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); | |
814525f4 | 54 | __u32 csum; |
b47820ed DJ |
55 | __u16 dummy_csum = 0; |
56 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
57 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 58 | |
b47820ed DJ |
59 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
60 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
61 | offset += csum_size; | |
62 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
63 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 64 | |
b47820ed DJ |
65 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
66 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
67 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
68 | EXT4_GOOD_OLD_INODE_SIZE, | |
69 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
70 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
71 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
72 | csum_size); | |
73 | offset += csum_size; | |
74 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
75 | EXT4_INODE_SIZE(inode->i_sb) - | |
76 | offset); | |
77 | } | |
814525f4 DW |
78 | } |
79 | ||
814525f4 DW |
80 | return csum; |
81 | } | |
82 | ||
83 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
84 | struct ext4_inode_info *ei) | |
85 | { | |
86 | __u32 provided, calculated; | |
87 | ||
88 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
89 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 90 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
91 | return 1; |
92 | ||
93 | provided = le16_to_cpu(raw->i_checksum_lo); | |
94 | calculated = ext4_inode_csum(inode, raw, ei); | |
95 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
96 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
97 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
98 | else | |
99 | calculated &= 0xFFFF; | |
100 | ||
101 | return provided == calculated; | |
102 | } | |
103 | ||
104 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
105 | struct ext4_inode_info *ei) | |
106 | { | |
107 | __u32 csum; | |
108 | ||
109 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
110 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 111 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
112 | return; |
113 | ||
114 | csum = ext4_inode_csum(inode, raw, ei); | |
115 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
116 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
117 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
118 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
119 | } | |
120 | ||
678aaf48 JK |
121 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
122 | loff_t new_size) | |
123 | { | |
7ff9c073 | 124 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
125 | /* |
126 | * If jinode is zero, then we never opened the file for | |
127 | * writing, so there's no need to call | |
128 | * jbd2_journal_begin_ordered_truncate() since there's no | |
129 | * outstanding writes we need to flush. | |
130 | */ | |
131 | if (!EXT4_I(inode)->jinode) | |
132 | return 0; | |
133 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
134 | EXT4_I(inode)->jinode, | |
135 | new_size); | |
678aaf48 JK |
136 | } |
137 | ||
d47992f8 LC |
138 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
139 | unsigned int length); | |
cb20d518 TT |
140 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
141 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
fffb2739 JK |
142 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
143 | int pextents); | |
64769240 | 144 | |
ac27a0ec DK |
145 | /* |
146 | * Test whether an inode is a fast symlink. | |
147 | */ | |
f348c252 | 148 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 149 | { |
65eddb56 YY |
150 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
151 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
ac27a0ec | 152 | |
bd9db175 ZL |
153 | if (ext4_has_inline_data(inode)) |
154 | return 0; | |
155 | ||
ac27a0ec DK |
156 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
157 | } | |
158 | ||
ac27a0ec DK |
159 | /* |
160 | * Restart the transaction associated with *handle. This does a commit, | |
161 | * so before we call here everything must be consistently dirtied against | |
162 | * this transaction. | |
163 | */ | |
fa5d1113 | 164 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 165 | int nblocks) |
ac27a0ec | 166 | { |
487caeef JK |
167 | int ret; |
168 | ||
169 | /* | |
e35fd660 | 170 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
171 | * moment, get_block can be called only for blocks inside i_size since |
172 | * page cache has been already dropped and writes are blocked by | |
173 | * i_mutex. So we can safely drop the i_data_sem here. | |
174 | */ | |
0390131b | 175 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 176 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 177 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 178 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 179 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 180 | ext4_discard_preallocations(inode); |
487caeef JK |
181 | |
182 | return ret; | |
ac27a0ec DK |
183 | } |
184 | ||
185 | /* | |
186 | * Called at the last iput() if i_nlink is zero. | |
187 | */ | |
0930fcc1 | 188 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
189 | { |
190 | handle_t *handle; | |
bc965ab3 | 191 | int err; |
ac27a0ec | 192 | |
7ff9c073 | 193 | trace_ext4_evict_inode(inode); |
2581fdc8 | 194 | |
0930fcc1 | 195 | if (inode->i_nlink) { |
2d859db3 JK |
196 | /* |
197 | * When journalling data dirty buffers are tracked only in the | |
198 | * journal. So although mm thinks everything is clean and | |
199 | * ready for reaping the inode might still have some pages to | |
200 | * write in the running transaction or waiting to be | |
201 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
202 | * (via truncate_inode_pages()) to discard these buffers can | |
203 | * cause data loss. Also even if we did not discard these | |
204 | * buffers, we would have no way to find them after the inode | |
205 | * is reaped and thus user could see stale data if he tries to | |
206 | * read them before the transaction is checkpointed. So be | |
207 | * careful and force everything to disk here... We use | |
208 | * ei->i_datasync_tid to store the newest transaction | |
209 | * containing inode's data. | |
210 | * | |
211 | * Note that directories do not have this problem because they | |
212 | * don't use page cache. | |
213 | */ | |
6a7fd522 VN |
214 | if (inode->i_ino != EXT4_JOURNAL_INO && |
215 | ext4_should_journal_data(inode) && | |
216 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { | |
2d859db3 JK |
217 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
218 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
219 | ||
d76a3a77 | 220 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
221 | filemap_write_and_wait(&inode->i_data); |
222 | } | |
91b0abe3 | 223 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 224 | |
0930fcc1 AV |
225 | goto no_delete; |
226 | } | |
227 | ||
e2bfb088 TT |
228 | if (is_bad_inode(inode)) |
229 | goto no_delete; | |
230 | dquot_initialize(inode); | |
907f4554 | 231 | |
678aaf48 JK |
232 | if (ext4_should_order_data(inode)) |
233 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 234 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 235 | |
8e8ad8a5 JK |
236 | /* |
237 | * Protect us against freezing - iput() caller didn't have to have any | |
238 | * protection against it | |
239 | */ | |
240 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
241 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
242 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 243 | if (IS_ERR(handle)) { |
bc965ab3 | 244 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
245 | /* |
246 | * If we're going to skip the normal cleanup, we still need to | |
247 | * make sure that the in-core orphan linked list is properly | |
248 | * cleaned up. | |
249 | */ | |
617ba13b | 250 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 251 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
252 | goto no_delete; |
253 | } | |
254 | ||
255 | if (IS_SYNC(inode)) | |
0390131b | 256 | ext4_handle_sync(handle); |
ac27a0ec | 257 | inode->i_size = 0; |
bc965ab3 TT |
258 | err = ext4_mark_inode_dirty(handle, inode); |
259 | if (err) { | |
12062ddd | 260 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
261 | "couldn't mark inode dirty (err %d)", err); |
262 | goto stop_handle; | |
263 | } | |
ac27a0ec | 264 | if (inode->i_blocks) |
617ba13b | 265 | ext4_truncate(inode); |
bc965ab3 TT |
266 | |
267 | /* | |
268 | * ext4_ext_truncate() doesn't reserve any slop when it | |
269 | * restarts journal transactions; therefore there may not be | |
270 | * enough credits left in the handle to remove the inode from | |
271 | * the orphan list and set the dtime field. | |
272 | */ | |
0390131b | 273 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
274 | err = ext4_journal_extend(handle, 3); |
275 | if (err > 0) | |
276 | err = ext4_journal_restart(handle, 3); | |
277 | if (err != 0) { | |
12062ddd | 278 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
279 | "couldn't extend journal (err %d)", err); |
280 | stop_handle: | |
281 | ext4_journal_stop(handle); | |
45388219 | 282 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 283 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
284 | goto no_delete; |
285 | } | |
286 | } | |
287 | ||
ac27a0ec | 288 | /* |
617ba13b | 289 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 290 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 291 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 292 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 293 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
294 | * (Well, we could do this if we need to, but heck - it works) |
295 | */ | |
617ba13b MC |
296 | ext4_orphan_del(handle, inode); |
297 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
298 | |
299 | /* | |
300 | * One subtle ordering requirement: if anything has gone wrong | |
301 | * (transaction abort, IO errors, whatever), then we can still | |
302 | * do these next steps (the fs will already have been marked as | |
303 | * having errors), but we can't free the inode if the mark_dirty | |
304 | * fails. | |
305 | */ | |
617ba13b | 306 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 307 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 308 | ext4_clear_inode(inode); |
ac27a0ec | 309 | else |
617ba13b MC |
310 | ext4_free_inode(handle, inode); |
311 | ext4_journal_stop(handle); | |
8e8ad8a5 | 312 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
313 | return; |
314 | no_delete: | |
0930fcc1 | 315 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
316 | } |
317 | ||
a9e7f447 DM |
318 | #ifdef CONFIG_QUOTA |
319 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 320 | { |
a9e7f447 | 321 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 322 | } |
a9e7f447 | 323 | #endif |
9d0be502 | 324 | |
0637c6f4 TT |
325 | /* |
326 | * Called with i_data_sem down, which is important since we can call | |
327 | * ext4_discard_preallocations() from here. | |
328 | */ | |
5f634d06 AK |
329 | void ext4_da_update_reserve_space(struct inode *inode, |
330 | int used, int quota_claim) | |
12219aea AK |
331 | { |
332 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 333 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
334 | |
335 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 336 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 337 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 338 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 339 | "with only %d reserved data blocks", |
0637c6f4 TT |
340 | __func__, inode->i_ino, used, |
341 | ei->i_reserved_data_blocks); | |
342 | WARN_ON(1); | |
343 | used = ei->i_reserved_data_blocks; | |
344 | } | |
12219aea | 345 | |
0637c6f4 TT |
346 | /* Update per-inode reservations */ |
347 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 348 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 349 | |
12219aea | 350 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 351 | |
72b8ab9d ES |
352 | /* Update quota subsystem for data blocks */ |
353 | if (quota_claim) | |
7b415bf6 | 354 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 355 | else { |
5f634d06 AK |
356 | /* |
357 | * We did fallocate with an offset that is already delayed | |
358 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 359 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 360 | */ |
7b415bf6 | 361 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 362 | } |
d6014301 AK |
363 | |
364 | /* | |
365 | * If we have done all the pending block allocations and if | |
366 | * there aren't any writers on the inode, we can discard the | |
367 | * inode's preallocations. | |
368 | */ | |
0637c6f4 TT |
369 | if ((ei->i_reserved_data_blocks == 0) && |
370 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 371 | ext4_discard_preallocations(inode); |
12219aea AK |
372 | } |
373 | ||
e29136f8 | 374 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
375 | unsigned int line, |
376 | struct ext4_map_blocks *map) | |
6fd058f7 | 377 | { |
24676da4 TT |
378 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
379 | map->m_len)) { | |
c398eda0 TT |
380 | ext4_error_inode(inode, func, line, map->m_pblk, |
381 | "lblock %lu mapped to illegal pblock " | |
382 | "(length %d)", (unsigned long) map->m_lblk, | |
383 | map->m_len); | |
6a797d27 | 384 | return -EFSCORRUPTED; |
6fd058f7 TT |
385 | } |
386 | return 0; | |
387 | } | |
388 | ||
53085fac JK |
389 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
390 | ext4_lblk_t len) | |
391 | { | |
392 | int ret; | |
393 | ||
394 | if (ext4_encrypted_inode(inode)) | |
a7550b30 | 395 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
396 | |
397 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
398 | if (ret > 0) | |
399 | ret = 0; | |
400 | ||
401 | return ret; | |
402 | } | |
403 | ||
e29136f8 | 404 | #define check_block_validity(inode, map) \ |
c398eda0 | 405 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 406 | |
921f266b DM |
407 | #ifdef ES_AGGRESSIVE_TEST |
408 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
409 | struct inode *inode, | |
410 | struct ext4_map_blocks *es_map, | |
411 | struct ext4_map_blocks *map, | |
412 | int flags) | |
413 | { | |
414 | int retval; | |
415 | ||
416 | map->m_flags = 0; | |
417 | /* | |
418 | * There is a race window that the result is not the same. | |
419 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
420 | * is that we lookup a block mapping in extent status tree with | |
421 | * out taking i_data_sem. So at the time the unwritten extent | |
422 | * could be converted. | |
423 | */ | |
2dcba478 | 424 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
425 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
426 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
427 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
428 | } else { | |
429 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
430 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
431 | } | |
2dcba478 | 432 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
433 | |
434 | /* | |
435 | * We don't check m_len because extent will be collpased in status | |
436 | * tree. So the m_len might not equal. | |
437 | */ | |
438 | if (es_map->m_lblk != map->m_lblk || | |
439 | es_map->m_flags != map->m_flags || | |
440 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 441 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
442 | "es_cached ex [%d/%d/%llu/%x] != " |
443 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
444 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
445 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
446 | map->m_len, map->m_pblk, map->m_flags, | |
447 | retval, flags); | |
448 | } | |
449 | } | |
450 | #endif /* ES_AGGRESSIVE_TEST */ | |
451 | ||
f5ab0d1f | 452 | /* |
e35fd660 | 453 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 454 | * and returns if the blocks are already mapped. |
f5ab0d1f | 455 | * |
f5ab0d1f MC |
456 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
457 | * and store the allocated blocks in the result buffer head and mark it | |
458 | * mapped. | |
459 | * | |
e35fd660 TT |
460 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
461 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
462 | * based files |
463 | * | |
facab4d9 JK |
464 | * On success, it returns the number of blocks being mapped or allocated. if |
465 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
466 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
467 | * |
468 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
469 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
470 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
471 | * |
472 | * It returns the error in case of allocation failure. | |
473 | */ | |
e35fd660 TT |
474 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
475 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 476 | { |
d100eef2 | 477 | struct extent_status es; |
0e855ac8 | 478 | int retval; |
b8a86845 | 479 | int ret = 0; |
921f266b DM |
480 | #ifdef ES_AGGRESSIVE_TEST |
481 | struct ext4_map_blocks orig_map; | |
482 | ||
483 | memcpy(&orig_map, map, sizeof(*map)); | |
484 | #endif | |
f5ab0d1f | 485 | |
e35fd660 TT |
486 | map->m_flags = 0; |
487 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
488 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
489 | (unsigned long) map->m_lblk); | |
d100eef2 | 490 | |
e861b5e9 TT |
491 | /* |
492 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
493 | */ | |
494 | if (unlikely(map->m_len > INT_MAX)) | |
495 | map->m_len = INT_MAX; | |
496 | ||
4adb6ab3 KM |
497 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
498 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 499 | return -EFSCORRUPTED; |
4adb6ab3 | 500 | |
d100eef2 ZL |
501 | /* Lookup extent status tree firstly */ |
502 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
503 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
504 | map->m_pblk = ext4_es_pblock(&es) + | |
505 | map->m_lblk - es.es_lblk; | |
506 | map->m_flags |= ext4_es_is_written(&es) ? | |
507 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
508 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
509 | if (retval > map->m_len) | |
510 | retval = map->m_len; | |
511 | map->m_len = retval; | |
512 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
513 | map->m_pblk = 0; |
514 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
515 | if (retval > map->m_len) | |
516 | retval = map->m_len; | |
517 | map->m_len = retval; | |
d100eef2 ZL |
518 | retval = 0; |
519 | } else { | |
520 | BUG_ON(1); | |
521 | } | |
921f266b DM |
522 | #ifdef ES_AGGRESSIVE_TEST |
523 | ext4_map_blocks_es_recheck(handle, inode, map, | |
524 | &orig_map, flags); | |
525 | #endif | |
d100eef2 ZL |
526 | goto found; |
527 | } | |
528 | ||
4df3d265 | 529 | /* |
b920c755 TT |
530 | * Try to see if we can get the block without requesting a new |
531 | * file system block. | |
4df3d265 | 532 | */ |
2dcba478 | 533 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 534 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
535 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
536 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 537 | } else { |
a4e5d88b DM |
538 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
539 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 540 | } |
f7fec032 | 541 | if (retval > 0) { |
3be78c73 | 542 | unsigned int status; |
f7fec032 | 543 | |
44fb851d ZL |
544 | if (unlikely(retval != map->m_len)) { |
545 | ext4_warning(inode->i_sb, | |
546 | "ES len assertion failed for inode " | |
547 | "%lu: retval %d != map->m_len %d", | |
548 | inode->i_ino, retval, map->m_len); | |
549 | WARN_ON(1); | |
921f266b | 550 | } |
921f266b | 551 | |
f7fec032 ZL |
552 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
553 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
554 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 555 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
556 | ext4_find_delalloc_range(inode, map->m_lblk, |
557 | map->m_lblk + map->m_len - 1)) | |
558 | status |= EXTENT_STATUS_DELAYED; | |
559 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
560 | map->m_len, map->m_pblk, status); | |
561 | if (ret < 0) | |
562 | retval = ret; | |
563 | } | |
2dcba478 | 564 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 565 | |
d100eef2 | 566 | found: |
e35fd660 | 567 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 568 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
569 | if (ret != 0) |
570 | return ret; | |
571 | } | |
572 | ||
f5ab0d1f | 573 | /* If it is only a block(s) look up */ |
c2177057 | 574 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
575 | return retval; |
576 | ||
577 | /* | |
578 | * Returns if the blocks have already allocated | |
579 | * | |
580 | * Note that if blocks have been preallocated | |
df3ab170 | 581 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
582 | * with buffer head unmapped. |
583 | */ | |
e35fd660 | 584 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
585 | /* |
586 | * If we need to convert extent to unwritten | |
587 | * we continue and do the actual work in | |
588 | * ext4_ext_map_blocks() | |
589 | */ | |
590 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
591 | return retval; | |
4df3d265 | 592 | |
2a8964d6 | 593 | /* |
a25a4e1a ZL |
594 | * Here we clear m_flags because after allocating an new extent, |
595 | * it will be set again. | |
2a8964d6 | 596 | */ |
a25a4e1a | 597 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 598 | |
4df3d265 | 599 | /* |
556615dc | 600 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 601 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 602 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 603 | * with create == 1 flag. |
4df3d265 | 604 | */ |
c8b459f4 | 605 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 606 | |
4df3d265 AK |
607 | /* |
608 | * We need to check for EXT4 here because migrate | |
609 | * could have changed the inode type in between | |
610 | */ | |
12e9b892 | 611 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 612 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 613 | } else { |
e35fd660 | 614 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 615 | |
e35fd660 | 616 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
617 | /* |
618 | * We allocated new blocks which will result in | |
619 | * i_data's format changing. Force the migrate | |
620 | * to fail by clearing migrate flags | |
621 | */ | |
19f5fb7a | 622 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 623 | } |
d2a17637 | 624 | |
5f634d06 AK |
625 | /* |
626 | * Update reserved blocks/metadata blocks after successful | |
627 | * block allocation which had been deferred till now. We don't | |
628 | * support fallocate for non extent files. So we can update | |
629 | * reserve space here. | |
630 | */ | |
631 | if ((retval > 0) && | |
1296cc85 | 632 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
633 | ext4_da_update_reserve_space(inode, retval, 1); |
634 | } | |
2ac3b6e0 | 635 | |
f7fec032 | 636 | if (retval > 0) { |
3be78c73 | 637 | unsigned int status; |
f7fec032 | 638 | |
44fb851d ZL |
639 | if (unlikely(retval != map->m_len)) { |
640 | ext4_warning(inode->i_sb, | |
641 | "ES len assertion failed for inode " | |
642 | "%lu: retval %d != map->m_len %d", | |
643 | inode->i_ino, retval, map->m_len); | |
644 | WARN_ON(1); | |
921f266b | 645 | } |
921f266b | 646 | |
c86d8db3 JK |
647 | /* |
648 | * We have to zeroout blocks before inserting them into extent | |
649 | * status tree. Otherwise someone could look them up there and | |
650 | * use them before they are really zeroed. | |
651 | */ | |
652 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
653 | map->m_flags & EXT4_MAP_MAPPED && | |
654 | map->m_flags & EXT4_MAP_NEW) { | |
655 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
656 | map->m_pblk, map->m_len); | |
657 | if (ret) { | |
658 | retval = ret; | |
659 | goto out_sem; | |
660 | } | |
661 | } | |
662 | ||
adb23551 ZL |
663 | /* |
664 | * If the extent has been zeroed out, we don't need to update | |
665 | * extent status tree. | |
666 | */ | |
667 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
668 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
669 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 670 | goto out_sem; |
adb23551 | 671 | } |
f7fec032 ZL |
672 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
673 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
674 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 675 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
676 | ext4_find_delalloc_range(inode, map->m_lblk, |
677 | map->m_lblk + map->m_len - 1)) | |
678 | status |= EXTENT_STATUS_DELAYED; | |
679 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
680 | map->m_pblk, status); | |
c86d8db3 | 681 | if (ret < 0) { |
f7fec032 | 682 | retval = ret; |
c86d8db3 JK |
683 | goto out_sem; |
684 | } | |
5356f261 AK |
685 | } |
686 | ||
c86d8db3 | 687 | out_sem: |
4df3d265 | 688 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 689 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 690 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
691 | if (ret != 0) |
692 | return ret; | |
06bd3c36 JK |
693 | |
694 | /* | |
695 | * Inodes with freshly allocated blocks where contents will be | |
696 | * visible after transaction commit must be on transaction's | |
697 | * ordered data list. | |
698 | */ | |
699 | if (map->m_flags & EXT4_MAP_NEW && | |
700 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
701 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
702 | !IS_NOQUOTA(inode) && | |
703 | ext4_should_order_data(inode)) { | |
ee0876bc JK |
704 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
705 | ret = ext4_jbd2_inode_add_wait(handle, inode); | |
706 | else | |
707 | ret = ext4_jbd2_inode_add_write(handle, inode); | |
06bd3c36 JK |
708 | if (ret) |
709 | return ret; | |
710 | } | |
6fd058f7 | 711 | } |
0e855ac8 AK |
712 | return retval; |
713 | } | |
714 | ||
ed8ad838 JK |
715 | /* |
716 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
717 | * we have to be careful as someone else may be manipulating b_state as well. | |
718 | */ | |
719 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
720 | { | |
721 | unsigned long old_state; | |
722 | unsigned long new_state; | |
723 | ||
724 | flags &= EXT4_MAP_FLAGS; | |
725 | ||
726 | /* Dummy buffer_head? Set non-atomically. */ | |
727 | if (!bh->b_page) { | |
728 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
729 | return; | |
730 | } | |
731 | /* | |
732 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
733 | * once we get rid of using bh as a container for mapping information | |
734 | * to pass to / from get_block functions, this can go away. | |
735 | */ | |
736 | do { | |
737 | old_state = READ_ONCE(bh->b_state); | |
738 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
739 | } while (unlikely( | |
740 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
741 | } | |
742 | ||
2ed88685 TT |
743 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
744 | struct buffer_head *bh, int flags) | |
ac27a0ec | 745 | { |
2ed88685 | 746 | struct ext4_map_blocks map; |
efe70c29 | 747 | int ret = 0; |
ac27a0ec | 748 | |
46c7f254 TM |
749 | if (ext4_has_inline_data(inode)) |
750 | return -ERANGE; | |
751 | ||
2ed88685 TT |
752 | map.m_lblk = iblock; |
753 | map.m_len = bh->b_size >> inode->i_blkbits; | |
754 | ||
efe70c29 JK |
755 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
756 | flags); | |
7fb5409d | 757 | if (ret > 0) { |
2ed88685 | 758 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 759 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 760 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 761 | ret = 0; |
ac27a0ec DK |
762 | } |
763 | return ret; | |
764 | } | |
765 | ||
2ed88685 TT |
766 | int ext4_get_block(struct inode *inode, sector_t iblock, |
767 | struct buffer_head *bh, int create) | |
768 | { | |
769 | return _ext4_get_block(inode, iblock, bh, | |
770 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
771 | } | |
772 | ||
705965bd JK |
773 | /* |
774 | * Get block function used when preparing for buffered write if we require | |
775 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
776 | * will be converted to written after the IO is complete. | |
777 | */ | |
778 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
779 | struct buffer_head *bh_result, int create) | |
780 | { | |
781 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
782 | inode->i_ino, create); | |
783 | return _ext4_get_block(inode, iblock, bh_result, | |
784 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
785 | } | |
786 | ||
efe70c29 JK |
787 | /* Maximum number of blocks we map for direct IO at once. */ |
788 | #define DIO_MAX_BLOCKS 4096 | |
789 | ||
e84dfbe2 JK |
790 | /* |
791 | * Get blocks function for the cases that need to start a transaction - | |
792 | * generally difference cases of direct IO and DAX IO. It also handles retries | |
793 | * in case of ENOSPC. | |
794 | */ | |
795 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, | |
796 | struct buffer_head *bh_result, int flags) | |
efe70c29 JK |
797 | { |
798 | int dio_credits; | |
e84dfbe2 JK |
799 | handle_t *handle; |
800 | int retries = 0; | |
801 | int ret; | |
efe70c29 JK |
802 | |
803 | /* Trim mapping request to maximum we can map at once for DIO */ | |
804 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) | |
805 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; | |
806 | dio_credits = ext4_chunk_trans_blocks(inode, | |
807 | bh_result->b_size >> inode->i_blkbits); | |
e84dfbe2 JK |
808 | retry: |
809 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); | |
810 | if (IS_ERR(handle)) | |
811 | return PTR_ERR(handle); | |
812 | ||
813 | ret = _ext4_get_block(inode, iblock, bh_result, flags); | |
814 | ext4_journal_stop(handle); | |
815 | ||
816 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
817 | goto retry; | |
818 | return ret; | |
efe70c29 JK |
819 | } |
820 | ||
705965bd JK |
821 | /* Get block function for DIO reads and writes to inodes without extents */ |
822 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, | |
823 | struct buffer_head *bh, int create) | |
824 | { | |
efe70c29 JK |
825 | /* We don't expect handle for direct IO */ |
826 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
827 | ||
e84dfbe2 JK |
828 | if (!create) |
829 | return _ext4_get_block(inode, iblock, bh, 0); | |
830 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); | |
705965bd JK |
831 | } |
832 | ||
833 | /* | |
109811c2 | 834 | * Get block function for AIO DIO writes when we create unwritten extent if |
705965bd JK |
835 | * blocks are not allocated yet. The extent will be converted to written |
836 | * after IO is complete. | |
837 | */ | |
109811c2 JK |
838 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
839 | sector_t iblock, struct buffer_head *bh_result, int create) | |
705965bd | 840 | { |
efe70c29 JK |
841 | int ret; |
842 | ||
efe70c29 JK |
843 | /* We don't expect handle for direct IO */ |
844 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
845 | ||
e84dfbe2 JK |
846 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
847 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
efe70c29 | 848 | |
109811c2 JK |
849 | /* |
850 | * When doing DIO using unwritten extents, we need io_end to convert | |
851 | * unwritten extents to written on IO completion. We allocate io_end | |
852 | * once we spot unwritten extent and store it in b_private. Generic | |
853 | * DIO code keeps b_private set and furthermore passes the value to | |
854 | * our completion callback in 'private' argument. | |
855 | */ | |
856 | if (!ret && buffer_unwritten(bh_result)) { | |
857 | if (!bh_result->b_private) { | |
858 | ext4_io_end_t *io_end; | |
859 | ||
860 | io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
861 | if (!io_end) | |
862 | return -ENOMEM; | |
863 | bh_result->b_private = io_end; | |
864 | ext4_set_io_unwritten_flag(inode, io_end); | |
865 | } | |
efe70c29 | 866 | set_buffer_defer_completion(bh_result); |
efe70c29 JK |
867 | } |
868 | ||
869 | return ret; | |
705965bd JK |
870 | } |
871 | ||
109811c2 JK |
872 | /* |
873 | * Get block function for non-AIO DIO writes when we create unwritten extent if | |
874 | * blocks are not allocated yet. The extent will be converted to written | |
875 | * after IO is complete from ext4_ext_direct_IO() function. | |
876 | */ | |
877 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, | |
878 | sector_t iblock, struct buffer_head *bh_result, int create) | |
879 | { | |
109811c2 JK |
880 | int ret; |
881 | ||
882 | /* We don't expect handle for direct IO */ | |
883 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
884 | ||
e84dfbe2 JK |
885 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
886 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
109811c2 JK |
887 | |
888 | /* | |
889 | * Mark inode as having pending DIO writes to unwritten extents. | |
890 | * ext4_ext_direct_IO() checks this flag and converts extents to | |
891 | * written. | |
892 | */ | |
893 | if (!ret && buffer_unwritten(bh_result)) | |
894 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
895 | ||
896 | return ret; | |
897 | } | |
898 | ||
705965bd JK |
899 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
900 | struct buffer_head *bh_result, int create) | |
901 | { | |
902 | int ret; | |
903 | ||
904 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", | |
905 | inode->i_ino, create); | |
efe70c29 JK |
906 | /* We don't expect handle for direct IO */ |
907 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
908 | ||
705965bd JK |
909 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
910 | /* | |
911 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
912 | * that. | |
913 | */ | |
efe70c29 | 914 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
705965bd JK |
915 | |
916 | return ret; | |
917 | } | |
918 | ||
919 | ||
ac27a0ec DK |
920 | /* |
921 | * `handle' can be NULL if create is zero | |
922 | */ | |
617ba13b | 923 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 924 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 925 | { |
2ed88685 TT |
926 | struct ext4_map_blocks map; |
927 | struct buffer_head *bh; | |
c5e298ae | 928 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 929 | int err; |
ac27a0ec DK |
930 | |
931 | J_ASSERT(handle != NULL || create == 0); | |
932 | ||
2ed88685 TT |
933 | map.m_lblk = block; |
934 | map.m_len = 1; | |
c5e298ae | 935 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 936 | |
10560082 TT |
937 | if (err == 0) |
938 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 939 | if (err < 0) |
10560082 | 940 | return ERR_PTR(err); |
2ed88685 TT |
941 | |
942 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
943 | if (unlikely(!bh)) |
944 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
945 | if (map.m_flags & EXT4_MAP_NEW) { |
946 | J_ASSERT(create != 0); | |
947 | J_ASSERT(handle != NULL); | |
ac27a0ec | 948 | |
2ed88685 TT |
949 | /* |
950 | * Now that we do not always journal data, we should | |
951 | * keep in mind whether this should always journal the | |
952 | * new buffer as metadata. For now, regular file | |
953 | * writes use ext4_get_block instead, so it's not a | |
954 | * problem. | |
955 | */ | |
956 | lock_buffer(bh); | |
957 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
958 | err = ext4_journal_get_create_access(handle, bh); |
959 | if (unlikely(err)) { | |
960 | unlock_buffer(bh); | |
961 | goto errout; | |
962 | } | |
963 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
964 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
965 | set_buffer_uptodate(bh); | |
ac27a0ec | 966 | } |
2ed88685 TT |
967 | unlock_buffer(bh); |
968 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
969 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
970 | if (unlikely(err)) |
971 | goto errout; | |
972 | } else | |
2ed88685 | 973 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 974 | return bh; |
10560082 TT |
975 | errout: |
976 | brelse(bh); | |
977 | return ERR_PTR(err); | |
ac27a0ec DK |
978 | } |
979 | ||
617ba13b | 980 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 981 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 982 | { |
af5bc92d | 983 | struct buffer_head *bh; |
ac27a0ec | 984 | |
c5e298ae | 985 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 986 | if (IS_ERR(bh)) |
ac27a0ec | 987 | return bh; |
1c215028 | 988 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 989 | return bh; |
dfec8a14 | 990 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
991 | wait_on_buffer(bh); |
992 | if (buffer_uptodate(bh)) | |
993 | return bh; | |
994 | put_bh(bh); | |
1c215028 | 995 | return ERR_PTR(-EIO); |
ac27a0ec DK |
996 | } |
997 | ||
f19d5870 TM |
998 | int ext4_walk_page_buffers(handle_t *handle, |
999 | struct buffer_head *head, | |
1000 | unsigned from, | |
1001 | unsigned to, | |
1002 | int *partial, | |
1003 | int (*fn)(handle_t *handle, | |
1004 | struct buffer_head *bh)) | |
ac27a0ec DK |
1005 | { |
1006 | struct buffer_head *bh; | |
1007 | unsigned block_start, block_end; | |
1008 | unsigned blocksize = head->b_size; | |
1009 | int err, ret = 0; | |
1010 | struct buffer_head *next; | |
1011 | ||
af5bc92d TT |
1012 | for (bh = head, block_start = 0; |
1013 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1014 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1015 | next = bh->b_this_page; |
1016 | block_end = block_start + blocksize; | |
1017 | if (block_end <= from || block_start >= to) { | |
1018 | if (partial && !buffer_uptodate(bh)) | |
1019 | *partial = 1; | |
1020 | continue; | |
1021 | } | |
1022 | err = (*fn)(handle, bh); | |
1023 | if (!ret) | |
1024 | ret = err; | |
1025 | } | |
1026 | return ret; | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * To preserve ordering, it is essential that the hole instantiation and | |
1031 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1032 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1033 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1034 | * prepare_write() is the right place. |
1035 | * | |
36ade451 JK |
1036 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1037 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1038 | * whole page. So we won't block on the journal in that case, which is good, | |
1039 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1040 | * |
617ba13b | 1041 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1042 | * quota file writes. If we were to commit the transaction while thus |
1043 | * reentered, there can be a deadlock - we would be holding a quota | |
1044 | * lock, and the commit would never complete if another thread had a | |
1045 | * transaction open and was blocking on the quota lock - a ranking | |
1046 | * violation. | |
1047 | * | |
dab291af | 1048 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1049 | * will _not_ run commit under these circumstances because handle->h_ref |
1050 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1051 | * write. | |
1052 | */ | |
f19d5870 TM |
1053 | int do_journal_get_write_access(handle_t *handle, |
1054 | struct buffer_head *bh) | |
ac27a0ec | 1055 | { |
56d35a4c JK |
1056 | int dirty = buffer_dirty(bh); |
1057 | int ret; | |
1058 | ||
ac27a0ec DK |
1059 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1060 | return 0; | |
56d35a4c | 1061 | /* |
ebdec241 | 1062 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1063 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1064 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1065 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1066 | * the bit before releasing a page lock and thus writeback cannot |
1067 | * ever write the buffer. | |
1068 | */ | |
1069 | if (dirty) | |
1070 | clear_buffer_dirty(bh); | |
5d601255 | 1071 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1072 | ret = ext4_journal_get_write_access(handle, bh); |
1073 | if (!ret && dirty) | |
1074 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1075 | return ret; | |
ac27a0ec DK |
1076 | } |
1077 | ||
2058f83a MH |
1078 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1079 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
1080 | get_block_t *get_block) | |
1081 | { | |
09cbfeaf | 1082 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1083 | unsigned to = from + len; |
1084 | struct inode *inode = page->mapping->host; | |
1085 | unsigned block_start, block_end; | |
1086 | sector_t block; | |
1087 | int err = 0; | |
1088 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1089 | unsigned bbits; | |
1090 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
1091 | bool decrypt = false; | |
1092 | ||
1093 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1094 | BUG_ON(from > PAGE_SIZE); |
1095 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1096 | BUG_ON(from > to); |
1097 | ||
1098 | if (!page_has_buffers(page)) | |
1099 | create_empty_buffers(page, blocksize, 0); | |
1100 | head = page_buffers(page); | |
1101 | bbits = ilog2(blocksize); | |
09cbfeaf | 1102 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1103 | |
1104 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1105 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1106 | block_end = block_start + blocksize; | |
1107 | if (block_end <= from || block_start >= to) { | |
1108 | if (PageUptodate(page)) { | |
1109 | if (!buffer_uptodate(bh)) | |
1110 | set_buffer_uptodate(bh); | |
1111 | } | |
1112 | continue; | |
1113 | } | |
1114 | if (buffer_new(bh)) | |
1115 | clear_buffer_new(bh); | |
1116 | if (!buffer_mapped(bh)) { | |
1117 | WARN_ON(bh->b_size != blocksize); | |
1118 | err = get_block(inode, block, bh, 1); | |
1119 | if (err) | |
1120 | break; | |
1121 | if (buffer_new(bh)) { | |
1122 | unmap_underlying_metadata(bh->b_bdev, | |
1123 | bh->b_blocknr); | |
1124 | if (PageUptodate(page)) { | |
1125 | clear_buffer_new(bh); | |
1126 | set_buffer_uptodate(bh); | |
1127 | mark_buffer_dirty(bh); | |
1128 | continue; | |
1129 | } | |
1130 | if (block_end > to || block_start < from) | |
1131 | zero_user_segments(page, to, block_end, | |
1132 | block_start, from); | |
1133 | continue; | |
1134 | } | |
1135 | } | |
1136 | if (PageUptodate(page)) { | |
1137 | if (!buffer_uptodate(bh)) | |
1138 | set_buffer_uptodate(bh); | |
1139 | continue; | |
1140 | } | |
1141 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1142 | !buffer_unwritten(bh) && | |
1143 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1144 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
2058f83a MH |
1145 | *wait_bh++ = bh; |
1146 | decrypt = ext4_encrypted_inode(inode) && | |
1147 | S_ISREG(inode->i_mode); | |
1148 | } | |
1149 | } | |
1150 | /* | |
1151 | * If we issued read requests, let them complete. | |
1152 | */ | |
1153 | while (wait_bh > wait) { | |
1154 | wait_on_buffer(*--wait_bh); | |
1155 | if (!buffer_uptodate(*wait_bh)) | |
1156 | err = -EIO; | |
1157 | } | |
1158 | if (unlikely(err)) | |
1159 | page_zero_new_buffers(page, from, to); | |
1160 | else if (decrypt) | |
a7550b30 | 1161 | err = fscrypt_decrypt_page(page); |
2058f83a MH |
1162 | return err; |
1163 | } | |
1164 | #endif | |
1165 | ||
bfc1af65 | 1166 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1167 | loff_t pos, unsigned len, unsigned flags, |
1168 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1169 | { |
af5bc92d | 1170 | struct inode *inode = mapping->host; |
1938a150 | 1171 | int ret, needed_blocks; |
ac27a0ec DK |
1172 | handle_t *handle; |
1173 | int retries = 0; | |
af5bc92d | 1174 | struct page *page; |
de9a55b8 | 1175 | pgoff_t index; |
af5bc92d | 1176 | unsigned from, to; |
bfc1af65 | 1177 | |
9bffad1e | 1178 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1179 | /* |
1180 | * Reserve one block more for addition to orphan list in case | |
1181 | * we allocate blocks but write fails for some reason | |
1182 | */ | |
1183 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1184 | index = pos >> PAGE_SHIFT; |
1185 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1186 | to = from + len; |
ac27a0ec | 1187 | |
f19d5870 TM |
1188 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1189 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1190 | flags, pagep); | |
1191 | if (ret < 0) | |
47564bfb TT |
1192 | return ret; |
1193 | if (ret == 1) | |
1194 | return 0; | |
f19d5870 TM |
1195 | } |
1196 | ||
47564bfb TT |
1197 | /* |
1198 | * grab_cache_page_write_begin() can take a long time if the | |
1199 | * system is thrashing due to memory pressure, or if the page | |
1200 | * is being written back. So grab it first before we start | |
1201 | * the transaction handle. This also allows us to allocate | |
1202 | * the page (if needed) without using GFP_NOFS. | |
1203 | */ | |
1204 | retry_grab: | |
1205 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1206 | if (!page) | |
1207 | return -ENOMEM; | |
1208 | unlock_page(page); | |
1209 | ||
1210 | retry_journal: | |
9924a92a | 1211 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1212 | if (IS_ERR(handle)) { |
09cbfeaf | 1213 | put_page(page); |
47564bfb | 1214 | return PTR_ERR(handle); |
7479d2b9 | 1215 | } |
ac27a0ec | 1216 | |
47564bfb TT |
1217 | lock_page(page); |
1218 | if (page->mapping != mapping) { | |
1219 | /* The page got truncated from under us */ | |
1220 | unlock_page(page); | |
09cbfeaf | 1221 | put_page(page); |
cf108bca | 1222 | ext4_journal_stop(handle); |
47564bfb | 1223 | goto retry_grab; |
cf108bca | 1224 | } |
7afe5aa5 DM |
1225 | /* In case writeback began while the page was unlocked */ |
1226 | wait_for_stable_page(page); | |
cf108bca | 1227 | |
2058f83a MH |
1228 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1229 | if (ext4_should_dioread_nolock(inode)) | |
1230 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1231 | ext4_get_block_unwritten); |
2058f83a MH |
1232 | else |
1233 | ret = ext4_block_write_begin(page, pos, len, | |
1234 | ext4_get_block); | |
1235 | #else | |
744692dc | 1236 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1237 | ret = __block_write_begin(page, pos, len, |
1238 | ext4_get_block_unwritten); | |
744692dc | 1239 | else |
6e1db88d | 1240 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1241 | #endif |
bfc1af65 | 1242 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1243 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1244 | from, to, NULL, | |
1245 | do_journal_get_write_access); | |
ac27a0ec | 1246 | } |
bfc1af65 NP |
1247 | |
1248 | if (ret) { | |
af5bc92d | 1249 | unlock_page(page); |
ae4d5372 | 1250 | /* |
6e1db88d | 1251 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1252 | * outside i_size. Trim these off again. Don't need |
1253 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1254 | * |
1255 | * Add inode to orphan list in case we crash before | |
1256 | * truncate finishes | |
ae4d5372 | 1257 | */ |
ffacfa7a | 1258 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1259 | ext4_orphan_add(handle, inode); |
1260 | ||
1261 | ext4_journal_stop(handle); | |
1262 | if (pos + len > inode->i_size) { | |
b9a4207d | 1263 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1264 | /* |
ffacfa7a | 1265 | * If truncate failed early the inode might |
1938a150 AK |
1266 | * still be on the orphan list; we need to |
1267 | * make sure the inode is removed from the | |
1268 | * orphan list in that case. | |
1269 | */ | |
1270 | if (inode->i_nlink) | |
1271 | ext4_orphan_del(NULL, inode); | |
1272 | } | |
bfc1af65 | 1273 | |
47564bfb TT |
1274 | if (ret == -ENOSPC && |
1275 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1276 | goto retry_journal; | |
09cbfeaf | 1277 | put_page(page); |
47564bfb TT |
1278 | return ret; |
1279 | } | |
1280 | *pagep = page; | |
ac27a0ec DK |
1281 | return ret; |
1282 | } | |
1283 | ||
bfc1af65 NP |
1284 | /* For write_end() in data=journal mode */ |
1285 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1286 | { |
13fca323 | 1287 | int ret; |
ac27a0ec DK |
1288 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1289 | return 0; | |
1290 | set_buffer_uptodate(bh); | |
13fca323 TT |
1291 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1292 | clear_buffer_meta(bh); | |
1293 | clear_buffer_prio(bh); | |
1294 | return ret; | |
ac27a0ec DK |
1295 | } |
1296 | ||
eed4333f ZL |
1297 | /* |
1298 | * We need to pick up the new inode size which generic_commit_write gave us | |
1299 | * `file' can be NULL - eg, when called from page_symlink(). | |
1300 | * | |
1301 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1302 | * buffers are managed internally. | |
1303 | */ | |
1304 | static int ext4_write_end(struct file *file, | |
1305 | struct address_space *mapping, | |
1306 | loff_t pos, unsigned len, unsigned copied, | |
1307 | struct page *page, void *fsdata) | |
f8514083 | 1308 | { |
f8514083 | 1309 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1310 | struct inode *inode = mapping->host; |
0572639f | 1311 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1312 | int ret = 0, ret2; |
1313 | int i_size_changed = 0; | |
1314 | ||
1315 | trace_ext4_write_end(inode, pos, len, copied); | |
42c832de TT |
1316 | if (ext4_has_inline_data(inode)) { |
1317 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1318 | copied, page); | |
1319 | if (ret < 0) | |
1320 | goto errout; | |
1321 | copied = ret; | |
1322 | } else | |
f19d5870 TM |
1323 | copied = block_write_end(file, mapping, pos, |
1324 | len, copied, page, fsdata); | |
f8514083 | 1325 | /* |
4631dbf6 | 1326 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1327 | * page writeout could otherwise come in and zero beyond i_size. |
1328 | */ | |
4631dbf6 | 1329 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 | 1330 | unlock_page(page); |
09cbfeaf | 1331 | put_page(page); |
f8514083 | 1332 | |
0572639f XW |
1333 | if (old_size < pos) |
1334 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1335 | /* |
1336 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1337 | * makes the holding time of page lock longer. Second, it forces lock | |
1338 | * ordering of page lock and transaction start for journaling | |
1339 | * filesystems. | |
1340 | */ | |
1341 | if (i_size_changed) | |
1342 | ext4_mark_inode_dirty(handle, inode); | |
1343 | ||
ffacfa7a | 1344 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1345 | /* if we have allocated more blocks and copied |
1346 | * less. We will have blocks allocated outside | |
1347 | * inode->i_size. So truncate them | |
1348 | */ | |
1349 | ext4_orphan_add(handle, inode); | |
74d553aa | 1350 | errout: |
617ba13b | 1351 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1352 | if (!ret) |
1353 | ret = ret2; | |
bfc1af65 | 1354 | |
f8514083 | 1355 | if (pos + len > inode->i_size) { |
b9a4207d | 1356 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1357 | /* |
ffacfa7a | 1358 | * If truncate failed early the inode might still be |
f8514083 AK |
1359 | * on the orphan list; we need to make sure the inode |
1360 | * is removed from the orphan list in that case. | |
1361 | */ | |
1362 | if (inode->i_nlink) | |
1363 | ext4_orphan_del(NULL, inode); | |
1364 | } | |
1365 | ||
bfc1af65 | 1366 | return ret ? ret : copied; |
ac27a0ec DK |
1367 | } |
1368 | ||
b90197b6 TT |
1369 | /* |
1370 | * This is a private version of page_zero_new_buffers() which doesn't | |
1371 | * set the buffer to be dirty, since in data=journalled mode we need | |
1372 | * to call ext4_handle_dirty_metadata() instead. | |
1373 | */ | |
1374 | static void zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1375 | { | |
1376 | unsigned int block_start = 0, block_end; | |
1377 | struct buffer_head *head, *bh; | |
1378 | ||
1379 | bh = head = page_buffers(page); | |
1380 | do { | |
1381 | block_end = block_start + bh->b_size; | |
1382 | if (buffer_new(bh)) { | |
1383 | if (block_end > from && block_start < to) { | |
1384 | if (!PageUptodate(page)) { | |
1385 | unsigned start, size; | |
1386 | ||
1387 | start = max(from, block_start); | |
1388 | size = min(to, block_end) - start; | |
1389 | ||
1390 | zero_user(page, start, size); | |
1391 | set_buffer_uptodate(bh); | |
1392 | } | |
1393 | clear_buffer_new(bh); | |
1394 | } | |
1395 | } | |
1396 | block_start = block_end; | |
1397 | bh = bh->b_this_page; | |
1398 | } while (bh != head); | |
1399 | } | |
1400 | ||
bfc1af65 | 1401 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1402 | struct address_space *mapping, |
1403 | loff_t pos, unsigned len, unsigned copied, | |
1404 | struct page *page, void *fsdata) | |
ac27a0ec | 1405 | { |
617ba13b | 1406 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1407 | struct inode *inode = mapping->host; |
0572639f | 1408 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1409 | int ret = 0, ret2; |
1410 | int partial = 0; | |
bfc1af65 | 1411 | unsigned from, to; |
4631dbf6 | 1412 | int size_changed = 0; |
ac27a0ec | 1413 | |
9bffad1e | 1414 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1415 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1416 | to = from + len; |
1417 | ||
441c8508 CW |
1418 | BUG_ON(!ext4_handle_valid(handle)); |
1419 | ||
3fdcfb66 TM |
1420 | if (ext4_has_inline_data(inode)) |
1421 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1422 | copied, page); | |
1423 | else { | |
1424 | if (copied < len) { | |
1425 | if (!PageUptodate(page)) | |
1426 | copied = 0; | |
b90197b6 | 1427 | zero_new_buffers(page, from+copied, to); |
3fdcfb66 | 1428 | } |
ac27a0ec | 1429 | |
3fdcfb66 TM |
1430 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1431 | to, &partial, write_end_fn); | |
1432 | if (!partial) | |
1433 | SetPageUptodate(page); | |
1434 | } | |
4631dbf6 | 1435 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1436 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1437 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1438 | unlock_page(page); |
09cbfeaf | 1439 | put_page(page); |
4631dbf6 | 1440 | |
0572639f XW |
1441 | if (old_size < pos) |
1442 | pagecache_isize_extended(inode, old_size, pos); | |
1443 | ||
4631dbf6 | 1444 | if (size_changed) { |
617ba13b | 1445 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1446 | if (!ret) |
1447 | ret = ret2; | |
1448 | } | |
bfc1af65 | 1449 | |
ffacfa7a | 1450 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1451 | /* if we have allocated more blocks and copied |
1452 | * less. We will have blocks allocated outside | |
1453 | * inode->i_size. So truncate them | |
1454 | */ | |
1455 | ext4_orphan_add(handle, inode); | |
1456 | ||
617ba13b | 1457 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1458 | if (!ret) |
1459 | ret = ret2; | |
f8514083 | 1460 | if (pos + len > inode->i_size) { |
b9a4207d | 1461 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1462 | /* |
ffacfa7a | 1463 | * If truncate failed early the inode might still be |
f8514083 AK |
1464 | * on the orphan list; we need to make sure the inode |
1465 | * is removed from the orphan list in that case. | |
1466 | */ | |
1467 | if (inode->i_nlink) | |
1468 | ext4_orphan_del(NULL, inode); | |
1469 | } | |
bfc1af65 NP |
1470 | |
1471 | return ret ? ret : copied; | |
ac27a0ec | 1472 | } |
d2a17637 | 1473 | |
9d0be502 | 1474 | /* |
c27e43a1 | 1475 | * Reserve space for a single cluster |
9d0be502 | 1476 | */ |
c27e43a1 | 1477 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1478 | { |
60e58e0f | 1479 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1480 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1481 | int ret; |
03179fe9 TT |
1482 | |
1483 | /* | |
1484 | * We will charge metadata quota at writeout time; this saves | |
1485 | * us from metadata over-estimation, though we may go over by | |
1486 | * a small amount in the end. Here we just reserve for data. | |
1487 | */ | |
1488 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1489 | if (ret) | |
1490 | return ret; | |
d2a17637 | 1491 | |
0637c6f4 | 1492 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1493 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1494 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1495 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1496 | return -ENOSPC; |
1497 | } | |
9d0be502 | 1498 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1499 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1500 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1501 | |
d2a17637 MC |
1502 | return 0; /* success */ |
1503 | } | |
1504 | ||
12219aea | 1505 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1506 | { |
1507 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1508 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1509 | |
cd213226 MC |
1510 | if (!to_free) |
1511 | return; /* Nothing to release, exit */ | |
1512 | ||
d2a17637 | 1513 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1514 | |
5a58ec87 | 1515 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1516 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1517 | /* |
0637c6f4 TT |
1518 | * if there aren't enough reserved blocks, then the |
1519 | * counter is messed up somewhere. Since this | |
1520 | * function is called from invalidate page, it's | |
1521 | * harmless to return without any action. | |
cd213226 | 1522 | */ |
8de5c325 | 1523 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1524 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1525 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1526 | ei->i_reserved_data_blocks); |
1527 | WARN_ON(1); | |
1528 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1529 | } |
0637c6f4 | 1530 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1531 | |
72b8ab9d | 1532 | /* update fs dirty data blocks counter */ |
57042651 | 1533 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1534 | |
d2a17637 | 1535 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1536 | |
7b415bf6 | 1537 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1538 | } |
1539 | ||
1540 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1541 | unsigned int offset, |
1542 | unsigned int length) | |
d2a17637 | 1543 | { |
9705acd6 | 1544 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1545 | struct buffer_head *head, *bh; |
1546 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1547 | struct inode *inode = page->mapping->host; |
1548 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1549 | unsigned int stop = offset + length; |
7b415bf6 | 1550 | int num_clusters; |
51865fda | 1551 | ext4_fsblk_t lblk; |
d2a17637 | 1552 | |
09cbfeaf | 1553 | BUG_ON(stop > PAGE_SIZE || stop < length); |
ca99fdd2 | 1554 | |
d2a17637 MC |
1555 | head = page_buffers(page); |
1556 | bh = head; | |
1557 | do { | |
1558 | unsigned int next_off = curr_off + bh->b_size; | |
1559 | ||
ca99fdd2 LC |
1560 | if (next_off > stop) |
1561 | break; | |
1562 | ||
d2a17637 MC |
1563 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1564 | to_release++; | |
9705acd6 | 1565 | contiguous_blks++; |
d2a17637 | 1566 | clear_buffer_delay(bh); |
9705acd6 LC |
1567 | } else if (contiguous_blks) { |
1568 | lblk = page->index << | |
09cbfeaf | 1569 | (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1570 | lblk += (curr_off >> inode->i_blkbits) - |
1571 | contiguous_blks; | |
1572 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1573 | contiguous_blks = 0; | |
d2a17637 MC |
1574 | } |
1575 | curr_off = next_off; | |
1576 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1577 | |
9705acd6 | 1578 | if (contiguous_blks) { |
09cbfeaf | 1579 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1580 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1581 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1582 | } |
1583 | ||
7b415bf6 AK |
1584 | /* If we have released all the blocks belonging to a cluster, then we |
1585 | * need to release the reserved space for that cluster. */ | |
1586 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1587 | while (num_clusters > 0) { | |
09cbfeaf | 1588 | lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
7b415bf6 AK |
1589 | ((num_clusters - 1) << sbi->s_cluster_bits); |
1590 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1591 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1592 | ext4_da_release_space(inode, 1); |
1593 | ||
1594 | num_clusters--; | |
1595 | } | |
d2a17637 | 1596 | } |
ac27a0ec | 1597 | |
64769240 AT |
1598 | /* |
1599 | * Delayed allocation stuff | |
1600 | */ | |
1601 | ||
4e7ea81d JK |
1602 | struct mpage_da_data { |
1603 | struct inode *inode; | |
1604 | struct writeback_control *wbc; | |
6b523df4 | 1605 | |
4e7ea81d JK |
1606 | pgoff_t first_page; /* The first page to write */ |
1607 | pgoff_t next_page; /* Current page to examine */ | |
1608 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1609 | /* |
4e7ea81d JK |
1610 | * Extent to map - this can be after first_page because that can be |
1611 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1612 | * is delalloc or unwritten. | |
791b7f08 | 1613 | */ |
4e7ea81d JK |
1614 | struct ext4_map_blocks map; |
1615 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1616 | }; | |
64769240 | 1617 | |
4e7ea81d JK |
1618 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1619 | bool invalidate) | |
c4a0c46e AK |
1620 | { |
1621 | int nr_pages, i; | |
1622 | pgoff_t index, end; | |
1623 | struct pagevec pvec; | |
1624 | struct inode *inode = mpd->inode; | |
1625 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1626 | |
1627 | /* This is necessary when next_page == 0. */ | |
1628 | if (mpd->first_page >= mpd->next_page) | |
1629 | return; | |
c4a0c46e | 1630 | |
c7f5938a CW |
1631 | index = mpd->first_page; |
1632 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1633 | if (invalidate) { |
1634 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1635 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1636 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1637 | ext4_es_remove_extent(inode, start, last - start + 1); |
1638 | } | |
51865fda | 1639 | |
66bea92c | 1640 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1641 | while (index <= end) { |
1642 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1643 | if (nr_pages == 0) | |
1644 | break; | |
1645 | for (i = 0; i < nr_pages; i++) { | |
1646 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1647 | if (page->index > end) |
c4a0c46e | 1648 | break; |
c4a0c46e AK |
1649 | BUG_ON(!PageLocked(page)); |
1650 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1651 | if (invalidate) { |
09cbfeaf | 1652 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1653 | ClearPageUptodate(page); |
1654 | } | |
c4a0c46e AK |
1655 | unlock_page(page); |
1656 | } | |
9b1d0998 JK |
1657 | index = pvec.pages[nr_pages - 1]->index + 1; |
1658 | pagevec_release(&pvec); | |
c4a0c46e | 1659 | } |
c4a0c46e AK |
1660 | } |
1661 | ||
df22291f AK |
1662 | static void ext4_print_free_blocks(struct inode *inode) |
1663 | { | |
1664 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1665 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1666 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1667 | |
1668 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1669 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1670 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1671 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1672 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1673 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1674 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1675 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1676 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1677 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1678 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1679 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1680 | ei->i_reserved_data_blocks); |
df22291f AK |
1681 | return; |
1682 | } | |
1683 | ||
c364b22c | 1684 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1685 | { |
c364b22c | 1686 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1687 | } |
1688 | ||
5356f261 AK |
1689 | /* |
1690 | * This function is grabs code from the very beginning of | |
1691 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1692 | * time. This function looks up the requested blocks and sets the | |
1693 | * buffer delay bit under the protection of i_data_sem. | |
1694 | */ | |
1695 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1696 | struct ext4_map_blocks *map, | |
1697 | struct buffer_head *bh) | |
1698 | { | |
d100eef2 | 1699 | struct extent_status es; |
5356f261 AK |
1700 | int retval; |
1701 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1702 | #ifdef ES_AGGRESSIVE_TEST |
1703 | struct ext4_map_blocks orig_map; | |
1704 | ||
1705 | memcpy(&orig_map, map, sizeof(*map)); | |
1706 | #endif | |
5356f261 AK |
1707 | |
1708 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1709 | invalid_block = ~0; | |
1710 | ||
1711 | map->m_flags = 0; | |
1712 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1713 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1714 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1715 | |
1716 | /* Lookup extent status tree firstly */ | |
1717 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1718 | if (ext4_es_is_hole(&es)) { |
1719 | retval = 0; | |
c8b459f4 | 1720 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1721 | goto add_delayed; |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * Delayed extent could be allocated by fallocate. | |
1726 | * So we need to check it. | |
1727 | */ | |
1728 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1729 | map_bh(bh, inode->i_sb, invalid_block); | |
1730 | set_buffer_new(bh); | |
1731 | set_buffer_delay(bh); | |
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1736 | retval = es.es_len - (iblock - es.es_lblk); | |
1737 | if (retval > map->m_len) | |
1738 | retval = map->m_len; | |
1739 | map->m_len = retval; | |
1740 | if (ext4_es_is_written(&es)) | |
1741 | map->m_flags |= EXT4_MAP_MAPPED; | |
1742 | else if (ext4_es_is_unwritten(&es)) | |
1743 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1744 | else | |
1745 | BUG_ON(1); | |
1746 | ||
921f266b DM |
1747 | #ifdef ES_AGGRESSIVE_TEST |
1748 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1749 | #endif | |
d100eef2 ZL |
1750 | return retval; |
1751 | } | |
1752 | ||
5356f261 AK |
1753 | /* |
1754 | * Try to see if we can get the block without requesting a new | |
1755 | * file system block. | |
1756 | */ | |
c8b459f4 | 1757 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1758 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1759 | retval = 0; |
cbd7584e | 1760 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1761 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1762 | else |
2f8e0a7c | 1763 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1764 | |
d100eef2 | 1765 | add_delayed: |
5356f261 | 1766 | if (retval == 0) { |
f7fec032 | 1767 | int ret; |
5356f261 AK |
1768 | /* |
1769 | * XXX: __block_prepare_write() unmaps passed block, | |
1770 | * is it OK? | |
1771 | */ | |
386ad67c LC |
1772 | /* |
1773 | * If the block was allocated from previously allocated cluster, | |
1774 | * then we don't need to reserve it again. However we still need | |
1775 | * to reserve metadata for every block we're going to write. | |
1776 | */ | |
c27e43a1 | 1777 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1778 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1779 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1780 | if (ret) { |
5356f261 | 1781 | /* not enough space to reserve */ |
f7fec032 | 1782 | retval = ret; |
5356f261 | 1783 | goto out_unlock; |
f7fec032 | 1784 | } |
5356f261 AK |
1785 | } |
1786 | ||
f7fec032 ZL |
1787 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1788 | ~0, EXTENT_STATUS_DELAYED); | |
1789 | if (ret) { | |
1790 | retval = ret; | |
51865fda | 1791 | goto out_unlock; |
f7fec032 | 1792 | } |
51865fda | 1793 | |
5356f261 AK |
1794 | map_bh(bh, inode->i_sb, invalid_block); |
1795 | set_buffer_new(bh); | |
1796 | set_buffer_delay(bh); | |
f7fec032 ZL |
1797 | } else if (retval > 0) { |
1798 | int ret; | |
3be78c73 | 1799 | unsigned int status; |
f7fec032 | 1800 | |
44fb851d ZL |
1801 | if (unlikely(retval != map->m_len)) { |
1802 | ext4_warning(inode->i_sb, | |
1803 | "ES len assertion failed for inode " | |
1804 | "%lu: retval %d != map->m_len %d", | |
1805 | inode->i_ino, retval, map->m_len); | |
1806 | WARN_ON(1); | |
921f266b | 1807 | } |
921f266b | 1808 | |
f7fec032 ZL |
1809 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1810 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1811 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1812 | map->m_pblk, status); | |
1813 | if (ret != 0) | |
1814 | retval = ret; | |
5356f261 AK |
1815 | } |
1816 | ||
1817 | out_unlock: | |
1818 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1819 | ||
1820 | return retval; | |
1821 | } | |
1822 | ||
64769240 | 1823 | /* |
d91bd2c1 | 1824 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1825 | * ext4_da_write_begin(). It will either return mapped block or |
1826 | * reserve space for a single block. | |
29fa89d0 AK |
1827 | * |
1828 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1829 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1830 | * | |
1831 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1832 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1833 | * initialized properly. | |
64769240 | 1834 | */ |
9c3569b5 TM |
1835 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1836 | struct buffer_head *bh, int create) | |
64769240 | 1837 | { |
2ed88685 | 1838 | struct ext4_map_blocks map; |
64769240 AT |
1839 | int ret = 0; |
1840 | ||
1841 | BUG_ON(create == 0); | |
2ed88685 TT |
1842 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1843 | ||
1844 | map.m_lblk = iblock; | |
1845 | map.m_len = 1; | |
64769240 AT |
1846 | |
1847 | /* | |
1848 | * first, we need to know whether the block is allocated already | |
1849 | * preallocated blocks are unmapped but should treated | |
1850 | * the same as allocated blocks. | |
1851 | */ | |
5356f261 AK |
1852 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1853 | if (ret <= 0) | |
2ed88685 | 1854 | return ret; |
64769240 | 1855 | |
2ed88685 | 1856 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1857 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1858 | |
1859 | if (buffer_unwritten(bh)) { | |
1860 | /* A delayed write to unwritten bh should be marked | |
1861 | * new and mapped. Mapped ensures that we don't do | |
1862 | * get_block multiple times when we write to the same | |
1863 | * offset and new ensures that we do proper zero out | |
1864 | * for partial write. | |
1865 | */ | |
1866 | set_buffer_new(bh); | |
c8205636 | 1867 | set_buffer_mapped(bh); |
2ed88685 TT |
1868 | } |
1869 | return 0; | |
64769240 | 1870 | } |
61628a3f | 1871 | |
62e086be AK |
1872 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1873 | { | |
1874 | get_bh(bh); | |
1875 | return 0; | |
1876 | } | |
1877 | ||
1878 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1879 | { | |
1880 | put_bh(bh); | |
1881 | return 0; | |
1882 | } | |
1883 | ||
1884 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1885 | unsigned int len) |
1886 | { | |
1887 | struct address_space *mapping = page->mapping; | |
1888 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1889 | struct buffer_head *page_bufs = NULL; |
62e086be | 1890 | handle_t *handle = NULL; |
3fdcfb66 TM |
1891 | int ret = 0, err = 0; |
1892 | int inline_data = ext4_has_inline_data(inode); | |
1893 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1894 | |
cb20d518 | 1895 | ClearPageChecked(page); |
3fdcfb66 TM |
1896 | |
1897 | if (inline_data) { | |
1898 | BUG_ON(page->index != 0); | |
1899 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1900 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1901 | if (inode_bh == NULL) | |
1902 | goto out; | |
1903 | } else { | |
1904 | page_bufs = page_buffers(page); | |
1905 | if (!page_bufs) { | |
1906 | BUG(); | |
1907 | goto out; | |
1908 | } | |
1909 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1910 | NULL, bget_one); | |
1911 | } | |
bdf96838 TT |
1912 | /* |
1913 | * We need to release the page lock before we start the | |
1914 | * journal, so grab a reference so the page won't disappear | |
1915 | * out from under us. | |
1916 | */ | |
1917 | get_page(page); | |
62e086be AK |
1918 | unlock_page(page); |
1919 | ||
9924a92a TT |
1920 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1921 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1922 | if (IS_ERR(handle)) { |
1923 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1924 | put_page(page); |
1925 | goto out_no_pagelock; | |
62e086be | 1926 | } |
441c8508 CW |
1927 | BUG_ON(!ext4_handle_valid(handle)); |
1928 | ||
bdf96838 TT |
1929 | lock_page(page); |
1930 | put_page(page); | |
1931 | if (page->mapping != mapping) { | |
1932 | /* The page got truncated from under us */ | |
1933 | ext4_journal_stop(handle); | |
1934 | ret = 0; | |
1935 | goto out; | |
1936 | } | |
1937 | ||
3fdcfb66 | 1938 | if (inline_data) { |
5d601255 | 1939 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1940 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1941 | |
3fdcfb66 TM |
1942 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1943 | ||
1944 | } else { | |
1945 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1946 | do_journal_get_write_access); | |
1947 | ||
1948 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1949 | write_end_fn); | |
1950 | } | |
62e086be AK |
1951 | if (ret == 0) |
1952 | ret = err; | |
2d859db3 | 1953 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1954 | err = ext4_journal_stop(handle); |
1955 | if (!ret) | |
1956 | ret = err; | |
1957 | ||
3fdcfb66 | 1958 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1959 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1960 | NULL, bput_one); |
19f5fb7a | 1961 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1962 | out: |
bdf96838 TT |
1963 | unlock_page(page); |
1964 | out_no_pagelock: | |
3fdcfb66 | 1965 | brelse(inode_bh); |
62e086be AK |
1966 | return ret; |
1967 | } | |
1968 | ||
61628a3f | 1969 | /* |
43ce1d23 AK |
1970 | * Note that we don't need to start a transaction unless we're journaling data |
1971 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1972 | * need to file the inode to the transaction's list in ordered mode because if | |
1973 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1974 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1975 | * transaction the data will hit the disk. In case we are journaling data, we |
1976 | * cannot start transaction directly because transaction start ranks above page | |
1977 | * lock so we have to do some magic. | |
1978 | * | |
b920c755 | 1979 | * This function can get called via... |
20970ba6 | 1980 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1981 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1982 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1983 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1984 | * |
1985 | * We don't do any block allocation in this function. If we have page with | |
1986 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1987 | * is important for mmaped based write. So if we do with blocksize 1K | |
1988 | * truncate(f, 1024); | |
1989 | * a = mmap(f, 0, 4096); | |
1990 | * a[0] = 'a'; | |
1991 | * truncate(f, 4096); | |
1992 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1993 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1994 | * do_wp_page). So writepage should write the first block. If we modify |
1995 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1996 | * page_mkwrite callback will do the block allocation and mark the | |
1997 | * buffer_heads mapped. | |
1998 | * | |
1999 | * We redirty the page if we have any buffer_heads that is either delay or | |
2000 | * unwritten in the page. | |
2001 | * | |
2002 | * We can get recursively called as show below. | |
2003 | * | |
2004 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2005 | * ext4_writepage() | |
2006 | * | |
2007 | * But since we don't do any block allocation we should not deadlock. | |
2008 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2009 | */ |
43ce1d23 | 2010 | static int ext4_writepage(struct page *page, |
62e086be | 2011 | struct writeback_control *wbc) |
64769240 | 2012 | { |
f8bec370 | 2013 | int ret = 0; |
61628a3f | 2014 | loff_t size; |
498e5f24 | 2015 | unsigned int len; |
744692dc | 2016 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2017 | struct inode *inode = page->mapping->host; |
36ade451 | 2018 | struct ext4_io_submit io_submit; |
1c8349a1 | 2019 | bool keep_towrite = false; |
61628a3f | 2020 | |
a9c667f8 | 2021 | trace_ext4_writepage(page); |
f0e6c985 | 2022 | size = i_size_read(inode); |
09cbfeaf KS |
2023 | if (page->index == size >> PAGE_SHIFT) |
2024 | len = size & ~PAGE_MASK; | |
f0e6c985 | 2025 | else |
09cbfeaf | 2026 | len = PAGE_SIZE; |
64769240 | 2027 | |
a42afc5f | 2028 | page_bufs = page_buffers(page); |
a42afc5f | 2029 | /* |
fe386132 JK |
2030 | * We cannot do block allocation or other extent handling in this |
2031 | * function. If there are buffers needing that, we have to redirty | |
2032 | * the page. But we may reach here when we do a journal commit via | |
2033 | * journal_submit_inode_data_buffers() and in that case we must write | |
2034 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2035 | * |
2036 | * Also, if there is only one buffer per page (the fs block | |
2037 | * size == the page size), if one buffer needs block | |
2038 | * allocation or needs to modify the extent tree to clear the | |
2039 | * unwritten flag, we know that the page can't be written at | |
2040 | * all, so we might as well refuse the write immediately. | |
2041 | * Unfortunately if the block size != page size, we can't as | |
2042 | * easily detect this case using ext4_walk_page_buffers(), but | |
2043 | * for the extremely common case, this is an optimization that | |
2044 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2045 | */ |
f19d5870 TM |
2046 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2047 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2048 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2049 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2050 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2051 | /* |
2052 | * For memory cleaning there's no point in writing only | |
2053 | * some buffers. So just bail out. Warn if we came here | |
2054 | * from direct reclaim. | |
2055 | */ | |
2056 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2057 | == PF_MEMALLOC); | |
f0e6c985 AK |
2058 | unlock_page(page); |
2059 | return 0; | |
2060 | } | |
1c8349a1 | 2061 | keep_towrite = true; |
a42afc5f | 2062 | } |
64769240 | 2063 | |
cb20d518 | 2064 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2065 | /* |
2066 | * It's mmapped pagecache. Add buffers and journal it. There | |
2067 | * doesn't seem much point in redirtying the page here. | |
2068 | */ | |
3f0ca309 | 2069 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2070 | |
97a851ed JK |
2071 | ext4_io_submit_init(&io_submit, wbc); |
2072 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2073 | if (!io_submit.io_end) { | |
2074 | redirty_page_for_writepage(wbc, page); | |
2075 | unlock_page(page); | |
2076 | return -ENOMEM; | |
2077 | } | |
1c8349a1 | 2078 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2079 | ext4_io_submit(&io_submit); |
97a851ed JK |
2080 | /* Drop io_end reference we got from init */ |
2081 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2082 | return ret; |
2083 | } | |
2084 | ||
5f1132b2 JK |
2085 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2086 | { | |
2087 | int len; | |
2088 | loff_t size = i_size_read(mpd->inode); | |
2089 | int err; | |
2090 | ||
2091 | BUG_ON(page->index != mpd->first_page); | |
09cbfeaf KS |
2092 | if (page->index == size >> PAGE_SHIFT) |
2093 | len = size & ~PAGE_MASK; | |
5f1132b2 | 2094 | else |
09cbfeaf | 2095 | len = PAGE_SIZE; |
5f1132b2 | 2096 | clear_page_dirty_for_io(page); |
1c8349a1 | 2097 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2098 | if (!err) |
2099 | mpd->wbc->nr_to_write--; | |
2100 | mpd->first_page++; | |
2101 | ||
2102 | return err; | |
2103 | } | |
2104 | ||
4e7ea81d JK |
2105 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2106 | ||
61628a3f | 2107 | /* |
fffb2739 JK |
2108 | * mballoc gives us at most this number of blocks... |
2109 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2110 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2111 | */ |
fffb2739 | 2112 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2113 | |
4e7ea81d JK |
2114 | /* |
2115 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2116 | * | |
2117 | * @mpd - extent of blocks | |
2118 | * @lblk - logical number of the block in the file | |
09930042 | 2119 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2120 | * |
09930042 JK |
2121 | * The function is used to collect contig. blocks in the same state. If the |
2122 | * buffer doesn't require mapping for writeback and we haven't started the | |
2123 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2124 | * caller can write the buffer right away. Otherwise the function returns true | |
2125 | * if the block has been added to the extent, false if the block couldn't be | |
2126 | * added. | |
4e7ea81d | 2127 | */ |
09930042 JK |
2128 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2129 | struct buffer_head *bh) | |
4e7ea81d JK |
2130 | { |
2131 | struct ext4_map_blocks *map = &mpd->map; | |
2132 | ||
09930042 JK |
2133 | /* Buffer that doesn't need mapping for writeback? */ |
2134 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2135 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2136 | /* So far no extent to map => we write the buffer right away */ | |
2137 | if (map->m_len == 0) | |
2138 | return true; | |
2139 | return false; | |
2140 | } | |
4e7ea81d JK |
2141 | |
2142 | /* First block in the extent? */ | |
2143 | if (map->m_len == 0) { | |
2144 | map->m_lblk = lblk; | |
2145 | map->m_len = 1; | |
09930042 JK |
2146 | map->m_flags = bh->b_state & BH_FLAGS; |
2147 | return true; | |
4e7ea81d JK |
2148 | } |
2149 | ||
09930042 JK |
2150 | /* Don't go larger than mballoc is willing to allocate */ |
2151 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2152 | return false; | |
2153 | ||
4e7ea81d JK |
2154 | /* Can we merge the block to our big extent? */ |
2155 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2156 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2157 | map->m_len++; |
09930042 | 2158 | return true; |
4e7ea81d | 2159 | } |
09930042 | 2160 | return false; |
4e7ea81d JK |
2161 | } |
2162 | ||
5f1132b2 JK |
2163 | /* |
2164 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2165 | * | |
2166 | * @mpd - extent of blocks for mapping | |
2167 | * @head - the first buffer in the page | |
2168 | * @bh - buffer we should start processing from | |
2169 | * @lblk - logical number of the block in the file corresponding to @bh | |
2170 | * | |
2171 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2172 | * the page for IO if all buffers in this page were mapped and there's no | |
2173 | * accumulated extent of buffers to map or add buffers in the page to the | |
2174 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2175 | * by processing the next page, 0 if it should stop adding buffers to the | |
2176 | * extent to map because we cannot extend it anymore. It can also return value | |
2177 | * < 0 in case of error during IO submission. | |
2178 | */ | |
2179 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2180 | struct buffer_head *head, | |
2181 | struct buffer_head *bh, | |
2182 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2183 | { |
2184 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2185 | int err; |
4e7ea81d JK |
2186 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
2187 | >> inode->i_blkbits; | |
2188 | ||
2189 | do { | |
2190 | BUG_ON(buffer_locked(bh)); | |
2191 | ||
09930042 | 2192 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2193 | /* Found extent to map? */ |
2194 | if (mpd->map.m_len) | |
5f1132b2 | 2195 | return 0; |
09930042 | 2196 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2197 | break; |
4e7ea81d | 2198 | } |
4e7ea81d | 2199 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2200 | /* So far everything mapped? Submit the page for IO. */ |
2201 | if (mpd->map.m_len == 0) { | |
2202 | err = mpage_submit_page(mpd, head->b_page); | |
2203 | if (err < 0) | |
2204 | return err; | |
2205 | } | |
2206 | return lblk < blocks; | |
4e7ea81d JK |
2207 | } |
2208 | ||
2209 | /* | |
2210 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2211 | * submit fully mapped pages for IO | |
2212 | * | |
2213 | * @mpd - description of extent to map, on return next extent to map | |
2214 | * | |
2215 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2216 | * to be already locked) and update buffer state according to new extent state. | |
2217 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2218 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2219 | * and do extent conversion after IO is finished. If the last page is not fully |
2220 | * mapped, we update @map to the next extent in the last page that needs | |
2221 | * mapping. Otherwise we submit the page for IO. | |
2222 | */ | |
2223 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2224 | { | |
2225 | struct pagevec pvec; | |
2226 | int nr_pages, i; | |
2227 | struct inode *inode = mpd->inode; | |
2228 | struct buffer_head *head, *bh; | |
09cbfeaf | 2229 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2230 | pgoff_t start, end; |
2231 | ext4_lblk_t lblk; | |
2232 | sector_t pblock; | |
2233 | int err; | |
2234 | ||
2235 | start = mpd->map.m_lblk >> bpp_bits; | |
2236 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2237 | lblk = start << bpp_bits; | |
2238 | pblock = mpd->map.m_pblk; | |
2239 | ||
2240 | pagevec_init(&pvec, 0); | |
2241 | while (start <= end) { | |
2242 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2243 | PAGEVEC_SIZE); | |
2244 | if (nr_pages == 0) | |
2245 | break; | |
2246 | for (i = 0; i < nr_pages; i++) { | |
2247 | struct page *page = pvec.pages[i]; | |
2248 | ||
2249 | if (page->index > end) | |
2250 | break; | |
70261f56 | 2251 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2252 | BUG_ON(page->index != start); |
2253 | bh = head = page_buffers(page); | |
2254 | do { | |
2255 | if (lblk < mpd->map.m_lblk) | |
2256 | continue; | |
2257 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2258 | /* | |
2259 | * Buffer after end of mapped extent. | |
2260 | * Find next buffer in the page to map. | |
2261 | */ | |
2262 | mpd->map.m_len = 0; | |
2263 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2264 | /* |
2265 | * FIXME: If dioread_nolock supports | |
2266 | * blocksize < pagesize, we need to make | |
2267 | * sure we add size mapped so far to | |
2268 | * io_end->size as the following call | |
2269 | * can submit the page for IO. | |
2270 | */ | |
2271 | err = mpage_process_page_bufs(mpd, head, | |
2272 | bh, lblk); | |
4e7ea81d | 2273 | pagevec_release(&pvec); |
5f1132b2 JK |
2274 | if (err > 0) |
2275 | err = 0; | |
2276 | return err; | |
4e7ea81d JK |
2277 | } |
2278 | if (buffer_delay(bh)) { | |
2279 | clear_buffer_delay(bh); | |
2280 | bh->b_blocknr = pblock++; | |
2281 | } | |
4e7ea81d | 2282 | clear_buffer_unwritten(bh); |
5f1132b2 | 2283 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2284 | |
2285 | /* | |
2286 | * FIXME: This is going to break if dioread_nolock | |
2287 | * supports blocksize < pagesize as we will try to | |
2288 | * convert potentially unmapped parts of inode. | |
2289 | */ | |
09cbfeaf | 2290 | mpd->io_submit.io_end->size += PAGE_SIZE; |
4e7ea81d JK |
2291 | /* Page fully mapped - let IO run! */ |
2292 | err = mpage_submit_page(mpd, page); | |
2293 | if (err < 0) { | |
2294 | pagevec_release(&pvec); | |
2295 | return err; | |
2296 | } | |
2297 | start++; | |
2298 | } | |
2299 | pagevec_release(&pvec); | |
2300 | } | |
2301 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2302 | mpd->map.m_len = 0; | |
2303 | mpd->map.m_flags = 0; | |
2304 | return 0; | |
2305 | } | |
2306 | ||
2307 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2308 | { | |
2309 | struct inode *inode = mpd->inode; | |
2310 | struct ext4_map_blocks *map = &mpd->map; | |
2311 | int get_blocks_flags; | |
090f32ee | 2312 | int err, dioread_nolock; |
4e7ea81d JK |
2313 | |
2314 | trace_ext4_da_write_pages_extent(inode, map); | |
2315 | /* | |
2316 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2317 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2318 | * where we have written into one or more preallocated blocks). It is |
2319 | * possible that we're going to need more metadata blocks than | |
2320 | * previously reserved. However we must not fail because we're in | |
2321 | * writeback and there is nothing we can do about it so it might result | |
2322 | * in data loss. So use reserved blocks to allocate metadata if | |
2323 | * possible. | |
2324 | * | |
754cfed6 TT |
2325 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2326 | * the blocks in question are delalloc blocks. This indicates | |
2327 | * that the blocks and quotas has already been checked when | |
2328 | * the data was copied into the page cache. | |
4e7ea81d JK |
2329 | */ |
2330 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2331 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2332 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2333 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2334 | if (dioread_nolock) | |
4e7ea81d JK |
2335 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2336 | if (map->m_flags & (1 << BH_Delay)) | |
2337 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2338 | ||
2339 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2340 | if (err < 0) | |
2341 | return err; | |
090f32ee | 2342 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2343 | if (!mpd->io_submit.io_end->handle && |
2344 | ext4_handle_valid(handle)) { | |
2345 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2346 | handle->h_rsv_handle = NULL; | |
2347 | } | |
3613d228 | 2348 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2349 | } |
4e7ea81d JK |
2350 | |
2351 | BUG_ON(map->m_len == 0); | |
2352 | if (map->m_flags & EXT4_MAP_NEW) { | |
2353 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2354 | int i; | |
2355 | ||
2356 | for (i = 0; i < map->m_len; i++) | |
2357 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2358 | } | |
2359 | return 0; | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2364 | * mpd->len and submit pages underlying it for IO | |
2365 | * | |
2366 | * @handle - handle for journal operations | |
2367 | * @mpd - extent to map | |
7534e854 JK |
2368 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2369 | * is no hope of writing the data. The caller should discard | |
2370 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2371 | * |
2372 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2373 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2374 | * them to initialized or split the described range from larger unwritten | |
2375 | * extent. Note that we need not map all the described range since allocation | |
2376 | * can return less blocks or the range is covered by more unwritten extents. We | |
2377 | * cannot map more because we are limited by reserved transaction credits. On | |
2378 | * the other hand we always make sure that the last touched page is fully | |
2379 | * mapped so that it can be written out (and thus forward progress is | |
2380 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2381 | */ | |
2382 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2383 | struct mpage_da_data *mpd, |
2384 | bool *give_up_on_write) | |
4e7ea81d JK |
2385 | { |
2386 | struct inode *inode = mpd->inode; | |
2387 | struct ext4_map_blocks *map = &mpd->map; | |
2388 | int err; | |
2389 | loff_t disksize; | |
6603120e | 2390 | int progress = 0; |
4e7ea81d JK |
2391 | |
2392 | mpd->io_submit.io_end->offset = | |
2393 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2394 | do { |
4e7ea81d JK |
2395 | err = mpage_map_one_extent(handle, mpd); |
2396 | if (err < 0) { | |
2397 | struct super_block *sb = inode->i_sb; | |
2398 | ||
cb530541 TT |
2399 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2400 | goto invalidate_dirty_pages; | |
4e7ea81d | 2401 | /* |
cb530541 TT |
2402 | * Let the uper layers retry transient errors. |
2403 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2404 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2405 | */ |
cb530541 | 2406 | if ((err == -ENOMEM) || |
6603120e DM |
2407 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2408 | if (progress) | |
2409 | goto update_disksize; | |
cb530541 | 2410 | return err; |
6603120e | 2411 | } |
cb530541 TT |
2412 | ext4_msg(sb, KERN_CRIT, |
2413 | "Delayed block allocation failed for " | |
2414 | "inode %lu at logical offset %llu with" | |
2415 | " max blocks %u with error %d", | |
2416 | inode->i_ino, | |
2417 | (unsigned long long)map->m_lblk, | |
2418 | (unsigned)map->m_len, -err); | |
2419 | ext4_msg(sb, KERN_CRIT, | |
2420 | "This should not happen!! Data will " | |
2421 | "be lost\n"); | |
2422 | if (err == -ENOSPC) | |
2423 | ext4_print_free_blocks(inode); | |
2424 | invalidate_dirty_pages: | |
2425 | *give_up_on_write = true; | |
4e7ea81d JK |
2426 | return err; |
2427 | } | |
6603120e | 2428 | progress = 1; |
4e7ea81d JK |
2429 | /* |
2430 | * Update buffer state, submit mapped pages, and get us new | |
2431 | * extent to map | |
2432 | */ | |
2433 | err = mpage_map_and_submit_buffers(mpd); | |
2434 | if (err < 0) | |
6603120e | 2435 | goto update_disksize; |
27d7c4ed | 2436 | } while (map->m_len); |
4e7ea81d | 2437 | |
6603120e | 2438 | update_disksize: |
622cad13 TT |
2439 | /* |
2440 | * Update on-disk size after IO is submitted. Races with | |
2441 | * truncate are avoided by checking i_size under i_data_sem. | |
2442 | */ | |
09cbfeaf | 2443 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
4e7ea81d JK |
2444 | if (disksize > EXT4_I(inode)->i_disksize) { |
2445 | int err2; | |
622cad13 TT |
2446 | loff_t i_size; |
2447 | ||
2448 | down_write(&EXT4_I(inode)->i_data_sem); | |
2449 | i_size = i_size_read(inode); | |
2450 | if (disksize > i_size) | |
2451 | disksize = i_size; | |
2452 | if (disksize > EXT4_I(inode)->i_disksize) | |
2453 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2454 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2455 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2456 | if (err2) |
2457 | ext4_error(inode->i_sb, | |
2458 | "Failed to mark inode %lu dirty", | |
2459 | inode->i_ino); | |
2460 | if (!err) | |
2461 | err = err2; | |
2462 | } | |
2463 | return err; | |
2464 | } | |
2465 | ||
fffb2739 JK |
2466 | /* |
2467 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2468 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2469 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2470 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2471 | * bpp - 1 blocks in bpp different extents. | |
2472 | */ | |
525f4ed8 MC |
2473 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2474 | { | |
fffb2739 | 2475 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2476 | |
fffb2739 JK |
2477 | return ext4_meta_trans_blocks(inode, |
2478 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2479 | } |
61628a3f | 2480 | |
8e48dcfb | 2481 | /* |
4e7ea81d JK |
2482 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2483 | * and underlying extent to map | |
2484 | * | |
2485 | * @mpd - where to look for pages | |
2486 | * | |
2487 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2488 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2489 | * extent of buffers underlying these pages that needs mapping (formed by | |
2490 | * either delayed or unwritten buffers). We also lock the pages containing | |
2491 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2492 | * mpd->lblk with length mpd->len blocks). | |
2493 | * | |
2494 | * Note that this function can attach bios to one io_end structure which are | |
2495 | * neither logically nor physically contiguous. Although it may seem as an | |
2496 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2497 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2498 | */ |
4e7ea81d | 2499 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2500 | { |
4e7ea81d JK |
2501 | struct address_space *mapping = mpd->inode->i_mapping; |
2502 | struct pagevec pvec; | |
2503 | unsigned int nr_pages; | |
aeac589a | 2504 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2505 | pgoff_t index = mpd->first_page; |
2506 | pgoff_t end = mpd->last_page; | |
2507 | int tag; | |
2508 | int i, err = 0; | |
2509 | int blkbits = mpd->inode->i_blkbits; | |
2510 | ext4_lblk_t lblk; | |
2511 | struct buffer_head *head; | |
8e48dcfb | 2512 | |
4e7ea81d | 2513 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2514 | tag = PAGECACHE_TAG_TOWRITE; |
2515 | else | |
2516 | tag = PAGECACHE_TAG_DIRTY; | |
2517 | ||
4e7ea81d JK |
2518 | pagevec_init(&pvec, 0); |
2519 | mpd->map.m_len = 0; | |
2520 | mpd->next_page = index; | |
4f01b02c | 2521 | while (index <= end) { |
5b41d924 | 2522 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2523 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2524 | if (nr_pages == 0) | |
4e7ea81d | 2525 | goto out; |
8e48dcfb TT |
2526 | |
2527 | for (i = 0; i < nr_pages; i++) { | |
2528 | struct page *page = pvec.pages[i]; | |
2529 | ||
2530 | /* | |
2531 | * At this point, the page may be truncated or | |
2532 | * invalidated (changing page->mapping to NULL), or | |
2533 | * even swizzled back from swapper_space to tmpfs file | |
2534 | * mapping. However, page->index will not change | |
2535 | * because we have a reference on the page. | |
2536 | */ | |
4f01b02c TT |
2537 | if (page->index > end) |
2538 | goto out; | |
8e48dcfb | 2539 | |
aeac589a ML |
2540 | /* |
2541 | * Accumulated enough dirty pages? This doesn't apply | |
2542 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2543 | * keep going because someone may be concurrently | |
2544 | * dirtying pages, and we might have synced a lot of | |
2545 | * newly appeared dirty pages, but have not synced all | |
2546 | * of the old dirty pages. | |
2547 | */ | |
2548 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2549 | goto out; | |
2550 | ||
4e7ea81d JK |
2551 | /* If we can't merge this page, we are done. */ |
2552 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2553 | goto out; | |
78aaced3 | 2554 | |
8e48dcfb | 2555 | lock_page(page); |
8e48dcfb | 2556 | /* |
4e7ea81d JK |
2557 | * If the page is no longer dirty, or its mapping no |
2558 | * longer corresponds to inode we are writing (which | |
2559 | * means it has been truncated or invalidated), or the | |
2560 | * page is already under writeback and we are not doing | |
2561 | * a data integrity writeback, skip the page | |
8e48dcfb | 2562 | */ |
4f01b02c TT |
2563 | if (!PageDirty(page) || |
2564 | (PageWriteback(page) && | |
4e7ea81d | 2565 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2566 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2567 | unlock_page(page); |
2568 | continue; | |
2569 | } | |
2570 | ||
7cb1a535 | 2571 | wait_on_page_writeback(page); |
8e48dcfb | 2572 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2573 | |
4e7ea81d | 2574 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2575 | mpd->first_page = page->index; |
8eb9e5ce | 2576 | mpd->next_page = page->index + 1; |
f8bec370 | 2577 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2578 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2579 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2580 | head = page_buffers(page); |
5f1132b2 JK |
2581 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2582 | if (err <= 0) | |
4e7ea81d | 2583 | goto out; |
5f1132b2 | 2584 | err = 0; |
aeac589a | 2585 | left--; |
8e48dcfb TT |
2586 | } |
2587 | pagevec_release(&pvec); | |
2588 | cond_resched(); | |
2589 | } | |
4f01b02c | 2590 | return 0; |
8eb9e5ce TT |
2591 | out: |
2592 | pagevec_release(&pvec); | |
4e7ea81d | 2593 | return err; |
8e48dcfb TT |
2594 | } |
2595 | ||
20970ba6 TT |
2596 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2597 | void *data) | |
2598 | { | |
2599 | struct address_space *mapping = data; | |
2600 | int ret = ext4_writepage(page, wbc); | |
2601 | mapping_set_error(mapping, ret); | |
2602 | return ret; | |
2603 | } | |
2604 | ||
2605 | static int ext4_writepages(struct address_space *mapping, | |
2606 | struct writeback_control *wbc) | |
64769240 | 2607 | { |
4e7ea81d JK |
2608 | pgoff_t writeback_index = 0; |
2609 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2610 | int range_whole = 0; |
4e7ea81d | 2611 | int cycled = 1; |
61628a3f | 2612 | handle_t *handle = NULL; |
df22291f | 2613 | struct mpage_da_data mpd; |
5e745b04 | 2614 | struct inode *inode = mapping->host; |
6b523df4 | 2615 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2616 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2617 | bool done; |
1bce63d1 | 2618 | struct blk_plug plug; |
cb530541 | 2619 | bool give_up_on_write = false; |
61628a3f | 2620 | |
c8585c6f | 2621 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
20970ba6 | 2622 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2623 | |
c8585c6f DJ |
2624 | if (dax_mapping(mapping)) { |
2625 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, | |
2626 | wbc); | |
2627 | goto out_writepages; | |
2628 | } | |
7f6d5b52 | 2629 | |
61628a3f MC |
2630 | /* |
2631 | * No pages to write? This is mainly a kludge to avoid starting | |
2632 | * a transaction for special inodes like journal inode on last iput() | |
2633 | * because that could violate lock ordering on umount | |
2634 | */ | |
a1d6cc56 | 2635 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2636 | goto out_writepages; |
2a21e37e | 2637 | |
20970ba6 TT |
2638 | if (ext4_should_journal_data(inode)) { |
2639 | struct blk_plug plug; | |
20970ba6 TT |
2640 | |
2641 | blk_start_plug(&plug); | |
2642 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2643 | blk_finish_plug(&plug); | |
bbf023c7 | 2644 | goto out_writepages; |
20970ba6 TT |
2645 | } |
2646 | ||
2a21e37e TT |
2647 | /* |
2648 | * If the filesystem has aborted, it is read-only, so return | |
2649 | * right away instead of dumping stack traces later on that | |
2650 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2651 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2652 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2653 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2654 | * *never* be called, so if that ever happens, we would want |
2655 | * the stack trace. | |
2656 | */ | |
bbf023c7 ML |
2657 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2658 | ret = -EROFS; | |
2659 | goto out_writepages; | |
2660 | } | |
2a21e37e | 2661 | |
6b523df4 JK |
2662 | if (ext4_should_dioread_nolock(inode)) { |
2663 | /* | |
70261f56 | 2664 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2665 | * the page and we may dirty the inode. |
2666 | */ | |
09cbfeaf | 2667 | rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits); |
6b523df4 JK |
2668 | } |
2669 | ||
4e7ea81d JK |
2670 | /* |
2671 | * If we have inline data and arrive here, it means that | |
2672 | * we will soon create the block for the 1st page, so | |
2673 | * we'd better clear the inline data here. | |
2674 | */ | |
2675 | if (ext4_has_inline_data(inode)) { | |
2676 | /* Just inode will be modified... */ | |
2677 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2678 | if (IS_ERR(handle)) { | |
2679 | ret = PTR_ERR(handle); | |
2680 | goto out_writepages; | |
2681 | } | |
2682 | BUG_ON(ext4_test_inode_state(inode, | |
2683 | EXT4_STATE_MAY_INLINE_DATA)); | |
2684 | ext4_destroy_inline_data(handle, inode); | |
2685 | ext4_journal_stop(handle); | |
2686 | } | |
2687 | ||
22208ded AK |
2688 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2689 | range_whole = 1; | |
61628a3f | 2690 | |
2acf2c26 | 2691 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2692 | writeback_index = mapping->writeback_index; |
2693 | if (writeback_index) | |
2acf2c26 | 2694 | cycled = 0; |
4e7ea81d JK |
2695 | mpd.first_page = writeback_index; |
2696 | mpd.last_page = -1; | |
5b41d924 | 2697 | } else { |
09cbfeaf KS |
2698 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2699 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2700 | } |
a1d6cc56 | 2701 | |
4e7ea81d JK |
2702 | mpd.inode = inode; |
2703 | mpd.wbc = wbc; | |
2704 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2705 | retry: |
6e6938b6 | 2706 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2707 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2708 | done = false; | |
1bce63d1 | 2709 | blk_start_plug(&plug); |
4e7ea81d JK |
2710 | while (!done && mpd.first_page <= mpd.last_page) { |
2711 | /* For each extent of pages we use new io_end */ | |
2712 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2713 | if (!mpd.io_submit.io_end) { | |
2714 | ret = -ENOMEM; | |
2715 | break; | |
2716 | } | |
a1d6cc56 AK |
2717 | |
2718 | /* | |
4e7ea81d JK |
2719 | * We have two constraints: We find one extent to map and we |
2720 | * must always write out whole page (makes a difference when | |
2721 | * blocksize < pagesize) so that we don't block on IO when we | |
2722 | * try to write out the rest of the page. Journalled mode is | |
2723 | * not supported by delalloc. | |
a1d6cc56 AK |
2724 | */ |
2725 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2726 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2727 | |
4e7ea81d | 2728 | /* start a new transaction */ |
6b523df4 JK |
2729 | handle = ext4_journal_start_with_reserve(inode, |
2730 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2731 | if (IS_ERR(handle)) { |
2732 | ret = PTR_ERR(handle); | |
1693918e | 2733 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2734 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2735 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2736 | /* Release allocated io_end */ |
2737 | ext4_put_io_end(mpd.io_submit.io_end); | |
2738 | break; | |
61628a3f | 2739 | } |
f63e6005 | 2740 | |
4e7ea81d JK |
2741 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2742 | ret = mpage_prepare_extent_to_map(&mpd); | |
2743 | if (!ret) { | |
2744 | if (mpd.map.m_len) | |
cb530541 TT |
2745 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2746 | &give_up_on_write); | |
4e7ea81d JK |
2747 | else { |
2748 | /* | |
2749 | * We scanned the whole range (or exhausted | |
2750 | * nr_to_write), submitted what was mapped and | |
2751 | * didn't find anything needing mapping. We are | |
2752 | * done. | |
2753 | */ | |
2754 | done = true; | |
2755 | } | |
f63e6005 | 2756 | } |
646caa9c JK |
2757 | /* |
2758 | * Caution: If the handle is synchronous, | |
2759 | * ext4_journal_stop() can wait for transaction commit | |
2760 | * to finish which may depend on writeback of pages to | |
2761 | * complete or on page lock to be released. In that | |
2762 | * case, we have to wait until after after we have | |
2763 | * submitted all the IO, released page locks we hold, | |
2764 | * and dropped io_end reference (for extent conversion | |
2765 | * to be able to complete) before stopping the handle. | |
2766 | */ | |
2767 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2768 | ext4_journal_stop(handle); | |
2769 | handle = NULL; | |
2770 | } | |
4e7ea81d JK |
2771 | /* Submit prepared bio */ |
2772 | ext4_io_submit(&mpd.io_submit); | |
2773 | /* Unlock pages we didn't use */ | |
cb530541 | 2774 | mpage_release_unused_pages(&mpd, give_up_on_write); |
646caa9c JK |
2775 | /* |
2776 | * Drop our io_end reference we got from init. We have | |
2777 | * to be careful and use deferred io_end finishing if | |
2778 | * we are still holding the transaction as we can | |
2779 | * release the last reference to io_end which may end | |
2780 | * up doing unwritten extent conversion. | |
2781 | */ | |
2782 | if (handle) { | |
2783 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2784 | ext4_journal_stop(handle); | |
2785 | } else | |
2786 | ext4_put_io_end(mpd.io_submit.io_end); | |
4e7ea81d JK |
2787 | |
2788 | if (ret == -ENOSPC && sbi->s_journal) { | |
2789 | /* | |
2790 | * Commit the transaction which would | |
22208ded AK |
2791 | * free blocks released in the transaction |
2792 | * and try again | |
2793 | */ | |
df22291f | 2794 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2795 | ret = 0; |
4e7ea81d JK |
2796 | continue; |
2797 | } | |
2798 | /* Fatal error - ENOMEM, EIO... */ | |
2799 | if (ret) | |
61628a3f | 2800 | break; |
a1d6cc56 | 2801 | } |
1bce63d1 | 2802 | blk_finish_plug(&plug); |
9c12a831 | 2803 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2804 | cycled = 1; |
4e7ea81d JK |
2805 | mpd.last_page = writeback_index - 1; |
2806 | mpd.first_page = 0; | |
2acf2c26 AK |
2807 | goto retry; |
2808 | } | |
22208ded AK |
2809 | |
2810 | /* Update index */ | |
22208ded AK |
2811 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2812 | /* | |
4e7ea81d | 2813 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2814 | * mode will write it back later |
2815 | */ | |
4e7ea81d | 2816 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2817 | |
61628a3f | 2818 | out_writepages: |
20970ba6 TT |
2819 | trace_ext4_writepages_result(inode, wbc, ret, |
2820 | nr_to_write - wbc->nr_to_write); | |
c8585c6f | 2821 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
61628a3f | 2822 | return ret; |
64769240 AT |
2823 | } |
2824 | ||
79f0be8d AK |
2825 | static int ext4_nonda_switch(struct super_block *sb) |
2826 | { | |
5c1ff336 | 2827 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2828 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2829 | ||
2830 | /* | |
2831 | * switch to non delalloc mode if we are running low | |
2832 | * on free block. The free block accounting via percpu | |
179f7ebf | 2833 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2834 | * accumulated on each CPU without updating global counters |
2835 | * Delalloc need an accurate free block accounting. So switch | |
2836 | * to non delalloc when we are near to error range. | |
2837 | */ | |
5c1ff336 EW |
2838 | free_clusters = |
2839 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2840 | dirty_clusters = | |
2841 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2842 | /* |
2843 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2844 | */ | |
5c1ff336 | 2845 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2846 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2847 | |
5c1ff336 EW |
2848 | if (2 * free_clusters < 3 * dirty_clusters || |
2849 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2850 | /* |
c8afb446 ES |
2851 | * free block count is less than 150% of dirty blocks |
2852 | * or free blocks is less than watermark | |
79f0be8d AK |
2853 | */ |
2854 | return 1; | |
2855 | } | |
2856 | return 0; | |
2857 | } | |
2858 | ||
0ff8947f ES |
2859 | /* We always reserve for an inode update; the superblock could be there too */ |
2860 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2861 | { | |
e2b911c5 | 2862 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2863 | return 1; |
2864 | ||
2865 | if (pos + len <= 0x7fffffffULL) | |
2866 | return 1; | |
2867 | ||
2868 | /* We might need to update the superblock to set LARGE_FILE */ | |
2869 | return 2; | |
2870 | } | |
2871 | ||
64769240 | 2872 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2873 | loff_t pos, unsigned len, unsigned flags, |
2874 | struct page **pagep, void **fsdata) | |
64769240 | 2875 | { |
72b8ab9d | 2876 | int ret, retries = 0; |
64769240 AT |
2877 | struct page *page; |
2878 | pgoff_t index; | |
64769240 AT |
2879 | struct inode *inode = mapping->host; |
2880 | handle_t *handle; | |
2881 | ||
09cbfeaf | 2882 | index = pos >> PAGE_SHIFT; |
79f0be8d AK |
2883 | |
2884 | if (ext4_nonda_switch(inode->i_sb)) { | |
2885 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2886 | return ext4_write_begin(file, mapping, pos, | |
2887 | len, flags, pagep, fsdata); | |
2888 | } | |
2889 | *fsdata = (void *)0; | |
9bffad1e | 2890 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2891 | |
2892 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2893 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2894 | pos, len, flags, | |
2895 | pagep, fsdata); | |
2896 | if (ret < 0) | |
47564bfb TT |
2897 | return ret; |
2898 | if (ret == 1) | |
2899 | return 0; | |
9c3569b5 TM |
2900 | } |
2901 | ||
47564bfb TT |
2902 | /* |
2903 | * grab_cache_page_write_begin() can take a long time if the | |
2904 | * system is thrashing due to memory pressure, or if the page | |
2905 | * is being written back. So grab it first before we start | |
2906 | * the transaction handle. This also allows us to allocate | |
2907 | * the page (if needed) without using GFP_NOFS. | |
2908 | */ | |
2909 | retry_grab: | |
2910 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2911 | if (!page) | |
2912 | return -ENOMEM; | |
2913 | unlock_page(page); | |
2914 | ||
64769240 AT |
2915 | /* |
2916 | * With delayed allocation, we don't log the i_disksize update | |
2917 | * if there is delayed block allocation. But we still need | |
2918 | * to journalling the i_disksize update if writes to the end | |
2919 | * of file which has an already mapped buffer. | |
2920 | */ | |
47564bfb | 2921 | retry_journal: |
0ff8947f ES |
2922 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2923 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2924 | if (IS_ERR(handle)) { |
09cbfeaf | 2925 | put_page(page); |
47564bfb | 2926 | return PTR_ERR(handle); |
64769240 AT |
2927 | } |
2928 | ||
47564bfb TT |
2929 | lock_page(page); |
2930 | if (page->mapping != mapping) { | |
2931 | /* The page got truncated from under us */ | |
2932 | unlock_page(page); | |
09cbfeaf | 2933 | put_page(page); |
d5a0d4f7 | 2934 | ext4_journal_stop(handle); |
47564bfb | 2935 | goto retry_grab; |
d5a0d4f7 | 2936 | } |
47564bfb | 2937 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2938 | wait_for_stable_page(page); |
64769240 | 2939 | |
2058f83a MH |
2940 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
2941 | ret = ext4_block_write_begin(page, pos, len, | |
2942 | ext4_da_get_block_prep); | |
2943 | #else | |
6e1db88d | 2944 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2945 | #endif |
64769240 AT |
2946 | if (ret < 0) { |
2947 | unlock_page(page); | |
2948 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2949 | /* |
2950 | * block_write_begin may have instantiated a few blocks | |
2951 | * outside i_size. Trim these off again. Don't need | |
2952 | * i_size_read because we hold i_mutex. | |
2953 | */ | |
2954 | if (pos + len > inode->i_size) | |
b9a4207d | 2955 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2956 | |
2957 | if (ret == -ENOSPC && | |
2958 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2959 | goto retry_journal; | |
2960 | ||
09cbfeaf | 2961 | put_page(page); |
47564bfb | 2962 | return ret; |
64769240 AT |
2963 | } |
2964 | ||
47564bfb | 2965 | *pagep = page; |
64769240 AT |
2966 | return ret; |
2967 | } | |
2968 | ||
632eaeab MC |
2969 | /* |
2970 | * Check if we should update i_disksize | |
2971 | * when write to the end of file but not require block allocation | |
2972 | */ | |
2973 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2974 | unsigned long offset) |
632eaeab MC |
2975 | { |
2976 | struct buffer_head *bh; | |
2977 | struct inode *inode = page->mapping->host; | |
2978 | unsigned int idx; | |
2979 | int i; | |
2980 | ||
2981 | bh = page_buffers(page); | |
2982 | idx = offset >> inode->i_blkbits; | |
2983 | ||
af5bc92d | 2984 | for (i = 0; i < idx; i++) |
632eaeab MC |
2985 | bh = bh->b_this_page; |
2986 | ||
29fa89d0 | 2987 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2988 | return 0; |
2989 | return 1; | |
2990 | } | |
2991 | ||
64769240 | 2992 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2993 | struct address_space *mapping, |
2994 | loff_t pos, unsigned len, unsigned copied, | |
2995 | struct page *page, void *fsdata) | |
64769240 AT |
2996 | { |
2997 | struct inode *inode = mapping->host; | |
2998 | int ret = 0, ret2; | |
2999 | handle_t *handle = ext4_journal_current_handle(); | |
3000 | loff_t new_i_size; | |
632eaeab | 3001 | unsigned long start, end; |
79f0be8d AK |
3002 | int write_mode = (int)(unsigned long)fsdata; |
3003 | ||
74d553aa TT |
3004 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3005 | return ext4_write_end(file, mapping, pos, | |
3006 | len, copied, page, fsdata); | |
632eaeab | 3007 | |
9bffad1e | 3008 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3009 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3010 | end = start + copied - 1; |
64769240 AT |
3011 | |
3012 | /* | |
3013 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3014 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3015 | * into that. | |
3016 | */ | |
64769240 | 3017 | new_i_size = pos + copied; |
ea51d132 | 3018 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3019 | if (ext4_has_inline_data(inode) || |
3020 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3021 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3022 | /* We need to mark inode dirty even if |
3023 | * new_i_size is less that inode->i_size | |
3024 | * bu greater than i_disksize.(hint delalloc) | |
3025 | */ | |
3026 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3027 | } |
632eaeab | 3028 | } |
9c3569b5 TM |
3029 | |
3030 | if (write_mode != CONVERT_INLINE_DATA && | |
3031 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3032 | ext4_has_inline_data(inode)) | |
3033 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3034 | page); | |
3035 | else | |
3036 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3037 | page, fsdata); |
9c3569b5 | 3038 | |
64769240 AT |
3039 | copied = ret2; |
3040 | if (ret2 < 0) | |
3041 | ret = ret2; | |
3042 | ret2 = ext4_journal_stop(handle); | |
3043 | if (!ret) | |
3044 | ret = ret2; | |
3045 | ||
3046 | return ret ? ret : copied; | |
3047 | } | |
3048 | ||
d47992f8 LC |
3049 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
3050 | unsigned int length) | |
64769240 | 3051 | { |
64769240 AT |
3052 | /* |
3053 | * Drop reserved blocks | |
3054 | */ | |
3055 | BUG_ON(!PageLocked(page)); | |
3056 | if (!page_has_buffers(page)) | |
3057 | goto out; | |
3058 | ||
ca99fdd2 | 3059 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
3060 | |
3061 | out: | |
d47992f8 | 3062 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
3063 | |
3064 | return; | |
3065 | } | |
3066 | ||
ccd2506b TT |
3067 | /* |
3068 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3069 | */ | |
3070 | int ext4_alloc_da_blocks(struct inode *inode) | |
3071 | { | |
fb40ba0d TT |
3072 | trace_ext4_alloc_da_blocks(inode); |
3073 | ||
71d4f7d0 | 3074 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3075 | return 0; |
3076 | ||
3077 | /* | |
3078 | * We do something simple for now. The filemap_flush() will | |
3079 | * also start triggering a write of the data blocks, which is | |
3080 | * not strictly speaking necessary (and for users of | |
3081 | * laptop_mode, not even desirable). However, to do otherwise | |
3082 | * would require replicating code paths in: | |
de9a55b8 | 3083 | * |
20970ba6 | 3084 | * ext4_writepages() -> |
ccd2506b TT |
3085 | * write_cache_pages() ---> (via passed in callback function) |
3086 | * __mpage_da_writepage() --> | |
3087 | * mpage_add_bh_to_extent() | |
3088 | * mpage_da_map_blocks() | |
3089 | * | |
3090 | * The problem is that write_cache_pages(), located in | |
3091 | * mm/page-writeback.c, marks pages clean in preparation for | |
3092 | * doing I/O, which is not desirable if we're not planning on | |
3093 | * doing I/O at all. | |
3094 | * | |
3095 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3096 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3097 | * would be ugly in the extreme. So instead we would need to |
3098 | * replicate parts of the code in the above functions, | |
25985edc | 3099 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3100 | * write out the pages, but rather only collect contiguous |
3101 | * logical block extents, call the multi-block allocator, and | |
3102 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3103 | * |
ccd2506b TT |
3104 | * For now, though, we'll cheat by calling filemap_flush(), |
3105 | * which will map the blocks, and start the I/O, but not | |
3106 | * actually wait for the I/O to complete. | |
3107 | */ | |
3108 | return filemap_flush(inode->i_mapping); | |
3109 | } | |
64769240 | 3110 | |
ac27a0ec DK |
3111 | /* |
3112 | * bmap() is special. It gets used by applications such as lilo and by | |
3113 | * the swapper to find the on-disk block of a specific piece of data. | |
3114 | * | |
3115 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3116 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3117 | * filesystem and enables swap, then they may get a nasty shock when the |
3118 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3119 | * the original zero's written out previously to the journal and | |
3120 | * awaiting writeback in the kernel's buffer cache. | |
3121 | * | |
3122 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3123 | * take extra steps to flush any blocks which might be in the cache. | |
3124 | */ | |
617ba13b | 3125 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3126 | { |
3127 | struct inode *inode = mapping->host; | |
3128 | journal_t *journal; | |
3129 | int err; | |
3130 | ||
46c7f254 TM |
3131 | /* |
3132 | * We can get here for an inline file via the FIBMAP ioctl | |
3133 | */ | |
3134 | if (ext4_has_inline_data(inode)) | |
3135 | return 0; | |
3136 | ||
64769240 AT |
3137 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3138 | test_opt(inode->i_sb, DELALLOC)) { | |
3139 | /* | |
3140 | * With delalloc we want to sync the file | |
3141 | * so that we can make sure we allocate | |
3142 | * blocks for file | |
3143 | */ | |
3144 | filemap_write_and_wait(mapping); | |
3145 | } | |
3146 | ||
19f5fb7a TT |
3147 | if (EXT4_JOURNAL(inode) && |
3148 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3149 | /* |
3150 | * This is a REALLY heavyweight approach, but the use of | |
3151 | * bmap on dirty files is expected to be extremely rare: | |
3152 | * only if we run lilo or swapon on a freshly made file | |
3153 | * do we expect this to happen. | |
3154 | * | |
3155 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3156 | * represent an unprivileged user DOS attack --- we'd be | |
3157 | * in trouble if mortal users could trigger this path at | |
3158 | * will.) | |
3159 | * | |
617ba13b | 3160 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3161 | * regular files. If somebody wants to bmap a directory |
3162 | * or symlink and gets confused because the buffer | |
3163 | * hasn't yet been flushed to disk, they deserve | |
3164 | * everything they get. | |
3165 | */ | |
3166 | ||
19f5fb7a | 3167 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3168 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3169 | jbd2_journal_lock_updates(journal); |
3170 | err = jbd2_journal_flush(journal); | |
3171 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3172 | |
3173 | if (err) | |
3174 | return 0; | |
3175 | } | |
3176 | ||
af5bc92d | 3177 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3178 | } |
3179 | ||
617ba13b | 3180 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3181 | { |
46c7f254 TM |
3182 | int ret = -EAGAIN; |
3183 | struct inode *inode = page->mapping->host; | |
3184 | ||
0562e0ba | 3185 | trace_ext4_readpage(page); |
46c7f254 TM |
3186 | |
3187 | if (ext4_has_inline_data(inode)) | |
3188 | ret = ext4_readpage_inline(inode, page); | |
3189 | ||
3190 | if (ret == -EAGAIN) | |
f64e02fe | 3191 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3192 | |
3193 | return ret; | |
ac27a0ec DK |
3194 | } |
3195 | ||
3196 | static int | |
617ba13b | 3197 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3198 | struct list_head *pages, unsigned nr_pages) |
3199 | { | |
46c7f254 TM |
3200 | struct inode *inode = mapping->host; |
3201 | ||
3202 | /* If the file has inline data, no need to do readpages. */ | |
3203 | if (ext4_has_inline_data(inode)) | |
3204 | return 0; | |
3205 | ||
f64e02fe | 3206 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3207 | } |
3208 | ||
d47992f8 LC |
3209 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3210 | unsigned int length) | |
ac27a0ec | 3211 | { |
ca99fdd2 | 3212 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3213 | |
4520fb3c JK |
3214 | /* No journalling happens on data buffers when this function is used */ |
3215 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3216 | ||
ca99fdd2 | 3217 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3218 | } |
3219 | ||
53e87268 | 3220 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3221 | unsigned int offset, |
3222 | unsigned int length) | |
4520fb3c JK |
3223 | { |
3224 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3225 | ||
ca99fdd2 | 3226 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3227 | |
ac27a0ec DK |
3228 | /* |
3229 | * If it's a full truncate we just forget about the pending dirtying | |
3230 | */ | |
09cbfeaf | 3231 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3232 | ClearPageChecked(page); |
3233 | ||
ca99fdd2 | 3234 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3235 | } |
3236 | ||
3237 | /* Wrapper for aops... */ | |
3238 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3239 | unsigned int offset, |
3240 | unsigned int length) | |
53e87268 | 3241 | { |
ca99fdd2 | 3242 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3243 | } |
3244 | ||
617ba13b | 3245 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3246 | { |
617ba13b | 3247 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3248 | |
0562e0ba JZ |
3249 | trace_ext4_releasepage(page); |
3250 | ||
e1c36595 JK |
3251 | /* Page has dirty journalled data -> cannot release */ |
3252 | if (PageChecked(page)) | |
ac27a0ec | 3253 | return 0; |
0390131b FM |
3254 | if (journal) |
3255 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3256 | else | |
3257 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3258 | } |
3259 | ||
ba5843f5 | 3260 | #ifdef CONFIG_FS_DAX |
12735f88 JK |
3261 | /* |
3262 | * Get block function for DAX IO and mmap faults. It takes care of converting | |
3263 | * unwritten extents to written ones and initializes new / converted blocks | |
3264 | * to zeros. | |
3265 | */ | |
3266 | int ext4_dax_get_block(struct inode *inode, sector_t iblock, | |
3267 | struct buffer_head *bh_result, int create) | |
ed923b57 | 3268 | { |
7cb476f8 | 3269 | int ret; |
ba5843f5 | 3270 | |
12735f88 | 3271 | ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create); |
7cb476f8 JK |
3272 | if (!create) |
3273 | return _ext4_get_block(inode, iblock, bh_result, 0); | |
ba5843f5 | 3274 | |
7cb476f8 JK |
3275 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
3276 | EXT4_GET_BLOCKS_PRE_IO | | |
3277 | EXT4_GET_BLOCKS_CREATE_ZERO); | |
3278 | if (ret < 0) | |
3279 | return ret; | |
ba5843f5 | 3280 | |
7cb476f8 | 3281 | if (buffer_unwritten(bh_result)) { |
ba5843f5 | 3282 | /* |
12735f88 JK |
3283 | * We are protected by i_mmap_sem or i_mutex so we know block |
3284 | * cannot go away from under us even though we dropped | |
3285 | * i_data_sem. Convert extent to written and write zeros there. | |
ba5843f5 | 3286 | */ |
7cb476f8 JK |
3287 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
3288 | EXT4_GET_BLOCKS_CONVERT | | |
3289 | EXT4_GET_BLOCKS_CREATE_ZERO); | |
3290 | if (ret < 0) | |
3291 | return ret; | |
ba5843f5 | 3292 | } |
7cb476f8 JK |
3293 | /* |
3294 | * At least for now we have to clear BH_New so that DAX code | |
3295 | * doesn't attempt to zero blocks again in a racy way. | |
3296 | */ | |
3297 | clear_buffer_new(bh_result); | |
3298 | return 0; | |
ed923b57 | 3299 | } |
12735f88 JK |
3300 | #else |
3301 | /* Just define empty function, it will never get called. */ | |
3302 | int ext4_dax_get_block(struct inode *inode, sector_t iblock, | |
3303 | struct buffer_head *bh_result, int create) | |
3304 | { | |
3305 | BUG(); | |
3306 | return 0; | |
ed923b57 | 3307 | } |
ba5843f5 | 3308 | #endif |
ed923b57 | 3309 | |
187372a3 | 3310 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3311 | ssize_t size, void *private) |
4c0425ff | 3312 | { |
109811c2 | 3313 | ext4_io_end_t *io_end = private; |
4c0425ff | 3314 | |
97a851ed | 3315 | /* if not async direct IO just return */ |
7b7a8665 | 3316 | if (!io_end) |
187372a3 | 3317 | return 0; |
4b70df18 | 3318 | |
88635ca2 | 3319 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3320 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
109811c2 | 3321 | io_end, io_end->inode->i_ino, iocb, offset, size); |
8d5d02e6 | 3322 | |
74c66bcb JK |
3323 | /* |
3324 | * Error during AIO DIO. We cannot convert unwritten extents as the | |
3325 | * data was not written. Just clear the unwritten flag and drop io_end. | |
3326 | */ | |
3327 | if (size <= 0) { | |
3328 | ext4_clear_io_unwritten_flag(io_end); | |
3329 | size = 0; | |
3330 | } | |
4c0425ff MC |
3331 | io_end->offset = offset; |
3332 | io_end->size = size; | |
7b7a8665 | 3333 | ext4_put_io_end(io_end); |
187372a3 CH |
3334 | |
3335 | return 0; | |
4c0425ff | 3336 | } |
c7064ef1 | 3337 | |
4c0425ff | 3338 | /* |
914f82a3 JK |
3339 | * Handling of direct IO writes. |
3340 | * | |
3341 | * For ext4 extent files, ext4 will do direct-io write even to holes, | |
4c0425ff MC |
3342 | * preallocated extents, and those write extend the file, no need to |
3343 | * fall back to buffered IO. | |
3344 | * | |
556615dc | 3345 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3346 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3347 | * still keep the range to write as unwritten. |
4c0425ff | 3348 | * |
69c499d1 | 3349 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3350 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3351 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3352 | * when async direct IO completed. |
4c0425ff MC |
3353 | * |
3354 | * If the O_DIRECT write will extend the file then add this inode to the | |
3355 | * orphan list. So recovery will truncate it back to the original size | |
3356 | * if the machine crashes during the write. | |
3357 | * | |
3358 | */ | |
0e01df10 | 3359 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3360 | { |
3361 | struct file *file = iocb->ki_filp; | |
3362 | struct inode *inode = file->f_mapping->host; | |
914f82a3 | 3363 | struct ext4_inode_info *ei = EXT4_I(inode); |
4c0425ff | 3364 | ssize_t ret; |
c8b8e32d | 3365 | loff_t offset = iocb->ki_pos; |
a6cbcd4a | 3366 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3367 | int overwrite = 0; |
3368 | get_block_t *get_block_func = NULL; | |
3369 | int dio_flags = 0; | |
4c0425ff | 3370 | loff_t final_size = offset + count; |
914f82a3 JK |
3371 | int orphan = 0; |
3372 | handle_t *handle; | |
729f52c6 | 3373 | |
914f82a3 JK |
3374 | if (final_size > inode->i_size) { |
3375 | /* Credits for sb + inode write */ | |
3376 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3377 | if (IS_ERR(handle)) { | |
3378 | ret = PTR_ERR(handle); | |
3379 | goto out; | |
3380 | } | |
3381 | ret = ext4_orphan_add(handle, inode); | |
3382 | if (ret) { | |
3383 | ext4_journal_stop(handle); | |
3384 | goto out; | |
3385 | } | |
3386 | orphan = 1; | |
3387 | ei->i_disksize = inode->i_size; | |
3388 | ext4_journal_stop(handle); | |
3389 | } | |
4bd809db | 3390 | |
69c499d1 | 3391 | BUG_ON(iocb->private == NULL); |
4bd809db | 3392 | |
e8340395 JK |
3393 | /* |
3394 | * Make all waiters for direct IO properly wait also for extent | |
3395 | * conversion. This also disallows race between truncate() and | |
3396 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3397 | */ | |
914f82a3 | 3398 | inode_dio_begin(inode); |
e8340395 | 3399 | |
69c499d1 TT |
3400 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3401 | overwrite = *((int *)iocb->private); | |
4bd809db | 3402 | |
2dcba478 | 3403 | if (overwrite) |
5955102c | 3404 | inode_unlock(inode); |
8d5d02e6 | 3405 | |
69c499d1 | 3406 | /* |
914f82a3 | 3407 | * For extent mapped files we could direct write to holes and fallocate. |
69c499d1 | 3408 | * |
109811c2 JK |
3409 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
3410 | * parallel buffered read to expose the stale data before DIO complete | |
3411 | * the data IO. | |
69c499d1 | 3412 | * |
109811c2 JK |
3413 | * As to previously fallocated extents, ext4 get_block will just simply |
3414 | * mark the buffer mapped but still keep the extents unwritten. | |
69c499d1 | 3415 | * |
109811c2 JK |
3416 | * For non AIO case, we will convert those unwritten extents to written |
3417 | * after return back from blockdev_direct_IO. That way we save us from | |
3418 | * allocating io_end structure and also the overhead of offloading | |
3419 | * the extent convertion to a workqueue. | |
69c499d1 TT |
3420 | * |
3421 | * For async DIO, the conversion needs to be deferred when the | |
3422 | * IO is completed. The ext4 end_io callback function will be | |
3423 | * called to take care of the conversion work. Here for async | |
3424 | * case, we allocate an io_end structure to hook to the iocb. | |
3425 | */ | |
3426 | iocb->private = NULL; | |
109811c2 | 3427 | if (overwrite) |
705965bd | 3428 | get_block_func = ext4_dio_get_block_overwrite; |
12735f88 JK |
3429 | else if (IS_DAX(inode)) { |
3430 | /* | |
3431 | * We can avoid zeroing for aligned DAX writes beyond EOF. Other | |
3432 | * writes need zeroing either because they can race with page | |
3433 | * faults or because they use partial blocks. | |
3434 | */ | |
3435 | if (round_down(offset, 1<<inode->i_blkbits) >= inode->i_size && | |
3436 | ext4_aligned_io(inode, offset, count)) | |
3437 | get_block_func = ext4_dio_get_block; | |
3438 | else | |
3439 | get_block_func = ext4_dax_get_block; | |
3440 | dio_flags = DIO_LOCKING; | |
3441 | } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || | |
3442 | round_down(offset, 1 << inode->i_blkbits) >= inode->i_size) { | |
914f82a3 JK |
3443 | get_block_func = ext4_dio_get_block; |
3444 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
3445 | } else if (is_sync_kiocb(iocb)) { | |
109811c2 JK |
3446 | get_block_func = ext4_dio_get_block_unwritten_sync; |
3447 | dio_flags = DIO_LOCKING; | |
69c499d1 | 3448 | } else { |
109811c2 | 3449 | get_block_func = ext4_dio_get_block_unwritten_async; |
69c499d1 TT |
3450 | dio_flags = DIO_LOCKING; |
3451 | } | |
2058f83a MH |
3452 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3453 | BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); | |
3454 | #endif | |
914f82a3 | 3455 | if (IS_DAX(inode)) { |
c8b8e32d | 3456 | ret = dax_do_io(iocb, inode, iter, get_block_func, |
923ae0ff | 3457 | ext4_end_io_dio, dio_flags); |
914f82a3 | 3458 | } else |
17f8c842 | 3459 | ret = __blockdev_direct_IO(iocb, inode, |
c8b8e32d | 3460 | inode->i_sb->s_bdev, iter, |
923ae0ff RZ |
3461 | get_block_func, |
3462 | ext4_end_io_dio, NULL, dio_flags); | |
69c499d1 | 3463 | |
97a851ed | 3464 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
69c499d1 TT |
3465 | EXT4_STATE_DIO_UNWRITTEN)) { |
3466 | int err; | |
3467 | /* | |
3468 | * for non AIO case, since the IO is already | |
3469 | * completed, we could do the conversion right here | |
3470 | */ | |
6b523df4 | 3471 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3472 | offset, ret); |
3473 | if (err < 0) | |
3474 | ret = err; | |
3475 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3476 | } | |
4bd809db | 3477 | |
914f82a3 | 3478 | inode_dio_end(inode); |
69c499d1 | 3479 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3480 | if (overwrite) |
5955102c | 3481 | inode_lock(inode); |
8d5d02e6 | 3482 | |
914f82a3 JK |
3483 | if (ret < 0 && final_size > inode->i_size) |
3484 | ext4_truncate_failed_write(inode); | |
3485 | ||
3486 | /* Handle extending of i_size after direct IO write */ | |
3487 | if (orphan) { | |
3488 | int err; | |
3489 | ||
3490 | /* Credits for sb + inode write */ | |
3491 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3492 | if (IS_ERR(handle)) { | |
3493 | /* This is really bad luck. We've written the data | |
3494 | * but cannot extend i_size. Bail out and pretend | |
3495 | * the write failed... */ | |
3496 | ret = PTR_ERR(handle); | |
3497 | if (inode->i_nlink) | |
3498 | ext4_orphan_del(NULL, inode); | |
3499 | ||
3500 | goto out; | |
3501 | } | |
3502 | if (inode->i_nlink) | |
3503 | ext4_orphan_del(handle, inode); | |
3504 | if (ret > 0) { | |
3505 | loff_t end = offset + ret; | |
3506 | if (end > inode->i_size) { | |
3507 | ei->i_disksize = end; | |
3508 | i_size_write(inode, end); | |
3509 | /* | |
3510 | * We're going to return a positive `ret' | |
3511 | * here due to non-zero-length I/O, so there's | |
3512 | * no way of reporting error returns from | |
3513 | * ext4_mark_inode_dirty() to userspace. So | |
3514 | * ignore it. | |
3515 | */ | |
3516 | ext4_mark_inode_dirty(handle, inode); | |
3517 | } | |
3518 | } | |
3519 | err = ext4_journal_stop(handle); | |
3520 | if (ret == 0) | |
3521 | ret = err; | |
3522 | } | |
3523 | out: | |
3524 | return ret; | |
3525 | } | |
3526 | ||
0e01df10 | 3527 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
914f82a3 JK |
3528 | { |
3529 | int unlocked = 0; | |
3530 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
3531 | ssize_t ret; | |
3532 | ||
3533 | if (ext4_should_dioread_nolock(inode)) { | |
3534 | /* | |
3535 | * Nolock dioread optimization may be dynamically disabled | |
3536 | * via ext4_inode_block_unlocked_dio(). Check inode's state | |
3537 | * while holding extra i_dio_count ref. | |
3538 | */ | |
3539 | inode_dio_begin(inode); | |
3540 | smp_mb(); | |
3541 | if (unlikely(ext4_test_inode_state(inode, | |
3542 | EXT4_STATE_DIOREAD_LOCK))) | |
3543 | inode_dio_end(inode); | |
3544 | else | |
3545 | unlocked = 1; | |
3546 | } | |
3547 | if (IS_DAX(inode)) { | |
0e01df10 | 3548 | ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block, |
914f82a3 JK |
3549 | NULL, unlocked ? 0 : DIO_LOCKING); |
3550 | } else { | |
3551 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, | |
0e01df10 | 3552 | iter, ext4_dio_get_block, |
914f82a3 JK |
3553 | NULL, NULL, |
3554 | unlocked ? 0 : DIO_LOCKING); | |
3555 | } | |
3556 | if (unlocked) | |
3557 | inode_dio_end(inode); | |
69c499d1 | 3558 | return ret; |
4c0425ff MC |
3559 | } |
3560 | ||
c8b8e32d | 3561 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3562 | { |
3563 | struct file *file = iocb->ki_filp; | |
3564 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3565 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3566 | loff_t offset = iocb->ki_pos; |
0562e0ba | 3567 | ssize_t ret; |
4c0425ff | 3568 | |
2058f83a MH |
3569 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3570 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3571 | return 0; | |
3572 | #endif | |
3573 | ||
84ebd795 TT |
3574 | /* |
3575 | * If we are doing data journalling we don't support O_DIRECT | |
3576 | */ | |
3577 | if (ext4_should_journal_data(inode)) | |
3578 | return 0; | |
3579 | ||
46c7f254 TM |
3580 | /* Let buffer I/O handle the inline data case. */ |
3581 | if (ext4_has_inline_data(inode)) | |
3582 | return 0; | |
3583 | ||
6f673763 | 3584 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
914f82a3 | 3585 | if (iov_iter_rw(iter) == READ) |
0e01df10 | 3586 | ret = ext4_direct_IO_read(iocb, iter); |
0562e0ba | 3587 | else |
0e01df10 | 3588 | ret = ext4_direct_IO_write(iocb, iter); |
6f673763 | 3589 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
0562e0ba | 3590 | return ret; |
4c0425ff MC |
3591 | } |
3592 | ||
ac27a0ec | 3593 | /* |
617ba13b | 3594 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3595 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3596 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3597 | * not necessarily locked. | |
3598 | * | |
3599 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3600 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3601 | * or jbddirty because all the journalling code will explode. | |
3602 | * | |
3603 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3604 | * is called, propagate that into the buffers appropriately. | |
3605 | */ | |
617ba13b | 3606 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3607 | { |
3608 | SetPageChecked(page); | |
3609 | return __set_page_dirty_nobuffers(page); | |
3610 | } | |
3611 | ||
74d553aa | 3612 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3613 | .readpage = ext4_readpage, |
3614 | .readpages = ext4_readpages, | |
43ce1d23 | 3615 | .writepage = ext4_writepage, |
20970ba6 | 3616 | .writepages = ext4_writepages, |
8ab22b9a | 3617 | .write_begin = ext4_write_begin, |
74d553aa | 3618 | .write_end = ext4_write_end, |
8ab22b9a HH |
3619 | .bmap = ext4_bmap, |
3620 | .invalidatepage = ext4_invalidatepage, | |
3621 | .releasepage = ext4_releasepage, | |
3622 | .direct_IO = ext4_direct_IO, | |
3623 | .migratepage = buffer_migrate_page, | |
3624 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3625 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3626 | }; |
3627 | ||
617ba13b | 3628 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3629 | .readpage = ext4_readpage, |
3630 | .readpages = ext4_readpages, | |
43ce1d23 | 3631 | .writepage = ext4_writepage, |
20970ba6 | 3632 | .writepages = ext4_writepages, |
8ab22b9a HH |
3633 | .write_begin = ext4_write_begin, |
3634 | .write_end = ext4_journalled_write_end, | |
3635 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3636 | .bmap = ext4_bmap, | |
4520fb3c | 3637 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3638 | .releasepage = ext4_releasepage, |
84ebd795 | 3639 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3640 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3641 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3642 | }; |
3643 | ||
64769240 | 3644 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3645 | .readpage = ext4_readpage, |
3646 | .readpages = ext4_readpages, | |
43ce1d23 | 3647 | .writepage = ext4_writepage, |
20970ba6 | 3648 | .writepages = ext4_writepages, |
8ab22b9a HH |
3649 | .write_begin = ext4_da_write_begin, |
3650 | .write_end = ext4_da_write_end, | |
3651 | .bmap = ext4_bmap, | |
3652 | .invalidatepage = ext4_da_invalidatepage, | |
3653 | .releasepage = ext4_releasepage, | |
3654 | .direct_IO = ext4_direct_IO, | |
3655 | .migratepage = buffer_migrate_page, | |
3656 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3657 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3658 | }; |
3659 | ||
617ba13b | 3660 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3661 | { |
3d2b1582 LC |
3662 | switch (ext4_inode_journal_mode(inode)) { |
3663 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3664 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3665 | break; |
3666 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3667 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3668 | return; |
3d2b1582 LC |
3669 | default: |
3670 | BUG(); | |
3671 | } | |
74d553aa TT |
3672 | if (test_opt(inode->i_sb, DELALLOC)) |
3673 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3674 | else | |
3675 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3676 | } |
3677 | ||
923ae0ff | 3678 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3679 | struct address_space *mapping, loff_t from, loff_t length) |
3680 | { | |
09cbfeaf KS |
3681 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3682 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3683 | unsigned blocksize, pos; |
d863dc36 LC |
3684 | ext4_lblk_t iblock; |
3685 | struct inode *inode = mapping->host; | |
3686 | struct buffer_head *bh; | |
3687 | struct page *page; | |
3688 | int err = 0; | |
3689 | ||
09cbfeaf | 3690 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3691 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3692 | if (!page) |
3693 | return -ENOMEM; | |
3694 | ||
3695 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3696 | |
09cbfeaf | 3697 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3698 | |
3699 | if (!page_has_buffers(page)) | |
3700 | create_empty_buffers(page, blocksize, 0); | |
3701 | ||
3702 | /* Find the buffer that contains "offset" */ | |
3703 | bh = page_buffers(page); | |
3704 | pos = blocksize; | |
3705 | while (offset >= pos) { | |
3706 | bh = bh->b_this_page; | |
3707 | iblock++; | |
3708 | pos += blocksize; | |
3709 | } | |
d863dc36 LC |
3710 | if (buffer_freed(bh)) { |
3711 | BUFFER_TRACE(bh, "freed: skip"); | |
3712 | goto unlock; | |
3713 | } | |
d863dc36 LC |
3714 | if (!buffer_mapped(bh)) { |
3715 | BUFFER_TRACE(bh, "unmapped"); | |
3716 | ext4_get_block(inode, iblock, bh, 0); | |
3717 | /* unmapped? It's a hole - nothing to do */ | |
3718 | if (!buffer_mapped(bh)) { | |
3719 | BUFFER_TRACE(bh, "still unmapped"); | |
3720 | goto unlock; | |
3721 | } | |
3722 | } | |
3723 | ||
3724 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3725 | if (PageUptodate(page)) | |
3726 | set_buffer_uptodate(bh); | |
3727 | ||
3728 | if (!buffer_uptodate(bh)) { | |
3729 | err = -EIO; | |
dfec8a14 | 3730 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
3731 | wait_on_buffer(bh); |
3732 | /* Uhhuh. Read error. Complain and punt. */ | |
3733 | if (!buffer_uptodate(bh)) | |
3734 | goto unlock; | |
c9c7429c MH |
3735 | if (S_ISREG(inode->i_mode) && |
3736 | ext4_encrypted_inode(inode)) { | |
3737 | /* We expect the key to be set. */ | |
a7550b30 | 3738 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
09cbfeaf | 3739 | BUG_ON(blocksize != PAGE_SIZE); |
a7550b30 | 3740 | WARN_ON_ONCE(fscrypt_decrypt_page(page)); |
c9c7429c | 3741 | } |
d863dc36 | 3742 | } |
d863dc36 LC |
3743 | if (ext4_should_journal_data(inode)) { |
3744 | BUFFER_TRACE(bh, "get write access"); | |
3745 | err = ext4_journal_get_write_access(handle, bh); | |
3746 | if (err) | |
3747 | goto unlock; | |
3748 | } | |
d863dc36 | 3749 | zero_user(page, offset, length); |
d863dc36 LC |
3750 | BUFFER_TRACE(bh, "zeroed end of block"); |
3751 | ||
d863dc36 LC |
3752 | if (ext4_should_journal_data(inode)) { |
3753 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3754 | } else { |
353eefd3 | 3755 | err = 0; |
d863dc36 | 3756 | mark_buffer_dirty(bh); |
3957ef53 | 3757 | if (ext4_should_order_data(inode)) |
ee0876bc | 3758 | err = ext4_jbd2_inode_add_write(handle, inode); |
0713ed0c | 3759 | } |
d863dc36 LC |
3760 | |
3761 | unlock: | |
3762 | unlock_page(page); | |
09cbfeaf | 3763 | put_page(page); |
d863dc36 LC |
3764 | return err; |
3765 | } | |
3766 | ||
923ae0ff RZ |
3767 | /* |
3768 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3769 | * starting from file offset 'from'. The range to be zero'd must | |
3770 | * be contained with in one block. If the specified range exceeds | |
3771 | * the end of the block it will be shortened to end of the block | |
3772 | * that cooresponds to 'from' | |
3773 | */ | |
3774 | static int ext4_block_zero_page_range(handle_t *handle, | |
3775 | struct address_space *mapping, loff_t from, loff_t length) | |
3776 | { | |
3777 | struct inode *inode = mapping->host; | |
09cbfeaf | 3778 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3779 | unsigned blocksize = inode->i_sb->s_blocksize; |
3780 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3781 | ||
3782 | /* | |
3783 | * correct length if it does not fall between | |
3784 | * 'from' and the end of the block | |
3785 | */ | |
3786 | if (length > max || length < 0) | |
3787 | length = max; | |
3788 | ||
3789 | if (IS_DAX(inode)) | |
3790 | return dax_zero_page_range(inode, from, length, ext4_get_block); | |
3791 | return __ext4_block_zero_page_range(handle, mapping, from, length); | |
3792 | } | |
3793 | ||
94350ab5 MW |
3794 | /* |
3795 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3796 | * up to the end of the block which corresponds to `from'. | |
3797 | * This required during truncate. We need to physically zero the tail end | |
3798 | * of that block so it doesn't yield old data if the file is later grown. | |
3799 | */ | |
c197855e | 3800 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3801 | struct address_space *mapping, loff_t from) |
3802 | { | |
09cbfeaf | 3803 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3804 | unsigned length; |
3805 | unsigned blocksize; | |
3806 | struct inode *inode = mapping->host; | |
3807 | ||
3808 | blocksize = inode->i_sb->s_blocksize; | |
3809 | length = blocksize - (offset & (blocksize - 1)); | |
3810 | ||
3811 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3812 | } | |
3813 | ||
a87dd18c LC |
3814 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3815 | loff_t lstart, loff_t length) | |
3816 | { | |
3817 | struct super_block *sb = inode->i_sb; | |
3818 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3819 | unsigned partial_start, partial_end; |
a87dd18c LC |
3820 | ext4_fsblk_t start, end; |
3821 | loff_t byte_end = (lstart + length - 1); | |
3822 | int err = 0; | |
3823 | ||
e1be3a92 LC |
3824 | partial_start = lstart & (sb->s_blocksize - 1); |
3825 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3826 | ||
a87dd18c LC |
3827 | start = lstart >> sb->s_blocksize_bits; |
3828 | end = byte_end >> sb->s_blocksize_bits; | |
3829 | ||
3830 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3831 | if (start == end && |
3832 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3833 | err = ext4_block_zero_page_range(handle, mapping, |
3834 | lstart, length); | |
3835 | return err; | |
3836 | } | |
3837 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3838 | if (partial_start) { |
a87dd18c LC |
3839 | err = ext4_block_zero_page_range(handle, mapping, |
3840 | lstart, sb->s_blocksize); | |
3841 | if (err) | |
3842 | return err; | |
3843 | } | |
3844 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3845 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3846 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3847 | byte_end - partial_end, |
3848 | partial_end + 1); | |
a87dd18c LC |
3849 | return err; |
3850 | } | |
3851 | ||
91ef4caf DG |
3852 | int ext4_can_truncate(struct inode *inode) |
3853 | { | |
91ef4caf DG |
3854 | if (S_ISREG(inode->i_mode)) |
3855 | return 1; | |
3856 | if (S_ISDIR(inode->i_mode)) | |
3857 | return 1; | |
3858 | if (S_ISLNK(inode->i_mode)) | |
3859 | return !ext4_inode_is_fast_symlink(inode); | |
3860 | return 0; | |
3861 | } | |
3862 | ||
01127848 JK |
3863 | /* |
3864 | * We have to make sure i_disksize gets properly updated before we truncate | |
3865 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3866 | * can get lost as it may have been postponed to submission of writeback but | |
3867 | * that will never happen after we truncate page cache. | |
3868 | */ | |
3869 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3870 | loff_t len) | |
3871 | { | |
3872 | handle_t *handle; | |
3873 | loff_t size = i_size_read(inode); | |
3874 | ||
5955102c | 3875 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3876 | if (offset > size || offset + len < size) |
3877 | return 0; | |
3878 | ||
3879 | if (EXT4_I(inode)->i_disksize >= size) | |
3880 | return 0; | |
3881 | ||
3882 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3883 | if (IS_ERR(handle)) | |
3884 | return PTR_ERR(handle); | |
3885 | ext4_update_i_disksize(inode, size); | |
3886 | ext4_mark_inode_dirty(handle, inode); | |
3887 | ext4_journal_stop(handle); | |
3888 | ||
3889 | return 0; | |
3890 | } | |
3891 | ||
a4bb6b64 AH |
3892 | /* |
3893 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3894 | * associated with the given offset and length | |
3895 | * | |
3896 | * @inode: File inode | |
3897 | * @offset: The offset where the hole will begin | |
3898 | * @len: The length of the hole | |
3899 | * | |
4907cb7b | 3900 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3901 | */ |
3902 | ||
aeb2817a | 3903 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3904 | { |
26a4c0c6 TT |
3905 | struct super_block *sb = inode->i_sb; |
3906 | ext4_lblk_t first_block, stop_block; | |
3907 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3908 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3909 | handle_t *handle; |
3910 | unsigned int credits; | |
3911 | int ret = 0; | |
3912 | ||
a4bb6b64 | 3913 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3914 | return -EOPNOTSUPP; |
a4bb6b64 | 3915 | |
b8a86845 | 3916 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3917 | |
26a4c0c6 TT |
3918 | /* |
3919 | * Write out all dirty pages to avoid race conditions | |
3920 | * Then release them. | |
3921 | */ | |
3922 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3923 | ret = filemap_write_and_wait_range(mapping, offset, | |
3924 | offset + length - 1); | |
3925 | if (ret) | |
3926 | return ret; | |
3927 | } | |
3928 | ||
5955102c | 3929 | inode_lock(inode); |
9ef06cec | 3930 | |
26a4c0c6 TT |
3931 | /* No need to punch hole beyond i_size */ |
3932 | if (offset >= inode->i_size) | |
3933 | goto out_mutex; | |
3934 | ||
3935 | /* | |
3936 | * If the hole extends beyond i_size, set the hole | |
3937 | * to end after the page that contains i_size | |
3938 | */ | |
3939 | if (offset + length > inode->i_size) { | |
3940 | length = inode->i_size + | |
09cbfeaf | 3941 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
3942 | offset; |
3943 | } | |
3944 | ||
a361293f JK |
3945 | if (offset & (sb->s_blocksize - 1) || |
3946 | (offset + length) & (sb->s_blocksize - 1)) { | |
3947 | /* | |
3948 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3949 | * partial block | |
3950 | */ | |
3951 | ret = ext4_inode_attach_jinode(inode); | |
3952 | if (ret < 0) | |
3953 | goto out_mutex; | |
3954 | ||
3955 | } | |
3956 | ||
ea3d7209 JK |
3957 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
3958 | ext4_inode_block_unlocked_dio(inode); | |
3959 | inode_dio_wait(inode); | |
3960 | ||
3961 | /* | |
3962 | * Prevent page faults from reinstantiating pages we have released from | |
3963 | * page cache. | |
3964 | */ | |
3965 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
3966 | first_block_offset = round_up(offset, sb->s_blocksize); |
3967 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3968 | |
a87dd18c | 3969 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
3970 | if (last_block_offset > first_block_offset) { |
3971 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
3972 | if (ret) | |
3973 | goto out_dio; | |
a87dd18c LC |
3974 | truncate_pagecache_range(inode, first_block_offset, |
3975 | last_block_offset); | |
01127848 | 3976 | } |
26a4c0c6 TT |
3977 | |
3978 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3979 | credits = ext4_writepage_trans_blocks(inode); | |
3980 | else | |
3981 | credits = ext4_blocks_for_truncate(inode); | |
3982 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3983 | if (IS_ERR(handle)) { | |
3984 | ret = PTR_ERR(handle); | |
3985 | ext4_std_error(sb, ret); | |
3986 | goto out_dio; | |
3987 | } | |
3988 | ||
a87dd18c LC |
3989 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3990 | length); | |
3991 | if (ret) | |
3992 | goto out_stop; | |
26a4c0c6 TT |
3993 | |
3994 | first_block = (offset + sb->s_blocksize - 1) >> | |
3995 | EXT4_BLOCK_SIZE_BITS(sb); | |
3996 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3997 | ||
3998 | /* If there are no blocks to remove, return now */ | |
3999 | if (first_block >= stop_block) | |
4000 | goto out_stop; | |
4001 | ||
4002 | down_write(&EXT4_I(inode)->i_data_sem); | |
4003 | ext4_discard_preallocations(inode); | |
4004 | ||
4005 | ret = ext4_es_remove_extent(inode, first_block, | |
4006 | stop_block - first_block); | |
4007 | if (ret) { | |
4008 | up_write(&EXT4_I(inode)->i_data_sem); | |
4009 | goto out_stop; | |
4010 | } | |
4011 | ||
4012 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4013 | ret = ext4_ext_remove_space(inode, first_block, | |
4014 | stop_block - 1); | |
4015 | else | |
4f579ae7 | 4016 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
4017 | stop_block); |
4018 | ||
819c4920 | 4019 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
4020 | if (IS_SYNC(inode)) |
4021 | ext4_handle_sync(handle); | |
e251f9bc | 4022 | |
26a4c0c6 TT |
4023 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4024 | ext4_mark_inode_dirty(handle, inode); | |
4025 | out_stop: | |
4026 | ext4_journal_stop(handle); | |
4027 | out_dio: | |
ea3d7209 | 4028 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 TT |
4029 | ext4_inode_resume_unlocked_dio(inode); |
4030 | out_mutex: | |
5955102c | 4031 | inode_unlock(inode); |
26a4c0c6 | 4032 | return ret; |
a4bb6b64 AH |
4033 | } |
4034 | ||
a361293f JK |
4035 | int ext4_inode_attach_jinode(struct inode *inode) |
4036 | { | |
4037 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4038 | struct jbd2_inode *jinode; | |
4039 | ||
4040 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4041 | return 0; | |
4042 | ||
4043 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4044 | spin_lock(&inode->i_lock); | |
4045 | if (!ei->jinode) { | |
4046 | if (!jinode) { | |
4047 | spin_unlock(&inode->i_lock); | |
4048 | return -ENOMEM; | |
4049 | } | |
4050 | ei->jinode = jinode; | |
4051 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4052 | jinode = NULL; | |
4053 | } | |
4054 | spin_unlock(&inode->i_lock); | |
4055 | if (unlikely(jinode != NULL)) | |
4056 | jbd2_free_inode(jinode); | |
4057 | return 0; | |
4058 | } | |
4059 | ||
ac27a0ec | 4060 | /* |
617ba13b | 4061 | * ext4_truncate() |
ac27a0ec | 4062 | * |
617ba13b MC |
4063 | * We block out ext4_get_block() block instantiations across the entire |
4064 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4065 | * simultaneously on behalf of the same inode. |
4066 | * | |
42b2aa86 | 4067 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4068 | * is one core, guiding principle: the file's tree must always be consistent on |
4069 | * disk. We must be able to restart the truncate after a crash. | |
4070 | * | |
4071 | * The file's tree may be transiently inconsistent in memory (although it | |
4072 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4073 | * the contents of (the filesystem + the journal) must be consistent and | |
4074 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4075 | * left-to-right works OK too). | |
4076 | * | |
4077 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4078 | * truncate against the orphan inode list. | |
4079 | * | |
4080 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4081 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4082 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4083 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4084 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4085 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4086 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4087 | */ |
617ba13b | 4088 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 4089 | { |
819c4920 TT |
4090 | struct ext4_inode_info *ei = EXT4_I(inode); |
4091 | unsigned int credits; | |
4092 | handle_t *handle; | |
4093 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4094 | |
19b5ef61 TT |
4095 | /* |
4096 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4097 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4098 | * have i_mutex locked because it's not necessary. |
4099 | */ | |
4100 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4101 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4102 | trace_ext4_truncate_enter(inode); |
4103 | ||
91ef4caf | 4104 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4105 | return; |
4106 | ||
12e9b892 | 4107 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4108 | |
5534fb5b | 4109 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4110 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4111 | |
aef1c851 TM |
4112 | if (ext4_has_inline_data(inode)) { |
4113 | int has_inline = 1; | |
4114 | ||
4115 | ext4_inline_data_truncate(inode, &has_inline); | |
4116 | if (has_inline) | |
4117 | return; | |
4118 | } | |
4119 | ||
a361293f JK |
4120 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4121 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4122 | if (ext4_inode_attach_jinode(inode) < 0) | |
4123 | return; | |
4124 | } | |
4125 | ||
819c4920 TT |
4126 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4127 | credits = ext4_writepage_trans_blocks(inode); | |
4128 | else | |
4129 | credits = ext4_blocks_for_truncate(inode); | |
4130 | ||
4131 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4132 | if (IS_ERR(handle)) { | |
4133 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
4134 | return; | |
4135 | } | |
4136 | ||
eb3544c6 LC |
4137 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4138 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4139 | |
4140 | /* | |
4141 | * We add the inode to the orphan list, so that if this | |
4142 | * truncate spans multiple transactions, and we crash, we will | |
4143 | * resume the truncate when the filesystem recovers. It also | |
4144 | * marks the inode dirty, to catch the new size. | |
4145 | * | |
4146 | * Implication: the file must always be in a sane, consistent | |
4147 | * truncatable state while each transaction commits. | |
4148 | */ | |
4149 | if (ext4_orphan_add(handle, inode)) | |
4150 | goto out_stop; | |
4151 | ||
4152 | down_write(&EXT4_I(inode)->i_data_sem); | |
4153 | ||
4154 | ext4_discard_preallocations(inode); | |
4155 | ||
ff9893dc | 4156 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 4157 | ext4_ext_truncate(handle, inode); |
ff9893dc | 4158 | else |
819c4920 TT |
4159 | ext4_ind_truncate(handle, inode); |
4160 | ||
4161 | up_write(&ei->i_data_sem); | |
4162 | ||
4163 | if (IS_SYNC(inode)) | |
4164 | ext4_handle_sync(handle); | |
4165 | ||
4166 | out_stop: | |
4167 | /* | |
4168 | * If this was a simple ftruncate() and the file will remain alive, | |
4169 | * then we need to clear up the orphan record which we created above. | |
4170 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4171 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4172 | * orphan info for us. |
4173 | */ | |
4174 | if (inode->i_nlink) | |
4175 | ext4_orphan_del(handle, inode); | |
4176 | ||
4177 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
4178 | ext4_mark_inode_dirty(handle, inode); | |
4179 | ext4_journal_stop(handle); | |
ac27a0ec | 4180 | |
0562e0ba | 4181 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
4182 | } |
4183 | ||
ac27a0ec | 4184 | /* |
617ba13b | 4185 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4186 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4187 | * data in memory that is needed to recreate the on-disk version of this | |
4188 | * inode. | |
4189 | */ | |
617ba13b MC |
4190 | static int __ext4_get_inode_loc(struct inode *inode, |
4191 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4192 | { |
240799cd TT |
4193 | struct ext4_group_desc *gdp; |
4194 | struct buffer_head *bh; | |
4195 | struct super_block *sb = inode->i_sb; | |
4196 | ext4_fsblk_t block; | |
4197 | int inodes_per_block, inode_offset; | |
4198 | ||
3a06d778 | 4199 | iloc->bh = NULL; |
240799cd | 4200 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 4201 | return -EFSCORRUPTED; |
ac27a0ec | 4202 | |
240799cd TT |
4203 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4204 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4205 | if (!gdp) | |
ac27a0ec DK |
4206 | return -EIO; |
4207 | ||
240799cd TT |
4208 | /* |
4209 | * Figure out the offset within the block group inode table | |
4210 | */ | |
00d09882 | 4211 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4212 | inode_offset = ((inode->i_ino - 1) % |
4213 | EXT4_INODES_PER_GROUP(sb)); | |
4214 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4215 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4216 | ||
4217 | bh = sb_getblk(sb, block); | |
aebf0243 | 4218 | if (unlikely(!bh)) |
860d21e2 | 4219 | return -ENOMEM; |
ac27a0ec DK |
4220 | if (!buffer_uptodate(bh)) { |
4221 | lock_buffer(bh); | |
9c83a923 HK |
4222 | |
4223 | /* | |
4224 | * If the buffer has the write error flag, we have failed | |
4225 | * to write out another inode in the same block. In this | |
4226 | * case, we don't have to read the block because we may | |
4227 | * read the old inode data successfully. | |
4228 | */ | |
4229 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4230 | set_buffer_uptodate(bh); | |
4231 | ||
ac27a0ec DK |
4232 | if (buffer_uptodate(bh)) { |
4233 | /* someone brought it uptodate while we waited */ | |
4234 | unlock_buffer(bh); | |
4235 | goto has_buffer; | |
4236 | } | |
4237 | ||
4238 | /* | |
4239 | * If we have all information of the inode in memory and this | |
4240 | * is the only valid inode in the block, we need not read the | |
4241 | * block. | |
4242 | */ | |
4243 | if (in_mem) { | |
4244 | struct buffer_head *bitmap_bh; | |
240799cd | 4245 | int i, start; |
ac27a0ec | 4246 | |
240799cd | 4247 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4248 | |
240799cd TT |
4249 | /* Is the inode bitmap in cache? */ |
4250 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4251 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4252 | goto make_io; |
4253 | ||
4254 | /* | |
4255 | * If the inode bitmap isn't in cache then the | |
4256 | * optimisation may end up performing two reads instead | |
4257 | * of one, so skip it. | |
4258 | */ | |
4259 | if (!buffer_uptodate(bitmap_bh)) { | |
4260 | brelse(bitmap_bh); | |
4261 | goto make_io; | |
4262 | } | |
240799cd | 4263 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4264 | if (i == inode_offset) |
4265 | continue; | |
617ba13b | 4266 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4267 | break; |
4268 | } | |
4269 | brelse(bitmap_bh); | |
240799cd | 4270 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4271 | /* all other inodes are free, so skip I/O */ |
4272 | memset(bh->b_data, 0, bh->b_size); | |
4273 | set_buffer_uptodate(bh); | |
4274 | unlock_buffer(bh); | |
4275 | goto has_buffer; | |
4276 | } | |
4277 | } | |
4278 | ||
4279 | make_io: | |
240799cd TT |
4280 | /* |
4281 | * If we need to do any I/O, try to pre-readahead extra | |
4282 | * blocks from the inode table. | |
4283 | */ | |
4284 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4285 | ext4_fsblk_t b, end, table; | |
4286 | unsigned num; | |
0d606e2c | 4287 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4288 | |
4289 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4290 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4291 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4292 | if (table > b) |
4293 | b = table; | |
0d606e2c | 4294 | end = b + ra_blks; |
240799cd | 4295 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4296 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4297 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4298 | table += num / inodes_per_block; |
4299 | if (end > table) | |
4300 | end = table; | |
4301 | while (b <= end) | |
4302 | sb_breadahead(sb, b++); | |
4303 | } | |
4304 | ||
ac27a0ec DK |
4305 | /* |
4306 | * There are other valid inodes in the buffer, this inode | |
4307 | * has in-inode xattrs, or we don't have this inode in memory. | |
4308 | * Read the block from disk. | |
4309 | */ | |
0562e0ba | 4310 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4311 | get_bh(bh); |
4312 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4313 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4314 | wait_on_buffer(bh); |
4315 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4316 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4317 | "unable to read itable block"); | |
ac27a0ec DK |
4318 | brelse(bh); |
4319 | return -EIO; | |
4320 | } | |
4321 | } | |
4322 | has_buffer: | |
4323 | iloc->bh = bh; | |
4324 | return 0; | |
4325 | } | |
4326 | ||
617ba13b | 4327 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4328 | { |
4329 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4330 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4331 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4332 | } |
4333 | ||
617ba13b | 4334 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4335 | { |
617ba13b | 4336 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4337 | unsigned int new_fl = 0; |
ac27a0ec | 4338 | |
617ba13b | 4339 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4340 | new_fl |= S_SYNC; |
617ba13b | 4341 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4342 | new_fl |= S_APPEND; |
617ba13b | 4343 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4344 | new_fl |= S_IMMUTABLE; |
617ba13b | 4345 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4346 | new_fl |= S_NOATIME; |
617ba13b | 4347 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4348 | new_fl |= S_DIRSYNC; |
0a6cf913 | 4349 | if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode)) |
923ae0ff | 4350 | new_fl |= S_DAX; |
5f16f322 | 4351 | inode_set_flags(inode, new_fl, |
923ae0ff | 4352 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
4353 | } |
4354 | ||
ff9ddf7e JK |
4355 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4356 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4357 | { | |
84a8dce2 DM |
4358 | unsigned int vfs_fl; |
4359 | unsigned long old_fl, new_fl; | |
4360 | ||
4361 | do { | |
4362 | vfs_fl = ei->vfs_inode.i_flags; | |
4363 | old_fl = ei->i_flags; | |
4364 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4365 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4366 | EXT4_DIRSYNC_FL); | |
4367 | if (vfs_fl & S_SYNC) | |
4368 | new_fl |= EXT4_SYNC_FL; | |
4369 | if (vfs_fl & S_APPEND) | |
4370 | new_fl |= EXT4_APPEND_FL; | |
4371 | if (vfs_fl & S_IMMUTABLE) | |
4372 | new_fl |= EXT4_IMMUTABLE_FL; | |
4373 | if (vfs_fl & S_NOATIME) | |
4374 | new_fl |= EXT4_NOATIME_FL; | |
4375 | if (vfs_fl & S_DIRSYNC) | |
4376 | new_fl |= EXT4_DIRSYNC_FL; | |
4377 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4378 | } |
de9a55b8 | 4379 | |
0fc1b451 | 4380 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4381 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4382 | { |
4383 | blkcnt_t i_blocks ; | |
8180a562 AK |
4384 | struct inode *inode = &(ei->vfs_inode); |
4385 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4386 | |
e2b911c5 | 4387 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4388 | /* we are using combined 48 bit field */ |
4389 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4390 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4391 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4392 | /* i_blocks represent file system block size */ |
4393 | return i_blocks << (inode->i_blkbits - 9); | |
4394 | } else { | |
4395 | return i_blocks; | |
4396 | } | |
0fc1b451 AK |
4397 | } else { |
4398 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4399 | } | |
4400 | } | |
ff9ddf7e | 4401 | |
152a7b0a TM |
4402 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4403 | struct ext4_inode *raw_inode, | |
4404 | struct ext4_inode_info *ei) | |
4405 | { | |
4406 | __le32 *magic = (void *)raw_inode + | |
4407 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4408 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4409 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4410 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4411 | } else |
4412 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4413 | } |
4414 | ||
040cb378 LX |
4415 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4416 | { | |
c2950744 | 4417 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4418 | return -EOPNOTSUPP; |
4419 | *projid = EXT4_I(inode)->i_projid; | |
4420 | return 0; | |
4421 | } | |
4422 | ||
1d1fe1ee | 4423 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4424 | { |
617ba13b MC |
4425 | struct ext4_iloc iloc; |
4426 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4427 | struct ext4_inode_info *ei; |
1d1fe1ee | 4428 | struct inode *inode; |
b436b9be | 4429 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4430 | long ret; |
ac27a0ec | 4431 | int block; |
08cefc7a EB |
4432 | uid_t i_uid; |
4433 | gid_t i_gid; | |
040cb378 | 4434 | projid_t i_projid; |
ac27a0ec | 4435 | |
1d1fe1ee DH |
4436 | inode = iget_locked(sb, ino); |
4437 | if (!inode) | |
4438 | return ERR_PTR(-ENOMEM); | |
4439 | if (!(inode->i_state & I_NEW)) | |
4440 | return inode; | |
4441 | ||
4442 | ei = EXT4_I(inode); | |
7dc57615 | 4443 | iloc.bh = NULL; |
ac27a0ec | 4444 | |
1d1fe1ee DH |
4445 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4446 | if (ret < 0) | |
ac27a0ec | 4447 | goto bad_inode; |
617ba13b | 4448 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4449 | |
4450 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4451 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4452 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4453 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4454 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4455 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4456 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4457 | ret = -EFSCORRUPTED; |
814525f4 DW |
4458 | goto bad_inode; |
4459 | } | |
4460 | } else | |
4461 | ei->i_extra_isize = 0; | |
4462 | ||
4463 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4464 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4465 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4466 | __u32 csum; | |
4467 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4468 | __le32 gen = raw_inode->i_generation; | |
4469 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4470 | sizeof(inum)); | |
4471 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4472 | sizeof(gen)); | |
4473 | } | |
4474 | ||
4475 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4476 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4477 | ret = -EFSBADCRC; |
814525f4 DW |
4478 | goto bad_inode; |
4479 | } | |
4480 | ||
ac27a0ec | 4481 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4482 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4483 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
c2950744 | 4484 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4485 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4486 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4487 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4488 | else | |
4489 | i_projid = EXT4_DEF_PROJID; | |
4490 | ||
af5bc92d | 4491 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4492 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4493 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4494 | } |
08cefc7a EB |
4495 | i_uid_write(inode, i_uid); |
4496 | i_gid_write(inode, i_gid); | |
040cb378 | 4497 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4498 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4499 | |
353eb83c | 4500 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4501 | ei->i_inline_off = 0; |
ac27a0ec DK |
4502 | ei->i_dir_start_lookup = 0; |
4503 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4504 | /* We now have enough fields to check if the inode was active or not. | |
4505 | * This is needed because nfsd might try to access dead inodes | |
4506 | * the test is that same one that e2fsck uses | |
4507 | * NeilBrown 1999oct15 | |
4508 | */ | |
4509 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4510 | if ((inode->i_mode == 0 || |
4511 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4512 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4513 | /* this inode is deleted */ |
1d1fe1ee | 4514 | ret = -ESTALE; |
ac27a0ec DK |
4515 | goto bad_inode; |
4516 | } | |
4517 | /* The only unlinked inodes we let through here have | |
4518 | * valid i_mode and are being read by the orphan | |
4519 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4520 | * the process of deleting those. |
4521 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4522 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4523 | } |
ac27a0ec | 4524 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4525 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4526 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4527 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4528 | ei->i_file_acl |= |
4529 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4530 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4531 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4532 | #ifdef CONFIG_QUOTA |
4533 | ei->i_reserved_quota = 0; | |
4534 | #endif | |
ac27a0ec DK |
4535 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4536 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4537 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4538 | /* |
4539 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4540 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4541 | */ | |
617ba13b | 4542 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4543 | ei->i_data[block] = raw_inode->i_block[block]; |
4544 | INIT_LIST_HEAD(&ei->i_orphan); | |
4545 | ||
b436b9be JK |
4546 | /* |
4547 | * Set transaction id's of transactions that have to be committed | |
4548 | * to finish f[data]sync. We set them to currently running transaction | |
4549 | * as we cannot be sure that the inode or some of its metadata isn't | |
4550 | * part of the transaction - the inode could have been reclaimed and | |
4551 | * now it is reread from disk. | |
4552 | */ | |
4553 | if (journal) { | |
4554 | transaction_t *transaction; | |
4555 | tid_t tid; | |
4556 | ||
a931da6a | 4557 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4558 | if (journal->j_running_transaction) |
4559 | transaction = journal->j_running_transaction; | |
4560 | else | |
4561 | transaction = journal->j_committing_transaction; | |
4562 | if (transaction) | |
4563 | tid = transaction->t_tid; | |
4564 | else | |
4565 | tid = journal->j_commit_sequence; | |
a931da6a | 4566 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4567 | ei->i_sync_tid = tid; |
4568 | ei->i_datasync_tid = tid; | |
4569 | } | |
4570 | ||
0040d987 | 4571 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4572 | if (ei->i_extra_isize == 0) { |
4573 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4574 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4575 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4576 | } else { |
152a7b0a | 4577 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4578 | } |
814525f4 | 4579 | } |
ac27a0ec | 4580 | |
ef7f3835 KS |
4581 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4582 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4583 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4584 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4585 | ||
ed3654eb | 4586 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4587 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4588 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4589 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4590 | inode->i_version |= | |
4591 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4592 | } | |
25ec56b5 JNC |
4593 | } |
4594 | ||
c4b5a614 | 4595 | ret = 0; |
485c26ec | 4596 | if (ei->i_file_acl && |
1032988c | 4597 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4598 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4599 | ei->i_file_acl); | |
6a797d27 | 4600 | ret = -EFSCORRUPTED; |
485c26ec | 4601 | goto bad_inode; |
f19d5870 TM |
4602 | } else if (!ext4_has_inline_data(inode)) { |
4603 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4604 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4605 | (S_ISLNK(inode->i_mode) && | |
4606 | !ext4_inode_is_fast_symlink(inode)))) | |
4607 | /* Validate extent which is part of inode */ | |
4608 | ret = ext4_ext_check_inode(inode); | |
4609 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4610 | (S_ISLNK(inode->i_mode) && | |
4611 | !ext4_inode_is_fast_symlink(inode))) { | |
4612 | /* Validate block references which are part of inode */ | |
4613 | ret = ext4_ind_check_inode(inode); | |
4614 | } | |
fe2c8191 | 4615 | } |
567f3e9a | 4616 | if (ret) |
de9a55b8 | 4617 | goto bad_inode; |
7a262f7c | 4618 | |
ac27a0ec | 4619 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4620 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4621 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4622 | ext4_set_aops(inode); |
ac27a0ec | 4623 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4624 | inode->i_op = &ext4_dir_inode_operations; |
4625 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4626 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4627 | if (ext4_encrypted_inode(inode)) { |
4628 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4629 | ext4_set_aops(inode); | |
4630 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4631 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4632 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4633 | nd_terminate_link(ei->i_data, inode->i_size, |
4634 | sizeof(ei->i_data) - 1); | |
4635 | } else { | |
617ba13b MC |
4636 | inode->i_op = &ext4_symlink_inode_operations; |
4637 | ext4_set_aops(inode); | |
ac27a0ec | 4638 | } |
21fc61c7 | 4639 | inode_nohighmem(inode); |
563bdd61 TT |
4640 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4641 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4642 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4643 | if (raw_inode->i_block[0]) |
4644 | init_special_inode(inode, inode->i_mode, | |
4645 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4646 | else | |
4647 | init_special_inode(inode, inode->i_mode, | |
4648 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4649 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4650 | make_bad_inode(inode); | |
563bdd61 | 4651 | } else { |
6a797d27 | 4652 | ret = -EFSCORRUPTED; |
24676da4 | 4653 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4654 | goto bad_inode; |
ac27a0ec | 4655 | } |
af5bc92d | 4656 | brelse(iloc.bh); |
617ba13b | 4657 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4658 | unlock_new_inode(inode); |
4659 | return inode; | |
ac27a0ec DK |
4660 | |
4661 | bad_inode: | |
567f3e9a | 4662 | brelse(iloc.bh); |
1d1fe1ee DH |
4663 | iget_failed(inode); |
4664 | return ERR_PTR(ret); | |
ac27a0ec DK |
4665 | } |
4666 | ||
f4bb2981 TT |
4667 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4668 | { | |
4669 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4670 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4671 | return ext4_iget(sb, ino); |
4672 | } | |
4673 | ||
0fc1b451 AK |
4674 | static int ext4_inode_blocks_set(handle_t *handle, |
4675 | struct ext4_inode *raw_inode, | |
4676 | struct ext4_inode_info *ei) | |
4677 | { | |
4678 | struct inode *inode = &(ei->vfs_inode); | |
4679 | u64 i_blocks = inode->i_blocks; | |
4680 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4681 | |
4682 | if (i_blocks <= ~0U) { | |
4683 | /* | |
4907cb7b | 4684 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4685 | * as multiple of 512 bytes |
4686 | */ | |
8180a562 | 4687 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4688 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4689 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4690 | return 0; |
4691 | } | |
e2b911c5 | 4692 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4693 | return -EFBIG; |
4694 | ||
4695 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4696 | /* |
4697 | * i_blocks can be represented in a 48 bit variable | |
4698 | * as multiple of 512 bytes | |
4699 | */ | |
8180a562 | 4700 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4701 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4702 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4703 | } else { |
84a8dce2 | 4704 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4705 | /* i_block is stored in file system block size */ |
4706 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4707 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4708 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4709 | } |
f287a1a5 | 4710 | return 0; |
0fc1b451 AK |
4711 | } |
4712 | ||
a26f4992 TT |
4713 | struct other_inode { |
4714 | unsigned long orig_ino; | |
4715 | struct ext4_inode *raw_inode; | |
4716 | }; | |
4717 | ||
4718 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4719 | void *data) | |
4720 | { | |
4721 | struct other_inode *oi = (struct other_inode *) data; | |
4722 | ||
4723 | if ((inode->i_ino != ino) || | |
4724 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4725 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4726 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4727 | return 0; | |
4728 | spin_lock(&inode->i_lock); | |
4729 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4730 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4731 | (inode->i_state & I_DIRTY_TIME)) { | |
4732 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4733 | ||
4734 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4735 | spin_unlock(&inode->i_lock); | |
4736 | ||
4737 | spin_lock(&ei->i_raw_lock); | |
4738 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4739 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4740 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4741 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4742 | spin_unlock(&ei->i_raw_lock); | |
4743 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4744 | return -1; | |
4745 | } | |
4746 | spin_unlock(&inode->i_lock); | |
4747 | return -1; | |
4748 | } | |
4749 | ||
4750 | /* | |
4751 | * Opportunistically update the other time fields for other inodes in | |
4752 | * the same inode table block. | |
4753 | */ | |
4754 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4755 | unsigned long orig_ino, char *buf) | |
4756 | { | |
4757 | struct other_inode oi; | |
4758 | unsigned long ino; | |
4759 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4760 | int inode_size = EXT4_INODE_SIZE(sb); | |
4761 | ||
4762 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
4763 | /* |
4764 | * Calculate the first inode in the inode table block. Inode | |
4765 | * numbers are one-based. That is, the first inode in a block | |
4766 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4767 | */ | |
4768 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
4769 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4770 | if (ino == orig_ino) | |
4771 | continue; | |
4772 | oi.raw_inode = (struct ext4_inode *) buf; | |
4773 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4774 | } | |
4775 | } | |
4776 | ||
ac27a0ec DK |
4777 | /* |
4778 | * Post the struct inode info into an on-disk inode location in the | |
4779 | * buffer-cache. This gobbles the caller's reference to the | |
4780 | * buffer_head in the inode location struct. | |
4781 | * | |
4782 | * The caller must have write access to iloc->bh. | |
4783 | */ | |
617ba13b | 4784 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4785 | struct inode *inode, |
830156c7 | 4786 | struct ext4_iloc *iloc) |
ac27a0ec | 4787 | { |
617ba13b MC |
4788 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4789 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4790 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4791 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4792 | int err = 0, rc, block; |
202ee5df | 4793 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4794 | uid_t i_uid; |
4795 | gid_t i_gid; | |
040cb378 | 4796 | projid_t i_projid; |
ac27a0ec | 4797 | |
202ee5df TT |
4798 | spin_lock(&ei->i_raw_lock); |
4799 | ||
4800 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4801 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4802 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4803 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4804 | |
ff9ddf7e | 4805 | ext4_get_inode_flags(ei); |
ac27a0ec | 4806 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4807 | i_uid = i_uid_read(inode); |
4808 | i_gid = i_gid_read(inode); | |
040cb378 | 4809 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 4810 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4811 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4812 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4813 | /* |
4814 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4815 | * re-used with the upper 16 bits of the uid/gid intact | |
4816 | */ | |
d343ed7b DJ |
4817 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
4818 | raw_inode->i_uid_high = 0; | |
4819 | raw_inode->i_gid_high = 0; | |
4820 | } else { | |
ac27a0ec | 4821 | raw_inode->i_uid_high = |
08cefc7a | 4822 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4823 | raw_inode->i_gid_high = |
08cefc7a | 4824 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4825 | } |
4826 | } else { | |
08cefc7a EB |
4827 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4828 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4829 | raw_inode->i_uid_high = 0; |
4830 | raw_inode->i_gid_high = 0; | |
4831 | } | |
4832 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4833 | |
4834 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4835 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4836 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4837 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4838 | ||
bce92d56 LX |
4839 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4840 | if (err) { | |
202ee5df | 4841 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4842 | goto out_brelse; |
202ee5df | 4843 | } |
ac27a0ec | 4844 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4845 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4846 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4847 | raw_inode->i_file_acl_high = |
4848 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4849 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4850 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4851 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4852 | need_datasync = 1; | |
4853 | } | |
a48380f7 | 4854 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 4855 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 4856 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
4857 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4858 | set_large_file = 1; | |
ac27a0ec DK |
4859 | } |
4860 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4861 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4862 | if (old_valid_dev(inode->i_rdev)) { | |
4863 | raw_inode->i_block[0] = | |
4864 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4865 | raw_inode->i_block[1] = 0; | |
4866 | } else { | |
4867 | raw_inode->i_block[0] = 0; | |
4868 | raw_inode->i_block[1] = | |
4869 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4870 | raw_inode->i_block[2] = 0; | |
4871 | } | |
f19d5870 | 4872 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4873 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4874 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4875 | } |
ac27a0ec | 4876 | |
ed3654eb | 4877 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4878 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4879 | if (ei->i_extra_isize) { | |
4880 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4881 | raw_inode->i_version_hi = | |
4882 | cpu_to_le32(inode->i_version >> 32); | |
4883 | raw_inode->i_extra_isize = | |
4884 | cpu_to_le16(ei->i_extra_isize); | |
4885 | } | |
25ec56b5 | 4886 | } |
040cb378 | 4887 | |
c2950744 | 4888 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
4889 | i_projid != EXT4_DEF_PROJID); |
4890 | ||
4891 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
4892 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4893 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
4894 | ||
814525f4 | 4895 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 4896 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
4897 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
4898 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
4899 | bh->b_data); | |
202ee5df | 4900 | |
830156c7 | 4901 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4902 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4903 | if (!err) |
4904 | err = rc; | |
19f5fb7a | 4905 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 4906 | if (set_large_file) { |
5d601255 | 4907 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4908 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4909 | if (err) | |
4910 | goto out_brelse; | |
4911 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 4912 | ext4_set_feature_large_file(sb); |
202ee5df TT |
4913 | ext4_handle_sync(handle); |
4914 | err = ext4_handle_dirty_super(handle, sb); | |
4915 | } | |
b71fc079 | 4916 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4917 | out_brelse: |
af5bc92d | 4918 | brelse(bh); |
617ba13b | 4919 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4920 | return err; |
4921 | } | |
4922 | ||
4923 | /* | |
617ba13b | 4924 | * ext4_write_inode() |
ac27a0ec DK |
4925 | * |
4926 | * We are called from a few places: | |
4927 | * | |
87f7e416 | 4928 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4929 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4930 | * transaction to commit. |
ac27a0ec | 4931 | * |
87f7e416 TT |
4932 | * - Within flush work (sys_sync(), kupdate and such). |
4933 | * We wait on commit, if told to. | |
ac27a0ec | 4934 | * |
87f7e416 TT |
4935 | * - Within iput_final() -> write_inode_now() |
4936 | * We wait on commit, if told to. | |
ac27a0ec DK |
4937 | * |
4938 | * In all cases it is actually safe for us to return without doing anything, | |
4939 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4940 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4941 | * writeback. | |
ac27a0ec DK |
4942 | * |
4943 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4944 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4945 | * which we are interested. | |
4946 | * | |
4947 | * It would be a bug for them to not do this. The code: | |
4948 | * | |
4949 | * mark_inode_dirty(inode) | |
4950 | * stuff(); | |
4951 | * inode->i_size = expr; | |
4952 | * | |
87f7e416 TT |
4953 | * is in error because write_inode() could occur while `stuff()' is running, |
4954 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4955 | * superblock's dirty inode list. | |
ac27a0ec | 4956 | */ |
a9185b41 | 4957 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4958 | { |
91ac6f43 FM |
4959 | int err; |
4960 | ||
87f7e416 | 4961 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4962 | return 0; |
4963 | ||
91ac6f43 FM |
4964 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4965 | if (ext4_journal_current_handle()) { | |
4966 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4967 | dump_stack(); | |
4968 | return -EIO; | |
4969 | } | |
ac27a0ec | 4970 | |
10542c22 JK |
4971 | /* |
4972 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4973 | * ext4_sync_fs() will force the commit after everything is | |
4974 | * written. | |
4975 | */ | |
4976 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4977 | return 0; |
4978 | ||
4979 | err = ext4_force_commit(inode->i_sb); | |
4980 | } else { | |
4981 | struct ext4_iloc iloc; | |
ac27a0ec | 4982 | |
8b472d73 | 4983 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4984 | if (err) |
4985 | return err; | |
10542c22 JK |
4986 | /* |
4987 | * sync(2) will flush the whole buffer cache. No need to do | |
4988 | * it here separately for each inode. | |
4989 | */ | |
4990 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4991 | sync_dirty_buffer(iloc.bh); |
4992 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4993 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4994 | "IO error syncing inode"); | |
830156c7 FM |
4995 | err = -EIO; |
4996 | } | |
fd2dd9fb | 4997 | brelse(iloc.bh); |
91ac6f43 FM |
4998 | } |
4999 | return err; | |
ac27a0ec DK |
5000 | } |
5001 | ||
53e87268 JK |
5002 | /* |
5003 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5004 | * buffers that are attached to a page stradding i_size and are undergoing | |
5005 | * commit. In that case we have to wait for commit to finish and try again. | |
5006 | */ | |
5007 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5008 | { | |
5009 | struct page *page; | |
5010 | unsigned offset; | |
5011 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5012 | tid_t commit_tid = 0; | |
5013 | int ret; | |
5014 | ||
09cbfeaf | 5015 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 JK |
5016 | /* |
5017 | * All buffers in the last page remain valid? Then there's nothing to | |
ea1754a0 | 5018 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
53e87268 JK |
5019 | * blocksize case |
5020 | */ | |
09cbfeaf | 5021 | if (offset > PAGE_SIZE - (1 << inode->i_blkbits)) |
53e87268 JK |
5022 | return; |
5023 | while (1) { | |
5024 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5025 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5026 | if (!page) |
5027 | return; | |
ca99fdd2 | 5028 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5029 | PAGE_SIZE - offset); |
53e87268 | 5030 | unlock_page(page); |
09cbfeaf | 5031 | put_page(page); |
53e87268 JK |
5032 | if (ret != -EBUSY) |
5033 | return; | |
5034 | commit_tid = 0; | |
5035 | read_lock(&journal->j_state_lock); | |
5036 | if (journal->j_committing_transaction) | |
5037 | commit_tid = journal->j_committing_transaction->t_tid; | |
5038 | read_unlock(&journal->j_state_lock); | |
5039 | if (commit_tid) | |
5040 | jbd2_log_wait_commit(journal, commit_tid); | |
5041 | } | |
5042 | } | |
5043 | ||
ac27a0ec | 5044 | /* |
617ba13b | 5045 | * ext4_setattr() |
ac27a0ec DK |
5046 | * |
5047 | * Called from notify_change. | |
5048 | * | |
5049 | * We want to trap VFS attempts to truncate the file as soon as | |
5050 | * possible. In particular, we want to make sure that when the VFS | |
5051 | * shrinks i_size, we put the inode on the orphan list and modify | |
5052 | * i_disksize immediately, so that during the subsequent flushing of | |
5053 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5054 | * commit will leave the blocks being flushed in an unused state on | |
5055 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5056 | * be freed, so we have a strong guarantee that no future commit will | |
5057 | * leave these blocks visible to the user.) | |
5058 | * | |
678aaf48 JK |
5059 | * Another thing we have to assure is that if we are in ordered mode |
5060 | * and inode is still attached to the committing transaction, we must | |
5061 | * we start writeout of all the dirty pages which are being truncated. | |
5062 | * This way we are sure that all the data written in the previous | |
5063 | * transaction are already on disk (truncate waits for pages under | |
5064 | * writeback). | |
5065 | * | |
5066 | * Called with inode->i_mutex down. | |
ac27a0ec | 5067 | */ |
617ba13b | 5068 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5069 | { |
2b0143b5 | 5070 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5071 | int error, rc = 0; |
3d287de3 | 5072 | int orphan = 0; |
ac27a0ec DK |
5073 | const unsigned int ia_valid = attr->ia_valid; |
5074 | ||
5075 | error = inode_change_ok(inode, attr); | |
5076 | if (error) | |
5077 | return error; | |
5078 | ||
a7cdadee JK |
5079 | if (is_quota_modification(inode, attr)) { |
5080 | error = dquot_initialize(inode); | |
5081 | if (error) | |
5082 | return error; | |
5083 | } | |
08cefc7a EB |
5084 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5085 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5086 | handle_t *handle; |
5087 | ||
5088 | /* (user+group)*(old+new) structure, inode write (sb, | |
5089 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5090 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5091 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5092 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5093 | if (IS_ERR(handle)) { |
5094 | error = PTR_ERR(handle); | |
5095 | goto err_out; | |
5096 | } | |
b43fa828 | 5097 | error = dquot_transfer(inode, attr); |
ac27a0ec | 5098 | if (error) { |
617ba13b | 5099 | ext4_journal_stop(handle); |
ac27a0ec DK |
5100 | return error; |
5101 | } | |
5102 | /* Update corresponding info in inode so that everything is in | |
5103 | * one transaction */ | |
5104 | if (attr->ia_valid & ATTR_UID) | |
5105 | inode->i_uid = attr->ia_uid; | |
5106 | if (attr->ia_valid & ATTR_GID) | |
5107 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5108 | error = ext4_mark_inode_dirty(handle, inode); |
5109 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5110 | } |
5111 | ||
3da40c7b | 5112 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5113 | handle_t *handle; |
3da40c7b JB |
5114 | loff_t oldsize = inode->i_size; |
5115 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 5116 | |
12e9b892 | 5117 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5118 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5119 | ||
0c095c7f TT |
5120 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5121 | return -EFBIG; | |
e2b46574 | 5122 | } |
3da40c7b JB |
5123 | if (!S_ISREG(inode->i_mode)) |
5124 | return -EINVAL; | |
dff6efc3 CH |
5125 | |
5126 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5127 | inode_inc_iversion(inode); | |
5128 | ||
3da40c7b | 5129 | if (ext4_should_order_data(inode) && |
5208386c | 5130 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 5131 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 5132 | attr->ia_size); |
3da40c7b JB |
5133 | if (error) |
5134 | goto err_out; | |
5135 | } | |
5136 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
5137 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5138 | if (IS_ERR(handle)) { | |
5139 | error = PTR_ERR(handle); | |
5140 | goto err_out; | |
5141 | } | |
3da40c7b | 5142 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5143 | error = ext4_orphan_add(handle, inode); |
5144 | orphan = 1; | |
5145 | } | |
911af577 EG |
5146 | /* |
5147 | * Update c/mtime on truncate up, ext4_truncate() will | |
5148 | * update c/mtime in shrink case below | |
5149 | */ | |
5150 | if (!shrink) { | |
5151 | inode->i_mtime = ext4_current_time(inode); | |
5152 | inode->i_ctime = inode->i_mtime; | |
5153 | } | |
90e775b7 | 5154 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5155 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5156 | rc = ext4_mark_inode_dirty(handle, inode); | |
5157 | if (!error) | |
5158 | error = rc; | |
90e775b7 JK |
5159 | /* |
5160 | * We have to update i_size under i_data_sem together | |
5161 | * with i_disksize to avoid races with writeback code | |
5162 | * running ext4_wb_update_i_disksize(). | |
5163 | */ | |
5164 | if (!error) | |
5165 | i_size_write(inode, attr->ia_size); | |
5166 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
5167 | ext4_journal_stop(handle); |
5168 | if (error) { | |
3da40c7b JB |
5169 | if (orphan) |
5170 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
5171 | goto err_out; |
5172 | } | |
d6320cbf | 5173 | } |
3da40c7b JB |
5174 | if (!shrink) |
5175 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 5176 | |
5208386c JK |
5177 | /* |
5178 | * Blocks are going to be removed from the inode. Wait | |
5179 | * for dio in flight. Temporarily disable | |
5180 | * dioread_nolock to prevent livelock. | |
5181 | */ | |
5182 | if (orphan) { | |
5183 | if (!ext4_should_journal_data(inode)) { | |
5184 | ext4_inode_block_unlocked_dio(inode); | |
5185 | inode_dio_wait(inode); | |
5186 | ext4_inode_resume_unlocked_dio(inode); | |
5187 | } else | |
5188 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 5189 | } |
ea3d7209 | 5190 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
5191 | /* |
5192 | * Truncate pagecache after we've waited for commit | |
5193 | * in data=journal mode to make pages freeable. | |
5194 | */ | |
923ae0ff | 5195 | truncate_pagecache(inode, inode->i_size); |
3da40c7b JB |
5196 | if (shrink) |
5197 | ext4_truncate(inode); | |
ea3d7209 | 5198 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5199 | } |
ac27a0ec | 5200 | |
1025774c CH |
5201 | if (!rc) { |
5202 | setattr_copy(inode, attr); | |
5203 | mark_inode_dirty(inode); | |
5204 | } | |
5205 | ||
5206 | /* | |
5207 | * If the call to ext4_truncate failed to get a transaction handle at | |
5208 | * all, we need to clean up the in-core orphan list manually. | |
5209 | */ | |
3d287de3 | 5210 | if (orphan && inode->i_nlink) |
617ba13b | 5211 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5212 | |
5213 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 5214 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5215 | |
5216 | err_out: | |
617ba13b | 5217 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5218 | if (!error) |
5219 | error = rc; | |
5220 | return error; | |
5221 | } | |
5222 | ||
3e3398a0 MC |
5223 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5224 | struct kstat *stat) | |
5225 | { | |
5226 | struct inode *inode; | |
8af8eecc | 5227 | unsigned long long delalloc_blocks; |
3e3398a0 | 5228 | |
2b0143b5 | 5229 | inode = d_inode(dentry); |
3e3398a0 MC |
5230 | generic_fillattr(inode, stat); |
5231 | ||
9206c561 AD |
5232 | /* |
5233 | * If there is inline data in the inode, the inode will normally not | |
5234 | * have data blocks allocated (it may have an external xattr block). | |
5235 | * Report at least one sector for such files, so tools like tar, rsync, | |
5236 | * others doen't incorrectly think the file is completely sparse. | |
5237 | */ | |
5238 | if (unlikely(ext4_has_inline_data(inode))) | |
5239 | stat->blocks += (stat->size + 511) >> 9; | |
5240 | ||
3e3398a0 MC |
5241 | /* |
5242 | * We can't update i_blocks if the block allocation is delayed | |
5243 | * otherwise in the case of system crash before the real block | |
5244 | * allocation is done, we will have i_blocks inconsistent with | |
5245 | * on-disk file blocks. | |
5246 | * We always keep i_blocks updated together with real | |
5247 | * allocation. But to not confuse with user, stat | |
5248 | * will return the blocks that include the delayed allocation | |
5249 | * blocks for this file. | |
5250 | */ | |
96607551 | 5251 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5252 | EXT4_I(inode)->i_reserved_data_blocks); |
5253 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5254 | return 0; |
5255 | } | |
ac27a0ec | 5256 | |
fffb2739 JK |
5257 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5258 | int pextents) | |
a02908f1 | 5259 | { |
12e9b892 | 5260 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5261 | return ext4_ind_trans_blocks(inode, lblocks); |
5262 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5263 | } |
ac51d837 | 5264 | |
ac27a0ec | 5265 | /* |
a02908f1 MC |
5266 | * Account for index blocks, block groups bitmaps and block group |
5267 | * descriptor blocks if modify datablocks and index blocks | |
5268 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5269 | * |
a02908f1 | 5270 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5271 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5272 | * they could still across block group boundary. |
ac27a0ec | 5273 | * |
a02908f1 MC |
5274 | * Also account for superblock, inode, quota and xattr blocks |
5275 | */ | |
fffb2739 JK |
5276 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
5277 | int pextents) | |
a02908f1 | 5278 | { |
8df9675f TT |
5279 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5280 | int gdpblocks; | |
a02908f1 MC |
5281 | int idxblocks; |
5282 | int ret = 0; | |
5283 | ||
5284 | /* | |
fffb2739 JK |
5285 | * How many index blocks need to touch to map @lblocks logical blocks |
5286 | * to @pextents physical extents? | |
a02908f1 | 5287 | */ |
fffb2739 | 5288 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5289 | |
5290 | ret = idxblocks; | |
5291 | ||
5292 | /* | |
5293 | * Now let's see how many group bitmaps and group descriptors need | |
5294 | * to account | |
5295 | */ | |
fffb2739 | 5296 | groups = idxblocks + pextents; |
a02908f1 | 5297 | gdpblocks = groups; |
8df9675f TT |
5298 | if (groups > ngroups) |
5299 | groups = ngroups; | |
a02908f1 MC |
5300 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5301 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5302 | ||
5303 | /* bitmaps and block group descriptor blocks */ | |
5304 | ret += groups + gdpblocks; | |
5305 | ||
5306 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5307 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5308 | ||
5309 | return ret; | |
5310 | } | |
5311 | ||
5312 | /* | |
25985edc | 5313 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5314 | * the modification of a single pages into a single transaction, |
5315 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5316 | * |
525f4ed8 | 5317 | * This could be called via ext4_write_begin() |
ac27a0ec | 5318 | * |
525f4ed8 | 5319 | * We need to consider the worse case, when |
a02908f1 | 5320 | * one new block per extent. |
ac27a0ec | 5321 | */ |
a86c6181 | 5322 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5323 | { |
617ba13b | 5324 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5325 | int ret; |
5326 | ||
fffb2739 | 5327 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5328 | |
a02908f1 | 5329 | /* Account for data blocks for journalled mode */ |
617ba13b | 5330 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5331 | ret += bpp; |
ac27a0ec DK |
5332 | return ret; |
5333 | } | |
f3bd1f3f MC |
5334 | |
5335 | /* | |
5336 | * Calculate the journal credits for a chunk of data modification. | |
5337 | * | |
5338 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5339 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5340 | * |
5341 | * journal buffers for data blocks are not included here, as DIO | |
5342 | * and fallocate do no need to journal data buffers. | |
5343 | */ | |
5344 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5345 | { | |
5346 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5347 | } | |
5348 | ||
ac27a0ec | 5349 | /* |
617ba13b | 5350 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5351 | * Give this, we know that the caller already has write access to iloc->bh. |
5352 | */ | |
617ba13b | 5353 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5354 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5355 | { |
5356 | int err = 0; | |
5357 | ||
c64db50e | 5358 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5359 | inode_inc_iversion(inode); |
5360 | ||
ac27a0ec DK |
5361 | /* the do_update_inode consumes one bh->b_count */ |
5362 | get_bh(iloc->bh); | |
5363 | ||
dab291af | 5364 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5365 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5366 | put_bh(iloc->bh); |
5367 | return err; | |
5368 | } | |
5369 | ||
5370 | /* | |
5371 | * On success, We end up with an outstanding reference count against | |
5372 | * iloc->bh. This _must_ be cleaned up later. | |
5373 | */ | |
5374 | ||
5375 | int | |
617ba13b MC |
5376 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5377 | struct ext4_iloc *iloc) | |
ac27a0ec | 5378 | { |
0390131b FM |
5379 | int err; |
5380 | ||
5381 | err = ext4_get_inode_loc(inode, iloc); | |
5382 | if (!err) { | |
5383 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5384 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5385 | if (err) { | |
5386 | brelse(iloc->bh); | |
5387 | iloc->bh = NULL; | |
ac27a0ec DK |
5388 | } |
5389 | } | |
617ba13b | 5390 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5391 | return err; |
5392 | } | |
5393 | ||
6dd4ee7c KS |
5394 | /* |
5395 | * Expand an inode by new_extra_isize bytes. | |
5396 | * Returns 0 on success or negative error number on failure. | |
5397 | */ | |
1d03ec98 AK |
5398 | static int ext4_expand_extra_isize(struct inode *inode, |
5399 | unsigned int new_extra_isize, | |
5400 | struct ext4_iloc iloc, | |
5401 | handle_t *handle) | |
6dd4ee7c KS |
5402 | { |
5403 | struct ext4_inode *raw_inode; | |
5404 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5405 | |
5406 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5407 | return 0; | |
5408 | ||
5409 | raw_inode = ext4_raw_inode(&iloc); | |
5410 | ||
5411 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5412 | |
5413 | /* No extended attributes present */ | |
19f5fb7a TT |
5414 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5415 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5416 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5417 | new_extra_isize); | |
5418 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5419 | return 0; | |
5420 | } | |
5421 | ||
5422 | /* try to expand with EAs present */ | |
5423 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5424 | raw_inode, handle); | |
5425 | } | |
5426 | ||
ac27a0ec DK |
5427 | /* |
5428 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5429 | * dirtiness (it may still be data-dirty). | |
5430 | * This means that the in-core inode may be reaped by prune_icache | |
5431 | * without having to perform any I/O. This is a very good thing, | |
5432 | * because *any* task may call prune_icache - even ones which | |
5433 | * have a transaction open against a different journal. | |
5434 | * | |
5435 | * Is this cheating? Not really. Sure, we haven't written the | |
5436 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5437 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5438 | * we start and wait on commits. | |
ac27a0ec | 5439 | */ |
617ba13b | 5440 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5441 | { |
617ba13b | 5442 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5443 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5444 | static unsigned int mnt_count; | |
5445 | int err, ret; | |
ac27a0ec DK |
5446 | |
5447 | might_sleep(); | |
7ff9c073 | 5448 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5449 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
5450 | if (err) |
5451 | return err; | |
0390131b FM |
5452 | if (ext4_handle_valid(handle) && |
5453 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5454 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5455 | /* |
5456 | * We need extra buffer credits since we may write into EA block | |
5457 | * with this same handle. If journal_extend fails, then it will | |
5458 | * only result in a minor loss of functionality for that inode. | |
5459 | * If this is felt to be critical, then e2fsck should be run to | |
5460 | * force a large enough s_min_extra_isize. | |
5461 | */ | |
5462 | if ((jbd2_journal_extend(handle, | |
5463 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5464 | ret = ext4_expand_extra_isize(inode, | |
5465 | sbi->s_want_extra_isize, | |
5466 | iloc, handle); | |
5467 | if (ret) { | |
c1bddad9 AK |
5468 | if (mnt_count != |
5469 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5470 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5471 | "Unable to expand inode %lu. Delete" |
5472 | " some EAs or run e2fsck.", | |
5473 | inode->i_ino); | |
c1bddad9 AK |
5474 | mnt_count = |
5475 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5476 | } |
5477 | } | |
5478 | } | |
5479 | } | |
5e1021f2 | 5480 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5481 | } |
5482 | ||
5483 | /* | |
617ba13b | 5484 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5485 | * |
5486 | * We're really interested in the case where a file is being extended. | |
5487 | * i_size has been changed by generic_commit_write() and we thus need | |
5488 | * to include the updated inode in the current transaction. | |
5489 | * | |
5dd4056d | 5490 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5491 | * are allocated to the file. |
5492 | * | |
5493 | * If the inode is marked synchronous, we don't honour that here - doing | |
5494 | * so would cause a commit on atime updates, which we don't bother doing. | |
5495 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5496 | * |
5497 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5498 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5499 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5500 | */ |
aa385729 | 5501 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5502 | { |
ac27a0ec DK |
5503 | handle_t *handle; |
5504 | ||
0ae45f63 TT |
5505 | if (flags == I_DIRTY_TIME) |
5506 | return; | |
9924a92a | 5507 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5508 | if (IS_ERR(handle)) |
5509 | goto out; | |
f3dc272f | 5510 | |
f3dc272f CW |
5511 | ext4_mark_inode_dirty(handle, inode); |
5512 | ||
617ba13b | 5513 | ext4_journal_stop(handle); |
ac27a0ec DK |
5514 | out: |
5515 | return; | |
5516 | } | |
5517 | ||
5518 | #if 0 | |
5519 | /* | |
5520 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5521 | * it from being flushed to disk early. Unlike | |
617ba13b | 5522 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5523 | * returns no iloc structure, so the caller needs to repeat the iloc |
5524 | * lookup to mark the inode dirty later. | |
5525 | */ | |
617ba13b | 5526 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5527 | { |
617ba13b | 5528 | struct ext4_iloc iloc; |
ac27a0ec DK |
5529 | |
5530 | int err = 0; | |
5531 | if (handle) { | |
617ba13b | 5532 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5533 | if (!err) { |
5534 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5535 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5536 | if (!err) |
0390131b | 5537 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5538 | NULL, |
0390131b | 5539 | iloc.bh); |
ac27a0ec DK |
5540 | brelse(iloc.bh); |
5541 | } | |
5542 | } | |
617ba13b | 5543 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5544 | return err; |
5545 | } | |
5546 | #endif | |
5547 | ||
617ba13b | 5548 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5549 | { |
5550 | journal_t *journal; | |
5551 | handle_t *handle; | |
5552 | int err; | |
c8585c6f | 5553 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5554 | |
5555 | /* | |
5556 | * We have to be very careful here: changing a data block's | |
5557 | * journaling status dynamically is dangerous. If we write a | |
5558 | * data block to the journal, change the status and then delete | |
5559 | * that block, we risk forgetting to revoke the old log record | |
5560 | * from the journal and so a subsequent replay can corrupt data. | |
5561 | * So, first we make sure that the journal is empty and that | |
5562 | * nobody is changing anything. | |
5563 | */ | |
5564 | ||
617ba13b | 5565 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5566 | if (!journal) |
5567 | return 0; | |
d699594d | 5568 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5569 | return -EROFS; |
5570 | ||
17335dcc DM |
5571 | /* Wait for all existing dio workers */ |
5572 | ext4_inode_block_unlocked_dio(inode); | |
5573 | inode_dio_wait(inode); | |
5574 | ||
4c546592 DJ |
5575 | /* |
5576 | * Before flushing the journal and switching inode's aops, we have | |
5577 | * to flush all dirty data the inode has. There can be outstanding | |
5578 | * delayed allocations, there can be unwritten extents created by | |
5579 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5580 | * dirty data which can be converted only after flushing the dirty | |
5581 | * data (and journalled aops don't know how to handle these cases). | |
5582 | */ | |
5583 | if (val) { | |
5584 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5585 | err = filemap_write_and_wait(inode->i_mapping); | |
5586 | if (err < 0) { | |
5587 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
5588 | ext4_inode_resume_unlocked_dio(inode); | |
5589 | return err; | |
5590 | } | |
5591 | } | |
5592 | ||
c8585c6f | 5593 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
dab291af | 5594 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5595 | |
5596 | /* | |
5597 | * OK, there are no updates running now, and all cached data is | |
5598 | * synced to disk. We are now in a completely consistent state | |
5599 | * which doesn't have anything in the journal, and we know that | |
5600 | * no filesystem updates are running, so it is safe to modify | |
5601 | * the inode's in-core data-journaling state flag now. | |
5602 | */ | |
5603 | ||
5604 | if (val) | |
12e9b892 | 5605 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5606 | else { |
4f879ca6 JK |
5607 | err = jbd2_journal_flush(journal); |
5608 | if (err < 0) { | |
5609 | jbd2_journal_unlock_updates(journal); | |
c8585c6f | 5610 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
4f879ca6 JK |
5611 | ext4_inode_resume_unlocked_dio(inode); |
5612 | return err; | |
5613 | } | |
12e9b892 | 5614 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5615 | } |
617ba13b | 5616 | ext4_set_aops(inode); |
ac27a0ec | 5617 | |
dab291af | 5618 | jbd2_journal_unlock_updates(journal); |
c8585c6f DJ |
5619 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
5620 | ||
4c546592 DJ |
5621 | if (val) |
5622 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
17335dcc | 5623 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5624 | |
5625 | /* Finally we can mark the inode as dirty. */ | |
5626 | ||
9924a92a | 5627 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5628 | if (IS_ERR(handle)) |
5629 | return PTR_ERR(handle); | |
5630 | ||
617ba13b | 5631 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5632 | ext4_handle_sync(handle); |
617ba13b MC |
5633 | ext4_journal_stop(handle); |
5634 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5635 | |
5636 | return err; | |
5637 | } | |
2e9ee850 AK |
5638 | |
5639 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5640 | { | |
5641 | return !buffer_mapped(bh); | |
5642 | } | |
5643 | ||
c2ec175c | 5644 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5645 | { |
c2ec175c | 5646 | struct page *page = vmf->page; |
2e9ee850 AK |
5647 | loff_t size; |
5648 | unsigned long len; | |
9ea7df53 | 5649 | int ret; |
2e9ee850 | 5650 | struct file *file = vma->vm_file; |
496ad9aa | 5651 | struct inode *inode = file_inode(file); |
2e9ee850 | 5652 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5653 | handle_t *handle; |
5654 | get_block_t *get_block; | |
5655 | int retries = 0; | |
2e9ee850 | 5656 | |
8e8ad8a5 | 5657 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5658 | file_update_time(vma->vm_file); |
ea3d7209 JK |
5659 | |
5660 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
9ea7df53 JK |
5661 | /* Delalloc case is easy... */ |
5662 | if (test_opt(inode->i_sb, DELALLOC) && | |
5663 | !ext4_should_journal_data(inode) && | |
5664 | !ext4_nonda_switch(inode->i_sb)) { | |
5665 | do { | |
5c500029 | 5666 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
5667 | ext4_da_get_block_prep); |
5668 | } while (ret == -ENOSPC && | |
5669 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5670 | goto out_ret; | |
2e9ee850 | 5671 | } |
0e499890 DW |
5672 | |
5673 | lock_page(page); | |
9ea7df53 JK |
5674 | size = i_size_read(inode); |
5675 | /* Page got truncated from under us? */ | |
5676 | if (page->mapping != mapping || page_offset(page) > size) { | |
5677 | unlock_page(page); | |
5678 | ret = VM_FAULT_NOPAGE; | |
5679 | goto out; | |
0e499890 | 5680 | } |
2e9ee850 | 5681 | |
09cbfeaf KS |
5682 | if (page->index == size >> PAGE_SHIFT) |
5683 | len = size & ~PAGE_MASK; | |
2e9ee850 | 5684 | else |
09cbfeaf | 5685 | len = PAGE_SIZE; |
a827eaff | 5686 | /* |
9ea7df53 JK |
5687 | * Return if we have all the buffers mapped. This avoids the need to do |
5688 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5689 | */ |
2e9ee850 | 5690 | if (page_has_buffers(page)) { |
f19d5870 TM |
5691 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5692 | 0, len, NULL, | |
5693 | ext4_bh_unmapped)) { | |
9ea7df53 | 5694 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5695 | wait_for_stable_page(page); |
9ea7df53 JK |
5696 | ret = VM_FAULT_LOCKED; |
5697 | goto out; | |
a827eaff | 5698 | } |
2e9ee850 | 5699 | } |
a827eaff | 5700 | unlock_page(page); |
9ea7df53 JK |
5701 | /* OK, we need to fill the hole... */ |
5702 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 5703 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
5704 | else |
5705 | get_block = ext4_get_block; | |
5706 | retry_alloc: | |
9924a92a TT |
5707 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5708 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5709 | if (IS_ERR(handle)) { |
c2ec175c | 5710 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5711 | goto out; |
5712 | } | |
5c500029 | 5713 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 5714 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 5715 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 5716 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
5717 | unlock_page(page); |
5718 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5719 | ext4_journal_stop(handle); |
9ea7df53 JK |
5720 | goto out; |
5721 | } | |
5722 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5723 | } | |
5724 | ext4_journal_stop(handle); | |
5725 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5726 | goto retry_alloc; | |
5727 | out_ret: | |
5728 | ret = block_page_mkwrite_return(ret); | |
5729 | out: | |
ea3d7209 | 5730 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 5731 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5732 | return ret; |
5733 | } | |
ea3d7209 JK |
5734 | |
5735 | int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
5736 | { | |
5737 | struct inode *inode = file_inode(vma->vm_file); | |
5738 | int err; | |
5739 | ||
5740 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
5741 | err = filemap_fault(vma, vmf); | |
5742 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
5743 | ||
5744 | return err; | |
5745 | } | |
2d90c160 JK |
5746 | |
5747 | /* | |
5748 | * Find the first extent at or after @lblk in an inode that is not a hole. | |
5749 | * Search for @map_len blocks at most. The extent is returned in @result. | |
5750 | * | |
5751 | * The function returns 1 if we found an extent. The function returns 0 in | |
5752 | * case there is no extent at or after @lblk and in that case also sets | |
5753 | * @result->es_len to 0. In case of error, the error code is returned. | |
5754 | */ | |
5755 | int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk, | |
5756 | unsigned int map_len, struct extent_status *result) | |
5757 | { | |
5758 | struct ext4_map_blocks map; | |
5759 | struct extent_status es = {}; | |
5760 | int ret; | |
5761 | ||
5762 | map.m_lblk = lblk; | |
5763 | map.m_len = map_len; | |
5764 | ||
5765 | /* | |
5766 | * For non-extent based files this loop may iterate several times since | |
5767 | * we do not determine full hole size. | |
5768 | */ | |
5769 | while (map.m_len > 0) { | |
5770 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
5771 | if (ret < 0) | |
5772 | return ret; | |
5773 | /* There's extent covering m_lblk? Just return it. */ | |
5774 | if (ret > 0) { | |
5775 | int status; | |
5776 | ||
5777 | ext4_es_store_pblock(result, map.m_pblk); | |
5778 | result->es_lblk = map.m_lblk; | |
5779 | result->es_len = map.m_len; | |
5780 | if (map.m_flags & EXT4_MAP_UNWRITTEN) | |
5781 | status = EXTENT_STATUS_UNWRITTEN; | |
5782 | else | |
5783 | status = EXTENT_STATUS_WRITTEN; | |
5784 | ext4_es_store_status(result, status); | |
5785 | return 1; | |
5786 | } | |
5787 | ext4_es_find_delayed_extent_range(inode, map.m_lblk, | |
5788 | map.m_lblk + map.m_len - 1, | |
5789 | &es); | |
5790 | /* Is delalloc data before next block in extent tree? */ | |
5791 | if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) { | |
5792 | ext4_lblk_t offset = 0; | |
5793 | ||
5794 | if (es.es_lblk < lblk) | |
5795 | offset = lblk - es.es_lblk; | |
5796 | result->es_lblk = es.es_lblk + offset; | |
5797 | ext4_es_store_pblock(result, | |
5798 | ext4_es_pblock(&es) + offset); | |
5799 | result->es_len = es.es_len - offset; | |
5800 | ext4_es_store_status(result, ext4_es_status(&es)); | |
5801 | ||
5802 | return 1; | |
5803 | } | |
5804 | /* There's a hole at m_lblk, advance us after it */ | |
5805 | map.m_lblk += map.m_len; | |
5806 | map_len -= map.m_len; | |
5807 | map.m_len = map_len; | |
5808 | cond_resched(); | |
5809 | } | |
5810 | result->es_len = 0; | |
5811 | return 0; | |
5812 | } |