTR(err); |
2ed88685 TT |
774 | |
775 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
776 | if (unlikely(!bh)) |
777 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
778 | if (map.m_flags & EXT4_MAP_NEW) { |
779 | J_ASSERT(create != 0); | |
780 | J_ASSERT(handle != NULL); | |
ac27a0ec | 781 | |
2ed88685 TT |
782 | /* |
783 | * Now that we do not always journal data, we should | |
784 | * keep in mind whether this should always journal the | |
785 | * new buffer as metadata. For now, regular file | |
786 | * writes use ext4_get_block instead, so it's not a | |
787 | * problem. | |
788 | */ | |
789 | lock_buffer(bh); | |
790 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
791 | err = ext4_journal_get_create_access(handle, bh); |
792 | if (unlikely(err)) { | |
793 | unlock_buffer(bh); | |
794 | goto errout; | |
795 | } | |
796 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
797 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
798 | set_buffer_uptodate(bh); | |
ac27a0ec | 799 | } |
2ed88685 TT |
800 | unlock_buffer(bh); |
801 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
802 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
803 | if (unlikely(err)) |
804 | goto errout; | |
805 | } else | |
2ed88685 | 806 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 807 | return bh; |
10560082 TT |
808 | errout: |
809 | brelse(bh); | |
810 | return ERR_PTR(err); | |
ac27a0ec DK |
811 | } |
812 | ||
617ba13b | 813 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 814 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 815 | { |
af5bc92d | 816 | struct buffer_head *bh; |
ac27a0ec | 817 | |
c5e298ae | 818 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 819 | if (IS_ERR(bh)) |
ac27a0ec | 820 | return bh; |
1c215028 | 821 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 822 | return bh; |
65299a3b | 823 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
824 | wait_on_buffer(bh); |
825 | if (buffer_uptodate(bh)) | |
826 | return bh; | |
827 | put_bh(bh); | |
1c215028 | 828 | return ERR_PTR(-EIO); |
ac27a0ec DK |
829 | } |
830 | ||
f19d5870 TM |
831 | int ext4_walk_page_buffers(handle_t *handle, |
832 | struct buffer_head *head, | |
833 | unsigned from, | |
834 | unsigned to, | |
835 | int *partial, | |
836 | int (*fn)(handle_t *handle, | |
837 | struct buffer_head *bh)) | |
ac27a0ec DK |
838 | { |
839 | struct buffer_head *bh; | |
840 | unsigned block_start, block_end; | |
841 | unsigned blocksize = head->b_size; | |
842 | int err, ret = 0; | |
843 | struct buffer_head *next; | |
844 | ||
af5bc92d TT |
845 | for (bh = head, block_start = 0; |
846 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 847 | block_start = block_end, bh = next) { |
ac27a0ec DK |
848 | next = bh->b_this_page; |
849 | block_end = block_start + blocksize; | |
850 | if (block_end <= from || block_start >= to) { | |
851 | if (partial && !buffer_uptodate(bh)) | |
852 | *partial = 1; | |
853 | continue; | |
854 | } | |
855 | err = (*fn)(handle, bh); | |
856 | if (!ret) | |
857 | ret = err; | |
858 | } | |
859 | return ret; | |
860 | } | |
861 | ||
862 | /* | |
863 | * To preserve ordering, it is essential that the hole instantiation and | |
864 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 865 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 866 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
867 | * prepare_write() is the right place. |
868 | * | |
36ade451 JK |
869 | * Also, this function can nest inside ext4_writepage(). In that case, we |
870 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
871 | * whole page. So we won't block on the journal in that case, which is good, | |
872 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 873 | * |
617ba13b | 874 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
875 | * quota file writes. If we were to commit the transaction while thus |
876 | * reentered, there can be a deadlock - we would be holding a quota | |
877 | * lock, and the commit would never complete if another thread had a | |
878 | * transaction open and was blocking on the quota lock - a ranking | |
879 | * violation. | |
880 | * | |
dab291af | 881 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
882 | * will _not_ run commit under these circumstances because handle->h_ref |
883 | * is elevated. We'll still have enough credits for the tiny quotafile | |
884 | * write. | |
885 | */ | |
f19d5870 TM |
886 | int do_journal_get_write_access(handle_t *handle, |
887 | struct buffer_head *bh) | |
ac27a0ec | 888 | { |
56d35a4c JK |
889 | int dirty = buffer_dirty(bh); |
890 | int ret; | |
891 | ||
ac27a0ec DK |
892 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
893 | return 0; | |
56d35a4c | 894 | /* |
ebdec241 | 895 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
896 | * the dirty bit as jbd2_journal_get_write_access() could complain |
897 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 898 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
899 | * the bit before releasing a page lock and thus writeback cannot |
900 | * ever write the buffer. | |
901 | */ | |
902 | if (dirty) | |
903 | clear_buffer_dirty(bh); | |
5d601255 | 904 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
905 | ret = ext4_journal_get_write_access(handle, bh); |
906 | if (!ret && dirty) | |
907 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
908 | return ret; | |
ac27a0ec DK |
909 | } |
910 | ||
2058f83a MH |
911 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
912 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
913 | get_block_t *get_block) | |
914 | { | |
915 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
916 | unsigned to = from + len; | |
917 | struct inode *inode = page->mapping->host; | |
918 | unsigned block_start, block_end; | |
919 | sector_t block; | |
920 | int err = 0; | |
921 | unsigned blocksize = inode->i_sb->s_blocksize; | |
922 | unsigned bbits; | |
923 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
924 | bool decrypt = false; | |
925 | ||
926 | BUG_ON(!PageLocked(page)); | |
927 | BUG_ON(from > PAGE_CACHE_SIZE); | |
928 | BUG_ON(to > PAGE_CACHE_SIZE); | |
929 | BUG_ON(from > to); | |
930 | ||
931 | if (!page_has_buffers(page)) | |
932 | create_empty_buffers(page, blocksize, 0); | |
933 | head = page_buffers(page); | |
934 | bbits = ilog2(blocksize); | |
935 | block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); | |
936 | ||
937 | for (bh = head, block_start = 0; bh != head || !block_start; | |
938 | block++, block_start = block_end, bh = bh->b_this_page) { | |
939 | block_end = block_start + blocksize; | |
940 | if (block_end <= from || block_start >= to) { | |
941 | if (PageUptodate(page)) { | |
942 | if (!buffer_uptodate(bh)) | |
943 | set_buffer_uptodate(bh); | |
944 | } | |
945 | continue; | |
946 | } | |
947 | if (buffer_new(bh)) | |
948 | clear_buffer_new(bh); | |
949 | if (!buffer_mapped(bh)) { | |
950 | WARN_ON(bh->b_size != blocksize); | |
951 | err = get_block(inode, block, bh, 1); | |
952 | if (err) | |
953 | break; | |
954 | if (buffer_new(bh)) { | |
955 | unmap_underlying_metadata(bh->b_bdev, | |
956 | bh->b_blocknr); | |
957 | if (PageUptodate(page)) { | |
958 | clear_buffer_new(bh); | |
959 | set_buffer_uptodate(bh); | |
960 | mark_buffer_dirty(bh); | |
961 | continue; | |
962 | } | |
963 | if (block_end > to || block_start < from) | |
964 | zero_user_segments(page, to, block_end, | |
965 | block_start, from); | |
966 | continue; | |
967 | } | |
968 | } | |
969 | if (PageUptodate(page)) { | |
970 | if (!buffer_uptodate(bh)) | |
971 | set_buffer_uptodate(bh); | |
972 | continue; | |
973 | } | |
974 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
975 | !buffer_unwritten(bh) && | |
976 | (block_start < from || block_end > to)) { | |
977 | ll_rw_block(READ, 1, &bh); | |
978 | *wait_bh++ = bh; | |
979 | decrypt = ext4_encrypted_inode(inode) && | |
980 | S_ISREG(inode->i_mode); | |
981 | } | |
982 | } | |
983 | /* | |
984 | * If we issued read requests, let them complete. | |
985 | */ | |
986 | while (wait_bh > wait) { | |
987 | wait_on_buffer(*--wait_bh); | |
988 | if (!buffer_uptodate(*wait_bh)) | |
989 | err = -EIO; | |
990 | } | |
991 | if (unlikely(err)) | |
992 | page_zero_new_buffers(page, from, to); | |
993 | else if (decrypt) | |
3684de8c | 994 | err = ext4_decrypt(page); |
2058f83a MH |
995 | return err; |
996 | } | |
997 | #endif | |
998 | ||
bfc1af65 | 999 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1000 | loff_t pos, unsigned len, unsigned flags, |
1001 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1002 | { |
af5bc92d | 1003 | struct inode *inode = mapping->host; |
1938a150 | 1004 | int ret, needed_blocks; |
ac27a0ec DK |
1005 | handle_t *handle; |
1006 | int retries = 0; | |
af5bc92d | 1007 | struct page *page; |
de9a55b8 | 1008 | pgoff_t index; |
af5bc92d | 1009 | unsigned from, to; |
bfc1af65 | 1010 | |
9bffad1e | 1011 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1012 | /* |
1013 | * Reserve one block more for addition to orphan list in case | |
1014 | * we allocate blocks but write fails for some reason | |
1015 | */ | |
1016 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1017 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1018 | from = pos & (PAGE_CACHE_SIZE - 1); |
1019 | to = from + len; | |
ac27a0ec | 1020 | |
f19d5870 TM |
1021 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1022 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1023 | flags, pagep); | |
1024 | if (ret < 0) | |
47564bfb TT |
1025 | return ret; |
1026 | if (ret == 1) | |
1027 | return 0; | |
f19d5870 TM |
1028 | } |
1029 | ||
47564bfb TT |
1030 | /* |
1031 | * grab_cache_page_write_begin() can take a long time if the | |
1032 | * system is thrashing due to memory pressure, or if the page | |
1033 | * is being written back. So grab it first before we start | |
1034 | * the transaction handle. This also allows us to allocate | |
1035 | * the page (if needed) without using GFP_NOFS. | |
1036 | */ | |
1037 | retry_grab: | |
1038 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1039 | if (!page) | |
1040 | return -ENOMEM; | |
1041 | unlock_page(page); | |
1042 | ||
1043 | retry_journal: | |
9924a92a | 1044 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1045 | if (IS_ERR(handle)) { |
47564bfb TT |
1046 | page_cache_release(page); |
1047 | return PTR_ERR(handle); | |
7479d2b9 | 1048 | } |
ac27a0ec | 1049 | |
47564bfb TT |
1050 | lock_page(page); |
1051 | if (page->mapping != mapping) { | |
1052 | /* The page got truncated from under us */ | |
1053 | unlock_page(page); | |
1054 | page_cache_release(page); | |
cf108bca | 1055 | ext4_journal_stop(handle); |
47564bfb | 1056 | goto retry_grab; |
cf108bca | 1057 | } |
7afe5aa5 DM |
1058 | /* In case writeback began while the page was unlocked */ |
1059 | wait_for_stable_page(page); | |
cf108bca | 1060 | |
2058f83a MH |
1061 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1062 | if (ext4_should_dioread_nolock(inode)) | |
1063 | ret = ext4_block_write_begin(page, pos, len, | |
1064 | ext4_get_block_write); | |
1065 | else | |
1066 | ret = ext4_block_write_begin(page, pos, len, | |
1067 | ext4_get_block); | |
1068 | #else | |
744692dc | 1069 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1070 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1071 | else |
6e1db88d | 1072 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1073 | #endif |
bfc1af65 | 1074 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1075 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1076 | from, to, NULL, | |
1077 | do_journal_get_write_access); | |
ac27a0ec | 1078 | } |
bfc1af65 NP |
1079 | |
1080 | if (ret) { | |
af5bc92d | 1081 | unlock_page(page); |
ae4d5372 | 1082 | /* |
6e1db88d | 1083 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1084 | * outside i_size. Trim these off again. Don't need |
1085 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1086 | * |
1087 | * Add inode to orphan list in case we crash before | |
1088 | * truncate finishes | |
ae4d5372 | 1089 | */ |
ffacfa7a | 1090 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1091 | ext4_orphan_add(handle, inode); |
1092 | ||
1093 | ext4_journal_stop(handle); | |
1094 | if (pos + len > inode->i_size) { | |
b9a4207d | 1095 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1096 | /* |
ffacfa7a | 1097 | * If truncate failed early the inode might |
1938a150 AK |
1098 | * still be on the orphan list; we need to |
1099 | * make sure the inode is removed from the | |
1100 | * orphan list in that case. | |
1101 | */ | |
1102 | if (inode->i_nlink) | |
1103 | ext4_orphan_del(NULL, inode); | |
1104 | } | |
bfc1af65 | 1105 | |
47564bfb TT |
1106 | if (ret == -ENOSPC && |
1107 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1108 | goto retry_journal; | |
1109 | page_cache_release(page); | |
1110 | return ret; | |
1111 | } | |
1112 | *pagep = page; | |
ac27a0ec DK |
1113 | return ret; |
1114 | } | |
1115 | ||
bfc1af65 NP |
1116 | /* For write_end() in data=journal mode */ |
1117 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1118 | { |
13fca323 | 1119 | int ret; |
ac27a0ec DK |
1120 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1121 | return 0; | |
1122 | set_buffer_uptodate(bh); | |
13fca323 TT |
1123 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1124 | clear_buffer_meta(bh); | |
1125 | clear_buffer_prio(bh); | |
1126 | return ret; | |
ac27a0ec DK |
1127 | } |
1128 | ||
eed4333f ZL |
1129 | /* |
1130 | * We need to pick up the new inode size which generic_commit_write gave us | |
1131 | * `file' can be NULL - eg, when called from page_symlink(). | |
1132 | * | |
1133 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1134 | * buffers are managed internally. | |
1135 | */ | |
1136 | static int ext4_write_end(struct file *file, | |
1137 | struct address_space *mapping, | |
1138 | loff_t pos, unsigned len, unsigned copied, | |
1139 | struct page *page, void *fsdata) | |
f8514083 | 1140 | { |
f8514083 | 1141 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1142 | struct inode *inode = mapping->host; |
0572639f | 1143 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1144 | int ret = 0, ret2; |
1145 | int i_size_changed = 0; | |
1146 | ||
1147 | trace_ext4_write_end(inode, pos, len, copied); | |
1148 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1149 | ret = ext4_jbd2_file_inode(handle, inode); | |
1150 | if (ret) { | |
1151 | unlock_page(page); | |
1152 | page_cache_release(page); | |
1153 | goto errout; | |
1154 | } | |
1155 | } | |
f8514083 | 1156 | |
42c832de TT |
1157 | if (ext4_has_inline_data(inode)) { |
1158 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1159 | copied, page); | |
1160 | if (ret < 0) | |
1161 | goto errout; | |
1162 | copied = ret; | |
1163 | } else | |
f19d5870 TM |
1164 | copied = block_write_end(file, mapping, pos, |
1165 | len, copied, page, fsdata); | |
f8514083 | 1166 | /* |
4631dbf6 | 1167 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1168 | * page writeout could otherwise come in and zero beyond i_size. |
1169 | */ | |
4631dbf6 | 1170 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 AK |
1171 | unlock_page(page); |
1172 | page_cache_release(page); | |
1173 | ||
0572639f XW |
1174 | if (old_size < pos) |
1175 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1176 | /* |
1177 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1178 | * makes the holding time of page lock longer. Second, it forces lock | |
1179 | * ordering of page lock and transaction start for journaling | |
1180 | * filesystems. | |
1181 | */ | |
1182 | if (i_size_changed) | |
1183 | ext4_mark_inode_dirty(handle, inode); | |
1184 | ||
ffacfa7a | 1185 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1186 | /* if we have allocated more blocks and copied |
1187 | * less. We will have blocks allocated outside | |
1188 | * inode->i_size. So truncate them | |
1189 | */ | |
1190 | ext4_orphan_add(handle, inode); | |
74d553aa | 1191 | errout: |
617ba13b | 1192 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1193 | if (!ret) |
1194 | ret = ret2; | |
bfc1af65 | 1195 | |
f8514083 | 1196 | if (pos + len > inode->i_size) { |
b9a4207d | 1197 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1198 | /* |
ffacfa7a | 1199 | * If truncate failed early the inode might still be |
f8514083 AK |
1200 | * on the orphan list; we need to make sure the inode |
1201 | * is removed from the orphan list in that case. | |
1202 | */ | |
1203 | if (inode->i_nlink) | |
1204 | ext4_orphan_del(NULL, inode); | |
1205 | } | |
1206 | ||
bfc1af65 | 1207 | return ret ? ret : copied; |
ac27a0ec DK |
1208 | } |
1209 | ||
b90197b6 TT |
1210 | /* |
1211 | * This is a private version of page_zero_new_buffers() which doesn't | |
1212 | * set the buffer to be dirty, since in data=journalled mode we need | |
1213 | * to call ext4_handle_dirty_metadata() instead. | |
1214 | */ | |
1215 | static void zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1216 | { | |
1217 | unsigned int block_start = 0, block_end; | |
1218 | struct buffer_head *head, *bh; | |
1219 | ||
1220 | bh = head = page_buffers(page); | |
1221 | do { | |
1222 | block_end = block_start + bh->b_size; | |
1223 | if (buffer_new(bh)) { | |
1224 | if (block_end > from && block_start < to) { | |
1225 | if (!PageUptodate(page)) { | |
1226 | unsigned start, size; | |
1227 | ||
1228 | start = max(from, block_start); | |
1229 | size = min(to, block_end) - start; | |
1230 | ||
1231 | zero_user(page, start, size); | |
1232 | set_buffer_uptodate(bh); | |
1233 | } | |
1234 | clear_buffer_new(bh); | |
1235 | } | |
1236 | } | |
1237 | block_start = block_end; | |
1238 | bh = bh->b_this_page; | |
1239 | } while (bh != head); | |
1240 | } | |
1241 | ||
bfc1af65 | 1242 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1243 | struct address_space *mapping, |
1244 | loff_t pos, unsigned len, unsigned copied, | |
1245 | struct page *page, void *fsdata) | |
ac27a0ec | 1246 | { |
617ba13b | 1247 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1248 | struct inode *inode = mapping->host; |
0572639f | 1249 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1250 | int ret = 0, ret2; |
1251 | int partial = 0; | |
bfc1af65 | 1252 | unsigned from, to; |
4631dbf6 | 1253 | int size_changed = 0; |
ac27a0ec | 1254 | |
9bffad1e | 1255 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1256 | from = pos & (PAGE_CACHE_SIZE - 1); |
1257 | to = from + len; | |
1258 | ||
441c8508 CW |
1259 | BUG_ON(!ext4_handle_valid(handle)); |
1260 | ||
3fdcfb66 TM |
1261 | if (ext4_has_inline_data(inode)) |
1262 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1263 | copied, page); | |
1264 | else { | |
1265 | if (copied < len) { | |
1266 | if (!PageUptodate(page)) | |
1267 | copied = 0; | |
b90197b6 | 1268 | zero_new_buffers(page, from+copied, to); |
3fdcfb66 | 1269 | } |
ac27a0ec | 1270 | |
3fdcfb66 TM |
1271 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1272 | to, &partial, write_end_fn); | |
1273 | if (!partial) | |
1274 | SetPageUptodate(page); | |
1275 | } | |
4631dbf6 | 1276 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1277 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1278 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 DM |
1279 | unlock_page(page); |
1280 | page_cache_release(page); | |
1281 | ||
0572639f XW |
1282 | if (old_size < pos) |
1283 | pagecache_isize_extended(inode, old_size, pos); | |
1284 | ||
4631dbf6 | 1285 | if (size_changed) { |
617ba13b | 1286 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1287 | if (!ret) |
1288 | ret = ret2; | |
1289 | } | |
bfc1af65 | 1290 | |
ffacfa7a | 1291 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1292 | /* if we have allocated more blocks and copied |
1293 | * less. We will have blocks allocated outside | |
1294 | * inode->i_size. So truncate them | |
1295 | */ | |
1296 | ext4_orphan_add(handle, inode); | |
1297 | ||
617ba13b | 1298 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1299 | if (!ret) |
1300 | ret = ret2; | |
f8514083 | 1301 | if (pos + len > inode->i_size) { |
b9a4207d | 1302 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1303 | /* |
ffacfa7a | 1304 | * If truncate failed early the inode might still be |
f8514083 AK |
1305 | * on the orphan list; we need to make sure the inode |
1306 | * is removed from the orphan list in that case. | |
1307 | */ | |
1308 | if (inode->i_nlink) | |
1309 | ext4_orphan_del(NULL, inode); | |
1310 | } | |
bfc1af65 NP |
1311 | |
1312 | return ret ? ret : copied; | |
ac27a0ec | 1313 | } |
d2a17637 | 1314 | |
9d0be502 | 1315 | /* |
c27e43a1 | 1316 | * Reserve space for a single cluster |
9d0be502 | 1317 | */ |
c27e43a1 | 1318 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1319 | { |
60e58e0f | 1320 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1321 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1322 | int ret; |
03179fe9 TT |
1323 | |
1324 | /* | |
1325 | * We will charge metadata quota at writeout time; this saves | |
1326 | * us from metadata over-estimation, though we may go over by | |
1327 | * a small amount in the end. Here we just reserve for data. | |
1328 | */ | |
1329 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1330 | if (ret) | |
1331 | return ret; | |
d2a17637 | 1332 | |
0637c6f4 | 1333 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1334 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1335 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1336 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1337 | return -ENOSPC; |
1338 | } | |
9d0be502 | 1339 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1340 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1341 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1342 | |
d2a17637 MC |
1343 | return 0; /* success */ |
1344 | } | |
1345 | ||
12219aea | 1346 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1347 | { |
1348 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1349 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1350 | |
cd213226 MC |
1351 | if (!to_free) |
1352 | return; /* Nothing to release, exit */ | |
1353 | ||
d2a17637 | 1354 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1355 | |
5a58ec87 | 1356 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1357 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1358 | /* |
0637c6f4 TT |
1359 | * if there aren't enough reserved blocks, then the |
1360 | * counter is messed up somewhere. Since this | |
1361 | * function is called from invalidate page, it's | |
1362 | * harmless to return without any action. | |
cd213226 | 1363 | */ |
8de5c325 | 1364 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1365 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1366 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1367 | ei->i_reserved_data_blocks); |
1368 | WARN_ON(1); | |
1369 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1370 | } |
0637c6f4 | 1371 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1372 | |
72b8ab9d | 1373 | /* update fs dirty data blocks counter */ |
57042651 | 1374 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1375 | |
d2a17637 | 1376 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1377 | |
7b415bf6 | 1378 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1379 | } |
1380 | ||
1381 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1382 | unsigned int offset, |
1383 | unsigned int length) | |
d2a17637 | 1384 | { |
9705acd6 | 1385 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1386 | struct buffer_head *head, *bh; |
1387 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1388 | struct inode *inode = page->mapping->host; |
1389 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1390 | unsigned int stop = offset + length; |
7b415bf6 | 1391 | int num_clusters; |
51865fda | 1392 | ext4_fsblk_t lblk; |
d2a17637 | 1393 | |
ca99fdd2 LC |
1394 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
1395 | ||
d2a17637 MC |
1396 | head = page_buffers(page); |
1397 | bh = head; | |
1398 | do { | |
1399 | unsigned int next_off = curr_off + bh->b_size; | |
1400 | ||
ca99fdd2 LC |
1401 | if (next_off > stop) |
1402 | break; | |
1403 | ||
d2a17637 MC |
1404 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1405 | to_release++; | |
9705acd6 | 1406 | contiguous_blks++; |
d2a17637 | 1407 | clear_buffer_delay(bh); |
9705acd6 LC |
1408 | } else if (contiguous_blks) { |
1409 | lblk = page->index << | |
1410 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1411 | lblk += (curr_off >> inode->i_blkbits) - | |
1412 | contiguous_blks; | |
1413 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1414 | contiguous_blks = 0; | |
d2a17637 MC |
1415 | } |
1416 | curr_off = next_off; | |
1417 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1418 | |
9705acd6 | 1419 | if (contiguous_blks) { |
51865fda | 1420 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1421 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1422 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1423 | } |
1424 | ||
7b415bf6 AK |
1425 | /* If we have released all the blocks belonging to a cluster, then we |
1426 | * need to release the reserved space for that cluster. */ | |
1427 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1428 | while (num_clusters > 0) { | |
7b415bf6 AK |
1429 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1430 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1431 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1432 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1433 | ext4_da_release_space(inode, 1); |
1434 | ||
1435 | num_clusters--; | |
1436 | } | |
d2a17637 | 1437 | } |
ac27a0ec | 1438 | |
64769240 AT |
1439 | /* |
1440 | * Delayed allocation stuff | |
1441 | */ | |
1442 | ||
4e7ea81d JK |
1443 | struct mpage_da_data { |
1444 | struct inode *inode; | |
1445 | struct writeback_control *wbc; | |
6b523df4 | 1446 | |
4e7ea81d JK |
1447 | pgoff_t first_page; /* The first page to write */ |
1448 | pgoff_t next_page; /* Current page to examine */ | |
1449 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1450 | /* |
4e7ea81d JK |
1451 | * Extent to map - this can be after first_page because that can be |
1452 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1453 | * is delalloc or unwritten. | |
791b7f08 | 1454 | */ |
4e7ea81d JK |
1455 | struct ext4_map_blocks map; |
1456 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1457 | }; | |
64769240 | 1458 | |
4e7ea81d JK |
1459 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1460 | bool invalidate) | |
c4a0c46e AK |
1461 | { |
1462 | int nr_pages, i; | |
1463 | pgoff_t index, end; | |
1464 | struct pagevec pvec; | |
1465 | struct inode *inode = mpd->inode; | |
1466 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1467 | |
1468 | /* This is necessary when next_page == 0. */ | |
1469 | if (mpd->first_page >= mpd->next_page) | |
1470 | return; | |
c4a0c46e | 1471 | |
c7f5938a CW |
1472 | index = mpd->first_page; |
1473 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1474 | if (invalidate) { |
1475 | ext4_lblk_t start, last; | |
1476 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1477 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1478 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1479 | } | |
51865fda | 1480 | |
66bea92c | 1481 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1482 | while (index <= end) { |
1483 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1484 | if (nr_pages == 0) | |
1485 | break; | |
1486 | for (i = 0; i < nr_pages; i++) { | |
1487 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1488 | if (page->index > end) |
c4a0c46e | 1489 | break; |
c4a0c46e AK |
1490 | BUG_ON(!PageLocked(page)); |
1491 | BUG_ON(PageWriteback(page)); | |
4e7ea81d JK |
1492 | if (invalidate) { |
1493 | block_invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
1494 | ClearPageUptodate(page); | |
1495 | } | |
c4a0c46e AK |
1496 | unlock_page(page); |
1497 | } | |
9b1d0998 JK |
1498 | index = pvec.pages[nr_pages - 1]->index + 1; |
1499 | pagevec_release(&pvec); | |
c4a0c46e | 1500 | } |
c4a0c46e AK |
1501 | } |
1502 | ||
df22291f AK |
1503 | static void ext4_print_free_blocks(struct inode *inode) |
1504 | { | |
1505 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1506 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1507 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1508 | |
1509 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1510 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1511 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1512 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1513 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1514 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1515 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1516 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1517 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1518 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1519 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1520 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1521 | ei->i_reserved_data_blocks); |
df22291f AK |
1522 | return; |
1523 | } | |
1524 | ||
c364b22c | 1525 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1526 | { |
c364b22c | 1527 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1528 | } |
1529 | ||
5356f261 AK |
1530 | /* |
1531 | * This function is grabs code from the very beginning of | |
1532 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1533 | * time. This function looks up the requested blocks and sets the | |
1534 | * buffer delay bit under the protection of i_data_sem. | |
1535 | */ | |
1536 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1537 | struct ext4_map_blocks *map, | |
1538 | struct buffer_head *bh) | |
1539 | { | |
d100eef2 | 1540 | struct extent_status es; |
5356f261 AK |
1541 | int retval; |
1542 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1543 | #ifdef ES_AGGRESSIVE_TEST |
1544 | struct ext4_map_blocks orig_map; | |
1545 | ||
1546 | memcpy(&orig_map, map, sizeof(*map)); | |
1547 | #endif | |
5356f261 AK |
1548 | |
1549 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1550 | invalid_block = ~0; | |
1551 | ||
1552 | map->m_flags = 0; | |
1553 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1554 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1555 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1556 | |
1557 | /* Lookup extent status tree firstly */ | |
1558 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1559 | if (ext4_es_is_hole(&es)) { |
1560 | retval = 0; | |
c8b459f4 | 1561 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1562 | goto add_delayed; |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * Delayed extent could be allocated by fallocate. | |
1567 | * So we need to check it. | |
1568 | */ | |
1569 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1570 | map_bh(bh, inode->i_sb, invalid_block); | |
1571 | set_buffer_new(bh); | |
1572 | set_buffer_delay(bh); | |
1573 | return 0; | |
1574 | } | |
1575 | ||
1576 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1577 | retval = es.es_len - (iblock - es.es_lblk); | |
1578 | if (retval > map->m_len) | |
1579 | retval = map->m_len; | |
1580 | map->m_len = retval; | |
1581 | if (ext4_es_is_written(&es)) | |
1582 | map->m_flags |= EXT4_MAP_MAPPED; | |
1583 | else if (ext4_es_is_unwritten(&es)) | |
1584 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1585 | else | |
1586 | BUG_ON(1); | |
1587 | ||
921f266b DM |
1588 | #ifdef ES_AGGRESSIVE_TEST |
1589 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1590 | #endif | |
d100eef2 ZL |
1591 | return retval; |
1592 | } | |
1593 | ||
5356f261 AK |
1594 | /* |
1595 | * Try to see if we can get the block without requesting a new | |
1596 | * file system block. | |
1597 | */ | |
c8b459f4 | 1598 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1599 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1600 | retval = 0; |
cbd7584e | 1601 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1602 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1603 | else |
2f8e0a7c | 1604 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1605 | |
d100eef2 | 1606 | add_delayed: |
5356f261 | 1607 | if (retval == 0) { |
f7fec032 | 1608 | int ret; |
5356f261 AK |
1609 | /* |
1610 | * XXX: __block_prepare_write() unmaps passed block, | |
1611 | * is it OK? | |
1612 | */ | |
386ad67c LC |
1613 | /* |
1614 | * If the block was allocated from previously allocated cluster, | |
1615 | * then we don't need to reserve it again. However we still need | |
1616 | * to reserve metadata for every block we're going to write. | |
1617 | */ | |
c27e43a1 | 1618 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1619 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1620 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1621 | if (ret) { |
5356f261 | 1622 | /* not enough space to reserve */ |
f7fec032 | 1623 | retval = ret; |
5356f261 | 1624 | goto out_unlock; |
f7fec032 | 1625 | } |
5356f261 AK |
1626 | } |
1627 | ||
f7fec032 ZL |
1628 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1629 | ~0, EXTENT_STATUS_DELAYED); | |
1630 | if (ret) { | |
1631 | retval = ret; | |
51865fda | 1632 | goto out_unlock; |
f7fec032 | 1633 | } |
51865fda | 1634 | |
5356f261 AK |
1635 | map_bh(bh, inode->i_sb, invalid_block); |
1636 | set_buffer_new(bh); | |
1637 | set_buffer_delay(bh); | |
f7fec032 ZL |
1638 | } else if (retval > 0) { |
1639 | int ret; | |
3be78c73 | 1640 | unsigned int status; |
f7fec032 | 1641 | |
44fb851d ZL |
1642 | if (unlikely(retval != map->m_len)) { |
1643 | ext4_warning(inode->i_sb, | |
1644 | "ES len assertion failed for inode " | |
1645 | "%lu: retval %d != map->m_len %d", | |
1646 | inode->i_ino, retval, map->m_len); | |
1647 | WARN_ON(1); | |
921f266b | 1648 | } |
921f266b | 1649 | |
f7fec032 ZL |
1650 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1651 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1652 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1653 | map->m_pblk, status); | |
1654 | if (ret != 0) | |
1655 | retval = ret; | |
5356f261 AK |
1656 | } |
1657 | ||
1658 | out_unlock: | |
1659 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1660 | ||
1661 | return retval; | |
1662 | } | |
1663 | ||
64769240 | 1664 | /* |
d91bd2c1 | 1665 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1666 | * ext4_da_write_begin(). It will either return mapped block or |
1667 | * reserve space for a single block. | |
29fa89d0 AK |
1668 | * |
1669 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1670 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1671 | * | |
1672 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1673 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1674 | * initialized properly. | |
64769240 | 1675 | */ |
9c3569b5 TM |
1676 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1677 | struct buffer_head *bh, int create) | |
64769240 | 1678 | { |
2ed88685 | 1679 | struct ext4_map_blocks map; |
64769240 AT |
1680 | int ret = 0; |
1681 | ||
1682 | BUG_ON(create == 0); | |
2ed88685 TT |
1683 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1684 | ||
1685 | map.m_lblk = iblock; | |
1686 | map.m_len = 1; | |
64769240 AT |
1687 | |
1688 | /* | |
1689 | * first, we need to know whether the block is allocated already | |
1690 | * preallocated blocks are unmapped but should treated | |
1691 | * the same as allocated blocks. | |
1692 | */ | |
5356f261 AK |
1693 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1694 | if (ret <= 0) | |
2ed88685 | 1695 | return ret; |
64769240 | 1696 | |
2ed88685 TT |
1697 | map_bh(bh, inode->i_sb, map.m_pblk); |
1698 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1699 | ||
1700 | if (buffer_unwritten(bh)) { | |
1701 | /* A delayed write to unwritten bh should be marked | |
1702 | * new and mapped. Mapped ensures that we don't do | |
1703 | * get_block multiple times when we write to the same | |
1704 | * offset and new ensures that we do proper zero out | |
1705 | * for partial write. | |
1706 | */ | |
1707 | set_buffer_new(bh); | |
c8205636 | 1708 | set_buffer_mapped(bh); |
2ed88685 TT |
1709 | } |
1710 | return 0; | |
64769240 | 1711 | } |
61628a3f | 1712 | |
62e086be AK |
1713 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1714 | { | |
1715 | get_bh(bh); | |
1716 | return 0; | |
1717 | } | |
1718 | ||
1719 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1720 | { | |
1721 | put_bh(bh); | |
1722 | return 0; | |
1723 | } | |
1724 | ||
1725 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1726 | unsigned int len) |
1727 | { | |
1728 | struct address_space *mapping = page->mapping; | |
1729 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1730 | struct buffer_head *page_bufs = NULL; |
62e086be | 1731 | handle_t *handle = NULL; |
3fdcfb66 TM |
1732 | int ret = 0, err = 0; |
1733 | int inline_data = ext4_has_inline_data(inode); | |
1734 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1735 | |
cb20d518 | 1736 | ClearPageChecked(page); |
3fdcfb66 TM |
1737 | |
1738 | if (inline_data) { | |
1739 | BUG_ON(page->index != 0); | |
1740 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1741 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1742 | if (inode_bh == NULL) | |
1743 | goto out; | |
1744 | } else { | |
1745 | page_bufs = page_buffers(page); | |
1746 | if (!page_bufs) { | |
1747 | BUG(); | |
1748 | goto out; | |
1749 | } | |
1750 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1751 | NULL, bget_one); | |
1752 | } | |
bdf96838 TT |
1753 | /* |
1754 | * We need to release the page lock before we start the | |
1755 | * journal, so grab a reference so the page won't disappear | |
1756 | * out from under us. | |
1757 | */ | |
1758 | get_page(page); | |
62e086be AK |
1759 | unlock_page(page); |
1760 | ||
9924a92a TT |
1761 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1762 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1763 | if (IS_ERR(handle)) { |
1764 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1765 | put_page(page); |
1766 | goto out_no_pagelock; | |
62e086be | 1767 | } |
441c8508 CW |
1768 | BUG_ON(!ext4_handle_valid(handle)); |
1769 | ||
bdf96838 TT |
1770 | lock_page(page); |
1771 | put_page(page); | |
1772 | if (page->mapping != mapping) { | |
1773 | /* The page got truncated from under us */ | |
1774 | ext4_journal_stop(handle); | |
1775 | ret = 0; | |
1776 | goto out; | |
1777 | } | |
1778 | ||
3fdcfb66 | 1779 | if (inline_data) { |
5d601255 | 1780 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1781 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1782 | |
3fdcfb66 TM |
1783 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1784 | ||
1785 | } else { | |
1786 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1787 | do_journal_get_write_access); | |
1788 | ||
1789 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1790 | write_end_fn); | |
1791 | } | |
62e086be AK |
1792 | if (ret == 0) |
1793 | ret = err; | |
2d859db3 | 1794 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1795 | err = ext4_journal_stop(handle); |
1796 | if (!ret) | |
1797 | ret = err; | |
1798 | ||
3fdcfb66 | 1799 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1800 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1801 | NULL, bput_one); |
19f5fb7a | 1802 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1803 | out: |
bdf96838 TT |
1804 | unlock_page(page); |
1805 | out_no_pagelock: | |
3fdcfb66 | 1806 | brelse(inode_bh); |
62e086be AK |
1807 | return ret; |
1808 | } | |
1809 | ||
61628a3f | 1810 | /* |
43ce1d23 AK |
1811 | * Note that we don't need to start a transaction unless we're journaling data |
1812 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1813 | * need to file the inode to the transaction's list in ordered mode because if | |
1814 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1815 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1816 | * transaction the data will hit the disk. In case we are journaling data, we |
1817 | * cannot start transaction directly because transaction start ranks above page | |
1818 | * lock so we have to do some magic. | |
1819 | * | |
b920c755 | 1820 | * This function can get called via... |
20970ba6 | 1821 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1822 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1823 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1824 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1825 | * |
1826 | * We don't do any block allocation in this function. If we have page with | |
1827 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1828 | * is important for mmaped based write. So if we do with blocksize 1K | |
1829 | * truncate(f, 1024); | |
1830 | * a = mmap(f, 0, 4096); | |
1831 | * a[0] = 'a'; | |
1832 | * truncate(f, 4096); | |
1833 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1834 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1835 | * do_wp_page). So writepage should write the first block. If we modify |
1836 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1837 | * page_mkwrite callback will do the block allocation and mark the | |
1838 | * buffer_heads mapped. | |
1839 | * | |
1840 | * We redirty the page if we have any buffer_heads that is either delay or | |
1841 | * unwritten in the page. | |
1842 | * | |
1843 | * We can get recursively called as show below. | |
1844 | * | |
1845 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1846 | * ext4_writepage() | |
1847 | * | |
1848 | * But since we don't do any block allocation we should not deadlock. | |
1849 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1850 | */ |
43ce1d23 | 1851 | static int ext4_writepage(struct page *page, |
62e086be | 1852 | struct writeback_control *wbc) |
64769240 | 1853 | { |
f8bec370 | 1854 | int ret = 0; |
61628a3f | 1855 | loff_t size; |
498e5f24 | 1856 | unsigned int len; |
744692dc | 1857 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1858 | struct inode *inode = page->mapping->host; |
36ade451 | 1859 | struct ext4_io_submit io_submit; |
1c8349a1 | 1860 | bool keep_towrite = false; |
61628a3f | 1861 | |
a9c667f8 | 1862 | trace_ext4_writepage(page); |
f0e6c985 AK |
1863 | size = i_size_read(inode); |
1864 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1865 | len = size & ~PAGE_CACHE_MASK; | |
1866 | else | |
1867 | len = PAGE_CACHE_SIZE; | |
64769240 | 1868 | |
a42afc5f | 1869 | page_bufs = page_buffers(page); |
a42afc5f | 1870 | /* |
fe386132 JK |
1871 | * We cannot do block allocation or other extent handling in this |
1872 | * function. If there are buffers needing that, we have to redirty | |
1873 | * the page. But we may reach here when we do a journal commit via | |
1874 | * journal_submit_inode_data_buffers() and in that case we must write | |
1875 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
1876 | * |
1877 | * Also, if there is only one buffer per page (the fs block | |
1878 | * size == the page size), if one buffer needs block | |
1879 | * allocation or needs to modify the extent tree to clear the | |
1880 | * unwritten flag, we know that the page can't be written at | |
1881 | * all, so we might as well refuse the write immediately. | |
1882 | * Unfortunately if the block size != page size, we can't as | |
1883 | * easily detect this case using ext4_walk_page_buffers(), but | |
1884 | * for the extremely common case, this is an optimization that | |
1885 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 1886 | */ |
f19d5870 TM |
1887 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
1888 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 1889 | redirty_page_for_writepage(wbc, page); |
cccd147a TT |
1890 | if ((current->flags & PF_MEMALLOC) || |
1891 | (inode->i_sb->s_blocksize == PAGE_CACHE_SIZE)) { | |
fe386132 JK |
1892 | /* |
1893 | * For memory cleaning there's no point in writing only | |
1894 | * some buffers. So just bail out. Warn if we came here | |
1895 | * from direct reclaim. | |
1896 | */ | |
1897 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
1898 | == PF_MEMALLOC); | |
f0e6c985 AK |
1899 | unlock_page(page); |
1900 | return 0; | |
1901 | } | |
1c8349a1 | 1902 | keep_towrite = true; |
a42afc5f | 1903 | } |
64769240 | 1904 | |
cb20d518 | 1905 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1906 | /* |
1907 | * It's mmapped pagecache. Add buffers and journal it. There | |
1908 | * doesn't seem much point in redirtying the page here. | |
1909 | */ | |
3f0ca309 | 1910 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1911 | |
97a851ed JK |
1912 | ext4_io_submit_init(&io_submit, wbc); |
1913 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
1914 | if (!io_submit.io_end) { | |
1915 | redirty_page_for_writepage(wbc, page); | |
1916 | unlock_page(page); | |
1917 | return -ENOMEM; | |
1918 | } | |
1c8349a1 | 1919 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 1920 | ext4_io_submit(&io_submit); |
97a851ed JK |
1921 | /* Drop io_end reference we got from init */ |
1922 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
1923 | return ret; |
1924 | } | |
1925 | ||
5f1132b2 JK |
1926 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
1927 | { | |
1928 | int len; | |
1929 | loff_t size = i_size_read(mpd->inode); | |
1930 | int err; | |
1931 | ||
1932 | BUG_ON(page->index != mpd->first_page); | |
1933 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1934 | len = size & ~PAGE_CACHE_MASK; | |
1935 | else | |
1936 | len = PAGE_CACHE_SIZE; | |
1937 | clear_page_dirty_for_io(page); | |
1c8349a1 | 1938 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
1939 | if (!err) |
1940 | mpd->wbc->nr_to_write--; | |
1941 | mpd->first_page++; | |
1942 | ||
1943 | return err; | |
1944 | } | |
1945 | ||
4e7ea81d JK |
1946 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
1947 | ||
61628a3f | 1948 | /* |
fffb2739 JK |
1949 | * mballoc gives us at most this number of blocks... |
1950 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 1951 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 1952 | */ |
fffb2739 | 1953 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 1954 | |
4e7ea81d JK |
1955 | /* |
1956 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
1957 | * | |
1958 | * @mpd - extent of blocks | |
1959 | * @lblk - logical number of the block in the file | |
09930042 | 1960 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 1961 | * |
09930042 JK |
1962 | * The function is used to collect contig. blocks in the same state. If the |
1963 | * buffer doesn't require mapping for writeback and we haven't started the | |
1964 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
1965 | * caller can write the buffer right away. Otherwise the function returns true | |
1966 | * if the block has been added to the extent, false if the block couldn't be | |
1967 | * added. | |
4e7ea81d | 1968 | */ |
09930042 JK |
1969 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
1970 | struct buffer_head *bh) | |
4e7ea81d JK |
1971 | { |
1972 | struct ext4_map_blocks *map = &mpd->map; | |
1973 | ||
09930042 JK |
1974 | /* Buffer that doesn't need mapping for writeback? */ |
1975 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
1976 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
1977 | /* So far no extent to map => we write the buffer right away */ | |
1978 | if (map->m_len == 0) | |
1979 | return true; | |
1980 | return false; | |
1981 | } | |
4e7ea81d JK |
1982 | |
1983 | /* First block in the extent? */ | |
1984 | if (map->m_len == 0) { | |
1985 | map->m_lblk = lblk; | |
1986 | map->m_len = 1; | |
09930042 JK |
1987 | map->m_flags = bh->b_state & BH_FLAGS; |
1988 | return true; | |
4e7ea81d JK |
1989 | } |
1990 | ||
09930042 JK |
1991 | /* Don't go larger than mballoc is willing to allocate */ |
1992 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
1993 | return false; | |
1994 | ||
4e7ea81d JK |
1995 | /* Can we merge the block to our big extent? */ |
1996 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 1997 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 1998 | map->m_len++; |
09930042 | 1999 | return true; |
4e7ea81d | 2000 | } |
09930042 | 2001 | return false; |
4e7ea81d JK |
2002 | } |
2003 | ||
5f1132b2 JK |
2004 | /* |
2005 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2006 | * | |
2007 | * @mpd - extent of blocks for mapping | |
2008 | * @head - the first buffer in the page | |
2009 | * @bh - buffer we should start processing from | |
2010 | * @lblk - logical number of the block in the file corresponding to @bh | |
2011 | * | |
2012 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2013 | * the page for IO if all buffers in this page were mapped and there's no | |
2014 | * accumulated extent of buffers to map or add buffers in the page to the | |
2015 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2016 | * by processing the next page, 0 if it should stop adding buffers to the | |
2017 | * extent to map because we cannot extend it anymore. It can also return value | |
2018 | * < 0 in case of error during IO submission. | |
2019 | */ | |
2020 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2021 | struct buffer_head *head, | |
2022 | struct buffer_head *bh, | |
2023 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2024 | { |
2025 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2026 | int err; |
4e7ea81d JK |
2027 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
2028 | >> inode->i_blkbits; | |
2029 | ||
2030 | do { | |
2031 | BUG_ON(buffer_locked(bh)); | |
2032 | ||
09930042 | 2033 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2034 | /* Found extent to map? */ |
2035 | if (mpd->map.m_len) | |
5f1132b2 | 2036 | return 0; |
09930042 | 2037 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2038 | break; |
4e7ea81d | 2039 | } |
4e7ea81d | 2040 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2041 | /* So far everything mapped? Submit the page for IO. */ |
2042 | if (mpd->map.m_len == 0) { | |
2043 | err = mpage_submit_page(mpd, head->b_page); | |
2044 | if (err < 0) | |
2045 | return err; | |
2046 | } | |
2047 | return lblk < blocks; | |
4e7ea81d JK |
2048 | } |
2049 | ||
2050 | /* | |
2051 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2052 | * submit fully mapped pages for IO | |
2053 | * | |
2054 | * @mpd - description of extent to map, on return next extent to map | |
2055 | * | |
2056 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2057 | * to be already locked) and update buffer state according to new extent state. | |
2058 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2059 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2060 | * and do extent conversion after IO is finished. If the last page is not fully |
2061 | * mapped, we update @map to the next extent in the last page that needs | |
2062 | * mapping. Otherwise we submit the page for IO. | |
2063 | */ | |
2064 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2065 | { | |
2066 | struct pagevec pvec; | |
2067 | int nr_pages, i; | |
2068 | struct inode *inode = mpd->inode; | |
2069 | struct buffer_head *head, *bh; | |
2070 | int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; | |
4e7ea81d JK |
2071 | pgoff_t start, end; |
2072 | ext4_lblk_t lblk; | |
2073 | sector_t pblock; | |
2074 | int err; | |
2075 | ||
2076 | start = mpd->map.m_lblk >> bpp_bits; | |
2077 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2078 | lblk = start << bpp_bits; | |
2079 | pblock = mpd->map.m_pblk; | |
2080 | ||
2081 | pagevec_init(&pvec, 0); | |
2082 | while (start <= end) { | |
2083 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2084 | PAGEVEC_SIZE); | |
2085 | if (nr_pages == 0) | |
2086 | break; | |
2087 | for (i = 0; i < nr_pages; i++) { | |
2088 | struct page *page = pvec.pages[i]; | |
2089 | ||
2090 | if (page->index > end) | |
2091 | break; | |
70261f56 | 2092 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2093 | BUG_ON(page->index != start); |
2094 | bh = head = page_buffers(page); | |
2095 | do { | |
2096 | if (lblk < mpd->map.m_lblk) | |
2097 | continue; | |
2098 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2099 | /* | |
2100 | * Buffer after end of mapped extent. | |
2101 | * Find next buffer in the page to map. | |
2102 | */ | |
2103 | mpd->map.m_len = 0; | |
2104 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2105 | /* |
2106 | * FIXME: If dioread_nolock supports | |
2107 | * blocksize < pagesize, we need to make | |
2108 | * sure we add size mapped so far to | |
2109 | * io_end->size as the following call | |
2110 | * can submit the page for IO. | |
2111 | */ | |
2112 | err = mpage_process_page_bufs(mpd, head, | |
2113 | bh, lblk); | |
4e7ea81d | 2114 | pagevec_release(&pvec); |
5f1132b2 JK |
2115 | if (err > 0) |
2116 | err = 0; | |
2117 | return err; | |
4e7ea81d JK |
2118 | } |
2119 | if (buffer_delay(bh)) { | |
2120 | clear_buffer_delay(bh); | |
2121 | bh->b_blocknr = pblock++; | |
2122 | } | |
4e7ea81d | 2123 | clear_buffer_unwritten(bh); |
5f1132b2 | 2124 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2125 | |
2126 | /* | |
2127 | * FIXME: This is going to break if dioread_nolock | |
2128 | * supports blocksize < pagesize as we will try to | |
2129 | * convert potentially unmapped parts of inode. | |
2130 | */ | |
2131 | mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; | |
2132 | /* Page fully mapped - let IO run! */ | |
2133 | err = mpage_submit_page(mpd, page); | |
2134 | if (err < 0) { | |
2135 | pagevec_release(&pvec); | |
2136 | return err; | |
2137 | } | |
2138 | start++; | |
2139 | } | |
2140 | pagevec_release(&pvec); | |
2141 | } | |
2142 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2143 | mpd->map.m_len = 0; | |
2144 | mpd->map.m_flags = 0; | |
2145 | return 0; | |
2146 | } | |
2147 | ||
2148 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2149 | { | |
2150 | struct inode *inode = mpd->inode; | |
2151 | struct ext4_map_blocks *map = &mpd->map; | |
2152 | int get_blocks_flags; | |
090f32ee | 2153 | int err, dioread_nolock; |
4e7ea81d JK |
2154 | |
2155 | trace_ext4_da_write_pages_extent(inode, map); | |
2156 | /* | |
2157 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2158 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2159 | * where we have written into one or more preallocated blocks). It is |
2160 | * possible that we're going to need more metadata blocks than | |
2161 | * previously reserved. However we must not fail because we're in | |
2162 | * writeback and there is nothing we can do about it so it might result | |
2163 | * in data loss. So use reserved blocks to allocate metadata if | |
2164 | * possible. | |
2165 | * | |
754cfed6 TT |
2166 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2167 | * the blocks in question are delalloc blocks. This indicates | |
2168 | * that the blocks and quotas has already been checked when | |
2169 | * the data was copied into the page cache. | |
4e7ea81d JK |
2170 | */ |
2171 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
2172 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
090f32ee LC |
2173 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2174 | if (dioread_nolock) | |
4e7ea81d JK |
2175 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2176 | if (map->m_flags & (1 << BH_Delay)) | |
2177 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2178 | ||
2179 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2180 | if (err < 0) | |
2181 | return err; | |
090f32ee | 2182 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2183 | if (!mpd->io_submit.io_end->handle && |
2184 | ext4_handle_valid(handle)) { | |
2185 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2186 | handle->h_rsv_handle = NULL; | |
2187 | } | |
3613d228 | 2188 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2189 | } |
4e7ea81d JK |
2190 | |
2191 | BUG_ON(map->m_len == 0); | |
2192 | if (map->m_flags & EXT4_MAP_NEW) { | |
2193 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2194 | int i; | |
2195 | ||
2196 | for (i = 0; i < map->m_len; i++) | |
2197 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2198 | } | |
2199 | return 0; | |
2200 | } | |
2201 | ||
2202 | /* | |
2203 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2204 | * mpd->len and submit pages underlying it for IO | |
2205 | * | |
2206 | * @handle - handle for journal operations | |
2207 | * @mpd - extent to map | |
7534e854 JK |
2208 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2209 | * is no hope of writing the data. The caller should discard | |
2210 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2211 | * |
2212 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2213 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2214 | * them to initialized or split the described range from larger unwritten | |
2215 | * extent. Note that we need not map all the described range since allocation | |
2216 | * can return less blocks or the range is covered by more unwritten extents. We | |
2217 | * cannot map more because we are limited by reserved transaction credits. On | |
2218 | * the other hand we always make sure that the last touched page is fully | |
2219 | * mapped so that it can be written out (and thus forward progress is | |
2220 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2221 | */ | |
2222 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2223 | struct mpage_da_data *mpd, |
2224 | bool *give_up_on_write) | |
4e7ea81d JK |
2225 | { |
2226 | struct inode *inode = mpd->inode; | |
2227 | struct ext4_map_blocks *map = &mpd->map; | |
2228 | int err; | |
2229 | loff_t disksize; | |
6603120e | 2230 | int progress = 0; |
4e7ea81d JK |
2231 | |
2232 | mpd->io_submit.io_end->offset = | |
2233 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2234 | do { |
4e7ea81d JK |
2235 | err = mpage_map_one_extent(handle, mpd); |
2236 | if (err < 0) { | |
2237 | struct super_block *sb = inode->i_sb; | |
2238 | ||
cb530541 TT |
2239 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2240 | goto invalidate_dirty_pages; | |
4e7ea81d | 2241 | /* |
cb530541 TT |
2242 | * Let the uper layers retry transient errors. |
2243 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2244 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2245 | */ |
cb530541 | 2246 | if ((err == -ENOMEM) || |
6603120e DM |
2247 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2248 | if (progress) | |
2249 | goto update_disksize; | |
cb530541 | 2250 | return err; |
6603120e | 2251 | } |
cb530541 TT |
2252 | ext4_msg(sb, KERN_CRIT, |
2253 | "Delayed block allocation failed for " | |
2254 | "inode %lu at logical offset %llu with" | |
2255 | " max blocks %u with error %d", | |
2256 | inode->i_ino, | |
2257 | (unsigned long long)map->m_lblk, | |
2258 | (unsigned)map->m_len, -err); | |
2259 | ext4_msg(sb, KERN_CRIT, | |
2260 | "This should not happen!! Data will " | |
2261 | "be lost\n"); | |
2262 | if (err == -ENOSPC) | |
2263 | ext4_print_free_blocks(inode); | |
2264 | invalidate_dirty_pages: | |
2265 | *give_up_on_write = true; | |
4e7ea81d JK |
2266 | return err; |
2267 | } | |
6603120e | 2268 | progress = 1; |
4e7ea81d JK |
2269 | /* |
2270 | * Update buffer state, submit mapped pages, and get us new | |
2271 | * extent to map | |
2272 | */ | |
2273 | err = mpage_map_and_submit_buffers(mpd); | |
2274 | if (err < 0) | |
6603120e | 2275 | goto update_disksize; |
27d7c4ed | 2276 | } while (map->m_len); |
4e7ea81d | 2277 | |
6603120e | 2278 | update_disksize: |
622cad13 TT |
2279 | /* |
2280 | * Update on-disk size after IO is submitted. Races with | |
2281 | * truncate are avoided by checking i_size under i_data_sem. | |
2282 | */ | |
4e7ea81d | 2283 | disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; |
4e7ea81d JK |
2284 | if (disksize > EXT4_I(inode)->i_disksize) { |
2285 | int err2; | |
622cad13 TT |
2286 | loff_t i_size; |
2287 | ||
2288 | down_write(&EXT4_I(inode)->i_data_sem); | |
2289 | i_size = i_size_read(inode); | |
2290 | if (disksize > i_size) | |
2291 | disksize = i_size; | |
2292 | if (disksize > EXT4_I(inode)->i_disksize) | |
2293 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2294 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2295 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2296 | if (err2) |
2297 | ext4_error(inode->i_sb, | |
2298 | "Failed to mark inode %lu dirty", | |
2299 | inode->i_ino); | |
2300 | if (!err) | |
2301 | err = err2; | |
2302 | } | |
2303 | return err; | |
2304 | } | |
2305 | ||
fffb2739 JK |
2306 | /* |
2307 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2308 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2309 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2310 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2311 | * bpp - 1 blocks in bpp different extents. | |
2312 | */ | |
525f4ed8 MC |
2313 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2314 | { | |
fffb2739 | 2315 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2316 | |
fffb2739 JK |
2317 | return ext4_meta_trans_blocks(inode, |
2318 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2319 | } |
61628a3f | 2320 | |
8e48dcfb | 2321 | /* |
4e7ea81d JK |
2322 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2323 | * and underlying extent to map | |
2324 | * | |
2325 | * @mpd - where to look for pages | |
2326 | * | |
2327 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2328 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2329 | * extent of buffers underlying these pages that needs mapping (formed by | |
2330 | * either delayed or unwritten buffers). We also lock the pages containing | |
2331 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2332 | * mpd->lblk with length mpd->len blocks). | |
2333 | * | |
2334 | * Note that this function can attach bios to one io_end structure which are | |
2335 | * neither logically nor physically contiguous. Although it may seem as an | |
2336 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2337 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2338 | */ |
4e7ea81d | 2339 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2340 | { |
4e7ea81d JK |
2341 | struct address_space *mapping = mpd->inode->i_mapping; |
2342 | struct pagevec pvec; | |
2343 | unsigned int nr_pages; | |
aeac589a | 2344 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2345 | pgoff_t index = mpd->first_page; |
2346 | pgoff_t end = mpd->last_page; | |
2347 | int tag; | |
2348 | int i, err = 0; | |
2349 | int blkbits = mpd->inode->i_blkbits; | |
2350 | ext4_lblk_t lblk; | |
2351 | struct buffer_head *head; | |
8e48dcfb | 2352 | |
4e7ea81d | 2353 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2354 | tag = PAGECACHE_TAG_TOWRITE; |
2355 | else | |
2356 | tag = PAGECACHE_TAG_DIRTY; | |
2357 | ||
4e7ea81d JK |
2358 | pagevec_init(&pvec, 0); |
2359 | mpd->map.m_len = 0; | |
2360 | mpd->next_page = index; | |
4f01b02c | 2361 | while (index <= end) { |
5b41d924 | 2362 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2363 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2364 | if (nr_pages == 0) | |
4e7ea81d | 2365 | goto out; |
8e48dcfb TT |
2366 | |
2367 | for (i = 0; i < nr_pages; i++) { | |
2368 | struct page *page = pvec.pages[i]; | |
2369 | ||
2370 | /* | |
2371 | * At this point, the page may be truncated or | |
2372 | * invalidated (changing page->mapping to NULL), or | |
2373 | * even swizzled back from swapper_space to tmpfs file | |
2374 | * mapping. However, page->index will not change | |
2375 | * because we have a reference on the page. | |
2376 | */ | |
4f01b02c TT |
2377 | if (page->index > end) |
2378 | goto out; | |
8e48dcfb | 2379 | |
aeac589a ML |
2380 | /* |
2381 | * Accumulated enough dirty pages? This doesn't apply | |
2382 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2383 | * keep going because someone may be concurrently | |
2384 | * dirtying pages, and we might have synced a lot of | |
2385 | * newly appeared dirty pages, but have not synced all | |
2386 | * of the old dirty pages. | |
2387 | */ | |
2388 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2389 | goto out; | |
2390 | ||
4e7ea81d JK |
2391 | /* If we can't merge this page, we are done. */ |
2392 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2393 | goto out; | |
78aaced3 | 2394 | |
8e48dcfb | 2395 | lock_page(page); |
8e48dcfb | 2396 | /* |
4e7ea81d JK |
2397 | * If the page is no longer dirty, or its mapping no |
2398 | * longer corresponds to inode we are writing (which | |
2399 | * means it has been truncated or invalidated), or the | |
2400 | * page is already under writeback and we are not doing | |
2401 | * a data integrity writeback, skip the page | |
8e48dcfb | 2402 | */ |
4f01b02c TT |
2403 | if (!PageDirty(page) || |
2404 | (PageWriteback(page) && | |
4e7ea81d | 2405 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2406 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2407 | unlock_page(page); |
2408 | continue; | |
2409 | } | |
2410 | ||
7cb1a535 | 2411 | wait_on_page_writeback(page); |
8e48dcfb | 2412 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2413 | |
4e7ea81d | 2414 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2415 | mpd->first_page = page->index; |
8eb9e5ce | 2416 | mpd->next_page = page->index + 1; |
f8bec370 | 2417 | /* Add all dirty buffers to mpd */ |
4e7ea81d JK |
2418 | lblk = ((ext4_lblk_t)page->index) << |
2419 | (PAGE_CACHE_SHIFT - blkbits); | |
f8bec370 | 2420 | head = page_buffers(page); |
5f1132b2 JK |
2421 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2422 | if (err <= 0) | |
4e7ea81d | 2423 | goto out; |
5f1132b2 | 2424 | err = 0; |
aeac589a | 2425 | left--; |
8e48dcfb TT |
2426 | } |
2427 | pagevec_release(&pvec); | |
2428 | cond_resched(); | |
2429 | } | |
4f01b02c | 2430 | return 0; |
8eb9e5ce TT |
2431 | out: |
2432 | pagevec_release(&pvec); | |
4e7ea81d | 2433 | return err; |
8e48dcfb TT |
2434 | } |
2435 | ||
20970ba6 TT |
2436 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2437 | void *data) | |
2438 | { | |
2439 | struct address_space *mapping = data; | |
2440 | int ret = ext4_writepage(page, wbc); | |
2441 | mapping_set_error(mapping, ret); | |
2442 | return ret; | |
2443 | } | |
2444 | ||
2445 | static int ext4_writepages(struct address_space *mapping, | |
2446 | struct writeback_control *wbc) | |
64769240 | 2447 | { |
4e7ea81d JK |
2448 | pgoff_t writeback_index = 0; |
2449 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2450 | int range_whole = 0; |
4e7ea81d | 2451 | int cycled = 1; |
61628a3f | 2452 | handle_t *handle = NULL; |
df22291f | 2453 | struct mpage_da_data mpd; |
5e745b04 | 2454 | struct inode *inode = mapping->host; |
6b523df4 | 2455 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2456 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2457 | bool done; |
1bce63d1 | 2458 | struct blk_plug plug; |
cb530541 | 2459 | bool give_up_on_write = false; |
61628a3f | 2460 | |
20970ba6 | 2461 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2462 | |
61628a3f MC |
2463 | /* |
2464 | * No pages to write? This is mainly a kludge to avoid starting | |
2465 | * a transaction for special inodes like journal inode on last iput() | |
2466 | * because that could violate lock ordering on umount | |
2467 | */ | |
a1d6cc56 | 2468 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2469 | goto out_writepages; |
2a21e37e | 2470 | |
20970ba6 TT |
2471 | if (ext4_should_journal_data(inode)) { |
2472 | struct blk_plug plug; | |
20970ba6 TT |
2473 | |
2474 | blk_start_plug(&plug); | |
2475 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2476 | blk_finish_plug(&plug); | |
bbf023c7 | 2477 | goto out_writepages; |
20970ba6 TT |
2478 | } |
2479 | ||
2a21e37e TT |
2480 | /* |
2481 | * If the filesystem has aborted, it is read-only, so return | |
2482 | * right away instead of dumping stack traces later on that | |
2483 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2484 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2485 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2486 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2487 | * *never* be called, so if that ever happens, we would want |
2488 | * the stack trace. | |
2489 | */ | |
bbf023c7 ML |
2490 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2491 | ret = -EROFS; | |
2492 | goto out_writepages; | |
2493 | } | |
2a21e37e | 2494 | |
6b523df4 JK |
2495 | if (ext4_should_dioread_nolock(inode)) { |
2496 | /* | |
70261f56 | 2497 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2498 | * the page and we may dirty the inode. |
2499 | */ | |
2500 | rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); | |
2501 | } | |
2502 | ||
4e7ea81d JK |
2503 | /* |
2504 | * If we have inline data and arrive here, it means that | |
2505 | * we will soon create the block for the 1st page, so | |
2506 | * we'd better clear the inline data here. | |
2507 | */ | |
2508 | if (ext4_has_inline_data(inode)) { | |
2509 | /* Just inode will be modified... */ | |
2510 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2511 | if (IS_ERR(handle)) { | |
2512 | ret = PTR_ERR(handle); | |
2513 | goto out_writepages; | |
2514 | } | |
2515 | BUG_ON(ext4_test_inode_state(inode, | |
2516 | EXT4_STATE_MAY_INLINE_DATA)); | |
2517 | ext4_destroy_inline_data(handle, inode); | |
2518 | ext4_journal_stop(handle); | |
2519 | } | |
2520 | ||
22208ded AK |
2521 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2522 | range_whole = 1; | |
61628a3f | 2523 | |
2acf2c26 | 2524 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2525 | writeback_index = mapping->writeback_index; |
2526 | if (writeback_index) | |
2acf2c26 | 2527 | cycled = 0; |
4e7ea81d JK |
2528 | mpd.first_page = writeback_index; |
2529 | mpd.last_page = -1; | |
5b41d924 | 2530 | } else { |
4e7ea81d JK |
2531 | mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; |
2532 | mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; | |
5b41d924 | 2533 | } |
a1d6cc56 | 2534 | |
4e7ea81d JK |
2535 | mpd.inode = inode; |
2536 | mpd.wbc = wbc; | |
2537 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2538 | retry: |
6e6938b6 | 2539 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2540 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2541 | done = false; | |
1bce63d1 | 2542 | blk_start_plug(&plug); |
4e7ea81d JK |
2543 | while (!done && mpd.first_page <= mpd.last_page) { |
2544 | /* For each extent of pages we use new io_end */ | |
2545 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2546 | if (!mpd.io_submit.io_end) { | |
2547 | ret = -ENOMEM; | |
2548 | break; | |
2549 | } | |
a1d6cc56 AK |
2550 | |
2551 | /* | |
4e7ea81d JK |
2552 | * We have two constraints: We find one extent to map and we |
2553 | * must always write out whole page (makes a difference when | |
2554 | * blocksize < pagesize) so that we don't block on IO when we | |
2555 | * try to write out the rest of the page. Journalled mode is | |
2556 | * not supported by delalloc. | |
a1d6cc56 AK |
2557 | */ |
2558 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2559 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2560 | |
4e7ea81d | 2561 | /* start a new transaction */ |
6b523df4 JK |
2562 | handle = ext4_journal_start_with_reserve(inode, |
2563 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2564 | if (IS_ERR(handle)) { |
2565 | ret = PTR_ERR(handle); | |
1693918e | 2566 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2567 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2568 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2569 | /* Release allocated io_end */ |
2570 | ext4_put_io_end(mpd.io_submit.io_end); | |
2571 | break; | |
61628a3f | 2572 | } |
f63e6005 | 2573 | |
4e7ea81d JK |
2574 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2575 | ret = mpage_prepare_extent_to_map(&mpd); | |
2576 | if (!ret) { | |
2577 | if (mpd.map.m_len) | |
cb530541 TT |
2578 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2579 | &give_up_on_write); | |
4e7ea81d JK |
2580 | else { |
2581 | /* | |
2582 | * We scanned the whole range (or exhausted | |
2583 | * nr_to_write), submitted what was mapped and | |
2584 | * didn't find anything needing mapping. We are | |
2585 | * done. | |
2586 | */ | |
2587 | done = true; | |
2588 | } | |
f63e6005 | 2589 | } |
61628a3f | 2590 | ext4_journal_stop(handle); |
4e7ea81d JK |
2591 | /* Submit prepared bio */ |
2592 | ext4_io_submit(&mpd.io_submit); | |
2593 | /* Unlock pages we didn't use */ | |
cb530541 | 2594 | mpage_release_unused_pages(&mpd, give_up_on_write); |
4e7ea81d JK |
2595 | /* Drop our io_end reference we got from init */ |
2596 | ext4_put_io_end(mpd.io_submit.io_end); | |
2597 | ||
2598 | if (ret == -ENOSPC && sbi->s_journal) { | |
2599 | /* | |
2600 | * Commit the transaction which would | |
22208ded AK |
2601 | * free blocks released in the transaction |
2602 | * and try again | |
2603 | */ | |
df22291f | 2604 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2605 | ret = 0; |
4e7ea81d JK |
2606 | continue; |
2607 | } | |
2608 | /* Fatal error - ENOMEM, EIO... */ | |
2609 | if (ret) | |
61628a3f | 2610 | break; |
a1d6cc56 | 2611 | } |
1bce63d1 | 2612 | blk_finish_plug(&plug); |
9c12a831 | 2613 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2614 | cycled = 1; |
4e7ea81d JK |
2615 | mpd.last_page = writeback_index - 1; |
2616 | mpd.first_page = 0; | |
2acf2c26 AK |
2617 | goto retry; |
2618 | } | |
22208ded AK |
2619 | |
2620 | /* Update index */ | |
22208ded AK |
2621 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2622 | /* | |
4e7ea81d | 2623 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2624 | * mode will write it back later |
2625 | */ | |
4e7ea81d | 2626 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2627 | |
61628a3f | 2628 | out_writepages: |
20970ba6 TT |
2629 | trace_ext4_writepages_result(inode, wbc, ret, |
2630 | nr_to_write - wbc->nr_to_write); | |
61628a3f | 2631 | return ret; |
64769240 AT |
2632 | } |
2633 | ||
79f0be8d AK |
2634 | static int ext4_nonda_switch(struct super_block *sb) |
2635 | { | |
5c1ff336 | 2636 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2637 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2638 | ||
2639 | /* | |
2640 | * switch to non delalloc mode if we are running low | |
2641 | * on free block. The free block accounting via percpu | |
179f7ebf | 2642 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2643 | * accumulated on each CPU without updating global counters |
2644 | * Delalloc need an accurate free block accounting. So switch | |
2645 | * to non delalloc when we are near to error range. | |
2646 | */ | |
5c1ff336 EW |
2647 | free_clusters = |
2648 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2649 | dirty_clusters = | |
2650 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2651 | /* |
2652 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2653 | */ | |
5c1ff336 | 2654 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2655 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2656 | |
5c1ff336 EW |
2657 | if (2 * free_clusters < 3 * dirty_clusters || |
2658 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2659 | /* |
c8afb446 ES |
2660 | * free block count is less than 150% of dirty blocks |
2661 | * or free blocks is less than watermark | |
79f0be8d AK |
2662 | */ |
2663 | return 1; | |
2664 | } | |
2665 | return 0; | |
2666 | } | |
2667 | ||
0ff8947f ES |
2668 | /* We always reserve for an inode update; the superblock could be there too */ |
2669 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2670 | { | |
e2b911c5 | 2671 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2672 | return 1; |
2673 | ||
2674 | if (pos + len <= 0x7fffffffULL) | |
2675 | return 1; | |
2676 | ||
2677 | /* We might need to update the superblock to set LARGE_FILE */ | |
2678 | return 2; | |
2679 | } | |
2680 | ||
64769240 | 2681 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2682 | loff_t pos, unsigned len, unsigned flags, |
2683 | struct page **pagep, void **fsdata) | |
64769240 | 2684 | { |
72b8ab9d | 2685 | int ret, retries = 0; |
64769240 AT |
2686 | struct page *page; |
2687 | pgoff_t index; | |
64769240 AT |
2688 | struct inode *inode = mapping->host; |
2689 | handle_t *handle; | |
2690 | ||
2691 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2692 | |
2693 | if (ext4_nonda_switch(inode->i_sb)) { | |
2694 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2695 | return ext4_write_begin(file, mapping, pos, | |
2696 | len, flags, pagep, fsdata); | |
2697 | } | |
2698 | *fsdata = (void *)0; | |
9bffad1e | 2699 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2700 | |
2701 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2702 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2703 | pos, len, flags, | |
2704 | pagep, fsdata); | |
2705 | if (ret < 0) | |
47564bfb TT |
2706 | return ret; |
2707 | if (ret == 1) | |
2708 | return 0; | |
9c3569b5 TM |
2709 | } |
2710 | ||
47564bfb TT |
2711 | /* |
2712 | * grab_cache_page_write_begin() can take a long time if the | |
2713 | * system is thrashing due to memory pressure, or if the page | |
2714 | * is being written back. So grab it first before we start | |
2715 | * the transaction handle. This also allows us to allocate | |
2716 | * the page (if needed) without using GFP_NOFS. | |
2717 | */ | |
2718 | retry_grab: | |
2719 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2720 | if (!page) | |
2721 | return -ENOMEM; | |
2722 | unlock_page(page); | |
2723 | ||
64769240 AT |
2724 | /* |
2725 | * With delayed allocation, we don't log the i_disksize update | |
2726 | * if there is delayed block allocation. But we still need | |
2727 | * to journalling the i_disksize update if writes to the end | |
2728 | * of file which has an already mapped buffer. | |
2729 | */ | |
47564bfb | 2730 | retry_journal: |
0ff8947f ES |
2731 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2732 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2733 | if (IS_ERR(handle)) { |
47564bfb TT |
2734 | page_cache_release(page); |
2735 | return PTR_ERR(handle); | |
64769240 AT |
2736 | } |
2737 | ||
47564bfb TT |
2738 | lock_page(page); |
2739 | if (page->mapping != mapping) { | |
2740 | /* The page got truncated from under us */ | |
2741 | unlock_page(page); | |
2742 | page_cache_release(page); | |
d5a0d4f7 | 2743 | ext4_journal_stop(handle); |
47564bfb | 2744 | goto retry_grab; |
d5a0d4f7 | 2745 | } |
47564bfb | 2746 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2747 | wait_for_stable_page(page); |
64769240 | 2748 | |
2058f83a MH |
2749 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
2750 | ret = ext4_block_write_begin(page, pos, len, | |
2751 | ext4_da_get_block_prep); | |
2752 | #else | |
6e1db88d | 2753 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2754 | #endif |
64769240 AT |
2755 | if (ret < 0) { |
2756 | unlock_page(page); | |
2757 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2758 | /* |
2759 | * block_write_begin may have instantiated a few blocks | |
2760 | * outside i_size. Trim these off again. Don't need | |
2761 | * i_size_read because we hold i_mutex. | |
2762 | */ | |
2763 | if (pos + len > inode->i_size) | |
b9a4207d | 2764 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2765 | |
2766 | if (ret == -ENOSPC && | |
2767 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2768 | goto retry_journal; | |
2769 | ||
2770 | page_cache_release(page); | |
2771 | return ret; | |
64769240 AT |
2772 | } |
2773 | ||
47564bfb | 2774 | *pagep = page; |
64769240 AT |
2775 | return ret; |
2776 | } | |
2777 | ||
632eaeab MC |
2778 | /* |
2779 | * Check if we should update i_disksize | |
2780 | * when write to the end of file but not require block allocation | |
2781 | */ | |
2782 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2783 | unsigned long offset) |
632eaeab MC |
2784 | { |
2785 | struct buffer_head *bh; | |
2786 | struct inode *inode = page->mapping->host; | |
2787 | unsigned int idx; | |
2788 | int i; | |
2789 | ||
2790 | bh = page_buffers(page); | |
2791 | idx = offset >> inode->i_blkbits; | |
2792 | ||
af5bc92d | 2793 | for (i = 0; i < idx; i++) |
632eaeab MC |
2794 | bh = bh->b_this_page; |
2795 | ||
29fa89d0 | 2796 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2797 | return 0; |
2798 | return 1; | |
2799 | } | |
2800 | ||
64769240 | 2801 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2802 | struct address_space *mapping, |
2803 | loff_t pos, unsigned len, unsigned copied, | |
2804 | struct page *page, void *fsdata) | |
64769240 AT |
2805 | { |
2806 | struct inode *inode = mapping->host; | |
2807 | int ret = 0, ret2; | |
2808 | handle_t *handle = ext4_journal_current_handle(); | |
2809 | loff_t new_i_size; | |
632eaeab | 2810 | unsigned long start, end; |
79f0be8d AK |
2811 | int write_mode = (int)(unsigned long)fsdata; |
2812 | ||
74d553aa TT |
2813 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2814 | return ext4_write_end(file, mapping, pos, | |
2815 | len, copied, page, fsdata); | |
632eaeab | 2816 | |
9bffad1e | 2817 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2818 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2819 | end = start + copied - 1; |
64769240 AT |
2820 | |
2821 | /* | |
2822 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2823 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2824 | * into that. | |
2825 | */ | |
64769240 | 2826 | new_i_size = pos + copied; |
ea51d132 | 2827 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2828 | if (ext4_has_inline_data(inode) || |
2829 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 2830 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
2831 | /* We need to mark inode dirty even if |
2832 | * new_i_size is less that inode->i_size | |
2833 | * bu greater than i_disksize.(hint delalloc) | |
2834 | */ | |
2835 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2836 | } |
632eaeab | 2837 | } |
9c3569b5 TM |
2838 | |
2839 | if (write_mode != CONVERT_INLINE_DATA && | |
2840 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2841 | ext4_has_inline_data(inode)) | |
2842 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2843 | page); | |
2844 | else | |
2845 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2846 | page, fsdata); |
9c3569b5 | 2847 | |
64769240 AT |
2848 | copied = ret2; |
2849 | if (ret2 < 0) | |
2850 | ret = ret2; | |
2851 | ret2 = ext4_journal_stop(handle); | |
2852 | if (!ret) | |
2853 | ret = ret2; | |
2854 | ||
2855 | return ret ? ret : copied; | |
2856 | } | |
2857 | ||
d47992f8 LC |
2858 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
2859 | unsigned int length) | |
64769240 | 2860 | { |
64769240 AT |
2861 | /* |
2862 | * Drop reserved blocks | |
2863 | */ | |
2864 | BUG_ON(!PageLocked(page)); | |
2865 | if (!page_has_buffers(page)) | |
2866 | goto out; | |
2867 | ||
ca99fdd2 | 2868 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
2869 | |
2870 | out: | |
d47992f8 | 2871 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
2872 | |
2873 | return; | |
2874 | } | |
2875 | ||
ccd2506b TT |
2876 | /* |
2877 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2878 | */ | |
2879 | int ext4_alloc_da_blocks(struct inode *inode) | |
2880 | { | |
fb40ba0d TT |
2881 | trace_ext4_alloc_da_blocks(inode); |
2882 | ||
71d4f7d0 | 2883 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
2884 | return 0; |
2885 | ||
2886 | /* | |
2887 | * We do something simple for now. The filemap_flush() will | |
2888 | * also start triggering a write of the data blocks, which is | |
2889 | * not strictly speaking necessary (and for users of | |
2890 | * laptop_mode, not even desirable). However, to do otherwise | |
2891 | * would require replicating code paths in: | |
de9a55b8 | 2892 | * |
20970ba6 | 2893 | * ext4_writepages() -> |
ccd2506b TT |
2894 | * write_cache_pages() ---> (via passed in callback function) |
2895 | * __mpage_da_writepage() --> | |
2896 | * mpage_add_bh_to_extent() | |
2897 | * mpage_da_map_blocks() | |
2898 | * | |
2899 | * The problem is that write_cache_pages(), located in | |
2900 | * mm/page-writeback.c, marks pages clean in preparation for | |
2901 | * doing I/O, which is not desirable if we're not planning on | |
2902 | * doing I/O at all. | |
2903 | * | |
2904 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2905 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2906 | * would be ugly in the extreme. So instead we would need to |
2907 | * replicate parts of the code in the above functions, | |
25985edc | 2908 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2909 | * write out the pages, but rather only collect contiguous |
2910 | * logical block extents, call the multi-block allocator, and | |
2911 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2912 | * |
ccd2506b TT |
2913 | * For now, though, we'll cheat by calling filemap_flush(), |
2914 | * which will map the blocks, and start the I/O, but not | |
2915 | * actually wait for the I/O to complete. | |
2916 | */ | |
2917 | return filemap_flush(inode->i_mapping); | |
2918 | } | |
64769240 | 2919 | |
ac27a0ec DK |
2920 | /* |
2921 | * bmap() is special. It gets used by applications such as lilo and by | |
2922 | * the swapper to find the on-disk block of a specific piece of data. | |
2923 | * | |
2924 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2925 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2926 | * filesystem and enables swap, then they may get a nasty shock when the |
2927 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2928 | * the original zero's written out previously to the journal and | |
2929 | * awaiting writeback in the kernel's buffer cache. | |
2930 | * | |
2931 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2932 | * take extra steps to flush any blocks which might be in the cache. | |
2933 | */ | |
617ba13b | 2934 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2935 | { |
2936 | struct inode *inode = mapping->host; | |
2937 | journal_t *journal; | |
2938 | int err; | |
2939 | ||
46c7f254 TM |
2940 | /* |
2941 | * We can get here for an inline file via the FIBMAP ioctl | |
2942 | */ | |
2943 | if (ext4_has_inline_data(inode)) | |
2944 | return 0; | |
2945 | ||
64769240 AT |
2946 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2947 | test_opt(inode->i_sb, DELALLOC)) { | |
2948 | /* | |
2949 | * With delalloc we want to sync the file | |
2950 | * so that we can make sure we allocate | |
2951 | * blocks for file | |
2952 | */ | |
2953 | filemap_write_and_wait(mapping); | |
2954 | } | |
2955 | ||
19f5fb7a TT |
2956 | if (EXT4_JOURNAL(inode) && |
2957 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2958 | /* |
2959 | * This is a REALLY heavyweight approach, but the use of | |
2960 | * bmap on dirty files is expected to be extremely rare: | |
2961 | * only if we run lilo or swapon on a freshly made file | |
2962 | * do we expect this to happen. | |
2963 | * | |
2964 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2965 | * represent an unprivileged user DOS attack --- we'd be | |
2966 | * in trouble if mortal users could trigger this path at | |
2967 | * will.) | |
2968 | * | |
617ba13b | 2969 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2970 | * regular files. If somebody wants to bmap a directory |
2971 | * or symlink and gets confused because the buffer | |
2972 | * hasn't yet been flushed to disk, they deserve | |
2973 | * everything they get. | |
2974 | */ | |
2975 | ||
19f5fb7a | 2976 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2977 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2978 | jbd2_journal_lock_updates(journal); |
2979 | err = jbd2_journal_flush(journal); | |
2980 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2981 | |
2982 | if (err) | |
2983 | return 0; | |
2984 | } | |
2985 | ||
af5bc92d | 2986 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2987 | } |
2988 | ||
617ba13b | 2989 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2990 | { |
46c7f254 TM |
2991 | int ret = -EAGAIN; |
2992 | struct inode *inode = page->mapping->host; | |
2993 | ||
0562e0ba | 2994 | trace_ext4_readpage(page); |
46c7f254 TM |
2995 | |
2996 | if (ext4_has_inline_data(inode)) | |
2997 | ret = ext4_readpage_inline(inode, page); | |
2998 | ||
2999 | if (ret == -EAGAIN) | |
f64e02fe | 3000 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3001 | |
3002 | return ret; | |
ac27a0ec DK |
3003 | } |
3004 | ||
3005 | static int | |
617ba13b | 3006 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3007 | struct list_head *pages, unsigned nr_pages) |
3008 | { | |
46c7f254 TM |
3009 | struct inode *inode = mapping->host; |
3010 | ||
3011 | /* If the file has inline data, no need to do readpages. */ | |
3012 | if (ext4_has_inline_data(inode)) | |
3013 | return 0; | |
3014 | ||
f64e02fe | 3015 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3016 | } |
3017 | ||
d47992f8 LC |
3018 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3019 | unsigned int length) | |
ac27a0ec | 3020 | { |
ca99fdd2 | 3021 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3022 | |
4520fb3c JK |
3023 | /* No journalling happens on data buffers when this function is used */ |
3024 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3025 | ||
ca99fdd2 | 3026 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3027 | } |
3028 | ||
53e87268 | 3029 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3030 | unsigned int offset, |
3031 | unsigned int length) | |
4520fb3c JK |
3032 | { |
3033 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3034 | ||
ca99fdd2 | 3035 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3036 | |
ac27a0ec DK |
3037 | /* |
3038 | * If it's a full truncate we just forget about the pending dirtying | |
3039 | */ | |
ca99fdd2 | 3040 | if (offset == 0 && length == PAGE_CACHE_SIZE) |
ac27a0ec DK |
3041 | ClearPageChecked(page); |
3042 | ||
ca99fdd2 | 3043 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3044 | } |
3045 | ||
3046 | /* Wrapper for aops... */ | |
3047 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3048 | unsigned int offset, |
3049 | unsigned int length) | |
53e87268 | 3050 | { |
ca99fdd2 | 3051 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3052 | } |
3053 | ||
617ba13b | 3054 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3055 | { |
617ba13b | 3056 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3057 | |
0562e0ba JZ |
3058 | trace_ext4_releasepage(page); |
3059 | ||
e1c36595 JK |
3060 | /* Page has dirty journalled data -> cannot release */ |
3061 | if (PageChecked(page)) | |
ac27a0ec | 3062 | return 0; |
0390131b FM |
3063 | if (journal) |
3064 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3065 | else | |
3066 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3067 | } |
3068 | ||
2ed88685 TT |
3069 | /* |
3070 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3071 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3072 | * The extent will be converted to initialized after the IO is complete. | |
3073 | */ | |
f19d5870 | 3074 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3075 | struct buffer_head *bh_result, int create) |
3076 | { | |
c7064ef1 | 3077 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3078 | inode->i_ino, create); |
2ed88685 TT |
3079 | return _ext4_get_block(inode, iblock, bh_result, |
3080 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3081 | } |
3082 | ||
2dcba478 | 3083 | static int ext4_get_block_overwrite(struct inode *inode, sector_t iblock, |
8b0f165f | 3084 | struct buffer_head *bh_result, int create) |
729f52c6 | 3085 | { |
2dcba478 JK |
3086 | int ret; |
3087 | ||
3088 | ext4_debug("ext4_get_block_overwrite: inode %lu, create flag %d\n", | |
8b0f165f | 3089 | inode->i_ino, create); |
2dcba478 JK |
3090 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
3091 | /* | |
3092 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
3093 | * that. | |
3094 | */ | |
3095 | WARN_ON_ONCE(!buffer_mapped(bh_result)); | |
3096 | ||
3097 | return ret; | |
729f52c6 ZL |
3098 | } |
3099 | ||
ed923b57 MW |
3100 | int ext4_get_block_dax(struct inode *inode, sector_t iblock, |
3101 | struct buffer_head *bh_result, int create) | |
3102 | { | |
3103 | int flags = EXT4_GET_BLOCKS_PRE_IO | EXT4_GET_BLOCKS_UNWRIT_EXT; | |
c86d8db3 | 3104 | |
ed923b57 MW |
3105 | if (create) |
3106 | flags |= EXT4_GET_BLOCKS_CREATE; | |
3107 | ext4_debug("ext4_get_block_dax: inode %lu, create flag %d\n", | |
3108 | inode->i_ino, create); | |
3109 | return _ext4_get_block(inode, iblock, bh_result, flags); | |
3110 | } | |
3111 | ||
4c0425ff | 3112 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3113 | ssize_t size, void *private) |
4c0425ff MC |
3114 | { |
3115 | ext4_io_end_t *io_end = iocb->private; | |
4c0425ff | 3116 | |
97a851ed | 3117 | /* if not async direct IO just return */ |
7b7a8665 | 3118 | if (!io_end) |
97a851ed | 3119 | return; |
4b70df18 | 3120 | |
88635ca2 | 3121 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3122 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3123 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3124 | size); | |
8d5d02e6 | 3125 | |
b5a7e970 | 3126 | iocb->private = NULL; |
4c0425ff MC |
3127 | io_end->offset = offset; |
3128 | io_end->size = size; | |
7b7a8665 | 3129 | ext4_put_io_end(io_end); |
4c0425ff | 3130 | } |
c7064ef1 | 3131 | |
4c0425ff MC |
3132 | /* |
3133 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3134 | * preallocated extents, and those write extend the file, no need to | |
3135 | * fall back to buffered IO. | |
3136 | * | |
556615dc | 3137 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3138 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3139 | * still keep the range to write as unwritten. |
4c0425ff | 3140 | * |
69c499d1 | 3141 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3142 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3143 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3144 | * when async direct IO completed. |
4c0425ff MC |
3145 | * |
3146 | * If the O_DIRECT write will extend the file then add this inode to the | |
3147 | * orphan list. So recovery will truncate it back to the original size | |
3148 | * if the machine crashes during the write. | |
3149 | * | |
3150 | */ | |
6f673763 OS |
3151 | static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
3152 | loff_t offset) | |
4c0425ff MC |
3153 | { |
3154 | struct file *file = iocb->ki_filp; | |
3155 | struct inode *inode = file->f_mapping->host; | |
3156 | ssize_t ret; | |
a6cbcd4a | 3157 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3158 | int overwrite = 0; |
3159 | get_block_t *get_block_func = NULL; | |
3160 | int dio_flags = 0; | |
4c0425ff | 3161 | loff_t final_size = offset + count; |
97a851ed | 3162 | ext4_io_end_t *io_end = NULL; |
729f52c6 | 3163 | |
69c499d1 | 3164 | /* Use the old path for reads and writes beyond i_size. */ |
6f673763 OS |
3165 | if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size) |
3166 | return ext4_ind_direct_IO(iocb, iter, offset); | |
4bd809db | 3167 | |
69c499d1 | 3168 | BUG_ON(iocb->private == NULL); |
4bd809db | 3169 | |
e8340395 JK |
3170 | /* |
3171 | * Make all waiters for direct IO properly wait also for extent | |
3172 | * conversion. This also disallows race between truncate() and | |
3173 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3174 | */ | |
6f673763 | 3175 | if (iov_iter_rw(iter) == WRITE) |
fe0f07d0 | 3176 | inode_dio_begin(inode); |
e8340395 | 3177 | |
69c499d1 TT |
3178 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3179 | overwrite = *((int *)iocb->private); | |
4bd809db | 3180 | |
2dcba478 | 3181 | if (overwrite) |
69c499d1 | 3182 | mutex_unlock(&inode->i_mutex); |
8d5d02e6 | 3183 | |
69c499d1 TT |
3184 | /* |
3185 | * We could direct write to holes and fallocate. | |
3186 | * | |
3187 | * Allocated blocks to fill the hole are marked as | |
556615dc | 3188 | * unwritten to prevent parallel buffered read to expose |
69c499d1 TT |
3189 | * the stale data before DIO complete the data IO. |
3190 | * | |
3191 | * As to previously fallocated extents, ext4 get_block will | |
3192 | * just simply mark the buffer mapped but still keep the | |
556615dc | 3193 | * extents unwritten. |
69c499d1 TT |
3194 | * |
3195 | * For non AIO case, we will convert those unwritten extents | |
3196 | * to written after return back from blockdev_direct_IO. | |
3197 | * | |
3198 | * For async DIO, the conversion needs to be deferred when the | |
3199 | * IO is completed. The ext4 end_io callback function will be | |
3200 | * called to take care of the conversion work. Here for async | |
3201 | * case, we allocate an io_end structure to hook to the iocb. | |
3202 | */ | |
3203 | iocb->private = NULL; | |
3204 | ext4_inode_aio_set(inode, NULL); | |
3205 | if (!is_sync_kiocb(iocb)) { | |
97a851ed | 3206 | io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3207 | if (!io_end) { |
3208 | ret = -ENOMEM; | |
3209 | goto retake_lock; | |
8b0f165f | 3210 | } |
97a851ed JK |
3211 | /* |
3212 | * Grab reference for DIO. Will be dropped in ext4_end_io_dio() | |
3213 | */ | |
3214 | iocb->private = ext4_get_io_end(io_end); | |
8d5d02e6 | 3215 | /* |
69c499d1 TT |
3216 | * we save the io structure for current async direct |
3217 | * IO, so that later ext4_map_blocks() could flag the | |
3218 | * io structure whether there is a unwritten extents | |
3219 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3220 | */ |
69c499d1 TT |
3221 | ext4_inode_aio_set(inode, io_end); |
3222 | } | |
4bd809db | 3223 | |
69c499d1 | 3224 | if (overwrite) { |
2dcba478 | 3225 | get_block_func = ext4_get_block_overwrite; |
69c499d1 TT |
3226 | } else { |
3227 | get_block_func = ext4_get_block_write; | |
3228 | dio_flags = DIO_LOCKING; | |
3229 | } | |
2058f83a MH |
3230 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3231 | BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); | |
3232 | #endif | |
923ae0ff | 3233 | if (IS_DAX(inode)) |
a95cd631 | 3234 | ret = dax_do_io(iocb, inode, iter, offset, get_block_func, |
923ae0ff RZ |
3235 | ext4_end_io_dio, dio_flags); |
3236 | else | |
17f8c842 | 3237 | ret = __blockdev_direct_IO(iocb, inode, |
923ae0ff RZ |
3238 | inode->i_sb->s_bdev, iter, offset, |
3239 | get_block_func, | |
3240 | ext4_end_io_dio, NULL, dio_flags); | |
69c499d1 | 3241 | |
69c499d1 | 3242 | /* |
97a851ed JK |
3243 | * Put our reference to io_end. This can free the io_end structure e.g. |
3244 | * in sync IO case or in case of error. It can even perform extent | |
3245 | * conversion if all bios we submitted finished before we got here. | |
3246 | * Note that in that case iocb->private can be already set to NULL | |
3247 | * here. | |
69c499d1 | 3248 | */ |
97a851ed JK |
3249 | if (io_end) { |
3250 | ext4_inode_aio_set(inode, NULL); | |
3251 | ext4_put_io_end(io_end); | |
3252 | /* | |
3253 | * When no IO was submitted ext4_end_io_dio() was not | |
3254 | * called so we have to put iocb's reference. | |
3255 | */ | |
3256 | if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { | |
3257 | WARN_ON(iocb->private != io_end); | |
3258 | WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
97a851ed JK |
3259 | ext4_put_io_end(io_end); |
3260 | iocb->private = NULL; | |
3261 | } | |
3262 | } | |
3263 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3264 | EXT4_STATE_DIO_UNWRITTEN)) { |
3265 | int err; | |
3266 | /* | |
3267 | * for non AIO case, since the IO is already | |
3268 | * completed, we could do the conversion right here | |
3269 | */ | |
6b523df4 | 3270 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3271 | offset, ret); |
3272 | if (err < 0) | |
3273 | ret = err; | |
3274 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3275 | } | |
4bd809db | 3276 | |
69c499d1 | 3277 | retake_lock: |
6f673763 | 3278 | if (iov_iter_rw(iter) == WRITE) |
fe0f07d0 | 3279 | inode_dio_end(inode); |
69c499d1 | 3280 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3281 | if (overwrite) |
69c499d1 | 3282 | mutex_lock(&inode->i_mutex); |
8d5d02e6 | 3283 | |
69c499d1 | 3284 | return ret; |
4c0425ff MC |
3285 | } |
3286 | ||
22c6186e OS |
3287 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
3288 | loff_t offset) | |
4c0425ff MC |
3289 | { |
3290 | struct file *file = iocb->ki_filp; | |
3291 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3292 | size_t count = iov_iter_count(iter); |
0562e0ba | 3293 | ssize_t ret; |
4c0425ff | 3294 | |
2058f83a MH |
3295 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3296 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3297 | return 0; | |
3298 | #endif | |
3299 | ||
84ebd795 TT |
3300 | /* |
3301 | * If we are doing data journalling we don't support O_DIRECT | |
3302 | */ | |
3303 | if (ext4_should_journal_data(inode)) | |
3304 | return 0; | |
3305 | ||
46c7f254 TM |
3306 | /* Let buffer I/O handle the inline data case. */ |
3307 | if (ext4_has_inline_data(inode)) | |
3308 | return 0; | |
3309 | ||
6f673763 | 3310 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
12e9b892 | 3311 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
6f673763 | 3312 | ret = ext4_ext_direct_IO(iocb, iter, offset); |
0562e0ba | 3313 | else |
6f673763 OS |
3314 | ret = ext4_ind_direct_IO(iocb, iter, offset); |
3315 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); | |
0562e0ba | 3316 | return ret; |
4c0425ff MC |
3317 | } |
3318 | ||
ac27a0ec | 3319 | /* |
617ba13b | 3320 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3321 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3322 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3323 | * not necessarily locked. | |
3324 | * | |
3325 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3326 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3327 | * or jbddirty because all the journalling code will explode. | |
3328 | * | |
3329 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3330 | * is called, propagate that into the buffers appropriately. | |
3331 | */ | |
617ba13b | 3332 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3333 | { |
3334 | SetPageChecked(page); | |
3335 | return __set_page_dirty_nobuffers(page); | |
3336 | } | |
3337 | ||
74d553aa | 3338 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3339 | .readpage = ext4_readpage, |
3340 | .readpages = ext4_readpages, | |
43ce1d23 | 3341 | .writepage = ext4_writepage, |
20970ba6 | 3342 | .writepages = ext4_writepages, |
8ab22b9a | 3343 | .write_begin = ext4_write_begin, |
74d553aa | 3344 | .write_end = ext4_write_end, |
8ab22b9a HH |
3345 | .bmap = ext4_bmap, |
3346 | .invalidatepage = ext4_invalidatepage, | |
3347 | .releasepage = ext4_releasepage, | |
3348 | .direct_IO = ext4_direct_IO, | |
3349 | .migratepage = buffer_migrate_page, | |
3350 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3351 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3352 | }; |
3353 | ||
617ba13b | 3354 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3355 | .readpage = ext4_readpage, |
3356 | .readpages = ext4_readpages, | |
43ce1d23 | 3357 | .writepage = ext4_writepage, |
20970ba6 | 3358 | .writepages = ext4_writepages, |
8ab22b9a HH |
3359 | .write_begin = ext4_write_begin, |
3360 | .write_end = ext4_journalled_write_end, | |
3361 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3362 | .bmap = ext4_bmap, | |
4520fb3c | 3363 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3364 | .releasepage = ext4_releasepage, |
84ebd795 | 3365 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3366 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3367 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3368 | }; |
3369 | ||
64769240 | 3370 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3371 | .readpage = ext4_readpage, |
3372 | .readpages = ext4_readpages, | |
43ce1d23 | 3373 | .writepage = ext4_writepage, |
20970ba6 | 3374 | .writepages = ext4_writepages, |
8ab22b9a HH |
3375 | .write_begin = ext4_da_write_begin, |
3376 | .write_end = ext4_da_write_end, | |
3377 | .bmap = ext4_bmap, | |
3378 | .invalidatepage = ext4_da_invalidatepage, | |
3379 | .releasepage = ext4_releasepage, | |
3380 | .direct_IO = ext4_direct_IO, | |
3381 | .migratepage = buffer_migrate_page, | |
3382 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3383 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3384 | }; |
3385 | ||
617ba13b | 3386 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3387 | { |
3d2b1582 LC |
3388 | switch (ext4_inode_journal_mode(inode)) { |
3389 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3390 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3391 | break; |
3392 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3393 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3394 | break; |
3395 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3396 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3397 | return; |
3d2b1582 LC |
3398 | default: |
3399 | BUG(); | |
3400 | } | |
74d553aa TT |
3401 | if (test_opt(inode->i_sb, DELALLOC)) |
3402 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3403 | else | |
3404 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3405 | } |
3406 | ||
923ae0ff | 3407 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3408 | struct address_space *mapping, loff_t from, loff_t length) |
3409 | { | |
3410 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3411 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
923ae0ff | 3412 | unsigned blocksize, pos; |
d863dc36 LC |
3413 | ext4_lblk_t iblock; |
3414 | struct inode *inode = mapping->host; | |
3415 | struct buffer_head *bh; | |
3416 | struct page *page; | |
3417 | int err = 0; | |
3418 | ||
3419 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
c62d2555 | 3420 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3421 | if (!page) |
3422 | return -ENOMEM; | |
3423 | ||
3424 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 LC |
3425 | |
3426 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3427 | ||
3428 | if (!page_has_buffers(page)) | |
3429 | create_empty_buffers(page, blocksize, 0); | |
3430 | ||
3431 | /* Find the buffer that contains "offset" */ | |
3432 | bh = page_buffers(page); | |
3433 | pos = blocksize; | |
3434 | while (offset >= pos) { | |
3435 | bh = bh->b_this_page; | |
3436 | iblock++; | |
3437 | pos += blocksize; | |
3438 | } | |
d863dc36 LC |
3439 | if (buffer_freed(bh)) { |
3440 | BUFFER_TRACE(bh, "freed: skip"); | |
3441 | goto unlock; | |
3442 | } | |
d863dc36 LC |
3443 | if (!buffer_mapped(bh)) { |
3444 | BUFFER_TRACE(bh, "unmapped"); | |
3445 | ext4_get_block(inode, iblock, bh, 0); | |
3446 | /* unmapped? It's a hole - nothing to do */ | |
3447 | if (!buffer_mapped(bh)) { | |
3448 | BUFFER_TRACE(bh, "still unmapped"); | |
3449 | goto unlock; | |
3450 | } | |
3451 | } | |
3452 | ||
3453 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3454 | if (PageUptodate(page)) | |
3455 | set_buffer_uptodate(bh); | |
3456 | ||
3457 | if (!buffer_uptodate(bh)) { | |
3458 | err = -EIO; | |
3459 | ll_rw_block(READ, 1, &bh); | |
3460 | wait_on_buffer(bh); | |
3461 | /* Uhhuh. Read error. Complain and punt. */ | |
3462 | if (!buffer_uptodate(bh)) | |
3463 | goto unlock; | |
c9c7429c MH |
3464 | if (S_ISREG(inode->i_mode) && |
3465 | ext4_encrypted_inode(inode)) { | |
3466 | /* We expect the key to be set. */ | |
3467 | BUG_ON(!ext4_has_encryption_key(inode)); | |
3468 | BUG_ON(blocksize != PAGE_CACHE_SIZE); | |
3684de8c | 3469 | WARN_ON_ONCE(ext4_decrypt(page)); |
c9c7429c | 3470 | } |
d863dc36 | 3471 | } |
d863dc36 LC |
3472 | if (ext4_should_journal_data(inode)) { |
3473 | BUFFER_TRACE(bh, "get write access"); | |
3474 | err = ext4_journal_get_write_access(handle, bh); | |
3475 | if (err) | |
3476 | goto unlock; | |
3477 | } | |
d863dc36 | 3478 | zero_user(page, offset, length); |
d863dc36 LC |
3479 | BUFFER_TRACE(bh, "zeroed end of block"); |
3480 | ||
d863dc36 LC |
3481 | if (ext4_should_journal_data(inode)) { |
3482 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3483 | } else { |
353eefd3 | 3484 | err = 0; |
d863dc36 | 3485 | mark_buffer_dirty(bh); |
0713ed0c LC |
3486 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) |
3487 | err = ext4_jbd2_file_inode(handle, inode); | |
3488 | } | |
d863dc36 LC |
3489 | |
3490 | unlock: | |
3491 | unlock_page(page); | |
3492 | page_cache_release(page); | |
3493 | return err; | |
3494 | } | |
3495 | ||
923ae0ff RZ |
3496 | /* |
3497 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3498 | * starting from file offset 'from'. The range to be zero'd must | |
3499 | * be contained with in one block. If the specified range exceeds | |
3500 | * the end of the block it will be shortened to end of the block | |
3501 | * that cooresponds to 'from' | |
3502 | */ | |
3503 | static int ext4_block_zero_page_range(handle_t *handle, | |
3504 | struct address_space *mapping, loff_t from, loff_t length) | |
3505 | { | |
3506 | struct inode *inode = mapping->host; | |
3507 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3508 | unsigned blocksize = inode->i_sb->s_blocksize; | |
3509 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3510 | ||
3511 | /* | |
3512 | * correct length if it does not fall between | |
3513 | * 'from' and the end of the block | |
3514 | */ | |
3515 | if (length > max || length < 0) | |
3516 | length = max; | |
3517 | ||
3518 | if (IS_DAX(inode)) | |
3519 | return dax_zero_page_range(inode, from, length, ext4_get_block); | |
3520 | return __ext4_block_zero_page_range(handle, mapping, from, length); | |
3521 | } | |
3522 | ||
94350ab5 MW |
3523 | /* |
3524 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3525 | * up to the end of the block which corresponds to `from'. | |
3526 | * This required during truncate. We need to physically zero the tail end | |
3527 | * of that block so it doesn't yield old data if the file is later grown. | |
3528 | */ | |
c197855e | 3529 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3530 | struct address_space *mapping, loff_t from) |
3531 | { | |
3532 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3533 | unsigned length; | |
3534 | unsigned blocksize; | |
3535 | struct inode *inode = mapping->host; | |
3536 | ||
3537 | blocksize = inode->i_sb->s_blocksize; | |
3538 | length = blocksize - (offset & (blocksize - 1)); | |
3539 | ||
3540 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3541 | } | |
3542 | ||
a87dd18c LC |
3543 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3544 | loff_t lstart, loff_t length) | |
3545 | { | |
3546 | struct super_block *sb = inode->i_sb; | |
3547 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3548 | unsigned partial_start, partial_end; |
a87dd18c LC |
3549 | ext4_fsblk_t start, end; |
3550 | loff_t byte_end = (lstart + length - 1); | |
3551 | int err = 0; | |
3552 | ||
e1be3a92 LC |
3553 | partial_start = lstart & (sb->s_blocksize - 1); |
3554 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3555 | ||
a87dd18c LC |
3556 | start = lstart >> sb->s_blocksize_bits; |
3557 | end = byte_end >> sb->s_blocksize_bits; | |
3558 | ||
3559 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3560 | if (start == end && |
3561 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3562 | err = ext4_block_zero_page_range(handle, mapping, |
3563 | lstart, length); | |
3564 | return err; | |
3565 | } | |
3566 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3567 | if (partial_start) { |
a87dd18c LC |
3568 | err = ext4_block_zero_page_range(handle, mapping, |
3569 | lstart, sb->s_blocksize); | |
3570 | if (err) | |
3571 | return err; | |
3572 | } | |
3573 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3574 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3575 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3576 | byte_end - partial_end, |
3577 | partial_end + 1); | |
a87dd18c LC |
3578 | return err; |
3579 | } | |
3580 | ||
91ef4caf DG |
3581 | int ext4_can_truncate(struct inode *inode) |
3582 | { | |
91ef4caf DG |
3583 | if (S_ISREG(inode->i_mode)) |
3584 | return 1; | |
3585 | if (S_ISDIR(inode->i_mode)) | |
3586 | return 1; | |
3587 | if (S_ISLNK(inode->i_mode)) | |
3588 | return !ext4_inode_is_fast_symlink(inode); | |
3589 | return 0; | |
3590 | } | |
3591 | ||
01127848 JK |
3592 | /* |
3593 | * We have to make sure i_disksize gets properly updated before we truncate | |
3594 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3595 | * can get lost as it may have been postponed to submission of writeback but | |
3596 | * that will never happen after we truncate page cache. | |
3597 | */ | |
3598 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3599 | loff_t len) | |
3600 | { | |
3601 | handle_t *handle; | |
3602 | loff_t size = i_size_read(inode); | |
3603 | ||
3604 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
3605 | if (offset > size || offset + len < size) | |
3606 | return 0; | |
3607 | ||
3608 | if (EXT4_I(inode)->i_disksize >= size) | |
3609 | return 0; | |
3610 | ||
3611 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3612 | if (IS_ERR(handle)) | |
3613 | return PTR_ERR(handle); | |
3614 | ext4_update_i_disksize(inode, size); | |
3615 | ext4_mark_inode_dirty(handle, inode); | |
3616 | ext4_journal_stop(handle); | |
3617 | ||
3618 | return 0; | |
3619 | } | |
3620 | ||
a4bb6b64 AH |
3621 | /* |
3622 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3623 | * associated with the given offset and length | |
3624 | * | |
3625 | * @inode: File inode | |
3626 | * @offset: The offset where the hole will begin | |
3627 | * @len: The length of the hole | |
3628 | * | |
4907cb7b | 3629 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3630 | */ |
3631 | ||
aeb2817a | 3632 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3633 | { |
26a4c0c6 TT |
3634 | struct super_block *sb = inode->i_sb; |
3635 | ext4_lblk_t first_block, stop_block; | |
3636 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3637 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3638 | handle_t *handle; |
3639 | unsigned int credits; | |
3640 | int ret = 0; | |
3641 | ||
a4bb6b64 | 3642 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3643 | return -EOPNOTSUPP; |
a4bb6b64 | 3644 | |
b8a86845 | 3645 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3646 | |
26a4c0c6 TT |
3647 | /* |
3648 | * Write out all dirty pages to avoid race conditions | |
3649 | * Then release them. | |
3650 | */ | |
3651 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3652 | ret = filemap_write_and_wait_range(mapping, offset, | |
3653 | offset + length - 1); | |
3654 | if (ret) | |
3655 | return ret; | |
3656 | } | |
3657 | ||
3658 | mutex_lock(&inode->i_mutex); | |
9ef06cec | 3659 | |
26a4c0c6 TT |
3660 | /* No need to punch hole beyond i_size */ |
3661 | if (offset >= inode->i_size) | |
3662 | goto out_mutex; | |
3663 | ||
3664 | /* | |
3665 | * If the hole extends beyond i_size, set the hole | |
3666 | * to end after the page that contains i_size | |
3667 | */ | |
3668 | if (offset + length > inode->i_size) { | |
3669 | length = inode->i_size + | |
3670 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3671 | offset; | |
3672 | } | |
3673 | ||
a361293f JK |
3674 | if (offset & (sb->s_blocksize - 1) || |
3675 | (offset + length) & (sb->s_blocksize - 1)) { | |
3676 | /* | |
3677 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3678 | * partial block | |
3679 | */ | |
3680 | ret = ext4_inode_attach_jinode(inode); | |
3681 | if (ret < 0) | |
3682 | goto out_mutex; | |
3683 | ||
3684 | } | |
3685 | ||
ea3d7209 JK |
3686 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
3687 | ext4_inode_block_unlocked_dio(inode); | |
3688 | inode_dio_wait(inode); | |
3689 | ||
3690 | /* | |
3691 | * Prevent page faults from reinstantiating pages we have released from | |
3692 | * page cache. | |
3693 | */ | |
3694 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
3695 | first_block_offset = round_up(offset, sb->s_blocksize); |
3696 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3697 | |
a87dd18c | 3698 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
3699 | if (last_block_offset > first_block_offset) { |
3700 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
3701 | if (ret) | |
3702 | goto out_dio; | |
a87dd18c LC |
3703 | truncate_pagecache_range(inode, first_block_offset, |
3704 | last_block_offset); | |
01127848 | 3705 | } |
26a4c0c6 | 3706 | |
26a4c0c6 TT |
3707 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3708 | credits = ext4_writepage_trans_blocks(inode); | |
3709 | else | |
3710 | credits = ext4_blocks_for_truncate(inode); | |
3711 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3712 | if (IS_ERR(handle)) { | |
3713 | ret = PTR_ERR(handle); | |
3714 | ext4_std_error(sb, ret); | |
3715 | goto out_dio; | |
3716 | } | |
3717 | ||
a87dd18c LC |
3718 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3719 | length); | |
3720 | if (ret) | |
3721 | goto out_stop; | |
26a4c0c6 TT |
3722 | |
3723 | first_block = (offset + sb->s_blocksize - 1) >> | |
3724 | EXT4_BLOCK_SIZE_BITS(sb); | |
3725 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3726 | ||
3727 | /* If there are no blocks to remove, return now */ | |
3728 | if (first_block >= stop_block) | |
3729 | goto out_stop; | |
3730 | ||
3731 | down_write(&EXT4_I(inode)->i_data_sem); | |
3732 | ext4_discard_preallocations(inode); | |
3733 | ||
3734 | ret = ext4_es_remove_extent(inode, first_block, | |
3735 | stop_block - first_block); | |
3736 | if (ret) { | |
3737 | up_write(&EXT4_I(inode)->i_data_sem); | |
3738 | goto out_stop; | |
3739 | } | |
3740 | ||
3741 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3742 | ret = ext4_ext_remove_space(inode, first_block, | |
3743 | stop_block - 1); | |
3744 | else | |
4f579ae7 | 3745 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
3746 | stop_block); |
3747 | ||
819c4920 | 3748 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3749 | if (IS_SYNC(inode)) |
3750 | ext4_handle_sync(handle); | |
e251f9bc | 3751 | |
26a4c0c6 TT |
3752 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3753 | ext4_mark_inode_dirty(handle, inode); | |
3754 | out_stop: | |
3755 | ext4_journal_stop(handle); | |
3756 | out_dio: | |
ea3d7209 | 3757 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 TT |
3758 | ext4_inode_resume_unlocked_dio(inode); |
3759 | out_mutex: | |
3760 | mutex_unlock(&inode->i_mutex); | |
3761 | return ret; | |
a4bb6b64 AH |
3762 | } |
3763 | ||
a361293f JK |
3764 | int ext4_inode_attach_jinode(struct inode *inode) |
3765 | { | |
3766 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3767 | struct jbd2_inode *jinode; | |
3768 | ||
3769 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
3770 | return 0; | |
3771 | ||
3772 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
3773 | spin_lock(&inode->i_lock); | |
3774 | if (!ei->jinode) { | |
3775 | if (!jinode) { | |
3776 | spin_unlock(&inode->i_lock); | |
3777 | return -ENOMEM; | |
3778 | } | |
3779 | ei->jinode = jinode; | |
3780 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
3781 | jinode = NULL; | |
3782 | } | |
3783 | spin_unlock(&inode->i_lock); | |
3784 | if (unlikely(jinode != NULL)) | |
3785 | jbd2_free_inode(jinode); | |
3786 | return 0; | |
3787 | } | |
3788 | ||
ac27a0ec | 3789 | /* |
617ba13b | 3790 | * ext4_truncate() |
ac27a0ec | 3791 | * |
617ba13b MC |
3792 | * We block out ext4_get_block() block instantiations across the entire |
3793 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3794 | * simultaneously on behalf of the same inode. |
3795 | * | |
42b2aa86 | 3796 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3797 | * is one core, guiding principle: the file's tree must always be consistent on |
3798 | * disk. We must be able to restart the truncate after a crash. | |
3799 | * | |
3800 | * The file's tree may be transiently inconsistent in memory (although it | |
3801 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3802 | * the contents of (the filesystem + the journal) must be consistent and | |
3803 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3804 | * left-to-right works OK too). | |
3805 | * | |
3806 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3807 | * truncate against the orphan inode list. | |
3808 | * | |
3809 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3810 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3811 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3812 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3813 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3814 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3815 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3816 | */ |
617ba13b | 3817 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3818 | { |
819c4920 TT |
3819 | struct ext4_inode_info *ei = EXT4_I(inode); |
3820 | unsigned int credits; | |
3821 | handle_t *handle; | |
3822 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 3823 | |
19b5ef61 TT |
3824 | /* |
3825 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 3826 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
3827 | * have i_mutex locked because it's not necessary. |
3828 | */ | |
3829 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3830 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3831 | trace_ext4_truncate_enter(inode); |
3832 | ||
91ef4caf | 3833 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3834 | return; |
3835 | ||
12e9b892 | 3836 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3837 | |
5534fb5b | 3838 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3839 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3840 | |
aef1c851 TM |
3841 | if (ext4_has_inline_data(inode)) { |
3842 | int has_inline = 1; | |
3843 | ||
3844 | ext4_inline_data_truncate(inode, &has_inline); | |
3845 | if (has_inline) | |
3846 | return; | |
3847 | } | |
3848 | ||
a361293f JK |
3849 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
3850 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
3851 | if (ext4_inode_attach_jinode(inode) < 0) | |
3852 | return; | |
3853 | } | |
3854 | ||
819c4920 TT |
3855 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3856 | credits = ext4_writepage_trans_blocks(inode); | |
3857 | else | |
3858 | credits = ext4_blocks_for_truncate(inode); | |
3859 | ||
3860 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3861 | if (IS_ERR(handle)) { | |
3862 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3863 | return; | |
3864 | } | |
3865 | ||
eb3544c6 LC |
3866 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
3867 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
3868 | |
3869 | /* | |
3870 | * We add the inode to the orphan list, so that if this | |
3871 | * truncate spans multiple transactions, and we crash, we will | |
3872 | * resume the truncate when the filesystem recovers. It also | |
3873 | * marks the inode dirty, to catch the new size. | |
3874 | * | |
3875 | * Implication: the file must always be in a sane, consistent | |
3876 | * truncatable state while each transaction commits. | |
3877 | */ | |
3878 | if (ext4_orphan_add(handle, inode)) | |
3879 | goto out_stop; | |
3880 | ||
3881 | down_write(&EXT4_I(inode)->i_data_sem); | |
3882 | ||
3883 | ext4_discard_preallocations(inode); | |
3884 | ||
ff9893dc | 3885 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3886 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3887 | else |
819c4920 TT |
3888 | ext4_ind_truncate(handle, inode); |
3889 | ||
3890 | up_write(&ei->i_data_sem); | |
3891 | ||
3892 | if (IS_SYNC(inode)) | |
3893 | ext4_handle_sync(handle); | |
3894 | ||
3895 | out_stop: | |
3896 | /* | |
3897 | * If this was a simple ftruncate() and the file will remain alive, | |
3898 | * then we need to clear up the orphan record which we created above. | |
3899 | * However, if this was a real unlink then we were called by | |
58d86a50 | 3900 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
3901 | * orphan info for us. |
3902 | */ | |
3903 | if (inode->i_nlink) | |
3904 | ext4_orphan_del(handle, inode); | |
3905 | ||
3906 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3907 | ext4_mark_inode_dirty(handle, inode); | |
3908 | ext4_journal_stop(handle); | |
ac27a0ec | 3909 | |
0562e0ba | 3910 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3911 | } |
3912 | ||
ac27a0ec | 3913 | /* |
617ba13b | 3914 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3915 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3916 | * data in memory that is needed to recreate the on-disk version of this | |
3917 | * inode. | |
3918 | */ | |
617ba13b MC |
3919 | static int __ext4_get_inode_loc(struct inode *inode, |
3920 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3921 | { |
240799cd TT |
3922 | struct ext4_group_desc *gdp; |
3923 | struct buffer_head *bh; | |
3924 | struct super_block *sb = inode->i_sb; | |
3925 | ext4_fsblk_t block; | |
3926 | int inodes_per_block, inode_offset; | |
3927 | ||
3a06d778 | 3928 | iloc->bh = NULL; |
240799cd | 3929 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 3930 | return -EFSCORRUPTED; |
ac27a0ec | 3931 | |
240799cd TT |
3932 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3933 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3934 | if (!gdp) | |
ac27a0ec DK |
3935 | return -EIO; |
3936 | ||
240799cd TT |
3937 | /* |
3938 | * Figure out the offset within the block group inode table | |
3939 | */ | |
00d09882 | 3940 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3941 | inode_offset = ((inode->i_ino - 1) % |
3942 | EXT4_INODES_PER_GROUP(sb)); | |
3943 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3944 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3945 | ||
3946 | bh = sb_getblk(sb, block); | |
aebf0243 | 3947 | if (unlikely(!bh)) |
860d21e2 | 3948 | return -ENOMEM; |
ac27a0ec DK |
3949 | if (!buffer_uptodate(bh)) { |
3950 | lock_buffer(bh); | |
9c83a923 HK |
3951 | |
3952 | /* | |
3953 | * If the buffer has the write error flag, we have failed | |
3954 | * to write out another inode in the same block. In this | |
3955 | * case, we don't have to read the block because we may | |
3956 | * read the old inode data successfully. | |
3957 | */ | |
3958 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3959 | set_buffer_uptodate(bh); | |
3960 | ||
ac27a0ec DK |
3961 | if (buffer_uptodate(bh)) { |
3962 | /* someone brought it uptodate while we waited */ | |
3963 | unlock_buffer(bh); | |
3964 | goto has_buffer; | |
3965 | } | |
3966 | ||
3967 | /* | |
3968 | * If we have all information of the inode in memory and this | |
3969 | * is the only valid inode in the block, we need not read the | |
3970 | * block. | |
3971 | */ | |
3972 | if (in_mem) { | |
3973 | struct buffer_head *bitmap_bh; | |
240799cd | 3974 | int i, start; |
ac27a0ec | 3975 | |
240799cd | 3976 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3977 | |
240799cd TT |
3978 | /* Is the inode bitmap in cache? */ |
3979 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3980 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3981 | goto make_io; |
3982 | ||
3983 | /* | |
3984 | * If the inode bitmap isn't in cache then the | |
3985 | * optimisation may end up performing two reads instead | |
3986 | * of one, so skip it. | |
3987 | */ | |
3988 | if (!buffer_uptodate(bitmap_bh)) { | |
3989 | brelse(bitmap_bh); | |
3990 | goto make_io; | |
3991 | } | |
240799cd | 3992 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3993 | if (i == inode_offset) |
3994 | continue; | |
617ba13b | 3995 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3996 | break; |
3997 | } | |
3998 | brelse(bitmap_bh); | |
240799cd | 3999 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4000 | /* all other inodes are free, so skip I/O */ |
4001 | memset(bh->b_data, 0, bh->b_size); | |
4002 | set_buffer_uptodate(bh); | |
4003 | unlock_buffer(bh); | |
4004 | goto has_buffer; | |
4005 | } | |
4006 | } | |
4007 | ||
4008 | make_io: | |
240799cd TT |
4009 | /* |
4010 | * If we need to do any I/O, try to pre-readahead extra | |
4011 | * blocks from the inode table. | |
4012 | */ | |
4013 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4014 | ext4_fsblk_t b, end, table; | |
4015 | unsigned num; | |
0d606e2c | 4016 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4017 | |
4018 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4019 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4020 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4021 | if (table > b) |
4022 | b = table; | |
0d606e2c | 4023 | end = b + ra_blks; |
240799cd | 4024 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4025 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4026 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4027 | table += num / inodes_per_block; |
4028 | if (end > table) | |
4029 | end = table; | |
4030 | while (b <= end) | |
4031 | sb_breadahead(sb, b++); | |
4032 | } | |
4033 | ||
ac27a0ec DK |
4034 | /* |
4035 | * There are other valid inodes in the buffer, this inode | |
4036 | * has in-inode xattrs, or we don't have this inode in memory. | |
4037 | * Read the block from disk. | |
4038 | */ | |
0562e0ba | 4039 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4040 | get_bh(bh); |
4041 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 4042 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4043 | wait_on_buffer(bh); |
4044 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4045 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4046 | "unable to read itable block"); | |
ac27a0ec DK |
4047 | brelse(bh); |
4048 | return -EIO; | |
4049 | } | |
4050 | } | |
4051 | has_buffer: | |
4052 | iloc->bh = bh; | |
4053 | return 0; | |
4054 | } | |
4055 | ||
617ba13b | 4056 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4057 | { |
4058 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4059 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4060 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4061 | } |
4062 | ||
617ba13b | 4063 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4064 | { |
617ba13b | 4065 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4066 | unsigned int new_fl = 0; |
ac27a0ec | 4067 | |
617ba13b | 4068 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4069 | new_fl |= S_SYNC; |
617ba13b | 4070 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4071 | new_fl |= S_APPEND; |
617ba13b | 4072 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4073 | new_fl |= S_IMMUTABLE; |
617ba13b | 4074 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4075 | new_fl |= S_NOATIME; |
617ba13b | 4076 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4077 | new_fl |= S_DIRSYNC; |
923ae0ff RZ |
4078 | if (test_opt(inode->i_sb, DAX)) |
4079 | new_fl |= S_DAX; | |
5f16f322 | 4080 | inode_set_flags(inode, new_fl, |
923ae0ff | 4081 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
4082 | } |
4083 | ||
ff9ddf7e JK |
4084 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4085 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4086 | { | |
84a8dce2 DM |
4087 | unsigned int vfs_fl; |
4088 | unsigned long old_fl, new_fl; | |
4089 | ||
4090 | do { | |
4091 | vfs_fl = ei->vfs_inode.i_flags; | |
4092 | old_fl = ei->i_flags; | |
4093 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4094 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4095 | EXT4_DIRSYNC_FL); | |
4096 | if (vfs_fl & S_SYNC) | |
4097 | new_fl |= EXT4_SYNC_FL; | |
4098 | if (vfs_fl & S_APPEND) | |
4099 | new_fl |= EXT4_APPEND_FL; | |
4100 | if (vfs_fl & S_IMMUTABLE) | |
4101 | new_fl |= EXT4_IMMUTABLE_FL; | |
4102 | if (vfs_fl & S_NOATIME) | |
4103 | new_fl |= EXT4_NOATIME_FL; | |
4104 | if (vfs_fl & S_DIRSYNC) | |
4105 | new_fl |= EXT4_DIRSYNC_FL; | |
4106 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4107 | } |
de9a55b8 | 4108 | |
0fc1b451 | 4109 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4110 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4111 | { |
4112 | blkcnt_t i_blocks ; | |
8180a562 AK |
4113 | struct inode *inode = &(ei->vfs_inode); |
4114 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4115 | |
e2b911c5 | 4116 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4117 | /* we are using combined 48 bit field */ |
4118 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4119 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4120 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4121 | /* i_blocks represent file system block size */ |
4122 | return i_blocks << (inode->i_blkbits - 9); | |
4123 | } else { | |
4124 | return i_blocks; | |
4125 | } | |
0fc1b451 AK |
4126 | } else { |
4127 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4128 | } | |
4129 | } | |
ff9ddf7e | 4130 | |
152a7b0a TM |
4131 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4132 | struct ext4_inode *raw_inode, | |
4133 | struct ext4_inode_info *ei) | |
4134 | { | |
4135 | __le32 *magic = (void *)raw_inode + | |
4136 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4137 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4138 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4139 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4140 | } else |
4141 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4142 | } |
4143 | ||
1d1fe1ee | 4144 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4145 | { |
617ba13b MC |
4146 | struct ext4_iloc iloc; |
4147 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4148 | struct ext4_inode_info *ei; |
1d1fe1ee | 4149 | struct inode *inode; |
b436b9be | 4150 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4151 | long ret; |
ac27a0ec | 4152 | int block; |
08cefc7a EB |
4153 | uid_t i_uid; |
4154 | gid_t i_gid; | |
ac27a0ec | 4155 | |
1d1fe1ee DH |
4156 | inode = iget_locked(sb, ino); |
4157 | if (!inode) | |
4158 | return ERR_PTR(-ENOMEM); | |
4159 | if (!(inode->i_state & I_NEW)) | |
4160 | return inode; | |
4161 | ||
4162 | ei = EXT4_I(inode); | |
7dc57615 | 4163 | iloc.bh = NULL; |
ac27a0ec | 4164 | |
1d1fe1ee DH |
4165 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4166 | if (ret < 0) | |
ac27a0ec | 4167 | goto bad_inode; |
617ba13b | 4168 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4169 | |
4170 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4171 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4172 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4173 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4174 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4175 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4176 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4177 | ret = -EFSCORRUPTED; |
814525f4 DW |
4178 | goto bad_inode; |
4179 | } | |
4180 | } else | |
4181 | ei->i_extra_isize = 0; | |
4182 | ||
4183 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4184 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4185 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4186 | __u32 csum; | |
4187 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4188 | __le32 gen = raw_inode->i_generation; | |
4189 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4190 | sizeof(inum)); | |
4191 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4192 | sizeof(gen)); | |
4193 | } | |
4194 | ||
4195 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4196 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4197 | ret = -EFSBADCRC; |
814525f4 DW |
4198 | goto bad_inode; |
4199 | } | |
4200 | ||
ac27a0ec | 4201 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4202 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4203 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4204 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4205 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4206 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4207 | } |
08cefc7a EB |
4208 | i_uid_write(inode, i_uid); |
4209 | i_gid_write(inode, i_gid); | |
bfe86848 | 4210 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4211 | |
353eb83c | 4212 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4213 | ei->i_inline_off = 0; |
ac27a0ec DK |
4214 | ei->i_dir_start_lookup = 0; |
4215 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4216 | /* We now have enough fields to check if the inode was active or not. | |
4217 | * This is needed because nfsd might try to access dead inodes | |
4218 | * the test is that same one that e2fsck uses | |
4219 | * NeilBrown 1999oct15 | |
4220 | */ | |
4221 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4222 | if ((inode->i_mode == 0 || |
4223 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4224 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4225 | /* this inode is deleted */ |
1d1fe1ee | 4226 | ret = -ESTALE; |
ac27a0ec DK |
4227 | goto bad_inode; |
4228 | } | |
4229 | /* The only unlinked inodes we let through here have | |
4230 | * valid i_mode and are being read by the orphan | |
4231 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4232 | * the process of deleting those. |
4233 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4234 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4235 | } |
ac27a0ec | 4236 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4237 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4238 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4239 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4240 | ei->i_file_acl |= |
4241 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4242 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4243 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4244 | #ifdef CONFIG_QUOTA |
4245 | ei->i_reserved_quota = 0; | |
4246 | #endif | |
ac27a0ec DK |
4247 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4248 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4249 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4250 | /* |
4251 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4252 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4253 | */ | |
617ba13b | 4254 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4255 | ei->i_data[block] = raw_inode->i_block[block]; |
4256 | INIT_LIST_HEAD(&ei->i_orphan); | |
4257 | ||
b436b9be JK |
4258 | /* |
4259 | * Set transaction id's of transactions that have to be committed | |
4260 | * to finish f[data]sync. We set them to currently running transaction | |
4261 | * as we cannot be sure that the inode or some of its metadata isn't | |
4262 | * part of the transaction - the inode could have been reclaimed and | |
4263 | * now it is reread from disk. | |
4264 | */ | |
4265 | if (journal) { | |
4266 | transaction_t *transaction; | |
4267 | tid_t tid; | |
4268 | ||
a931da6a | 4269 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4270 | if (journal->j_running_transaction) |
4271 | transaction = journal->j_running_transaction; | |
4272 | else | |
4273 | transaction = journal->j_committing_transaction; | |
4274 | if (transaction) | |
4275 | tid = transaction->t_tid; | |
4276 | else | |
4277 | tid = journal->j_commit_sequence; | |
a931da6a | 4278 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4279 | ei->i_sync_tid = tid; |
4280 | ei->i_datasync_tid = tid; | |
4281 | } | |
4282 | ||
0040d987 | 4283 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4284 | if (ei->i_extra_isize == 0) { |
4285 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4286 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4287 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4288 | } else { |
152a7b0a | 4289 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4290 | } |
814525f4 | 4291 | } |
ac27a0ec | 4292 | |
ef7f3835 KS |
4293 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4294 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4295 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4296 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4297 | ||
ed3654eb | 4298 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4299 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4300 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4301 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4302 | inode->i_version |= | |
4303 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4304 | } | |
25ec56b5 JNC |
4305 | } |
4306 | ||
c4b5a614 | 4307 | ret = 0; |
485c26ec | 4308 | if (ei->i_file_acl && |
1032988c | 4309 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4310 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4311 | ei->i_file_acl); | |
6a797d27 | 4312 | ret = -EFSCORRUPTED; |
485c26ec | 4313 | goto bad_inode; |
f19d5870 TM |
4314 | } else if (!ext4_has_inline_data(inode)) { |
4315 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4316 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4317 | (S_ISLNK(inode->i_mode) && | |
4318 | !ext4_inode_is_fast_symlink(inode)))) | |
4319 | /* Validate extent which is part of inode */ | |
4320 | ret = ext4_ext_check_inode(inode); | |
4321 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4322 | (S_ISLNK(inode->i_mode) && | |
4323 | !ext4_inode_is_fast_symlink(inode))) { | |
4324 | /* Validate block references which are part of inode */ | |
4325 | ret = ext4_ind_check_inode(inode); | |
4326 | } | |
fe2c8191 | 4327 | } |
567f3e9a | 4328 | if (ret) |
de9a55b8 | 4329 | goto bad_inode; |
7a262f7c | 4330 | |
ac27a0ec | 4331 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4332 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4333 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4334 | ext4_set_aops(inode); |
ac27a0ec | 4335 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4336 | inode->i_op = &ext4_dir_inode_operations; |
4337 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4338 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4339 | if (ext4_encrypted_inode(inode)) { |
4340 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4341 | ext4_set_aops(inode); | |
4342 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4343 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4344 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4345 | nd_terminate_link(ei->i_data, inode->i_size, |
4346 | sizeof(ei->i_data) - 1); | |
4347 | } else { | |
617ba13b MC |
4348 | inode->i_op = &ext4_symlink_inode_operations; |
4349 | ext4_set_aops(inode); | |
ac27a0ec | 4350 | } |
563bdd61 TT |
4351 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4352 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4353 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4354 | if (raw_inode->i_block[0]) |
4355 | init_special_inode(inode, inode->i_mode, | |
4356 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4357 | else | |
4358 | init_special_inode(inode, inode->i_mode, | |
4359 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4360 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4361 | make_bad_inode(inode); | |
563bdd61 | 4362 | } else { |
6a797d27 | 4363 | ret = -EFSCORRUPTED; |
24676da4 | 4364 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4365 | goto bad_inode; |
ac27a0ec | 4366 | } |
af5bc92d | 4367 | brelse(iloc.bh); |
617ba13b | 4368 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4369 | unlock_new_inode(inode); |
4370 | return inode; | |
ac27a0ec DK |
4371 | |
4372 | bad_inode: | |
567f3e9a | 4373 | brelse(iloc.bh); |
1d1fe1ee DH |
4374 | iget_failed(inode); |
4375 | return ERR_PTR(ret); | |
ac27a0ec DK |
4376 | } |
4377 | ||
f4bb2981 TT |
4378 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4379 | { | |
4380 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4381 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4382 | return ext4_iget(sb, ino); |
4383 | } | |
4384 | ||
0fc1b451 AK |
4385 | static int ext4_inode_blocks_set(handle_t *handle, |
4386 | struct ext4_inode *raw_inode, | |
4387 | struct ext4_inode_info *ei) | |
4388 | { | |
4389 | struct inode *inode = &(ei->vfs_inode); | |
4390 | u64 i_blocks = inode->i_blocks; | |
4391 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4392 | |
4393 | if (i_blocks <= ~0U) { | |
4394 | /* | |
4907cb7b | 4395 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4396 | * as multiple of 512 bytes |
4397 | */ | |
8180a562 | 4398 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4399 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4400 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4401 | return 0; |
4402 | } | |
e2b911c5 | 4403 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4404 | return -EFBIG; |
4405 | ||
4406 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4407 | /* |
4408 | * i_blocks can be represented in a 48 bit variable | |
4409 | * as multiple of 512 bytes | |
4410 | */ | |
8180a562 | 4411 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4412 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4413 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4414 | } else { |
84a8dce2 | 4415 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4416 | /* i_block is stored in file system block size */ |
4417 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4418 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4419 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4420 | } |
f287a1a5 | 4421 | return 0; |
0fc1b451 AK |
4422 | } |
4423 | ||
a26f4992 TT |
4424 | struct other_inode { |
4425 | unsigned long orig_ino; | |
4426 | struct ext4_inode *raw_inode; | |
4427 | }; | |
4428 | ||
4429 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4430 | void *data) | |
4431 | { | |
4432 | struct other_inode *oi = (struct other_inode *) data; | |
4433 | ||
4434 | if ((inode->i_ino != ino) || | |
4435 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4436 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4437 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4438 | return 0; | |
4439 | spin_lock(&inode->i_lock); | |
4440 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4441 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4442 | (inode->i_state & I_DIRTY_TIME)) { | |
4443 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4444 | ||
4445 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4446 | spin_unlock(&inode->i_lock); | |
4447 | ||
4448 | spin_lock(&ei->i_raw_lock); | |
4449 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4450 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4451 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4452 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4453 | spin_unlock(&ei->i_raw_lock); | |
4454 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4455 | return -1; | |
4456 | } | |
4457 | spin_unlock(&inode->i_lock); | |
4458 | return -1; | |
4459 | } | |
4460 | ||
4461 | /* | |
4462 | * Opportunistically update the other time fields for other inodes in | |
4463 | * the same inode table block. | |
4464 | */ | |
4465 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4466 | unsigned long orig_ino, char *buf) | |
4467 | { | |
4468 | struct other_inode oi; | |
4469 | unsigned long ino; | |
4470 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4471 | int inode_size = EXT4_INODE_SIZE(sb); | |
4472 | ||
4473 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
4474 | /* |
4475 | * Calculate the first inode in the inode table block. Inode | |
4476 | * numbers are one-based. That is, the first inode in a block | |
4477 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4478 | */ | |
4479 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
4480 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4481 | if (ino == orig_ino) | |
4482 | continue; | |
4483 | oi.raw_inode = (struct ext4_inode *) buf; | |
4484 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4485 | } | |
4486 | } | |
4487 | ||
ac27a0ec DK |
4488 | /* |
4489 | * Post the struct inode info into an on-disk inode location in the | |
4490 | * buffer-cache. This gobbles the caller's reference to the | |
4491 | * buffer_head in the inode location struct. | |
4492 | * | |
4493 | * The caller must have write access to iloc->bh. | |
4494 | */ | |
617ba13b | 4495 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4496 | struct inode *inode, |
830156c7 | 4497 | struct ext4_iloc *iloc) |
ac27a0ec | 4498 | { |
617ba13b MC |
4499 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4500 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4501 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4502 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4503 | int err = 0, rc, block; |
202ee5df | 4504 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4505 | uid_t i_uid; |
4506 | gid_t i_gid; | |
ac27a0ec | 4507 | |
202ee5df TT |
4508 | spin_lock(&ei->i_raw_lock); |
4509 | ||
4510 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4511 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4512 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4513 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4514 | |
ff9ddf7e | 4515 | ext4_get_inode_flags(ei); |
ac27a0ec | 4516 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4517 | i_uid = i_uid_read(inode); |
4518 | i_gid = i_gid_read(inode); | |
af5bc92d | 4519 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4520 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4521 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4522 | /* |
4523 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4524 | * re-used with the upper 16 bits of the uid/gid intact | |
4525 | */ | |
af5bc92d | 4526 | if (!ei->i_dtime) { |
ac27a0ec | 4527 | raw_inode->i_uid_high = |
08cefc7a | 4528 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4529 | raw_inode->i_gid_high = |
08cefc7a | 4530 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4531 | } else { |
4532 | raw_inode->i_uid_high = 0; | |
4533 | raw_inode->i_gid_high = 0; | |
4534 | } | |
4535 | } else { | |
08cefc7a EB |
4536 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4537 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4538 | raw_inode->i_uid_high = 0; |
4539 | raw_inode->i_gid_high = 0; | |
4540 | } | |
4541 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4542 | |
4543 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4544 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4545 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4546 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4547 | ||
bce92d56 LX |
4548 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4549 | if (err) { | |
202ee5df | 4550 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4551 | goto out_brelse; |
202ee5df | 4552 | } |
ac27a0ec | 4553 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4554 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4555 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4556 | raw_inode->i_file_acl_high = |
4557 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4558 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4559 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4560 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4561 | need_datasync = 1; | |
4562 | } | |
a48380f7 | 4563 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 4564 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 4565 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
4566 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4567 | set_large_file = 1; | |
ac27a0ec DK |
4568 | } |
4569 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4570 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4571 | if (old_valid_dev(inode->i_rdev)) { | |
4572 | raw_inode->i_block[0] = | |
4573 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4574 | raw_inode->i_block[1] = 0; | |
4575 | } else { | |
4576 | raw_inode->i_block[0] = 0; | |
4577 | raw_inode->i_block[1] = | |
4578 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4579 | raw_inode->i_block[2] = 0; | |
4580 | } | |
f19d5870 | 4581 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4582 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4583 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4584 | } |
ac27a0ec | 4585 | |
ed3654eb | 4586 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4587 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4588 | if (ei->i_extra_isize) { | |
4589 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4590 | raw_inode->i_version_hi = | |
4591 | cpu_to_le32(inode->i_version >> 32); | |
4592 | raw_inode->i_extra_isize = | |
4593 | cpu_to_le16(ei->i_extra_isize); | |
4594 | } | |
25ec56b5 | 4595 | } |
814525f4 | 4596 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 4597 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
4598 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
4599 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
4600 | bh->b_data); | |
202ee5df | 4601 | |
830156c7 | 4602 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4603 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4604 | if (!err) |
4605 | err = rc; | |
19f5fb7a | 4606 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 4607 | if (set_large_file) { |
5d601255 | 4608 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4609 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4610 | if (err) | |
4611 | goto out_brelse; | |
4612 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 4613 | ext4_set_feature_large_file(sb); |
202ee5df TT |
4614 | ext4_handle_sync(handle); |
4615 | err = ext4_handle_dirty_super(handle, sb); | |
4616 | } | |
b71fc079 | 4617 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4618 | out_brelse: |
af5bc92d | 4619 | brelse(bh); |
617ba13b | 4620 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4621 | return err; |
4622 | } | |
4623 | ||
4624 | /* | |
617ba13b | 4625 | * ext4_write_inode() |
ac27a0ec DK |
4626 | * |
4627 | * We are called from a few places: | |
4628 | * | |
87f7e416 | 4629 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4630 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4631 | * transaction to commit. |
ac27a0ec | 4632 | * |
87f7e416 TT |
4633 | * - Within flush work (sys_sync(), kupdate and such). |
4634 | * We wait on commit, if told to. | |
ac27a0ec | 4635 | * |
87f7e416 TT |
4636 | * - Within iput_final() -> write_inode_now() |
4637 | * We wait on commit, if told to. | |
ac27a0ec DK |
4638 | * |
4639 | * In all cases it is actually safe for us to return without doing anything, | |
4640 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4641 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4642 | * writeback. | |
ac27a0ec DK |
4643 | * |
4644 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4645 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4646 | * which we are interested. | |
4647 | * | |
4648 | * It would be a bug for them to not do this. The code: | |
4649 | * | |
4650 | * mark_inode_dirty(inode) | |
4651 | * stuff(); | |
4652 | * inode->i_size = expr; | |
4653 | * | |
87f7e416 TT |
4654 | * is in error because write_inode() could occur while `stuff()' is running, |
4655 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4656 | * superblock's dirty inode list. | |
ac27a0ec | 4657 | */ |
a9185b41 | 4658 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4659 | { |
91ac6f43 FM |
4660 | int err; |
4661 | ||
87f7e416 | 4662 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4663 | return 0; |
4664 | ||
91ac6f43 FM |
4665 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4666 | if (ext4_journal_current_handle()) { | |
4667 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4668 | dump_stack(); | |
4669 | return -EIO; | |
4670 | } | |
ac27a0ec | 4671 | |
10542c22 JK |
4672 | /* |
4673 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4674 | * ext4_sync_fs() will force the commit after everything is | |
4675 | * written. | |
4676 | */ | |
4677 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4678 | return 0; |
4679 | ||
4680 | err = ext4_force_commit(inode->i_sb); | |
4681 | } else { | |
4682 | struct ext4_iloc iloc; | |
ac27a0ec | 4683 | |
8b472d73 | 4684 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4685 | if (err) |
4686 | return err; | |
10542c22 JK |
4687 | /* |
4688 | * sync(2) will flush the whole buffer cache. No need to do | |
4689 | * it here separately for each inode. | |
4690 | */ | |
4691 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4692 | sync_dirty_buffer(iloc.bh); |
4693 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4694 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4695 | "IO error syncing inode"); | |
830156c7 FM |
4696 | err = -EIO; |
4697 | } | |
fd2dd9fb | 4698 | brelse(iloc.bh); |
91ac6f43 FM |
4699 | } |
4700 | return err; | |
ac27a0ec DK |
4701 | } |
4702 | ||
53e87268 JK |
4703 | /* |
4704 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4705 | * buffers that are attached to a page stradding i_size and are undergoing | |
4706 | * commit. In that case we have to wait for commit to finish and try again. | |
4707 | */ | |
4708 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4709 | { | |
4710 | struct page *page; | |
4711 | unsigned offset; | |
4712 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4713 | tid_t commit_tid = 0; | |
4714 | int ret; | |
4715 | ||
4716 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4717 | /* | |
4718 | * All buffers in the last page remain valid? Then there's nothing to | |
4719 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4720 | * blocksize case | |
4721 | */ | |
4722 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4723 | return; | |
4724 | while (1) { | |
4725 | page = find_lock_page(inode->i_mapping, | |
4726 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4727 | if (!page) | |
4728 | return; | |
ca99fdd2 LC |
4729 | ret = __ext4_journalled_invalidatepage(page, offset, |
4730 | PAGE_CACHE_SIZE - offset); | |
53e87268 JK |
4731 | unlock_page(page); |
4732 | page_cache_release(page); | |
4733 | if (ret != -EBUSY) | |
4734 | return; | |
4735 | commit_tid = 0; | |
4736 | read_lock(&journal->j_state_lock); | |
4737 | if (journal->j_committing_transaction) | |
4738 | commit_tid = journal->j_committing_transaction->t_tid; | |
4739 | read_unlock(&journal->j_state_lock); | |
4740 | if (commit_tid) | |
4741 | jbd2_log_wait_commit(journal, commit_tid); | |
4742 | } | |
4743 | } | |
4744 | ||
ac27a0ec | 4745 | /* |
617ba13b | 4746 | * ext4_setattr() |
ac27a0ec DK |
4747 | * |
4748 | * Called from notify_change. | |
4749 | * | |
4750 | * We want to trap VFS attempts to truncate the file as soon as | |
4751 | * possible. In particular, we want to make sure that when the VFS | |
4752 | * shrinks i_size, we put the inode on the orphan list and modify | |
4753 | * i_disksize immediately, so that during the subsequent flushing of | |
4754 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4755 | * commit will leave the blocks being flushed in an unused state on | |
4756 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4757 | * be freed, so we have a strong guarantee that no future commit will | |
4758 | * leave these blocks visible to the user.) | |
4759 | * | |
678aaf48 JK |
4760 | * Another thing we have to assure is that if we are in ordered mode |
4761 | * and inode is still attached to the committing transaction, we must | |
4762 | * we start writeout of all the dirty pages which are being truncated. | |
4763 | * This way we are sure that all the data written in the previous | |
4764 | * transaction are already on disk (truncate waits for pages under | |
4765 | * writeback). | |
4766 | * | |
4767 | * Called with inode->i_mutex down. | |
ac27a0ec | 4768 | */ |
617ba13b | 4769 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 4770 | { |
2b0143b5 | 4771 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 4772 | int error, rc = 0; |
3d287de3 | 4773 | int orphan = 0; |
ac27a0ec DK |
4774 | const unsigned int ia_valid = attr->ia_valid; |
4775 | ||
4776 | error = inode_change_ok(inode, attr); | |
4777 | if (error) | |
4778 | return error; | |
4779 | ||
a7cdadee JK |
4780 | if (is_quota_modification(inode, attr)) { |
4781 | error = dquot_initialize(inode); | |
4782 | if (error) | |
4783 | return error; | |
4784 | } | |
08cefc7a EB |
4785 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4786 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4787 | handle_t *handle; |
4788 | ||
4789 | /* (user+group)*(old+new) structure, inode write (sb, | |
4790 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4791 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4792 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4793 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4794 | if (IS_ERR(handle)) { |
4795 | error = PTR_ERR(handle); | |
4796 | goto err_out; | |
4797 | } | |
b43fa828 | 4798 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4799 | if (error) { |
617ba13b | 4800 | ext4_journal_stop(handle); |
ac27a0ec DK |
4801 | return error; |
4802 | } | |
4803 | /* Update corresponding info in inode so that everything is in | |
4804 | * one transaction */ | |
4805 | if (attr->ia_valid & ATTR_UID) | |
4806 | inode->i_uid = attr->ia_uid; | |
4807 | if (attr->ia_valid & ATTR_GID) | |
4808 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4809 | error = ext4_mark_inode_dirty(handle, inode); |
4810 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4811 | } |
4812 | ||
3da40c7b | 4813 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 4814 | handle_t *handle; |
3da40c7b JB |
4815 | loff_t oldsize = inode->i_size; |
4816 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 4817 | |
12e9b892 | 4818 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4819 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4820 | ||
0c095c7f TT |
4821 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4822 | return -EFBIG; | |
e2b46574 | 4823 | } |
3da40c7b JB |
4824 | if (!S_ISREG(inode->i_mode)) |
4825 | return -EINVAL; | |
dff6efc3 CH |
4826 | |
4827 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
4828 | inode_inc_iversion(inode); | |
4829 | ||
3da40c7b | 4830 | if (ext4_should_order_data(inode) && |
5208386c | 4831 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 4832 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 4833 | attr->ia_size); |
3da40c7b JB |
4834 | if (error) |
4835 | goto err_out; | |
4836 | } | |
4837 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
4838 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
4839 | if (IS_ERR(handle)) { | |
4840 | error = PTR_ERR(handle); | |
4841 | goto err_out; | |
4842 | } | |
3da40c7b | 4843 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
4844 | error = ext4_orphan_add(handle, inode); |
4845 | orphan = 1; | |
4846 | } | |
911af577 EG |
4847 | /* |
4848 | * Update c/mtime on truncate up, ext4_truncate() will | |
4849 | * update c/mtime in shrink case below | |
4850 | */ | |
4851 | if (!shrink) { | |
4852 | inode->i_mtime = ext4_current_time(inode); | |
4853 | inode->i_ctime = inode->i_mtime; | |
4854 | } | |
90e775b7 | 4855 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
4856 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4857 | rc = ext4_mark_inode_dirty(handle, inode); | |
4858 | if (!error) | |
4859 | error = rc; | |
90e775b7 JK |
4860 | /* |
4861 | * We have to update i_size under i_data_sem together | |
4862 | * with i_disksize to avoid races with writeback code | |
4863 | * running ext4_wb_update_i_disksize(). | |
4864 | */ | |
4865 | if (!error) | |
4866 | i_size_write(inode, attr->ia_size); | |
4867 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
4868 | ext4_journal_stop(handle); |
4869 | if (error) { | |
3da40c7b JB |
4870 | if (orphan) |
4871 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
4872 | goto err_out; |
4873 | } | |
d6320cbf | 4874 | } |
3da40c7b JB |
4875 | if (!shrink) |
4876 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 4877 | |
5208386c JK |
4878 | /* |
4879 | * Blocks are going to be removed from the inode. Wait | |
4880 | * for dio in flight. Temporarily disable | |
4881 | * dioread_nolock to prevent livelock. | |
4882 | */ | |
4883 | if (orphan) { | |
4884 | if (!ext4_should_journal_data(inode)) { | |
4885 | ext4_inode_block_unlocked_dio(inode); | |
4886 | inode_dio_wait(inode); | |
4887 | ext4_inode_resume_unlocked_dio(inode); | |
4888 | } else | |
4889 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 4890 | } |
ea3d7209 | 4891 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
4892 | /* |
4893 | * Truncate pagecache after we've waited for commit | |
4894 | * in data=journal mode to make pages freeable. | |
4895 | */ | |
923ae0ff | 4896 | truncate_pagecache(inode, inode->i_size); |
3da40c7b JB |
4897 | if (shrink) |
4898 | ext4_truncate(inode); | |
ea3d7209 | 4899 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 4900 | } |
ac27a0ec | 4901 | |
1025774c CH |
4902 | if (!rc) { |
4903 | setattr_copy(inode, attr); | |
4904 | mark_inode_dirty(inode); | |
4905 | } | |
4906 | ||
4907 | /* | |
4908 | * If the call to ext4_truncate failed to get a transaction handle at | |
4909 | * all, we need to clean up the in-core orphan list manually. | |
4910 | */ | |
3d287de3 | 4911 | if (orphan && inode->i_nlink) |
617ba13b | 4912 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4913 | |
4914 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 4915 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
4916 | |
4917 | err_out: | |
617ba13b | 4918 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4919 | if (!error) |
4920 | error = rc; | |
4921 | return error; | |
4922 | } | |
4923 | ||
3e3398a0 MC |
4924 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4925 | struct kstat *stat) | |
4926 | { | |
4927 | struct inode *inode; | |
8af8eecc | 4928 | unsigned long long delalloc_blocks; |
3e3398a0 | 4929 | |
2b0143b5 | 4930 | inode = d_inode(dentry); |
3e3398a0 MC |
4931 | generic_fillattr(inode, stat); |
4932 | ||
9206c561 AD |
4933 | /* |
4934 | * If there is inline data in the inode, the inode will normally not | |
4935 | * have data blocks allocated (it may have an external xattr block). | |
4936 | * Report at least one sector for such files, so tools like tar, rsync, | |
4937 | * others doen't incorrectly think the file is completely sparse. | |
4938 | */ | |
4939 | if (unlikely(ext4_has_inline_data(inode))) | |
4940 | stat->blocks += (stat->size + 511) >> 9; | |
4941 | ||
3e3398a0 MC |
4942 | /* |
4943 | * We can't update i_blocks if the block allocation is delayed | |
4944 | * otherwise in the case of system crash before the real block | |
4945 | * allocation is done, we will have i_blocks inconsistent with | |
4946 | * on-disk file blocks. | |
4947 | * We always keep i_blocks updated together with real | |
4948 | * allocation. But to not confuse with user, stat | |
4949 | * will return the blocks that include the delayed allocation | |
4950 | * blocks for this file. | |
4951 | */ | |
96607551 | 4952 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
4953 | EXT4_I(inode)->i_reserved_data_blocks); |
4954 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
4955 | return 0; |
4956 | } | |
ac27a0ec | 4957 | |
fffb2739 JK |
4958 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
4959 | int pextents) | |
a02908f1 | 4960 | { |
12e9b892 | 4961 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
4962 | return ext4_ind_trans_blocks(inode, lblocks); |
4963 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 4964 | } |
ac51d837 | 4965 | |
ac27a0ec | 4966 | /* |
a02908f1 MC |
4967 | * Account for index blocks, block groups bitmaps and block group |
4968 | * descriptor blocks if modify datablocks and index blocks | |
4969 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4970 | * |
a02908f1 | 4971 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4972 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4973 | * they could still across block group boundary. |
ac27a0ec | 4974 | * |
a02908f1 MC |
4975 | * Also account for superblock, inode, quota and xattr blocks |
4976 | */ | |
fffb2739 JK |
4977 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
4978 | int pextents) | |
a02908f1 | 4979 | { |
8df9675f TT |
4980 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4981 | int gdpblocks; | |
a02908f1 MC |
4982 | int idxblocks; |
4983 | int ret = 0; | |
4984 | ||
4985 | /* | |
fffb2739 JK |
4986 | * How many index blocks need to touch to map @lblocks logical blocks |
4987 | * to @pextents physical extents? | |
a02908f1 | 4988 | */ |
fffb2739 | 4989 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
4990 | |
4991 | ret = idxblocks; | |
4992 | ||
4993 | /* | |
4994 | * Now let's see how many group bitmaps and group descriptors need | |
4995 | * to account | |
4996 | */ | |
fffb2739 | 4997 | groups = idxblocks + pextents; |
a02908f1 | 4998 | gdpblocks = groups; |
8df9675f TT |
4999 | if (groups > ngroups) |
5000 | groups = ngroups; | |
a02908f1 MC |
5001 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5002 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5003 | ||
5004 | /* bitmaps and block group descriptor blocks */ | |
5005 | ret += groups + gdpblocks; | |
5006 | ||
5007 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5008 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5009 | ||
5010 | return ret; | |
5011 | } | |
5012 | ||
5013 | /* | |
25985edc | 5014 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5015 | * the modification of a single pages into a single transaction, |
5016 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5017 | * |
525f4ed8 | 5018 | * This could be called via ext4_write_begin() |
ac27a0ec | 5019 | * |
525f4ed8 | 5020 | * We need to consider the worse case, when |
a02908f1 | 5021 | * one new block per extent. |
ac27a0ec | 5022 | */ |
a86c6181 | 5023 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5024 | { |
617ba13b | 5025 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5026 | int ret; |
5027 | ||
fffb2739 | 5028 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5029 | |
a02908f1 | 5030 | /* Account for data blocks for journalled mode */ |
617ba13b | 5031 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5032 | ret += bpp; |
ac27a0ec DK |
5033 | return ret; |
5034 | } | |
f3bd1f3f MC |
5035 | |
5036 | /* | |
5037 | * Calculate the journal credits for a chunk of data modification. | |
5038 | * | |
5039 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5040 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5041 | * |
5042 | * journal buffers for data blocks are not included here, as DIO | |
5043 | * and fallocate do no need to journal data buffers. | |
5044 | */ | |
5045 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5046 | { | |
5047 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5048 | } | |
5049 | ||
ac27a0ec | 5050 | /* |
617ba13b | 5051 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5052 | * Give this, we know that the caller already has write access to iloc->bh. |
5053 | */ | |
617ba13b | 5054 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5055 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5056 | { |
5057 | int err = 0; | |
5058 | ||
c64db50e | 5059 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5060 | inode_inc_iversion(inode); |
5061 | ||
ac27a0ec DK |
5062 | /* the do_update_inode consumes one bh->b_count */ |
5063 | get_bh(iloc->bh); | |
5064 | ||
dab291af | 5065 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5066 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5067 | put_bh(iloc->bh); |
5068 | return err; | |
5069 | } | |
5070 | ||
5071 | /* | |
5072 | * On success, We end up with an outstanding reference count against | |
5073 | * iloc->bh. This _must_ be cleaned up later. | |
5074 | */ | |
5075 | ||
5076 | int | |
617ba13b MC |
5077 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5078 | struct ext4_iloc *iloc) | |
ac27a0ec | 5079 | { |
0390131b FM |
5080 | int err; |
5081 | ||
5082 | err = ext4_get_inode_loc(inode, iloc); | |
5083 | if (!err) { | |
5084 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5085 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5086 | if (err) { | |
5087 | brelse(iloc->bh); | |
5088 | iloc->bh = NULL; | |
ac27a0ec DK |
5089 | } |
5090 | } | |
617ba13b | 5091 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5092 | return err; |
5093 | } | |
5094 | ||
6dd4ee7c KS |
5095 | /* |
5096 | * Expand an inode by new_extra_isize bytes. | |
5097 | * Returns 0 on success or negative error number on failure. | |
5098 | */ | |
1d03ec98 AK |
5099 | static int ext4_expand_extra_isize(struct inode *inode, |
5100 | unsigned int new_extra_isize, | |
5101 | struct ext4_iloc iloc, | |
5102 | handle_t *handle) | |
6dd4ee7c KS |
5103 | { |
5104 | struct ext4_inode *raw_inode; | |
5105 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5106 | |
5107 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5108 | return 0; | |
5109 | ||
5110 | raw_inode = ext4_raw_inode(&iloc); | |
5111 | ||
5112 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5113 | |
5114 | /* No extended attributes present */ | |
19f5fb7a TT |
5115 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5116 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5117 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5118 | new_extra_isize); | |
5119 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5120 | return 0; | |
5121 | } | |
5122 | ||
5123 | /* try to expand with EAs present */ | |
5124 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5125 | raw_inode, handle); | |
5126 | } | |
5127 | ||
ac27a0ec DK |
5128 | /* |
5129 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5130 | * dirtiness (it may still be data-dirty). | |
5131 | * This means that the in-core inode may be reaped by prune_icache | |
5132 | * without having to perform any I/O. This is a very good thing, | |
5133 | * because *any* task may call prune_icache - even ones which | |
5134 | * have a transaction open against a different journal. | |
5135 | * | |
5136 | * Is this cheating? Not really. Sure, we haven't written the | |
5137 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5138 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5139 | * we start and wait on commits. | |
ac27a0ec | 5140 | */ |
617ba13b | 5141 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5142 | { |
617ba13b | 5143 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5144 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5145 | static unsigned int mnt_count; | |
5146 | int err, ret; | |
ac27a0ec DK |
5147 | |
5148 | might_sleep(); | |
7ff9c073 | 5149 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5150 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5151 | if (ext4_handle_valid(handle) && |
5152 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5153 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5154 | /* |
5155 | * We need extra buffer credits since we may write into EA block | |
5156 | * with this same handle. If journal_extend fails, then it will | |
5157 | * only result in a minor loss of functionality for that inode. | |
5158 | * If this is felt to be critical, then e2fsck should be run to | |
5159 | * force a large enough s_min_extra_isize. | |
5160 | */ | |
5161 | if ((jbd2_journal_extend(handle, | |
5162 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5163 | ret = ext4_expand_extra_isize(inode, | |
5164 | sbi->s_want_extra_isize, | |
5165 | iloc, handle); | |
5166 | if (ret) { | |
19f5fb7a TT |
5167 | ext4_set_inode_state(inode, |
5168 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5169 | if (mnt_count != |
5170 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5171 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5172 | "Unable to expand inode %lu. Delete" |
5173 | " some EAs or run e2fsck.", | |
5174 | inode->i_ino); | |
c1bddad9 AK |
5175 | mnt_count = |
5176 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5177 | } |
5178 | } | |
5179 | } | |
5180 | } | |
ac27a0ec | 5181 | if (!err) |
617ba13b | 5182 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5183 | return err; |
5184 | } | |
5185 | ||
5186 | /* | |
617ba13b | 5187 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5188 | * |
5189 | * We're really interested in the case where a file is being extended. | |
5190 | * i_size has been changed by generic_commit_write() and we thus need | |
5191 | * to include the updated inode in the current transaction. | |
5192 | * | |
5dd4056d | 5193 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5194 | * are allocated to the file. |
5195 | * | |
5196 | * If the inode is marked synchronous, we don't honour that here - doing | |
5197 | * so would cause a commit on atime updates, which we don't bother doing. | |
5198 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5199 | * |
5200 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5201 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5202 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5203 | */ |
aa385729 | 5204 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5205 | { |
ac27a0ec DK |
5206 | handle_t *handle; |
5207 | ||
0ae45f63 TT |
5208 | if (flags == I_DIRTY_TIME) |
5209 | return; | |
9924a92a | 5210 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5211 | if (IS_ERR(handle)) |
5212 | goto out; | |
f3dc272f | 5213 | |
f3dc272f CW |
5214 | ext4_mark_inode_dirty(handle, inode); |
5215 | ||
617ba13b | 5216 | ext4_journal_stop(handle); |
ac27a0ec DK |
5217 | out: |
5218 | return; | |
5219 | } | |
5220 | ||
5221 | #if 0 | |
5222 | /* | |
5223 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5224 | * it from being flushed to disk early. Unlike | |
617ba13b | 5225 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5226 | * returns no iloc structure, so the caller needs to repeat the iloc |
5227 | * lookup to mark the inode dirty later. | |
5228 | */ | |
617ba13b | 5229 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5230 | { |
617ba13b | 5231 | struct ext4_iloc iloc; |
ac27a0ec DK |
5232 | |
5233 | int err = 0; | |
5234 | if (handle) { | |
617ba13b | 5235 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5236 | if (!err) { |
5237 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5238 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5239 | if (!err) |
0390131b | 5240 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5241 | NULL, |
0390131b | 5242 | iloc.bh); |
ac27a0ec DK |
5243 | brelse(iloc.bh); |
5244 | } | |
5245 | } | |
617ba13b | 5246 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5247 | return err; |
5248 | } | |
5249 | #endif | |
5250 | ||
617ba13b | 5251 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5252 | { |
5253 | journal_t *journal; | |
5254 | handle_t *handle; | |
5255 | int err; | |
5256 | ||
5257 | /* | |
5258 | * We have to be very careful here: changing a data block's | |
5259 | * journaling status dynamically is dangerous. If we write a | |
5260 | * data block to the journal, change the status and then delete | |
5261 | * that block, we risk forgetting to revoke the old log record | |
5262 | * from the journal and so a subsequent replay can corrupt data. | |
5263 | * So, first we make sure that the journal is empty and that | |
5264 | * nobody is changing anything. | |
5265 | */ | |
5266 | ||
617ba13b | 5267 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5268 | if (!journal) |
5269 | return 0; | |
d699594d | 5270 | if (is_journal_aborted(journal)) |
ac27a0ec | 5271 | return -EROFS; |
2aff57b0 YY |
5272 | /* We have to allocate physical blocks for delalloc blocks |
5273 | * before flushing journal. otherwise delalloc blocks can not | |
5274 | * be allocated any more. even more truncate on delalloc blocks | |
5275 | * could trigger BUG by flushing delalloc blocks in journal. | |
5276 | * There is no delalloc block in non-journal data mode. | |
5277 | */ | |
5278 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5279 | err = ext4_alloc_da_blocks(inode); | |
5280 | if (err < 0) | |
5281 | return err; | |
5282 | } | |
ac27a0ec | 5283 | |
17335dcc DM |
5284 | /* Wait for all existing dio workers */ |
5285 | ext4_inode_block_unlocked_dio(inode); | |
5286 | inode_dio_wait(inode); | |
5287 | ||
dab291af | 5288 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5289 | |
5290 | /* | |
5291 | * OK, there are no updates running now, and all cached data is | |
5292 | * synced to disk. We are now in a completely consistent state | |
5293 | * which doesn't have anything in the journal, and we know that | |
5294 | * no filesystem updates are running, so it is safe to modify | |
5295 | * the inode's in-core data-journaling state flag now. | |
5296 | */ | |
5297 | ||
5298 | if (val) | |
12e9b892 | 5299 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5300 | else { |
4f879ca6 JK |
5301 | err = jbd2_journal_flush(journal); |
5302 | if (err < 0) { | |
5303 | jbd2_journal_unlock_updates(journal); | |
5304 | ext4_inode_resume_unlocked_dio(inode); | |
5305 | return err; | |
5306 | } | |
12e9b892 | 5307 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5308 | } |
617ba13b | 5309 | ext4_set_aops(inode); |
ac27a0ec | 5310 | |
dab291af | 5311 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5312 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5313 | |
5314 | /* Finally we can mark the inode as dirty. */ | |
5315 | ||
9924a92a | 5316 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5317 | if (IS_ERR(handle)) |
5318 | return PTR_ERR(handle); | |
5319 | ||
617ba13b | 5320 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5321 | ext4_handle_sync(handle); |
617ba13b MC |
5322 | ext4_journal_stop(handle); |
5323 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5324 | |
5325 | return err; | |
5326 | } | |
2e9ee850 AK |
5327 | |
5328 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5329 | { | |
5330 | return !buffer_mapped(bh); | |
5331 | } | |
5332 | ||
c2ec175c | 5333 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5334 | { |
c2ec175c | 5335 | struct page *page = vmf->page; |
2e9ee850 AK |
5336 | loff_t size; |
5337 | unsigned long len; | |
9ea7df53 | 5338 | int ret; |
2e9ee850 | 5339 | struct file *file = vma->vm_file; |
496ad9aa | 5340 | struct inode *inode = file_inode(file); |
2e9ee850 | 5341 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5342 | handle_t *handle; |
5343 | get_block_t *get_block; | |
5344 | int retries = 0; | |
2e9ee850 | 5345 | |
8e8ad8a5 | 5346 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5347 | file_update_time(vma->vm_file); |
ea3d7209 JK |
5348 | |
5349 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
9ea7df53 JK |
5350 | /* Delalloc case is easy... */ |
5351 | if (test_opt(inode->i_sb, DELALLOC) && | |
5352 | !ext4_should_journal_data(inode) && | |
5353 | !ext4_nonda_switch(inode->i_sb)) { | |
5354 | do { | |
5c500029 | 5355 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
5356 | ext4_da_get_block_prep); |
5357 | } while (ret == -ENOSPC && | |
5358 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5359 | goto out_ret; | |
2e9ee850 | 5360 | } |
0e499890 DW |
5361 | |
5362 | lock_page(page); | |
9ea7df53 JK |
5363 | size = i_size_read(inode); |
5364 | /* Page got truncated from under us? */ | |
5365 | if (page->mapping != mapping || page_offset(page) > size) { | |
5366 | unlock_page(page); | |
5367 | ret = VM_FAULT_NOPAGE; | |
5368 | goto out; | |
0e499890 | 5369 | } |
2e9ee850 AK |
5370 | |
5371 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5372 | len = size & ~PAGE_CACHE_MASK; | |
5373 | else | |
5374 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5375 | /* |
9ea7df53 JK |
5376 | * Return if we have all the buffers mapped. This avoids the need to do |
5377 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5378 | */ |
2e9ee850 | 5379 | if (page_has_buffers(page)) { |
f19d5870 TM |
5380 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5381 | 0, len, NULL, | |
5382 | ext4_bh_unmapped)) { | |
9ea7df53 | 5383 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5384 | wait_for_stable_page(page); |
9ea7df53 JK |
5385 | ret = VM_FAULT_LOCKED; |
5386 | goto out; | |
a827eaff | 5387 | } |
2e9ee850 | 5388 | } |
a827eaff | 5389 | unlock_page(page); |
9ea7df53 JK |
5390 | /* OK, we need to fill the hole... */ |
5391 | if (ext4_should_dioread_nolock(inode)) | |
5392 | get_block = ext4_get_block_write; | |
5393 | else | |
5394 | get_block = ext4_get_block; | |
5395 | retry_alloc: | |
9924a92a TT |
5396 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5397 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5398 | if (IS_ERR(handle)) { |
c2ec175c | 5399 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5400 | goto out; |
5401 | } | |
5c500029 | 5402 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 5403 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 5404 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5405 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5406 | unlock_page(page); | |
5407 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5408 | ext4_journal_stop(handle); |
9ea7df53 JK |
5409 | goto out; |
5410 | } | |
5411 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5412 | } | |
5413 | ext4_journal_stop(handle); | |
5414 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5415 | goto retry_alloc; | |
5416 | out_ret: | |
5417 | ret = block_page_mkwrite_return(ret); | |
5418 | out: | |
ea3d7209 | 5419 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 5420 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5421 | return ret; |
5422 | } | |
ea3d7209 JK |
5423 | |
5424 | int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
5425 | { | |
5426 | struct inode *inode = file_inode(vma->vm_file); | |
5427 | int err; | |
5428 | ||
5429 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
5430 | err = filemap_fault(vma, vmf); | |
5431 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
5432 | ||
5433 | return err; | |
5434 | } |