]>
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 | |
9b623df6 JK |
650 | * use them before they are really zeroed. We also have to |
651 | * unmap metadata before zeroing as otherwise writeback can | |
652 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
653 | */ |
654 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
655 | map->m_flags & EXT4_MAP_MAPPED && | |
656 | map->m_flags & EXT4_MAP_NEW) { | |
9b623df6 JK |
657 | ext4_lblk_t i; |
658 | ||
659 | for (i = 0; i < map->m_len; i++) { | |
660 | unmap_underlying_metadata(inode->i_sb->s_bdev, | |
661 | map->m_pblk + i); | |
662 | } | |
c86d8db3 JK |
663 | ret = ext4_issue_zeroout(inode, map->m_lblk, |
664 | map->m_pblk, map->m_len); | |
665 | if (ret) { | |
666 | retval = ret; | |
667 | goto out_sem; | |
668 | } | |
669 | } | |
670 | ||
adb23551 ZL |
671 | /* |
672 | * If the extent has been zeroed out, we don't need to update | |
673 | * extent status tree. | |
674 | */ | |
675 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
676 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
677 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 678 | goto out_sem; |
adb23551 | 679 | } |
f7fec032 ZL |
680 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
681 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
682 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 683 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
684 | ext4_find_delalloc_range(inode, map->m_lblk, |
685 | map->m_lblk + map->m_len - 1)) | |
686 | status |= EXTENT_STATUS_DELAYED; | |
687 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
688 | map->m_pblk, status); | |
c86d8db3 | 689 | if (ret < 0) { |
f7fec032 | 690 | retval = ret; |
c86d8db3 JK |
691 | goto out_sem; |
692 | } | |
5356f261 AK |
693 | } |
694 | ||
c86d8db3 | 695 | out_sem: |
4df3d265 | 696 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 697 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 698 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
699 | if (ret != 0) |
700 | return ret; | |
06bd3c36 JK |
701 | |
702 | /* | |
703 | * Inodes with freshly allocated blocks where contents will be | |
704 | * visible after transaction commit must be on transaction's | |
705 | * ordered data list. | |
706 | */ | |
707 | if (map->m_flags & EXT4_MAP_NEW && | |
708 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
709 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
710 | !IS_NOQUOTA(inode) && | |
711 | ext4_should_order_data(inode)) { | |
ee0876bc JK |
712 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
713 | ret = ext4_jbd2_inode_add_wait(handle, inode); | |
714 | else | |
715 | ret = ext4_jbd2_inode_add_write(handle, inode); | |
06bd3c36 JK |
716 | if (ret) |
717 | return ret; | |
718 | } | |
6fd058f7 | 719 | } |
0e855ac8 AK |
720 | return retval; |
721 | } | |
722 | ||
ed8ad838 JK |
723 | /* |
724 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
725 | * we have to be careful as someone else may be manipulating b_state as well. | |
726 | */ | |
727 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
728 | { | |
729 | unsigned long old_state; | |
730 | unsigned long new_state; | |
731 | ||
732 | flags &= EXT4_MAP_FLAGS; | |
733 | ||
734 | /* Dummy buffer_head? Set non-atomically. */ | |
735 | if (!bh->b_page) { | |
736 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
737 | return; | |
738 | } | |
739 | /* | |
740 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
741 | * once we get rid of using bh as a container for mapping information | |
742 | * to pass to / from get_block functions, this can go away. | |
743 | */ | |
744 | do { | |
745 | old_state = READ_ONCE(bh->b_state); | |
746 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
747 | } while (unlikely( | |
748 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
749 | } | |
750 | ||
2ed88685 TT |
751 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
752 | struct buffer_head *bh, int flags) | |
ac27a0ec | 753 | { |
2ed88685 | 754 | struct ext4_map_blocks map; |
efe70c29 | 755 | int ret = 0; |
ac27a0ec | 756 | |
46c7f254 TM |
757 | if (ext4_has_inline_data(inode)) |
758 | return -ERANGE; | |
759 | ||
2ed88685 TT |
760 | map.m_lblk = iblock; |
761 | map.m_len = bh->b_size >> inode->i_blkbits; | |
762 | ||
efe70c29 JK |
763 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
764 | flags); | |
7fb5409d | 765 | if (ret > 0) { |
2ed88685 | 766 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 767 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 768 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 769 | ret = 0; |
ac27a0ec DK |
770 | } |
771 | return ret; | |
772 | } | |
773 | ||
2ed88685 TT |
774 | int ext4_get_block(struct inode *inode, sector_t iblock, |
775 | struct buffer_head *bh, int create) | |
776 | { | |
777 | return _ext4_get_block(inode, iblock, bh, | |
778 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
779 | } | |
780 | ||
705965bd JK |
781 | /* |
782 | * Get block function used when preparing for buffered write if we require | |
783 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
784 | * will be converted to written after the IO is complete. | |
785 | */ | |
786 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
787 | struct buffer_head *bh_result, int create) | |
788 | { | |
789 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
790 | inode->i_ino, create); | |
791 | return _ext4_get_block(inode, iblock, bh_result, | |
792 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
793 | } | |
794 | ||
efe70c29 JK |
795 | /* Maximum number of blocks we map for direct IO at once. */ |
796 | #define DIO_MAX_BLOCKS 4096 | |
797 | ||
e84dfbe2 JK |
798 | /* |
799 | * Get blocks function for the cases that need to start a transaction - | |
800 | * generally difference cases of direct IO and DAX IO. It also handles retries | |
801 | * in case of ENOSPC. | |
802 | */ | |
803 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, | |
804 | struct buffer_head *bh_result, int flags) | |
efe70c29 JK |
805 | { |
806 | int dio_credits; | |
e84dfbe2 JK |
807 | handle_t *handle; |
808 | int retries = 0; | |
809 | int ret; | |
efe70c29 JK |
810 | |
811 | /* Trim mapping request to maximum we can map at once for DIO */ | |
812 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) | |
813 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; | |
814 | dio_credits = ext4_chunk_trans_blocks(inode, | |
815 | bh_result->b_size >> inode->i_blkbits); | |
e84dfbe2 JK |
816 | retry: |
817 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); | |
818 | if (IS_ERR(handle)) | |
819 | return PTR_ERR(handle); | |
820 | ||
821 | ret = _ext4_get_block(inode, iblock, bh_result, flags); | |
822 | ext4_journal_stop(handle); | |
823 | ||
824 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
825 | goto retry; | |
826 | return ret; | |
efe70c29 JK |
827 | } |
828 | ||
705965bd JK |
829 | /* Get block function for DIO reads and writes to inodes without extents */ |
830 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, | |
831 | struct buffer_head *bh, int create) | |
832 | { | |
efe70c29 JK |
833 | /* We don't expect handle for direct IO */ |
834 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
835 | ||
e84dfbe2 JK |
836 | if (!create) |
837 | return _ext4_get_block(inode, iblock, bh, 0); | |
838 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); | |
705965bd JK |
839 | } |
840 | ||
841 | /* | |
109811c2 | 842 | * Get block function for AIO DIO writes when we create unwritten extent if |
705965bd JK |
843 | * blocks are not allocated yet. The extent will be converted to written |
844 | * after IO is complete. | |
845 | */ | |
109811c2 JK |
846 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
847 | sector_t iblock, struct buffer_head *bh_result, int create) | |
705965bd | 848 | { |
efe70c29 JK |
849 | int ret; |
850 | ||
efe70c29 JK |
851 | /* We don't expect handle for direct IO */ |
852 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
853 | ||
e84dfbe2 JK |
854 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
855 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
efe70c29 | 856 | |
109811c2 JK |
857 | /* |
858 | * When doing DIO using unwritten extents, we need io_end to convert | |
859 | * unwritten extents to written on IO completion. We allocate io_end | |
860 | * once we spot unwritten extent and store it in b_private. Generic | |
861 | * DIO code keeps b_private set and furthermore passes the value to | |
862 | * our completion callback in 'private' argument. | |
863 | */ | |
864 | if (!ret && buffer_unwritten(bh_result)) { | |
865 | if (!bh_result->b_private) { | |
866 | ext4_io_end_t *io_end; | |
867 | ||
868 | io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
869 | if (!io_end) | |
870 | return -ENOMEM; | |
871 | bh_result->b_private = io_end; | |
872 | ext4_set_io_unwritten_flag(inode, io_end); | |
873 | } | |
efe70c29 | 874 | set_buffer_defer_completion(bh_result); |
efe70c29 JK |
875 | } |
876 | ||
877 | return ret; | |
705965bd JK |
878 | } |
879 | ||
109811c2 JK |
880 | /* |
881 | * Get block function for non-AIO DIO writes when we create unwritten extent if | |
882 | * blocks are not allocated yet. The extent will be converted to written | |
883 | * after IO is complete from ext4_ext_direct_IO() function. | |
884 | */ | |
885 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, | |
886 | sector_t iblock, struct buffer_head *bh_result, int create) | |
887 | { | |
109811c2 JK |
888 | int ret; |
889 | ||
890 | /* We don't expect handle for direct IO */ | |
891 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
892 | ||
e84dfbe2 JK |
893 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
894 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
109811c2 JK |
895 | |
896 | /* | |
897 | * Mark inode as having pending DIO writes to unwritten extents. | |
898 | * ext4_ext_direct_IO() checks this flag and converts extents to | |
899 | * written. | |
900 | */ | |
901 | if (!ret && buffer_unwritten(bh_result)) | |
902 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
903 | ||
904 | return ret; | |
905 | } | |
906 | ||
705965bd JK |
907 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
908 | struct buffer_head *bh_result, int create) | |
909 | { | |
910 | int ret; | |
911 | ||
912 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", | |
913 | inode->i_ino, create); | |
efe70c29 JK |
914 | /* We don't expect handle for direct IO */ |
915 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
916 | ||
705965bd JK |
917 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
918 | /* | |
919 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
920 | * that. | |
921 | */ | |
efe70c29 | 922 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
705965bd JK |
923 | |
924 | return ret; | |
925 | } | |
926 | ||
927 | ||
ac27a0ec DK |
928 | /* |
929 | * `handle' can be NULL if create is zero | |
930 | */ | |
617ba13b | 931 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 932 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 933 | { |
2ed88685 TT |
934 | struct ext4_map_blocks map; |
935 | struct buffer_head *bh; | |
c5e298ae | 936 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 937 | int err; |
ac27a0ec DK |
938 | |
939 | J_ASSERT(handle != NULL || create == 0); | |
940 | ||
2ed88685 TT |
941 | map.m_lblk = block; |
942 | map.m_len = 1; | |
c5e298ae | 943 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 944 | |
10560082 TT |
945 | if (err == 0) |
946 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 947 | if (err < 0) |
10560082 | 948 | return ERR_PTR(err); |
2ed88685 TT |
949 | |
950 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
951 | if (unlikely(!bh)) |
952 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
953 | if (map.m_flags & EXT4_MAP_NEW) { |
954 | J_ASSERT(create != 0); | |
955 | J_ASSERT(handle != NULL); | |
ac27a0ec | 956 | |
2ed88685 TT |
957 | /* |
958 | * Now that we do not always journal data, we should | |
959 | * keep in mind whether this should always journal the | |
960 | * new buffer as metadata. For now, regular file | |
961 | * writes use ext4_get_block instead, so it's not a | |
962 | * problem. | |
963 | */ | |
964 | lock_buffer(bh); | |
965 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
966 | err = ext4_journal_get_create_access(handle, bh); |
967 | if (unlikely(err)) { | |
968 | unlock_buffer(bh); | |
969 | goto errout; | |
970 | } | |
971 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
972 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
973 | set_buffer_uptodate(bh); | |
ac27a0ec | 974 | } |
2ed88685 TT |
975 | unlock_buffer(bh); |
976 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
977 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
978 | if (unlikely(err)) |
979 | goto errout; | |
980 | } else | |
2ed88685 | 981 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 982 | return bh; |
10560082 TT |
983 | errout: |
984 | brelse(bh); | |
985 | return ERR_PTR(err); | |
ac27a0ec DK |
986 | } |
987 | ||
617ba13b | 988 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 989 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 990 | { |
af5bc92d | 991 | struct buffer_head *bh; |
ac27a0ec | 992 | |
c5e298ae | 993 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 994 | if (IS_ERR(bh)) |
ac27a0ec | 995 | return bh; |
1c215028 | 996 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 997 | return bh; |
dfec8a14 | 998 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
999 | wait_on_buffer(bh); |
1000 | if (buffer_uptodate(bh)) | |
1001 | return bh; | |
1002 | put_bh(bh); | |
1c215028 | 1003 | return ERR_PTR(-EIO); |
ac27a0ec DK |
1004 | } |
1005 | ||
f19d5870 TM |
1006 | int ext4_walk_page_buffers(handle_t *handle, |
1007 | struct buffer_head *head, | |
1008 | unsigned from, | |
1009 | unsigned to, | |
1010 | int *partial, | |
1011 | int (*fn)(handle_t *handle, | |
1012 | struct buffer_head *bh)) | |
ac27a0ec DK |
1013 | { |
1014 | struct buffer_head *bh; | |
1015 | unsigned block_start, block_end; | |
1016 | unsigned blocksize = head->b_size; | |
1017 | int err, ret = 0; | |
1018 | struct buffer_head *next; | |
1019 | ||
af5bc92d TT |
1020 | for (bh = head, block_start = 0; |
1021 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1022 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1023 | next = bh->b_this_page; |
1024 | block_end = block_start + blocksize; | |
1025 | if (block_end <= from || block_start >= to) { | |
1026 | if (partial && !buffer_uptodate(bh)) | |
1027 | *partial = 1; | |
1028 | continue; | |
1029 | } | |
1030 | err = (*fn)(handle, bh); | |
1031 | if (!ret) | |
1032 | ret = err; | |
1033 | } | |
1034 | return ret; | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * To preserve ordering, it is essential that the hole instantiation and | |
1039 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1040 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1041 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1042 | * prepare_write() is the right place. |
1043 | * | |
36ade451 JK |
1044 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1045 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1046 | * whole page. So we won't block on the journal in that case, which is good, | |
1047 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1048 | * |
617ba13b | 1049 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1050 | * quota file writes. If we were to commit the transaction while thus |
1051 | * reentered, there can be a deadlock - we would be holding a quota | |
1052 | * lock, and the commit would never complete if another thread had a | |
1053 | * transaction open and was blocking on the quota lock - a ranking | |
1054 | * violation. | |
1055 | * | |
dab291af | 1056 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1057 | * will _not_ run commit under these circumstances because handle->h_ref |
1058 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1059 | * write. | |
1060 | */ | |
f19d5870 TM |
1061 | int do_journal_get_write_access(handle_t *handle, |
1062 | struct buffer_head *bh) | |
ac27a0ec | 1063 | { |
56d35a4c JK |
1064 | int dirty = buffer_dirty(bh); |
1065 | int ret; | |
1066 | ||
ac27a0ec DK |
1067 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1068 | return 0; | |
56d35a4c | 1069 | /* |
ebdec241 | 1070 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1071 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1072 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1073 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1074 | * the bit before releasing a page lock and thus writeback cannot |
1075 | * ever write the buffer. | |
1076 | */ | |
1077 | if (dirty) | |
1078 | clear_buffer_dirty(bh); | |
5d601255 | 1079 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1080 | ret = ext4_journal_get_write_access(handle, bh); |
1081 | if (!ret && dirty) | |
1082 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1083 | return ret; | |
ac27a0ec DK |
1084 | } |
1085 | ||
2058f83a MH |
1086 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1087 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
1088 | get_block_t *get_block) | |
1089 | { | |
09cbfeaf | 1090 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1091 | unsigned to = from + len; |
1092 | struct inode *inode = page->mapping->host; | |
1093 | unsigned block_start, block_end; | |
1094 | sector_t block; | |
1095 | int err = 0; | |
1096 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1097 | unsigned bbits; | |
1098 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
1099 | bool decrypt = false; | |
1100 | ||
1101 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1102 | BUG_ON(from > PAGE_SIZE); |
1103 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1104 | BUG_ON(from > to); |
1105 | ||
1106 | if (!page_has_buffers(page)) | |
1107 | create_empty_buffers(page, blocksize, 0); | |
1108 | head = page_buffers(page); | |
1109 | bbits = ilog2(blocksize); | |
09cbfeaf | 1110 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1111 | |
1112 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1113 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1114 | block_end = block_start + blocksize; | |
1115 | if (block_end <= from || block_start >= to) { | |
1116 | if (PageUptodate(page)) { | |
1117 | if (!buffer_uptodate(bh)) | |
1118 | set_buffer_uptodate(bh); | |
1119 | } | |
1120 | continue; | |
1121 | } | |
1122 | if (buffer_new(bh)) | |
1123 | clear_buffer_new(bh); | |
1124 | if (!buffer_mapped(bh)) { | |
1125 | WARN_ON(bh->b_size != blocksize); | |
1126 | err = get_block(inode, block, bh, 1); | |
1127 | if (err) | |
1128 | break; | |
1129 | if (buffer_new(bh)) { | |
1130 | unmap_underlying_metadata(bh->b_bdev, | |
1131 | bh->b_blocknr); | |
1132 | if (PageUptodate(page)) { | |
1133 | clear_buffer_new(bh); | |
1134 | set_buffer_uptodate(bh); | |
1135 | mark_buffer_dirty(bh); | |
1136 | continue; | |
1137 | } | |
1138 | if (block_end > to || block_start < from) | |
1139 | zero_user_segments(page, to, block_end, | |
1140 | block_start, from); | |
1141 | continue; | |
1142 | } | |
1143 | } | |
1144 | if (PageUptodate(page)) { | |
1145 | if (!buffer_uptodate(bh)) | |
1146 | set_buffer_uptodate(bh); | |
1147 | continue; | |
1148 | } | |
1149 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1150 | !buffer_unwritten(bh) && | |
1151 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1152 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
2058f83a MH |
1153 | *wait_bh++ = bh; |
1154 | decrypt = ext4_encrypted_inode(inode) && | |
1155 | S_ISREG(inode->i_mode); | |
1156 | } | |
1157 | } | |
1158 | /* | |
1159 | * If we issued read requests, let them complete. | |
1160 | */ | |
1161 | while (wait_bh > wait) { | |
1162 | wait_on_buffer(*--wait_bh); | |
1163 | if (!buffer_uptodate(*wait_bh)) | |
1164 | err = -EIO; | |
1165 | } | |
1166 | if (unlikely(err)) | |
1167 | page_zero_new_buffers(page, from, to); | |
1168 | else if (decrypt) | |
7821d4dd | 1169 | err = fscrypt_decrypt_page(page->mapping->host, page, |
9c4bb8a3 | 1170 | PAGE_SIZE, 0, page->index); |
2058f83a MH |
1171 | return err; |
1172 | } | |
1173 | #endif | |
1174 | ||
bfc1af65 | 1175 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1176 | loff_t pos, unsigned len, unsigned flags, |
1177 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1178 | { |
af5bc92d | 1179 | struct inode *inode = mapping->host; |
1938a150 | 1180 | int ret, needed_blocks; |
ac27a0ec DK |
1181 | handle_t *handle; |
1182 | int retries = 0; | |
af5bc92d | 1183 | struct page *page; |
de9a55b8 | 1184 | pgoff_t index; |
af5bc92d | 1185 | unsigned from, to; |
bfc1af65 | 1186 | |
9bffad1e | 1187 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1188 | /* |
1189 | * Reserve one block more for addition to orphan list in case | |
1190 | * we allocate blocks but write fails for some reason | |
1191 | */ | |
1192 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1193 | index = pos >> PAGE_SHIFT; |
1194 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1195 | to = from + len; |
ac27a0ec | 1196 | |
f19d5870 TM |
1197 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1198 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1199 | flags, pagep); | |
1200 | if (ret < 0) | |
47564bfb TT |
1201 | return ret; |
1202 | if (ret == 1) | |
1203 | return 0; | |
f19d5870 TM |
1204 | } |
1205 | ||
47564bfb TT |
1206 | /* |
1207 | * grab_cache_page_write_begin() can take a long time if the | |
1208 | * system is thrashing due to memory pressure, or if the page | |
1209 | * is being written back. So grab it first before we start | |
1210 | * the transaction handle. This also allows us to allocate | |
1211 | * the page (if needed) without using GFP_NOFS. | |
1212 | */ | |
1213 | retry_grab: | |
1214 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1215 | if (!page) | |
1216 | return -ENOMEM; | |
1217 | unlock_page(page); | |
1218 | ||
1219 | retry_journal: | |
9924a92a | 1220 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1221 | if (IS_ERR(handle)) { |
09cbfeaf | 1222 | put_page(page); |
47564bfb | 1223 | return PTR_ERR(handle); |
7479d2b9 | 1224 | } |
ac27a0ec | 1225 | |
47564bfb TT |
1226 | lock_page(page); |
1227 | if (page->mapping != mapping) { | |
1228 | /* The page got truncated from under us */ | |
1229 | unlock_page(page); | |
09cbfeaf | 1230 | put_page(page); |
cf108bca | 1231 | ext4_journal_stop(handle); |
47564bfb | 1232 | goto retry_grab; |
cf108bca | 1233 | } |
7afe5aa5 DM |
1234 | /* In case writeback began while the page was unlocked */ |
1235 | wait_for_stable_page(page); | |
cf108bca | 1236 | |
2058f83a MH |
1237 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1238 | if (ext4_should_dioread_nolock(inode)) | |
1239 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1240 | ext4_get_block_unwritten); |
2058f83a MH |
1241 | else |
1242 | ret = ext4_block_write_begin(page, pos, len, | |
1243 | ext4_get_block); | |
1244 | #else | |
744692dc | 1245 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1246 | ret = __block_write_begin(page, pos, len, |
1247 | ext4_get_block_unwritten); | |
744692dc | 1248 | else |
6e1db88d | 1249 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1250 | #endif |
bfc1af65 | 1251 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1252 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1253 | from, to, NULL, | |
1254 | do_journal_get_write_access); | |
ac27a0ec | 1255 | } |
bfc1af65 NP |
1256 | |
1257 | if (ret) { | |
af5bc92d | 1258 | unlock_page(page); |
ae4d5372 | 1259 | /* |
6e1db88d | 1260 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1261 | * outside i_size. Trim these off again. Don't need |
1262 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1263 | * |
1264 | * Add inode to orphan list in case we crash before | |
1265 | * truncate finishes | |
ae4d5372 | 1266 | */ |
ffacfa7a | 1267 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1268 | ext4_orphan_add(handle, inode); |
1269 | ||
1270 | ext4_journal_stop(handle); | |
1271 | if (pos + len > inode->i_size) { | |
b9a4207d | 1272 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1273 | /* |
ffacfa7a | 1274 | * If truncate failed early the inode might |
1938a150 AK |
1275 | * still be on the orphan list; we need to |
1276 | * make sure the inode is removed from the | |
1277 | * orphan list in that case. | |
1278 | */ | |
1279 | if (inode->i_nlink) | |
1280 | ext4_orphan_del(NULL, inode); | |
1281 | } | |
bfc1af65 | 1282 | |
47564bfb TT |
1283 | if (ret == -ENOSPC && |
1284 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1285 | goto retry_journal; | |
09cbfeaf | 1286 | put_page(page); |
47564bfb TT |
1287 | return ret; |
1288 | } | |
1289 | *pagep = page; | |
ac27a0ec DK |
1290 | return ret; |
1291 | } | |
1292 | ||
bfc1af65 NP |
1293 | /* For write_end() in data=journal mode */ |
1294 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1295 | { |
13fca323 | 1296 | int ret; |
ac27a0ec DK |
1297 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1298 | return 0; | |
1299 | set_buffer_uptodate(bh); | |
13fca323 TT |
1300 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1301 | clear_buffer_meta(bh); | |
1302 | clear_buffer_prio(bh); | |
1303 | return ret; | |
ac27a0ec DK |
1304 | } |
1305 | ||
eed4333f ZL |
1306 | /* |
1307 | * We need to pick up the new inode size which generic_commit_write gave us | |
1308 | * `file' can be NULL - eg, when called from page_symlink(). | |
1309 | * | |
1310 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1311 | * buffers are managed internally. | |
1312 | */ | |
1313 | static int ext4_write_end(struct file *file, | |
1314 | struct address_space *mapping, | |
1315 | loff_t pos, unsigned len, unsigned copied, | |
1316 | struct page *page, void *fsdata) | |
f8514083 | 1317 | { |
f8514083 | 1318 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1319 | struct inode *inode = mapping->host; |
0572639f | 1320 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1321 | int ret = 0, ret2; |
1322 | int i_size_changed = 0; | |
1323 | ||
1324 | trace_ext4_write_end(inode, pos, len, copied); | |
42c832de TT |
1325 | if (ext4_has_inline_data(inode)) { |
1326 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1327 | copied, page); | |
1328 | if (ret < 0) | |
1329 | goto errout; | |
1330 | copied = ret; | |
1331 | } else | |
f19d5870 TM |
1332 | copied = block_write_end(file, mapping, pos, |
1333 | len, copied, page, fsdata); | |
f8514083 | 1334 | /* |
4631dbf6 | 1335 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1336 | * page writeout could otherwise come in and zero beyond i_size. |
1337 | */ | |
4631dbf6 | 1338 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 | 1339 | unlock_page(page); |
09cbfeaf | 1340 | put_page(page); |
f8514083 | 1341 | |
0572639f XW |
1342 | if (old_size < pos) |
1343 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1344 | /* |
1345 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1346 | * makes the holding time of page lock longer. Second, it forces lock | |
1347 | * ordering of page lock and transaction start for journaling | |
1348 | * filesystems. | |
1349 | */ | |
1350 | if (i_size_changed) | |
1351 | ext4_mark_inode_dirty(handle, inode); | |
1352 | ||
ffacfa7a | 1353 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1354 | /* if we have allocated more blocks and copied |
1355 | * less. We will have blocks allocated outside | |
1356 | * inode->i_size. So truncate them | |
1357 | */ | |
1358 | ext4_orphan_add(handle, inode); | |
74d553aa | 1359 | errout: |
617ba13b | 1360 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1361 | if (!ret) |
1362 | ret = ret2; | |
bfc1af65 | 1363 | |
f8514083 | 1364 | if (pos + len > inode->i_size) { |
b9a4207d | 1365 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1366 | /* |
ffacfa7a | 1367 | * If truncate failed early the inode might still be |
f8514083 AK |
1368 | * on the orphan list; we need to make sure the inode |
1369 | * is removed from the orphan list in that case. | |
1370 | */ | |
1371 | if (inode->i_nlink) | |
1372 | ext4_orphan_del(NULL, inode); | |
1373 | } | |
1374 | ||
bfc1af65 | 1375 | return ret ? ret : copied; |
ac27a0ec DK |
1376 | } |
1377 | ||
b90197b6 TT |
1378 | /* |
1379 | * This is a private version of page_zero_new_buffers() which doesn't | |
1380 | * set the buffer to be dirty, since in data=journalled mode we need | |
1381 | * to call ext4_handle_dirty_metadata() instead. | |
1382 | */ | |
1383 | static void zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1384 | { | |
1385 | unsigned int block_start = 0, block_end; | |
1386 | struct buffer_head *head, *bh; | |
1387 | ||
1388 | bh = head = page_buffers(page); | |
1389 | do { | |
1390 | block_end = block_start + bh->b_size; | |
1391 | if (buffer_new(bh)) { | |
1392 | if (block_end > from && block_start < to) { | |
1393 | if (!PageUptodate(page)) { | |
1394 | unsigned start, size; | |
1395 | ||
1396 | start = max(from, block_start); | |
1397 | size = min(to, block_end) - start; | |
1398 | ||
1399 | zero_user(page, start, size); | |
1400 | set_buffer_uptodate(bh); | |
1401 | } | |
1402 | clear_buffer_new(bh); | |
1403 | } | |
1404 | } | |
1405 | block_start = block_end; | |
1406 | bh = bh->b_this_page; | |
1407 | } while (bh != head); | |
1408 | } | |
1409 | ||
bfc1af65 | 1410 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1411 | struct address_space *mapping, |
1412 | loff_t pos, unsigned len, unsigned copied, | |
1413 | struct page *page, void *fsdata) | |
ac27a0ec | 1414 | { |
617ba13b | 1415 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1416 | struct inode *inode = mapping->host; |
0572639f | 1417 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1418 | int ret = 0, ret2; |
1419 | int partial = 0; | |
bfc1af65 | 1420 | unsigned from, to; |
4631dbf6 | 1421 | int size_changed = 0; |
ac27a0ec | 1422 | |
9bffad1e | 1423 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1424 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1425 | to = from + len; |
1426 | ||
441c8508 CW |
1427 | BUG_ON(!ext4_handle_valid(handle)); |
1428 | ||
3fdcfb66 TM |
1429 | if (ext4_has_inline_data(inode)) |
1430 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1431 | copied, page); | |
1432 | else { | |
1433 | if (copied < len) { | |
1434 | if (!PageUptodate(page)) | |
1435 | copied = 0; | |
b90197b6 | 1436 | zero_new_buffers(page, from+copied, to); |
3fdcfb66 | 1437 | } |
ac27a0ec | 1438 | |
3fdcfb66 TM |
1439 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1440 | to, &partial, write_end_fn); | |
1441 | if (!partial) | |
1442 | SetPageUptodate(page); | |
1443 | } | |
4631dbf6 | 1444 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1445 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1446 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1447 | unlock_page(page); |
09cbfeaf | 1448 | put_page(page); |
4631dbf6 | 1449 | |
0572639f XW |
1450 | if (old_size < pos) |
1451 | pagecache_isize_extended(inode, old_size, pos); | |
1452 | ||
4631dbf6 | 1453 | if (size_changed) { |
617ba13b | 1454 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1455 | if (!ret) |
1456 | ret = ret2; | |
1457 | } | |
bfc1af65 | 1458 | |
ffacfa7a | 1459 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1460 | /* if we have allocated more blocks and copied |
1461 | * less. We will have blocks allocated outside | |
1462 | * inode->i_size. So truncate them | |
1463 | */ | |
1464 | ext4_orphan_add(handle, inode); | |
1465 | ||
617ba13b | 1466 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1467 | if (!ret) |
1468 | ret = ret2; | |
f8514083 | 1469 | if (pos + len > inode->i_size) { |
b9a4207d | 1470 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1471 | /* |
ffacfa7a | 1472 | * If truncate failed early the inode might still be |
f8514083 AK |
1473 | * on the orphan list; we need to make sure the inode |
1474 | * is removed from the orphan list in that case. | |
1475 | */ | |
1476 | if (inode->i_nlink) | |
1477 | ext4_orphan_del(NULL, inode); | |
1478 | } | |
bfc1af65 NP |
1479 | |
1480 | return ret ? ret : copied; | |
ac27a0ec | 1481 | } |
d2a17637 | 1482 | |
9d0be502 | 1483 | /* |
c27e43a1 | 1484 | * Reserve space for a single cluster |
9d0be502 | 1485 | */ |
c27e43a1 | 1486 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1487 | { |
60e58e0f | 1488 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1489 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1490 | int ret; |
03179fe9 TT |
1491 | |
1492 | /* | |
1493 | * We will charge metadata quota at writeout time; this saves | |
1494 | * us from metadata over-estimation, though we may go over by | |
1495 | * a small amount in the end. Here we just reserve for data. | |
1496 | */ | |
1497 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1498 | if (ret) | |
1499 | return ret; | |
d2a17637 | 1500 | |
0637c6f4 | 1501 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1502 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1503 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1504 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1505 | return -ENOSPC; |
1506 | } | |
9d0be502 | 1507 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1508 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1509 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1510 | |
d2a17637 MC |
1511 | return 0; /* success */ |
1512 | } | |
1513 | ||
12219aea | 1514 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1515 | { |
1516 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1517 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1518 | |
cd213226 MC |
1519 | if (!to_free) |
1520 | return; /* Nothing to release, exit */ | |
1521 | ||
d2a17637 | 1522 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1523 | |
5a58ec87 | 1524 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1525 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1526 | /* |
0637c6f4 TT |
1527 | * if there aren't enough reserved blocks, then the |
1528 | * counter is messed up somewhere. Since this | |
1529 | * function is called from invalidate page, it's | |
1530 | * harmless to return without any action. | |
cd213226 | 1531 | */ |
8de5c325 | 1532 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1533 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1534 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1535 | ei->i_reserved_data_blocks); |
1536 | WARN_ON(1); | |
1537 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1538 | } |
0637c6f4 | 1539 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1540 | |
72b8ab9d | 1541 | /* update fs dirty data blocks counter */ |
57042651 | 1542 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1543 | |
d2a17637 | 1544 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1545 | |
7b415bf6 | 1546 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1547 | } |
1548 | ||
1549 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1550 | unsigned int offset, |
1551 | unsigned int length) | |
d2a17637 | 1552 | { |
9705acd6 | 1553 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1554 | struct buffer_head *head, *bh; |
1555 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1556 | struct inode *inode = page->mapping->host; |
1557 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1558 | unsigned int stop = offset + length; |
7b415bf6 | 1559 | int num_clusters; |
51865fda | 1560 | ext4_fsblk_t lblk; |
d2a17637 | 1561 | |
09cbfeaf | 1562 | BUG_ON(stop > PAGE_SIZE || stop < length); |
ca99fdd2 | 1563 | |
d2a17637 MC |
1564 | head = page_buffers(page); |
1565 | bh = head; | |
1566 | do { | |
1567 | unsigned int next_off = curr_off + bh->b_size; | |
1568 | ||
ca99fdd2 LC |
1569 | if (next_off > stop) |
1570 | break; | |
1571 | ||
d2a17637 MC |
1572 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1573 | to_release++; | |
9705acd6 | 1574 | contiguous_blks++; |
d2a17637 | 1575 | clear_buffer_delay(bh); |
9705acd6 LC |
1576 | } else if (contiguous_blks) { |
1577 | lblk = page->index << | |
09cbfeaf | 1578 | (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1579 | lblk += (curr_off >> inode->i_blkbits) - |
1580 | contiguous_blks; | |
1581 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1582 | contiguous_blks = 0; | |
d2a17637 MC |
1583 | } |
1584 | curr_off = next_off; | |
1585 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1586 | |
9705acd6 | 1587 | if (contiguous_blks) { |
09cbfeaf | 1588 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1589 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1590 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1591 | } |
1592 | ||
7b415bf6 AK |
1593 | /* If we have released all the blocks belonging to a cluster, then we |
1594 | * need to release the reserved space for that cluster. */ | |
1595 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1596 | while (num_clusters > 0) { | |
09cbfeaf | 1597 | lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
7b415bf6 AK |
1598 | ((num_clusters - 1) << sbi->s_cluster_bits); |
1599 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1600 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1601 | ext4_da_release_space(inode, 1); |
1602 | ||
1603 | num_clusters--; | |
1604 | } | |
d2a17637 | 1605 | } |
ac27a0ec | 1606 | |
64769240 AT |
1607 | /* |
1608 | * Delayed allocation stuff | |
1609 | */ | |
1610 | ||
4e7ea81d JK |
1611 | struct mpage_da_data { |
1612 | struct inode *inode; | |
1613 | struct writeback_control *wbc; | |
6b523df4 | 1614 | |
4e7ea81d JK |
1615 | pgoff_t first_page; /* The first page to write */ |
1616 | pgoff_t next_page; /* Current page to examine */ | |
1617 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1618 | /* |
4e7ea81d JK |
1619 | * Extent to map - this can be after first_page because that can be |
1620 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1621 | * is delalloc or unwritten. | |
791b7f08 | 1622 | */ |
4e7ea81d JK |
1623 | struct ext4_map_blocks map; |
1624 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1625 | }; | |
64769240 | 1626 | |
4e7ea81d JK |
1627 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1628 | bool invalidate) | |
c4a0c46e AK |
1629 | { |
1630 | int nr_pages, i; | |
1631 | pgoff_t index, end; | |
1632 | struct pagevec pvec; | |
1633 | struct inode *inode = mpd->inode; | |
1634 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1635 | |
1636 | /* This is necessary when next_page == 0. */ | |
1637 | if (mpd->first_page >= mpd->next_page) | |
1638 | return; | |
c4a0c46e | 1639 | |
c7f5938a CW |
1640 | index = mpd->first_page; |
1641 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1642 | if (invalidate) { |
1643 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1644 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1645 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1646 | ext4_es_remove_extent(inode, start, last - start + 1); |
1647 | } | |
51865fda | 1648 | |
66bea92c | 1649 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1650 | while (index <= end) { |
1651 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1652 | if (nr_pages == 0) | |
1653 | break; | |
1654 | for (i = 0; i < nr_pages; i++) { | |
1655 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1656 | if (page->index > end) |
c4a0c46e | 1657 | break; |
c4a0c46e AK |
1658 | BUG_ON(!PageLocked(page)); |
1659 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1660 | if (invalidate) { |
4e800c03 | 1661 | if (page_mapped(page)) |
1662 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1663 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1664 | ClearPageUptodate(page); |
1665 | } | |
c4a0c46e AK |
1666 | unlock_page(page); |
1667 | } | |
9b1d0998 JK |
1668 | index = pvec.pages[nr_pages - 1]->index + 1; |
1669 | pagevec_release(&pvec); | |
c4a0c46e | 1670 | } |
c4a0c46e AK |
1671 | } |
1672 | ||
df22291f AK |
1673 | static void ext4_print_free_blocks(struct inode *inode) |
1674 | { | |
1675 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1676 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1677 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1678 | |
1679 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1680 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1681 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1682 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1683 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1684 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1685 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1686 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1687 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1688 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1689 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1690 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1691 | ei->i_reserved_data_blocks); |
df22291f AK |
1692 | return; |
1693 | } | |
1694 | ||
c364b22c | 1695 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1696 | { |
c364b22c | 1697 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1698 | } |
1699 | ||
5356f261 AK |
1700 | /* |
1701 | * This function is grabs code from the very beginning of | |
1702 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1703 | * time. This function looks up the requested blocks and sets the | |
1704 | * buffer delay bit under the protection of i_data_sem. | |
1705 | */ | |
1706 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1707 | struct ext4_map_blocks *map, | |
1708 | struct buffer_head *bh) | |
1709 | { | |
d100eef2 | 1710 | struct extent_status es; |
5356f261 AK |
1711 | int retval; |
1712 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1713 | #ifdef ES_AGGRESSIVE_TEST |
1714 | struct ext4_map_blocks orig_map; | |
1715 | ||
1716 | memcpy(&orig_map, map, sizeof(*map)); | |
1717 | #endif | |
5356f261 AK |
1718 | |
1719 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1720 | invalid_block = ~0; | |
1721 | ||
1722 | map->m_flags = 0; | |
1723 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1724 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1725 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1726 | |
1727 | /* Lookup extent status tree firstly */ | |
1728 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1729 | if (ext4_es_is_hole(&es)) { |
1730 | retval = 0; | |
c8b459f4 | 1731 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1732 | goto add_delayed; |
1733 | } | |
1734 | ||
1735 | /* | |
1736 | * Delayed extent could be allocated by fallocate. | |
1737 | * So we need to check it. | |
1738 | */ | |
1739 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1740 | map_bh(bh, inode->i_sb, invalid_block); | |
1741 | set_buffer_new(bh); | |
1742 | set_buffer_delay(bh); | |
1743 | return 0; | |
1744 | } | |
1745 | ||
1746 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1747 | retval = es.es_len - (iblock - es.es_lblk); | |
1748 | if (retval > map->m_len) | |
1749 | retval = map->m_len; | |
1750 | map->m_len = retval; | |
1751 | if (ext4_es_is_written(&es)) | |
1752 | map->m_flags |= EXT4_MAP_MAPPED; | |
1753 | else if (ext4_es_is_unwritten(&es)) | |
1754 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1755 | else | |
1756 | BUG_ON(1); | |
1757 | ||
921f266b DM |
1758 | #ifdef ES_AGGRESSIVE_TEST |
1759 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1760 | #endif | |
d100eef2 ZL |
1761 | return retval; |
1762 | } | |
1763 | ||
5356f261 AK |
1764 | /* |
1765 | * Try to see if we can get the block without requesting a new | |
1766 | * file system block. | |
1767 | */ | |
c8b459f4 | 1768 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1769 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1770 | retval = 0; |
cbd7584e | 1771 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1772 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1773 | else |
2f8e0a7c | 1774 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1775 | |
d100eef2 | 1776 | add_delayed: |
5356f261 | 1777 | if (retval == 0) { |
f7fec032 | 1778 | int ret; |
5356f261 AK |
1779 | /* |
1780 | * XXX: __block_prepare_write() unmaps passed block, | |
1781 | * is it OK? | |
1782 | */ | |
386ad67c LC |
1783 | /* |
1784 | * If the block was allocated from previously allocated cluster, | |
1785 | * then we don't need to reserve it again. However we still need | |
1786 | * to reserve metadata for every block we're going to write. | |
1787 | */ | |
c27e43a1 | 1788 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1789 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1790 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1791 | if (ret) { |
5356f261 | 1792 | /* not enough space to reserve */ |
f7fec032 | 1793 | retval = ret; |
5356f261 | 1794 | goto out_unlock; |
f7fec032 | 1795 | } |
5356f261 AK |
1796 | } |
1797 | ||
f7fec032 ZL |
1798 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1799 | ~0, EXTENT_STATUS_DELAYED); | |
1800 | if (ret) { | |
1801 | retval = ret; | |
51865fda | 1802 | goto out_unlock; |
f7fec032 | 1803 | } |
51865fda | 1804 | |
5356f261 AK |
1805 | map_bh(bh, inode->i_sb, invalid_block); |
1806 | set_buffer_new(bh); | |
1807 | set_buffer_delay(bh); | |
f7fec032 ZL |
1808 | } else if (retval > 0) { |
1809 | int ret; | |
3be78c73 | 1810 | unsigned int status; |
f7fec032 | 1811 | |
44fb851d ZL |
1812 | if (unlikely(retval != map->m_len)) { |
1813 | ext4_warning(inode->i_sb, | |
1814 | "ES len assertion failed for inode " | |
1815 | "%lu: retval %d != map->m_len %d", | |
1816 | inode->i_ino, retval, map->m_len); | |
1817 | WARN_ON(1); | |
921f266b | 1818 | } |
921f266b | 1819 | |
f7fec032 ZL |
1820 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1821 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1822 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1823 | map->m_pblk, status); | |
1824 | if (ret != 0) | |
1825 | retval = ret; | |
5356f261 AK |
1826 | } |
1827 | ||
1828 | out_unlock: | |
1829 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1830 | ||
1831 | return retval; | |
1832 | } | |
1833 | ||
64769240 | 1834 | /* |
d91bd2c1 | 1835 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1836 | * ext4_da_write_begin(). It will either return mapped block or |
1837 | * reserve space for a single block. | |
29fa89d0 AK |
1838 | * |
1839 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1840 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1841 | * | |
1842 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1843 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1844 | * initialized properly. | |
64769240 | 1845 | */ |
9c3569b5 TM |
1846 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1847 | struct buffer_head *bh, int create) | |
64769240 | 1848 | { |
2ed88685 | 1849 | struct ext4_map_blocks map; |
64769240 AT |
1850 | int ret = 0; |
1851 | ||
1852 | BUG_ON(create == 0); | |
2ed88685 TT |
1853 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1854 | ||
1855 | map.m_lblk = iblock; | |
1856 | map.m_len = 1; | |
64769240 AT |
1857 | |
1858 | /* | |
1859 | * first, we need to know whether the block is allocated already | |
1860 | * preallocated blocks are unmapped but should treated | |
1861 | * the same as allocated blocks. | |
1862 | */ | |
5356f261 AK |
1863 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1864 | if (ret <= 0) | |
2ed88685 | 1865 | return ret; |
64769240 | 1866 | |
2ed88685 | 1867 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1868 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1869 | |
1870 | if (buffer_unwritten(bh)) { | |
1871 | /* A delayed write to unwritten bh should be marked | |
1872 | * new and mapped. Mapped ensures that we don't do | |
1873 | * get_block multiple times when we write to the same | |
1874 | * offset and new ensures that we do proper zero out | |
1875 | * for partial write. | |
1876 | */ | |
1877 | set_buffer_new(bh); | |
c8205636 | 1878 | set_buffer_mapped(bh); |
2ed88685 TT |
1879 | } |
1880 | return 0; | |
64769240 | 1881 | } |
61628a3f | 1882 | |
62e086be AK |
1883 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1884 | { | |
1885 | get_bh(bh); | |
1886 | return 0; | |
1887 | } | |
1888 | ||
1889 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1890 | { | |
1891 | put_bh(bh); | |
1892 | return 0; | |
1893 | } | |
1894 | ||
1895 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1896 | unsigned int len) |
1897 | { | |
1898 | struct address_space *mapping = page->mapping; | |
1899 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1900 | struct buffer_head *page_bufs = NULL; |
62e086be | 1901 | handle_t *handle = NULL; |
3fdcfb66 TM |
1902 | int ret = 0, err = 0; |
1903 | int inline_data = ext4_has_inline_data(inode); | |
1904 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1905 | |
cb20d518 | 1906 | ClearPageChecked(page); |
3fdcfb66 TM |
1907 | |
1908 | if (inline_data) { | |
1909 | BUG_ON(page->index != 0); | |
1910 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1911 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1912 | if (inode_bh == NULL) | |
1913 | goto out; | |
1914 | } else { | |
1915 | page_bufs = page_buffers(page); | |
1916 | if (!page_bufs) { | |
1917 | BUG(); | |
1918 | goto out; | |
1919 | } | |
1920 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1921 | NULL, bget_one); | |
1922 | } | |
bdf96838 TT |
1923 | /* |
1924 | * We need to release the page lock before we start the | |
1925 | * journal, so grab a reference so the page won't disappear | |
1926 | * out from under us. | |
1927 | */ | |
1928 | get_page(page); | |
62e086be AK |
1929 | unlock_page(page); |
1930 | ||
9924a92a TT |
1931 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1932 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1933 | if (IS_ERR(handle)) { |
1934 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1935 | put_page(page); |
1936 | goto out_no_pagelock; | |
62e086be | 1937 | } |
441c8508 CW |
1938 | BUG_ON(!ext4_handle_valid(handle)); |
1939 | ||
bdf96838 TT |
1940 | lock_page(page); |
1941 | put_page(page); | |
1942 | if (page->mapping != mapping) { | |
1943 | /* The page got truncated from under us */ | |
1944 | ext4_journal_stop(handle); | |
1945 | ret = 0; | |
1946 | goto out; | |
1947 | } | |
1948 | ||
3fdcfb66 | 1949 | if (inline_data) { |
5d601255 | 1950 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1951 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1952 | |
3fdcfb66 TM |
1953 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1954 | ||
1955 | } else { | |
1956 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1957 | do_journal_get_write_access); | |
1958 | ||
1959 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1960 | write_end_fn); | |
1961 | } | |
62e086be AK |
1962 | if (ret == 0) |
1963 | ret = err; | |
2d859db3 | 1964 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1965 | err = ext4_journal_stop(handle); |
1966 | if (!ret) | |
1967 | ret = err; | |
1968 | ||
3fdcfb66 | 1969 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1970 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1971 | NULL, bput_one); |
19f5fb7a | 1972 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1973 | out: |
bdf96838 TT |
1974 | unlock_page(page); |
1975 | out_no_pagelock: | |
3fdcfb66 | 1976 | brelse(inode_bh); |
62e086be AK |
1977 | return ret; |
1978 | } | |
1979 | ||
61628a3f | 1980 | /* |
43ce1d23 AK |
1981 | * Note that we don't need to start a transaction unless we're journaling data |
1982 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1983 | * need to file the inode to the transaction's list in ordered mode because if | |
1984 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1985 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1986 | * transaction the data will hit the disk. In case we are journaling data, we |
1987 | * cannot start transaction directly because transaction start ranks above page | |
1988 | * lock so we have to do some magic. | |
1989 | * | |
b920c755 | 1990 | * This function can get called via... |
20970ba6 | 1991 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1992 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1993 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1994 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1995 | * |
1996 | * We don't do any block allocation in this function. If we have page with | |
1997 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1998 | * is important for mmaped based write. So if we do with blocksize 1K | |
1999 | * truncate(f, 1024); | |
2000 | * a = mmap(f, 0, 4096); | |
2001 | * a[0] = 'a'; | |
2002 | * truncate(f, 4096); | |
2003 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2004 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2005 | * do_wp_page). So writepage should write the first block. If we modify |
2006 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2007 | * page_mkwrite callback will do the block allocation and mark the | |
2008 | * buffer_heads mapped. | |
2009 | * | |
2010 | * We redirty the page if we have any buffer_heads that is either delay or | |
2011 | * unwritten in the page. | |
2012 | * | |
2013 | * We can get recursively called as show below. | |
2014 | * | |
2015 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2016 | * ext4_writepage() | |
2017 | * | |
2018 | * But since we don't do any block allocation we should not deadlock. | |
2019 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2020 | */ |
43ce1d23 | 2021 | static int ext4_writepage(struct page *page, |
62e086be | 2022 | struct writeback_control *wbc) |
64769240 | 2023 | { |
f8bec370 | 2024 | int ret = 0; |
61628a3f | 2025 | loff_t size; |
498e5f24 | 2026 | unsigned int len; |
744692dc | 2027 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2028 | struct inode *inode = page->mapping->host; |
36ade451 | 2029 | struct ext4_io_submit io_submit; |
1c8349a1 | 2030 | bool keep_towrite = false; |
61628a3f | 2031 | |
a9c667f8 | 2032 | trace_ext4_writepage(page); |
f0e6c985 | 2033 | size = i_size_read(inode); |
09cbfeaf KS |
2034 | if (page->index == size >> PAGE_SHIFT) |
2035 | len = size & ~PAGE_MASK; | |
f0e6c985 | 2036 | else |
09cbfeaf | 2037 | len = PAGE_SIZE; |
64769240 | 2038 | |
a42afc5f | 2039 | page_bufs = page_buffers(page); |
a42afc5f | 2040 | /* |
fe386132 JK |
2041 | * We cannot do block allocation or other extent handling in this |
2042 | * function. If there are buffers needing that, we have to redirty | |
2043 | * the page. But we may reach here when we do a journal commit via | |
2044 | * journal_submit_inode_data_buffers() and in that case we must write | |
2045 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2046 | * |
2047 | * Also, if there is only one buffer per page (the fs block | |
2048 | * size == the page size), if one buffer needs block | |
2049 | * allocation or needs to modify the extent tree to clear the | |
2050 | * unwritten flag, we know that the page can't be written at | |
2051 | * all, so we might as well refuse the write immediately. | |
2052 | * Unfortunately if the block size != page size, we can't as | |
2053 | * easily detect this case using ext4_walk_page_buffers(), but | |
2054 | * for the extremely common case, this is an optimization that | |
2055 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2056 | */ |
f19d5870 TM |
2057 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2058 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2059 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2060 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2061 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2062 | /* |
2063 | * For memory cleaning there's no point in writing only | |
2064 | * some buffers. So just bail out. Warn if we came here | |
2065 | * from direct reclaim. | |
2066 | */ | |
2067 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2068 | == PF_MEMALLOC); | |
f0e6c985 AK |
2069 | unlock_page(page); |
2070 | return 0; | |
2071 | } | |
1c8349a1 | 2072 | keep_towrite = true; |
a42afc5f | 2073 | } |
64769240 | 2074 | |
cb20d518 | 2075 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2076 | /* |
2077 | * It's mmapped pagecache. Add buffers and journal it. There | |
2078 | * doesn't seem much point in redirtying the page here. | |
2079 | */ | |
3f0ca309 | 2080 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2081 | |
97a851ed JK |
2082 | ext4_io_submit_init(&io_submit, wbc); |
2083 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2084 | if (!io_submit.io_end) { | |
2085 | redirty_page_for_writepage(wbc, page); | |
2086 | unlock_page(page); | |
2087 | return -ENOMEM; | |
2088 | } | |
1c8349a1 | 2089 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2090 | ext4_io_submit(&io_submit); |
97a851ed JK |
2091 | /* Drop io_end reference we got from init */ |
2092 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2093 | return ret; |
2094 | } | |
2095 | ||
5f1132b2 JK |
2096 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2097 | { | |
2098 | int len; | |
2099 | loff_t size = i_size_read(mpd->inode); | |
2100 | int err; | |
2101 | ||
2102 | BUG_ON(page->index != mpd->first_page); | |
09cbfeaf KS |
2103 | if (page->index == size >> PAGE_SHIFT) |
2104 | len = size & ~PAGE_MASK; | |
5f1132b2 | 2105 | else |
09cbfeaf | 2106 | len = PAGE_SIZE; |
5f1132b2 | 2107 | clear_page_dirty_for_io(page); |
1c8349a1 | 2108 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2109 | if (!err) |
2110 | mpd->wbc->nr_to_write--; | |
2111 | mpd->first_page++; | |
2112 | ||
2113 | return err; | |
2114 | } | |
2115 | ||
4e7ea81d JK |
2116 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2117 | ||
61628a3f | 2118 | /* |
fffb2739 JK |
2119 | * mballoc gives us at most this number of blocks... |
2120 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2121 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2122 | */ |
fffb2739 | 2123 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2124 | |
4e7ea81d JK |
2125 | /* |
2126 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2127 | * | |
2128 | * @mpd - extent of blocks | |
2129 | * @lblk - logical number of the block in the file | |
09930042 | 2130 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2131 | * |
09930042 JK |
2132 | * The function is used to collect contig. blocks in the same state. If the |
2133 | * buffer doesn't require mapping for writeback and we haven't started the | |
2134 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2135 | * caller can write the buffer right away. Otherwise the function returns true | |
2136 | * if the block has been added to the extent, false if the block couldn't be | |
2137 | * added. | |
4e7ea81d | 2138 | */ |
09930042 JK |
2139 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2140 | struct buffer_head *bh) | |
4e7ea81d JK |
2141 | { |
2142 | struct ext4_map_blocks *map = &mpd->map; | |
2143 | ||
09930042 JK |
2144 | /* Buffer that doesn't need mapping for writeback? */ |
2145 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2146 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2147 | /* So far no extent to map => we write the buffer right away */ | |
2148 | if (map->m_len == 0) | |
2149 | return true; | |
2150 | return false; | |
2151 | } | |
4e7ea81d JK |
2152 | |
2153 | /* First block in the extent? */ | |
2154 | if (map->m_len == 0) { | |
2155 | map->m_lblk = lblk; | |
2156 | map->m_len = 1; | |
09930042 JK |
2157 | map->m_flags = bh->b_state & BH_FLAGS; |
2158 | return true; | |
4e7ea81d JK |
2159 | } |
2160 | ||
09930042 JK |
2161 | /* Don't go larger than mballoc is willing to allocate */ |
2162 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2163 | return false; | |
2164 | ||
4e7ea81d JK |
2165 | /* Can we merge the block to our big extent? */ |
2166 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2167 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2168 | map->m_len++; |
09930042 | 2169 | return true; |
4e7ea81d | 2170 | } |
09930042 | 2171 | return false; |
4e7ea81d JK |
2172 | } |
2173 | ||
5f1132b2 JK |
2174 | /* |
2175 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2176 | * | |
2177 | * @mpd - extent of blocks for mapping | |
2178 | * @head - the first buffer in the page | |
2179 | * @bh - buffer we should start processing from | |
2180 | * @lblk - logical number of the block in the file corresponding to @bh | |
2181 | * | |
2182 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2183 | * the page for IO if all buffers in this page were mapped and there's no | |
2184 | * accumulated extent of buffers to map or add buffers in the page to the | |
2185 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2186 | * by processing the next page, 0 if it should stop adding buffers to the | |
2187 | * extent to map because we cannot extend it anymore. It can also return value | |
2188 | * < 0 in case of error during IO submission. | |
2189 | */ | |
2190 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2191 | struct buffer_head *head, | |
2192 | struct buffer_head *bh, | |
2193 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2194 | { |
2195 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2196 | int err; |
4e7ea81d JK |
2197 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
2198 | >> inode->i_blkbits; | |
2199 | ||
2200 | do { | |
2201 | BUG_ON(buffer_locked(bh)); | |
2202 | ||
09930042 | 2203 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2204 | /* Found extent to map? */ |
2205 | if (mpd->map.m_len) | |
5f1132b2 | 2206 | return 0; |
09930042 | 2207 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2208 | break; |
4e7ea81d | 2209 | } |
4e7ea81d | 2210 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2211 | /* So far everything mapped? Submit the page for IO. */ |
2212 | if (mpd->map.m_len == 0) { | |
2213 | err = mpage_submit_page(mpd, head->b_page); | |
2214 | if (err < 0) | |
2215 | return err; | |
2216 | } | |
2217 | return lblk < blocks; | |
4e7ea81d JK |
2218 | } |
2219 | ||
2220 | /* | |
2221 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2222 | * submit fully mapped pages for IO | |
2223 | * | |
2224 | * @mpd - description of extent to map, on return next extent to map | |
2225 | * | |
2226 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2227 | * to be already locked) and update buffer state according to new extent state. | |
2228 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2229 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2230 | * and do extent conversion after IO is finished. If the last page is not fully |
2231 | * mapped, we update @map to the next extent in the last page that needs | |
2232 | * mapping. Otherwise we submit the page for IO. | |
2233 | */ | |
2234 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2235 | { | |
2236 | struct pagevec pvec; | |
2237 | int nr_pages, i; | |
2238 | struct inode *inode = mpd->inode; | |
2239 | struct buffer_head *head, *bh; | |
09cbfeaf | 2240 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2241 | pgoff_t start, end; |
2242 | ext4_lblk_t lblk; | |
2243 | sector_t pblock; | |
2244 | int err; | |
2245 | ||
2246 | start = mpd->map.m_lblk >> bpp_bits; | |
2247 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2248 | lblk = start << bpp_bits; | |
2249 | pblock = mpd->map.m_pblk; | |
2250 | ||
2251 | pagevec_init(&pvec, 0); | |
2252 | while (start <= end) { | |
2253 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2254 | PAGEVEC_SIZE); | |
2255 | if (nr_pages == 0) | |
2256 | break; | |
2257 | for (i = 0; i < nr_pages; i++) { | |
2258 | struct page *page = pvec.pages[i]; | |
2259 | ||
2260 | if (page->index > end) | |
2261 | break; | |
70261f56 | 2262 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2263 | BUG_ON(page->index != start); |
2264 | bh = head = page_buffers(page); | |
2265 | do { | |
2266 | if (lblk < mpd->map.m_lblk) | |
2267 | continue; | |
2268 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2269 | /* | |
2270 | * Buffer after end of mapped extent. | |
2271 | * Find next buffer in the page to map. | |
2272 | */ | |
2273 | mpd->map.m_len = 0; | |
2274 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2275 | /* |
2276 | * FIXME: If dioread_nolock supports | |
2277 | * blocksize < pagesize, we need to make | |
2278 | * sure we add size mapped so far to | |
2279 | * io_end->size as the following call | |
2280 | * can submit the page for IO. | |
2281 | */ | |
2282 | err = mpage_process_page_bufs(mpd, head, | |
2283 | bh, lblk); | |
4e7ea81d | 2284 | pagevec_release(&pvec); |
5f1132b2 JK |
2285 | if (err > 0) |
2286 | err = 0; | |
2287 | return err; | |
4e7ea81d JK |
2288 | } |
2289 | if (buffer_delay(bh)) { | |
2290 | clear_buffer_delay(bh); | |
2291 | bh->b_blocknr = pblock++; | |
2292 | } | |
4e7ea81d | 2293 | clear_buffer_unwritten(bh); |
5f1132b2 | 2294 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2295 | |
2296 | /* | |
2297 | * FIXME: This is going to break if dioread_nolock | |
2298 | * supports blocksize < pagesize as we will try to | |
2299 | * convert potentially unmapped parts of inode. | |
2300 | */ | |
09cbfeaf | 2301 | mpd->io_submit.io_end->size += PAGE_SIZE; |
4e7ea81d JK |
2302 | /* Page fully mapped - let IO run! */ |
2303 | err = mpage_submit_page(mpd, page); | |
2304 | if (err < 0) { | |
2305 | pagevec_release(&pvec); | |
2306 | return err; | |
2307 | } | |
2308 | start++; | |
2309 | } | |
2310 | pagevec_release(&pvec); | |
2311 | } | |
2312 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2313 | mpd->map.m_len = 0; | |
2314 | mpd->map.m_flags = 0; | |
2315 | return 0; | |
2316 | } | |
2317 | ||
2318 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2319 | { | |
2320 | struct inode *inode = mpd->inode; | |
2321 | struct ext4_map_blocks *map = &mpd->map; | |
2322 | int get_blocks_flags; | |
090f32ee | 2323 | int err, dioread_nolock; |
4e7ea81d JK |
2324 | |
2325 | trace_ext4_da_write_pages_extent(inode, map); | |
2326 | /* | |
2327 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2328 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2329 | * where we have written into one or more preallocated blocks). It is |
2330 | * possible that we're going to need more metadata blocks than | |
2331 | * previously reserved. However we must not fail because we're in | |
2332 | * writeback and there is nothing we can do about it so it might result | |
2333 | * in data loss. So use reserved blocks to allocate metadata if | |
2334 | * possible. | |
2335 | * | |
754cfed6 TT |
2336 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2337 | * the blocks in question are delalloc blocks. This indicates | |
2338 | * that the blocks and quotas has already been checked when | |
2339 | * the data was copied into the page cache. | |
4e7ea81d JK |
2340 | */ |
2341 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2342 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2343 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2344 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2345 | if (dioread_nolock) | |
4e7ea81d JK |
2346 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2347 | if (map->m_flags & (1 << BH_Delay)) | |
2348 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2349 | ||
2350 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2351 | if (err < 0) | |
2352 | return err; | |
090f32ee | 2353 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2354 | if (!mpd->io_submit.io_end->handle && |
2355 | ext4_handle_valid(handle)) { | |
2356 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2357 | handle->h_rsv_handle = NULL; | |
2358 | } | |
3613d228 | 2359 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2360 | } |
4e7ea81d JK |
2361 | |
2362 | BUG_ON(map->m_len == 0); | |
2363 | if (map->m_flags & EXT4_MAP_NEW) { | |
2364 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2365 | int i; | |
2366 | ||
2367 | for (i = 0; i < map->m_len; i++) | |
2368 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2369 | } | |
2370 | return 0; | |
2371 | } | |
2372 | ||
2373 | /* | |
2374 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2375 | * mpd->len and submit pages underlying it for IO | |
2376 | * | |
2377 | * @handle - handle for journal operations | |
2378 | * @mpd - extent to map | |
7534e854 JK |
2379 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2380 | * is no hope of writing the data. The caller should discard | |
2381 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2382 | * |
2383 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2384 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2385 | * them to initialized or split the described range from larger unwritten | |
2386 | * extent. Note that we need not map all the described range since allocation | |
2387 | * can return less blocks or the range is covered by more unwritten extents. We | |
2388 | * cannot map more because we are limited by reserved transaction credits. On | |
2389 | * the other hand we always make sure that the last touched page is fully | |
2390 | * mapped so that it can be written out (and thus forward progress is | |
2391 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2392 | */ | |
2393 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2394 | struct mpage_da_data *mpd, |
2395 | bool *give_up_on_write) | |
4e7ea81d JK |
2396 | { |
2397 | struct inode *inode = mpd->inode; | |
2398 | struct ext4_map_blocks *map = &mpd->map; | |
2399 | int err; | |
2400 | loff_t disksize; | |
6603120e | 2401 | int progress = 0; |
4e7ea81d JK |
2402 | |
2403 | mpd->io_submit.io_end->offset = | |
2404 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2405 | do { |
4e7ea81d JK |
2406 | err = mpage_map_one_extent(handle, mpd); |
2407 | if (err < 0) { | |
2408 | struct super_block *sb = inode->i_sb; | |
2409 | ||
cb530541 TT |
2410 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2411 | goto invalidate_dirty_pages; | |
4e7ea81d | 2412 | /* |
cb530541 TT |
2413 | * Let the uper layers retry transient errors. |
2414 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2415 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2416 | */ |
cb530541 | 2417 | if ((err == -ENOMEM) || |
6603120e DM |
2418 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2419 | if (progress) | |
2420 | goto update_disksize; | |
cb530541 | 2421 | return err; |
6603120e | 2422 | } |
cb530541 TT |
2423 | ext4_msg(sb, KERN_CRIT, |
2424 | "Delayed block allocation failed for " | |
2425 | "inode %lu at logical offset %llu with" | |
2426 | " max blocks %u with error %d", | |
2427 | inode->i_ino, | |
2428 | (unsigned long long)map->m_lblk, | |
2429 | (unsigned)map->m_len, -err); | |
2430 | ext4_msg(sb, KERN_CRIT, | |
2431 | "This should not happen!! Data will " | |
2432 | "be lost\n"); | |
2433 | if (err == -ENOSPC) | |
2434 | ext4_print_free_blocks(inode); | |
2435 | invalidate_dirty_pages: | |
2436 | *give_up_on_write = true; | |
4e7ea81d JK |
2437 | return err; |
2438 | } | |
6603120e | 2439 | progress = 1; |
4e7ea81d JK |
2440 | /* |
2441 | * Update buffer state, submit mapped pages, and get us new | |
2442 | * extent to map | |
2443 | */ | |
2444 | err = mpage_map_and_submit_buffers(mpd); | |
2445 | if (err < 0) | |
6603120e | 2446 | goto update_disksize; |
27d7c4ed | 2447 | } while (map->m_len); |
4e7ea81d | 2448 | |
6603120e | 2449 | update_disksize: |
622cad13 TT |
2450 | /* |
2451 | * Update on-disk size after IO is submitted. Races with | |
2452 | * truncate are avoided by checking i_size under i_data_sem. | |
2453 | */ | |
09cbfeaf | 2454 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
4e7ea81d JK |
2455 | if (disksize > EXT4_I(inode)->i_disksize) { |
2456 | int err2; | |
622cad13 TT |
2457 | loff_t i_size; |
2458 | ||
2459 | down_write(&EXT4_I(inode)->i_data_sem); | |
2460 | i_size = i_size_read(inode); | |
2461 | if (disksize > i_size) | |
2462 | disksize = i_size; | |
2463 | if (disksize > EXT4_I(inode)->i_disksize) | |
2464 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2465 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2466 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2467 | if (err2) |
2468 | ext4_error(inode->i_sb, | |
2469 | "Failed to mark inode %lu dirty", | |
2470 | inode->i_ino); | |
2471 | if (!err) | |
2472 | err = err2; | |
2473 | } | |
2474 | return err; | |
2475 | } | |
2476 | ||
fffb2739 JK |
2477 | /* |
2478 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2479 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2480 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2481 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2482 | * bpp - 1 blocks in bpp different extents. | |
2483 | */ | |
525f4ed8 MC |
2484 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2485 | { | |
fffb2739 | 2486 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2487 | |
fffb2739 JK |
2488 | return ext4_meta_trans_blocks(inode, |
2489 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2490 | } |
61628a3f | 2491 | |
8e48dcfb | 2492 | /* |
4e7ea81d JK |
2493 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2494 | * and underlying extent to map | |
2495 | * | |
2496 | * @mpd - where to look for pages | |
2497 | * | |
2498 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2499 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2500 | * extent of buffers underlying these pages that needs mapping (formed by | |
2501 | * either delayed or unwritten buffers). We also lock the pages containing | |
2502 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2503 | * mpd->lblk with length mpd->len blocks). | |
2504 | * | |
2505 | * Note that this function can attach bios to one io_end structure which are | |
2506 | * neither logically nor physically contiguous. Although it may seem as an | |
2507 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2508 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2509 | */ |
4e7ea81d | 2510 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2511 | { |
4e7ea81d JK |
2512 | struct address_space *mapping = mpd->inode->i_mapping; |
2513 | struct pagevec pvec; | |
2514 | unsigned int nr_pages; | |
aeac589a | 2515 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2516 | pgoff_t index = mpd->first_page; |
2517 | pgoff_t end = mpd->last_page; | |
2518 | int tag; | |
2519 | int i, err = 0; | |
2520 | int blkbits = mpd->inode->i_blkbits; | |
2521 | ext4_lblk_t lblk; | |
2522 | struct buffer_head *head; | |
8e48dcfb | 2523 | |
4e7ea81d | 2524 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2525 | tag = PAGECACHE_TAG_TOWRITE; |
2526 | else | |
2527 | tag = PAGECACHE_TAG_DIRTY; | |
2528 | ||
4e7ea81d JK |
2529 | pagevec_init(&pvec, 0); |
2530 | mpd->map.m_len = 0; | |
2531 | mpd->next_page = index; | |
4f01b02c | 2532 | while (index <= end) { |
5b41d924 | 2533 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2534 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2535 | if (nr_pages == 0) | |
4e7ea81d | 2536 | goto out; |
8e48dcfb TT |
2537 | |
2538 | for (i = 0; i < nr_pages; i++) { | |
2539 | struct page *page = pvec.pages[i]; | |
2540 | ||
2541 | /* | |
2542 | * At this point, the page may be truncated or | |
2543 | * invalidated (changing page->mapping to NULL), or | |
2544 | * even swizzled back from swapper_space to tmpfs file | |
2545 | * mapping. However, page->index will not change | |
2546 | * because we have a reference on the page. | |
2547 | */ | |
4f01b02c TT |
2548 | if (page->index > end) |
2549 | goto out; | |
8e48dcfb | 2550 | |
aeac589a ML |
2551 | /* |
2552 | * Accumulated enough dirty pages? This doesn't apply | |
2553 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2554 | * keep going because someone may be concurrently | |
2555 | * dirtying pages, and we might have synced a lot of | |
2556 | * newly appeared dirty pages, but have not synced all | |
2557 | * of the old dirty pages. | |
2558 | */ | |
2559 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2560 | goto out; | |
2561 | ||
4e7ea81d JK |
2562 | /* If we can't merge this page, we are done. */ |
2563 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2564 | goto out; | |
78aaced3 | 2565 | |
8e48dcfb | 2566 | lock_page(page); |
8e48dcfb | 2567 | /* |
4e7ea81d JK |
2568 | * If the page is no longer dirty, or its mapping no |
2569 | * longer corresponds to inode we are writing (which | |
2570 | * means it has been truncated or invalidated), or the | |
2571 | * page is already under writeback and we are not doing | |
2572 | * a data integrity writeback, skip the page | |
8e48dcfb | 2573 | */ |
4f01b02c TT |
2574 | if (!PageDirty(page) || |
2575 | (PageWriteback(page) && | |
4e7ea81d | 2576 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2577 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2578 | unlock_page(page); |
2579 | continue; | |
2580 | } | |
2581 | ||
7cb1a535 | 2582 | wait_on_page_writeback(page); |
8e48dcfb | 2583 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2584 | |
4e7ea81d | 2585 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2586 | mpd->first_page = page->index; |
8eb9e5ce | 2587 | mpd->next_page = page->index + 1; |
f8bec370 | 2588 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2589 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2590 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2591 | head = page_buffers(page); |
5f1132b2 JK |
2592 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2593 | if (err <= 0) | |
4e7ea81d | 2594 | goto out; |
5f1132b2 | 2595 | err = 0; |
aeac589a | 2596 | left--; |
8e48dcfb TT |
2597 | } |
2598 | pagevec_release(&pvec); | |
2599 | cond_resched(); | |
2600 | } | |
4f01b02c | 2601 | return 0; |
8eb9e5ce TT |
2602 | out: |
2603 | pagevec_release(&pvec); | |
4e7ea81d | 2604 | return err; |
8e48dcfb TT |
2605 | } |
2606 | ||
20970ba6 TT |
2607 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2608 | void *data) | |
2609 | { | |
2610 | struct address_space *mapping = data; | |
2611 | int ret = ext4_writepage(page, wbc); | |
2612 | mapping_set_error(mapping, ret); | |
2613 | return ret; | |
2614 | } | |
2615 | ||
2616 | static int ext4_writepages(struct address_space *mapping, | |
2617 | struct writeback_control *wbc) | |
64769240 | 2618 | { |
4e7ea81d JK |
2619 | pgoff_t writeback_index = 0; |
2620 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2621 | int range_whole = 0; |
4e7ea81d | 2622 | int cycled = 1; |
61628a3f | 2623 | handle_t *handle = NULL; |
df22291f | 2624 | struct mpage_da_data mpd; |
5e745b04 | 2625 | struct inode *inode = mapping->host; |
6b523df4 | 2626 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2627 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2628 | bool done; |
1bce63d1 | 2629 | struct blk_plug plug; |
cb530541 | 2630 | bool give_up_on_write = false; |
61628a3f | 2631 | |
c8585c6f | 2632 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
20970ba6 | 2633 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2634 | |
c8585c6f DJ |
2635 | if (dax_mapping(mapping)) { |
2636 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, | |
2637 | wbc); | |
2638 | goto out_writepages; | |
2639 | } | |
7f6d5b52 | 2640 | |
61628a3f MC |
2641 | /* |
2642 | * No pages to write? This is mainly a kludge to avoid starting | |
2643 | * a transaction for special inodes like journal inode on last iput() | |
2644 | * because that could violate lock ordering on umount | |
2645 | */ | |
a1d6cc56 | 2646 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2647 | goto out_writepages; |
2a21e37e | 2648 | |
20970ba6 TT |
2649 | if (ext4_should_journal_data(inode)) { |
2650 | struct blk_plug plug; | |
20970ba6 TT |
2651 | |
2652 | blk_start_plug(&plug); | |
2653 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2654 | blk_finish_plug(&plug); | |
bbf023c7 | 2655 | goto out_writepages; |
20970ba6 TT |
2656 | } |
2657 | ||
2a21e37e TT |
2658 | /* |
2659 | * If the filesystem has aborted, it is read-only, so return | |
2660 | * right away instead of dumping stack traces later on that | |
2661 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2662 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2663 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2664 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2665 | * *never* be called, so if that ever happens, we would want |
2666 | * the stack trace. | |
2667 | */ | |
bbf023c7 ML |
2668 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2669 | ret = -EROFS; | |
2670 | goto out_writepages; | |
2671 | } | |
2a21e37e | 2672 | |
6b523df4 JK |
2673 | if (ext4_should_dioread_nolock(inode)) { |
2674 | /* | |
70261f56 | 2675 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2676 | * the page and we may dirty the inode. |
2677 | */ | |
09cbfeaf | 2678 | rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits); |
6b523df4 JK |
2679 | } |
2680 | ||
4e7ea81d JK |
2681 | /* |
2682 | * If we have inline data and arrive here, it means that | |
2683 | * we will soon create the block for the 1st page, so | |
2684 | * we'd better clear the inline data here. | |
2685 | */ | |
2686 | if (ext4_has_inline_data(inode)) { | |
2687 | /* Just inode will be modified... */ | |
2688 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2689 | if (IS_ERR(handle)) { | |
2690 | ret = PTR_ERR(handle); | |
2691 | goto out_writepages; | |
2692 | } | |
2693 | BUG_ON(ext4_test_inode_state(inode, | |
2694 | EXT4_STATE_MAY_INLINE_DATA)); | |
2695 | ext4_destroy_inline_data(handle, inode); | |
2696 | ext4_journal_stop(handle); | |
2697 | } | |
2698 | ||
22208ded AK |
2699 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2700 | range_whole = 1; | |
61628a3f | 2701 | |
2acf2c26 | 2702 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2703 | writeback_index = mapping->writeback_index; |
2704 | if (writeback_index) | |
2acf2c26 | 2705 | cycled = 0; |
4e7ea81d JK |
2706 | mpd.first_page = writeback_index; |
2707 | mpd.last_page = -1; | |
5b41d924 | 2708 | } else { |
09cbfeaf KS |
2709 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2710 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2711 | } |
a1d6cc56 | 2712 | |
4e7ea81d JK |
2713 | mpd.inode = inode; |
2714 | mpd.wbc = wbc; | |
2715 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2716 | retry: |
6e6938b6 | 2717 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2718 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2719 | done = false; | |
1bce63d1 | 2720 | blk_start_plug(&plug); |
4e7ea81d JK |
2721 | while (!done && mpd.first_page <= mpd.last_page) { |
2722 | /* For each extent of pages we use new io_end */ | |
2723 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2724 | if (!mpd.io_submit.io_end) { | |
2725 | ret = -ENOMEM; | |
2726 | break; | |
2727 | } | |
a1d6cc56 AK |
2728 | |
2729 | /* | |
4e7ea81d JK |
2730 | * We have two constraints: We find one extent to map and we |
2731 | * must always write out whole page (makes a difference when | |
2732 | * blocksize < pagesize) so that we don't block on IO when we | |
2733 | * try to write out the rest of the page. Journalled mode is | |
2734 | * not supported by delalloc. | |
a1d6cc56 AK |
2735 | */ |
2736 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2737 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2738 | |
4e7ea81d | 2739 | /* start a new transaction */ |
6b523df4 JK |
2740 | handle = ext4_journal_start_with_reserve(inode, |
2741 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2742 | if (IS_ERR(handle)) { |
2743 | ret = PTR_ERR(handle); | |
1693918e | 2744 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2745 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2746 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2747 | /* Release allocated io_end */ |
2748 | ext4_put_io_end(mpd.io_submit.io_end); | |
2749 | break; | |
61628a3f | 2750 | } |
f63e6005 | 2751 | |
4e7ea81d JK |
2752 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2753 | ret = mpage_prepare_extent_to_map(&mpd); | |
2754 | if (!ret) { | |
2755 | if (mpd.map.m_len) | |
cb530541 TT |
2756 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2757 | &give_up_on_write); | |
4e7ea81d JK |
2758 | else { |
2759 | /* | |
2760 | * We scanned the whole range (or exhausted | |
2761 | * nr_to_write), submitted what was mapped and | |
2762 | * didn't find anything needing mapping. We are | |
2763 | * done. | |
2764 | */ | |
2765 | done = true; | |
2766 | } | |
f63e6005 | 2767 | } |
646caa9c JK |
2768 | /* |
2769 | * Caution: If the handle is synchronous, | |
2770 | * ext4_journal_stop() can wait for transaction commit | |
2771 | * to finish which may depend on writeback of pages to | |
2772 | * complete or on page lock to be released. In that | |
2773 | * case, we have to wait until after after we have | |
2774 | * submitted all the IO, released page locks we hold, | |
2775 | * and dropped io_end reference (for extent conversion | |
2776 | * to be able to complete) before stopping the handle. | |
2777 | */ | |
2778 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2779 | ext4_journal_stop(handle); | |
2780 | handle = NULL; | |
2781 | } | |
4e7ea81d JK |
2782 | /* Submit prepared bio */ |
2783 | ext4_io_submit(&mpd.io_submit); | |
2784 | /* Unlock pages we didn't use */ | |
cb530541 | 2785 | mpage_release_unused_pages(&mpd, give_up_on_write); |
646caa9c JK |
2786 | /* |
2787 | * Drop our io_end reference we got from init. We have | |
2788 | * to be careful and use deferred io_end finishing if | |
2789 | * we are still holding the transaction as we can | |
2790 | * release the last reference to io_end which may end | |
2791 | * up doing unwritten extent conversion. | |
2792 | */ | |
2793 | if (handle) { | |
2794 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2795 | ext4_journal_stop(handle); | |
2796 | } else | |
2797 | ext4_put_io_end(mpd.io_submit.io_end); | |
4e7ea81d JK |
2798 | |
2799 | if (ret == -ENOSPC && sbi->s_journal) { | |
2800 | /* | |
2801 | * Commit the transaction which would | |
22208ded AK |
2802 | * free blocks released in the transaction |
2803 | * and try again | |
2804 | */ | |
df22291f | 2805 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2806 | ret = 0; |
4e7ea81d JK |
2807 | continue; |
2808 | } | |
2809 | /* Fatal error - ENOMEM, EIO... */ | |
2810 | if (ret) | |
61628a3f | 2811 | break; |
a1d6cc56 | 2812 | } |
1bce63d1 | 2813 | blk_finish_plug(&plug); |
9c12a831 | 2814 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2815 | cycled = 1; |
4e7ea81d JK |
2816 | mpd.last_page = writeback_index - 1; |
2817 | mpd.first_page = 0; | |
2acf2c26 AK |
2818 | goto retry; |
2819 | } | |
22208ded AK |
2820 | |
2821 | /* Update index */ | |
22208ded AK |
2822 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2823 | /* | |
4e7ea81d | 2824 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2825 | * mode will write it back later |
2826 | */ | |
4e7ea81d | 2827 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2828 | |
61628a3f | 2829 | out_writepages: |
20970ba6 TT |
2830 | trace_ext4_writepages_result(inode, wbc, ret, |
2831 | nr_to_write - wbc->nr_to_write); | |
c8585c6f | 2832 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
61628a3f | 2833 | return ret; |
64769240 AT |
2834 | } |
2835 | ||
79f0be8d AK |
2836 | static int ext4_nonda_switch(struct super_block *sb) |
2837 | { | |
5c1ff336 | 2838 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2839 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2840 | ||
2841 | /* | |
2842 | * switch to non delalloc mode if we are running low | |
2843 | * on free block. The free block accounting via percpu | |
179f7ebf | 2844 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2845 | * accumulated on each CPU without updating global counters |
2846 | * Delalloc need an accurate free block accounting. So switch | |
2847 | * to non delalloc when we are near to error range. | |
2848 | */ | |
5c1ff336 EW |
2849 | free_clusters = |
2850 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2851 | dirty_clusters = | |
2852 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2853 | /* |
2854 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2855 | */ | |
5c1ff336 | 2856 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2857 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2858 | |
5c1ff336 EW |
2859 | if (2 * free_clusters < 3 * dirty_clusters || |
2860 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2861 | /* |
c8afb446 ES |
2862 | * free block count is less than 150% of dirty blocks |
2863 | * or free blocks is less than watermark | |
79f0be8d AK |
2864 | */ |
2865 | return 1; | |
2866 | } | |
2867 | return 0; | |
2868 | } | |
2869 | ||
0ff8947f ES |
2870 | /* We always reserve for an inode update; the superblock could be there too */ |
2871 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2872 | { | |
e2b911c5 | 2873 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2874 | return 1; |
2875 | ||
2876 | if (pos + len <= 0x7fffffffULL) | |
2877 | return 1; | |
2878 | ||
2879 | /* We might need to update the superblock to set LARGE_FILE */ | |
2880 | return 2; | |
2881 | } | |
2882 | ||
64769240 | 2883 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2884 | loff_t pos, unsigned len, unsigned flags, |
2885 | struct page **pagep, void **fsdata) | |
64769240 | 2886 | { |
72b8ab9d | 2887 | int ret, retries = 0; |
64769240 AT |
2888 | struct page *page; |
2889 | pgoff_t index; | |
64769240 AT |
2890 | struct inode *inode = mapping->host; |
2891 | handle_t *handle; | |
2892 | ||
09cbfeaf | 2893 | index = pos >> PAGE_SHIFT; |
79f0be8d AK |
2894 | |
2895 | if (ext4_nonda_switch(inode->i_sb)) { | |
2896 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2897 | return ext4_write_begin(file, mapping, pos, | |
2898 | len, flags, pagep, fsdata); | |
2899 | } | |
2900 | *fsdata = (void *)0; | |
9bffad1e | 2901 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2902 | |
2903 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2904 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2905 | pos, len, flags, | |
2906 | pagep, fsdata); | |
2907 | if (ret < 0) | |
47564bfb TT |
2908 | return ret; |
2909 | if (ret == 1) | |
2910 | return 0; | |
9c3569b5 TM |
2911 | } |
2912 | ||
47564bfb TT |
2913 | /* |
2914 | * grab_cache_page_write_begin() can take a long time if the | |
2915 | * system is thrashing due to memory pressure, or if the page | |
2916 | * is being written back. So grab it first before we start | |
2917 | * the transaction handle. This also allows us to allocate | |
2918 | * the page (if needed) without using GFP_NOFS. | |
2919 | */ | |
2920 | retry_grab: | |
2921 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2922 | if (!page) | |
2923 | return -ENOMEM; | |
2924 | unlock_page(page); | |
2925 | ||
64769240 AT |
2926 | /* |
2927 | * With delayed allocation, we don't log the i_disksize update | |
2928 | * if there is delayed block allocation. But we still need | |
2929 | * to journalling the i_disksize update if writes to the end | |
2930 | * of file which has an already mapped buffer. | |
2931 | */ | |
47564bfb | 2932 | retry_journal: |
0ff8947f ES |
2933 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2934 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2935 | if (IS_ERR(handle)) { |
09cbfeaf | 2936 | put_page(page); |
47564bfb | 2937 | return PTR_ERR(handle); |
64769240 AT |
2938 | } |
2939 | ||
47564bfb TT |
2940 | lock_page(page); |
2941 | if (page->mapping != mapping) { | |
2942 | /* The page got truncated from under us */ | |
2943 | unlock_page(page); | |
09cbfeaf | 2944 | put_page(page); |
d5a0d4f7 | 2945 | ext4_journal_stop(handle); |
47564bfb | 2946 | goto retry_grab; |
d5a0d4f7 | 2947 | } |
47564bfb | 2948 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2949 | wait_for_stable_page(page); |
64769240 | 2950 | |
2058f83a MH |
2951 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
2952 | ret = ext4_block_write_begin(page, pos, len, | |
2953 | ext4_da_get_block_prep); | |
2954 | #else | |
6e1db88d | 2955 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2956 | #endif |
64769240 AT |
2957 | if (ret < 0) { |
2958 | unlock_page(page); | |
2959 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2960 | /* |
2961 | * block_write_begin may have instantiated a few blocks | |
2962 | * outside i_size. Trim these off again. Don't need | |
2963 | * i_size_read because we hold i_mutex. | |
2964 | */ | |
2965 | if (pos + len > inode->i_size) | |
b9a4207d | 2966 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2967 | |
2968 | if (ret == -ENOSPC && | |
2969 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2970 | goto retry_journal; | |
2971 | ||
09cbfeaf | 2972 | put_page(page); |
47564bfb | 2973 | return ret; |
64769240 AT |
2974 | } |
2975 | ||
47564bfb | 2976 | *pagep = page; |
64769240 AT |
2977 | return ret; |
2978 | } | |
2979 | ||
632eaeab MC |
2980 | /* |
2981 | * Check if we should update i_disksize | |
2982 | * when write to the end of file but not require block allocation | |
2983 | */ | |
2984 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2985 | unsigned long offset) |
632eaeab MC |
2986 | { |
2987 | struct buffer_head *bh; | |
2988 | struct inode *inode = page->mapping->host; | |
2989 | unsigned int idx; | |
2990 | int i; | |
2991 | ||
2992 | bh = page_buffers(page); | |
2993 | idx = offset >> inode->i_blkbits; | |
2994 | ||
af5bc92d | 2995 | for (i = 0; i < idx; i++) |
632eaeab MC |
2996 | bh = bh->b_this_page; |
2997 | ||
29fa89d0 | 2998 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2999 | return 0; |
3000 | return 1; | |
3001 | } | |
3002 | ||
64769240 | 3003 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3004 | struct address_space *mapping, |
3005 | loff_t pos, unsigned len, unsigned copied, | |
3006 | struct page *page, void *fsdata) | |
64769240 AT |
3007 | { |
3008 | struct inode *inode = mapping->host; | |
3009 | int ret = 0, ret2; | |
3010 | handle_t *handle = ext4_journal_current_handle(); | |
3011 | loff_t new_i_size; | |
632eaeab | 3012 | unsigned long start, end; |
79f0be8d AK |
3013 | int write_mode = (int)(unsigned long)fsdata; |
3014 | ||
74d553aa TT |
3015 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3016 | return ext4_write_end(file, mapping, pos, | |
3017 | len, copied, page, fsdata); | |
632eaeab | 3018 | |
9bffad1e | 3019 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3020 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3021 | end = start + copied - 1; |
64769240 AT |
3022 | |
3023 | /* | |
3024 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3025 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3026 | * into that. | |
3027 | */ | |
64769240 | 3028 | new_i_size = pos + copied; |
ea51d132 | 3029 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3030 | if (ext4_has_inline_data(inode) || |
3031 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3032 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3033 | /* We need to mark inode dirty even if |
3034 | * new_i_size is less that inode->i_size | |
3035 | * bu greater than i_disksize.(hint delalloc) | |
3036 | */ | |
3037 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3038 | } |
632eaeab | 3039 | } |
9c3569b5 TM |
3040 | |
3041 | if (write_mode != CONVERT_INLINE_DATA && | |
3042 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3043 | ext4_has_inline_data(inode)) | |
3044 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3045 | page); | |
3046 | else | |
3047 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3048 | page, fsdata); |
9c3569b5 | 3049 | |
64769240 AT |
3050 | copied = ret2; |
3051 | if (ret2 < 0) | |
3052 | ret = ret2; | |
3053 | ret2 = ext4_journal_stop(handle); | |
3054 | if (!ret) | |
3055 | ret = ret2; | |
3056 | ||
3057 | return ret ? ret : copied; | |
3058 | } | |
3059 | ||
d47992f8 LC |
3060 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
3061 | unsigned int length) | |
64769240 | 3062 | { |
64769240 AT |
3063 | /* |
3064 | * Drop reserved blocks | |
3065 | */ | |
3066 | BUG_ON(!PageLocked(page)); | |
3067 | if (!page_has_buffers(page)) | |
3068 | goto out; | |
3069 | ||
ca99fdd2 | 3070 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
3071 | |
3072 | out: | |
d47992f8 | 3073 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
3074 | |
3075 | return; | |
3076 | } | |
3077 | ||
ccd2506b TT |
3078 | /* |
3079 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3080 | */ | |
3081 | int ext4_alloc_da_blocks(struct inode *inode) | |
3082 | { | |
fb40ba0d TT |
3083 | trace_ext4_alloc_da_blocks(inode); |
3084 | ||
71d4f7d0 | 3085 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3086 | return 0; |
3087 | ||
3088 | /* | |
3089 | * We do something simple for now. The filemap_flush() will | |
3090 | * also start triggering a write of the data blocks, which is | |
3091 | * not strictly speaking necessary (and for users of | |
3092 | * laptop_mode, not even desirable). However, to do otherwise | |
3093 | * would require replicating code paths in: | |
de9a55b8 | 3094 | * |
20970ba6 | 3095 | * ext4_writepages() -> |
ccd2506b TT |
3096 | * write_cache_pages() ---> (via passed in callback function) |
3097 | * __mpage_da_writepage() --> | |
3098 | * mpage_add_bh_to_extent() | |
3099 | * mpage_da_map_blocks() | |
3100 | * | |
3101 | * The problem is that write_cache_pages(), located in | |
3102 | * mm/page-writeback.c, marks pages clean in preparation for | |
3103 | * doing I/O, which is not desirable if we're not planning on | |
3104 | * doing I/O at all. | |
3105 | * | |
3106 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3107 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3108 | * would be ugly in the extreme. So instead we would need to |
3109 | * replicate parts of the code in the above functions, | |
25985edc | 3110 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3111 | * write out the pages, but rather only collect contiguous |
3112 | * logical block extents, call the multi-block allocator, and | |
3113 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3114 | * |
ccd2506b TT |
3115 | * For now, though, we'll cheat by calling filemap_flush(), |
3116 | * which will map the blocks, and start the I/O, but not | |
3117 | * actually wait for the I/O to complete. | |
3118 | */ | |
3119 | return filemap_flush(inode->i_mapping); | |
3120 | } | |
64769240 | 3121 | |
ac27a0ec DK |
3122 | /* |
3123 | * bmap() is special. It gets used by applications such as lilo and by | |
3124 | * the swapper to find the on-disk block of a specific piece of data. | |
3125 | * | |
3126 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3127 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3128 | * filesystem and enables swap, then they may get a nasty shock when the |
3129 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3130 | * the original zero's written out previously to the journal and | |
3131 | * awaiting writeback in the kernel's buffer cache. | |
3132 | * | |
3133 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3134 | * take extra steps to flush any blocks which might be in the cache. | |
3135 | */ | |
617ba13b | 3136 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3137 | { |
3138 | struct inode *inode = mapping->host; | |
3139 | journal_t *journal; | |
3140 | int err; | |
3141 | ||
46c7f254 TM |
3142 | /* |
3143 | * We can get here for an inline file via the FIBMAP ioctl | |
3144 | */ | |
3145 | if (ext4_has_inline_data(inode)) | |
3146 | return 0; | |
3147 | ||
64769240 AT |
3148 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3149 | test_opt(inode->i_sb, DELALLOC)) { | |
3150 | /* | |
3151 | * With delalloc we want to sync the file | |
3152 | * so that we can make sure we allocate | |
3153 | * blocks for file | |
3154 | */ | |
3155 | filemap_write_and_wait(mapping); | |
3156 | } | |
3157 | ||
19f5fb7a TT |
3158 | if (EXT4_JOURNAL(inode) && |
3159 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3160 | /* |
3161 | * This is a REALLY heavyweight approach, but the use of | |
3162 | * bmap on dirty files is expected to be extremely rare: | |
3163 | * only if we run lilo or swapon on a freshly made file | |
3164 | * do we expect this to happen. | |
3165 | * | |
3166 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3167 | * represent an unprivileged user DOS attack --- we'd be | |
3168 | * in trouble if mortal users could trigger this path at | |
3169 | * will.) | |
3170 | * | |
617ba13b | 3171 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3172 | * regular files. If somebody wants to bmap a directory |
3173 | * or symlink and gets confused because the buffer | |
3174 | * hasn't yet been flushed to disk, they deserve | |
3175 | * everything they get. | |
3176 | */ | |
3177 | ||
19f5fb7a | 3178 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3179 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3180 | jbd2_journal_lock_updates(journal); |
3181 | err = jbd2_journal_flush(journal); | |
3182 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3183 | |
3184 | if (err) | |
3185 | return 0; | |
3186 | } | |
3187 | ||
af5bc92d | 3188 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3189 | } |
3190 | ||
617ba13b | 3191 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3192 | { |
46c7f254 TM |
3193 | int ret = -EAGAIN; |
3194 | struct inode *inode = page->mapping->host; | |
3195 | ||
0562e0ba | 3196 | trace_ext4_readpage(page); |
46c7f254 TM |
3197 | |
3198 | if (ext4_has_inline_data(inode)) | |
3199 | ret = ext4_readpage_inline(inode, page); | |
3200 | ||
3201 | if (ret == -EAGAIN) | |
f64e02fe | 3202 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3203 | |
3204 | return ret; | |
ac27a0ec DK |
3205 | } |
3206 | ||
3207 | static int | |
617ba13b | 3208 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3209 | struct list_head *pages, unsigned nr_pages) |
3210 | { | |
46c7f254 TM |
3211 | struct inode *inode = mapping->host; |
3212 | ||
3213 | /* If the file has inline data, no need to do readpages. */ | |
3214 | if (ext4_has_inline_data(inode)) | |
3215 | return 0; | |
3216 | ||
f64e02fe | 3217 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3218 | } |
3219 | ||
d47992f8 LC |
3220 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3221 | unsigned int length) | |
ac27a0ec | 3222 | { |
ca99fdd2 | 3223 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3224 | |
4520fb3c JK |
3225 | /* No journalling happens on data buffers when this function is used */ |
3226 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3227 | ||
ca99fdd2 | 3228 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3229 | } |
3230 | ||
53e87268 | 3231 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3232 | unsigned int offset, |
3233 | unsigned int length) | |
4520fb3c JK |
3234 | { |
3235 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3236 | ||
ca99fdd2 | 3237 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3238 | |
ac27a0ec DK |
3239 | /* |
3240 | * If it's a full truncate we just forget about the pending dirtying | |
3241 | */ | |
09cbfeaf | 3242 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3243 | ClearPageChecked(page); |
3244 | ||
ca99fdd2 | 3245 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3246 | } |
3247 | ||
3248 | /* Wrapper for aops... */ | |
3249 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3250 | unsigned int offset, |
3251 | unsigned int length) | |
53e87268 | 3252 | { |
ca99fdd2 | 3253 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3254 | } |
3255 | ||
617ba13b | 3256 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3257 | { |
617ba13b | 3258 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3259 | |
0562e0ba JZ |
3260 | trace_ext4_releasepage(page); |
3261 | ||
e1c36595 JK |
3262 | /* Page has dirty journalled data -> cannot release */ |
3263 | if (PageChecked(page)) | |
ac27a0ec | 3264 | return 0; |
0390131b FM |
3265 | if (journal) |
3266 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3267 | else | |
3268 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3269 | } |
3270 | ||
ba5843f5 | 3271 | #ifdef CONFIG_FS_DAX |
12735f88 JK |
3272 | /* |
3273 | * Get block function for DAX IO and mmap faults. It takes care of converting | |
3274 | * unwritten extents to written ones and initializes new / converted blocks | |
3275 | * to zeros. | |
3276 | */ | |
3277 | int ext4_dax_get_block(struct inode *inode, sector_t iblock, | |
3278 | struct buffer_head *bh_result, int create) | |
ed923b57 | 3279 | { |
7cb476f8 | 3280 | int ret; |
ba5843f5 | 3281 | |
12735f88 | 3282 | ext4_debug("inode %lu, create flag %d\n", inode->i_ino, create); |
7cb476f8 JK |
3283 | if (!create) |
3284 | return _ext4_get_block(inode, iblock, bh_result, 0); | |
ba5843f5 | 3285 | |
7cb476f8 JK |
3286 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
3287 | EXT4_GET_BLOCKS_PRE_IO | | |
3288 | EXT4_GET_BLOCKS_CREATE_ZERO); | |
3289 | if (ret < 0) | |
3290 | return ret; | |
ba5843f5 | 3291 | |
7cb476f8 | 3292 | if (buffer_unwritten(bh_result)) { |
ba5843f5 | 3293 | /* |
12735f88 JK |
3294 | * We are protected by i_mmap_sem or i_mutex so we know block |
3295 | * cannot go away from under us even though we dropped | |
3296 | * i_data_sem. Convert extent to written and write zeros there. | |
ba5843f5 | 3297 | */ |
7cb476f8 JK |
3298 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
3299 | EXT4_GET_BLOCKS_CONVERT | | |
3300 | EXT4_GET_BLOCKS_CREATE_ZERO); | |
3301 | if (ret < 0) | |
3302 | return ret; | |
ba5843f5 | 3303 | } |
7cb476f8 JK |
3304 | /* |
3305 | * At least for now we have to clear BH_New so that DAX code | |
3306 | * doesn't attempt to zero blocks again in a racy way. | |
3307 | */ | |
3308 | clear_buffer_new(bh_result); | |
3309 | return 0; | |
ed923b57 | 3310 | } |
12735f88 JK |
3311 | #else |
3312 | /* Just define empty function, it will never get called. */ | |
3313 | int ext4_dax_get_block(struct inode *inode, sector_t iblock, | |
3314 | struct buffer_head *bh_result, int create) | |
3315 | { | |
3316 | BUG(); | |
3317 | return 0; | |
ed923b57 | 3318 | } |
ba5843f5 | 3319 | #endif |
ed923b57 | 3320 | |
187372a3 | 3321 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3322 | ssize_t size, void *private) |
4c0425ff | 3323 | { |
109811c2 | 3324 | ext4_io_end_t *io_end = private; |
4c0425ff | 3325 | |
97a851ed | 3326 | /* if not async direct IO just return */ |
7b7a8665 | 3327 | if (!io_end) |
187372a3 | 3328 | return 0; |
4b70df18 | 3329 | |
88635ca2 | 3330 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3331 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
109811c2 | 3332 | io_end, io_end->inode->i_ino, iocb, offset, size); |
8d5d02e6 | 3333 | |
74c66bcb JK |
3334 | /* |
3335 | * Error during AIO DIO. We cannot convert unwritten extents as the | |
3336 | * data was not written. Just clear the unwritten flag and drop io_end. | |
3337 | */ | |
3338 | if (size <= 0) { | |
3339 | ext4_clear_io_unwritten_flag(io_end); | |
3340 | size = 0; | |
3341 | } | |
4c0425ff MC |
3342 | io_end->offset = offset; |
3343 | io_end->size = size; | |
7b7a8665 | 3344 | ext4_put_io_end(io_end); |
187372a3 CH |
3345 | |
3346 | return 0; | |
4c0425ff | 3347 | } |
c7064ef1 | 3348 | |
4c0425ff | 3349 | /* |
914f82a3 JK |
3350 | * Handling of direct IO writes. |
3351 | * | |
3352 | * For ext4 extent files, ext4 will do direct-io write even to holes, | |
4c0425ff MC |
3353 | * preallocated extents, and those write extend the file, no need to |
3354 | * fall back to buffered IO. | |
3355 | * | |
556615dc | 3356 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3357 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3358 | * still keep the range to write as unwritten. |
4c0425ff | 3359 | * |
69c499d1 | 3360 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3361 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3362 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3363 | * when async direct IO completed. |
4c0425ff MC |
3364 | * |
3365 | * If the O_DIRECT write will extend the file then add this inode to the | |
3366 | * orphan list. So recovery will truncate it back to the original size | |
3367 | * if the machine crashes during the write. | |
3368 | * | |
3369 | */ | |
0e01df10 | 3370 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3371 | { |
3372 | struct file *file = iocb->ki_filp; | |
3373 | struct inode *inode = file->f_mapping->host; | |
914f82a3 | 3374 | struct ext4_inode_info *ei = EXT4_I(inode); |
4c0425ff | 3375 | ssize_t ret; |
c8b8e32d | 3376 | loff_t offset = iocb->ki_pos; |
a6cbcd4a | 3377 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3378 | int overwrite = 0; |
3379 | get_block_t *get_block_func = NULL; | |
3380 | int dio_flags = 0; | |
4c0425ff | 3381 | loff_t final_size = offset + count; |
914f82a3 JK |
3382 | int orphan = 0; |
3383 | handle_t *handle; | |
729f52c6 | 3384 | |
914f82a3 JK |
3385 | if (final_size > inode->i_size) { |
3386 | /* Credits for sb + inode write */ | |
3387 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3388 | if (IS_ERR(handle)) { | |
3389 | ret = PTR_ERR(handle); | |
3390 | goto out; | |
3391 | } | |
3392 | ret = ext4_orphan_add(handle, inode); | |
3393 | if (ret) { | |
3394 | ext4_journal_stop(handle); | |
3395 | goto out; | |
3396 | } | |
3397 | orphan = 1; | |
3398 | ei->i_disksize = inode->i_size; | |
3399 | ext4_journal_stop(handle); | |
3400 | } | |
4bd809db | 3401 | |
69c499d1 | 3402 | BUG_ON(iocb->private == NULL); |
4bd809db | 3403 | |
e8340395 JK |
3404 | /* |
3405 | * Make all waiters for direct IO properly wait also for extent | |
3406 | * conversion. This also disallows race between truncate() and | |
3407 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3408 | */ | |
914f82a3 | 3409 | inode_dio_begin(inode); |
e8340395 | 3410 | |
69c499d1 TT |
3411 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3412 | overwrite = *((int *)iocb->private); | |
4bd809db | 3413 | |
2dcba478 | 3414 | if (overwrite) |
5955102c | 3415 | inode_unlock(inode); |
8d5d02e6 | 3416 | |
69c499d1 | 3417 | /* |
914f82a3 | 3418 | * For extent mapped files we could direct write to holes and fallocate. |
69c499d1 | 3419 | * |
109811c2 JK |
3420 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
3421 | * parallel buffered read to expose the stale data before DIO complete | |
3422 | * the data IO. | |
69c499d1 | 3423 | * |
109811c2 JK |
3424 | * As to previously fallocated extents, ext4 get_block will just simply |
3425 | * mark the buffer mapped but still keep the extents unwritten. | |
69c499d1 | 3426 | * |
109811c2 JK |
3427 | * For non AIO case, we will convert those unwritten extents to written |
3428 | * after return back from blockdev_direct_IO. That way we save us from | |
3429 | * allocating io_end structure and also the overhead of offloading | |
3430 | * the extent convertion to a workqueue. | |
69c499d1 TT |
3431 | * |
3432 | * For async DIO, the conversion needs to be deferred when the | |
3433 | * IO is completed. The ext4 end_io callback function will be | |
3434 | * called to take care of the conversion work. Here for async | |
3435 | * case, we allocate an io_end structure to hook to the iocb. | |
3436 | */ | |
3437 | iocb->private = NULL; | |
109811c2 | 3438 | if (overwrite) |
705965bd | 3439 | get_block_func = ext4_dio_get_block_overwrite; |
12735f88 JK |
3440 | else if (IS_DAX(inode)) { |
3441 | /* | |
3442 | * We can avoid zeroing for aligned DAX writes beyond EOF. Other | |
3443 | * writes need zeroing either because they can race with page | |
3444 | * faults or because they use partial blocks. | |
3445 | */ | |
3446 | if (round_down(offset, 1<<inode->i_blkbits) >= inode->i_size && | |
3447 | ext4_aligned_io(inode, offset, count)) | |
3448 | get_block_func = ext4_dio_get_block; | |
3449 | else | |
3450 | get_block_func = ext4_dax_get_block; | |
3451 | dio_flags = DIO_LOCKING; | |
3452 | } else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || | |
3453 | round_down(offset, 1 << inode->i_blkbits) >= inode->i_size) { | |
914f82a3 JK |
3454 | get_block_func = ext4_dio_get_block; |
3455 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
3456 | } else if (is_sync_kiocb(iocb)) { | |
109811c2 JK |
3457 | get_block_func = ext4_dio_get_block_unwritten_sync; |
3458 | dio_flags = DIO_LOCKING; | |
69c499d1 | 3459 | } else { |
109811c2 | 3460 | get_block_func = ext4_dio_get_block_unwritten_async; |
69c499d1 TT |
3461 | dio_flags = DIO_LOCKING; |
3462 | } | |
2058f83a MH |
3463 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3464 | BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); | |
3465 | #endif | |
914f82a3 | 3466 | if (IS_DAX(inode)) { |
c8b8e32d | 3467 | ret = dax_do_io(iocb, inode, iter, get_block_func, |
923ae0ff | 3468 | ext4_end_io_dio, dio_flags); |
914f82a3 | 3469 | } else |
17f8c842 | 3470 | ret = __blockdev_direct_IO(iocb, inode, |
c8b8e32d | 3471 | inode->i_sb->s_bdev, iter, |
923ae0ff RZ |
3472 | get_block_func, |
3473 | ext4_end_io_dio, NULL, dio_flags); | |
69c499d1 | 3474 | |
97a851ed | 3475 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
69c499d1 TT |
3476 | EXT4_STATE_DIO_UNWRITTEN)) { |
3477 | int err; | |
3478 | /* | |
3479 | * for non AIO case, since the IO is already | |
3480 | * completed, we could do the conversion right here | |
3481 | */ | |
6b523df4 | 3482 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3483 | offset, ret); |
3484 | if (err < 0) | |
3485 | ret = err; | |
3486 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3487 | } | |
4bd809db | 3488 | |
914f82a3 | 3489 | inode_dio_end(inode); |
69c499d1 | 3490 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3491 | if (overwrite) |
5955102c | 3492 | inode_lock(inode); |
8d5d02e6 | 3493 | |
914f82a3 JK |
3494 | if (ret < 0 && final_size > inode->i_size) |
3495 | ext4_truncate_failed_write(inode); | |
3496 | ||
3497 | /* Handle extending of i_size after direct IO write */ | |
3498 | if (orphan) { | |
3499 | int err; | |
3500 | ||
3501 | /* Credits for sb + inode write */ | |
3502 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3503 | if (IS_ERR(handle)) { | |
3504 | /* This is really bad luck. We've written the data | |
3505 | * but cannot extend i_size. Bail out and pretend | |
3506 | * the write failed... */ | |
3507 | ret = PTR_ERR(handle); | |
3508 | if (inode->i_nlink) | |
3509 | ext4_orphan_del(NULL, inode); | |
3510 | ||
3511 | goto out; | |
3512 | } | |
3513 | if (inode->i_nlink) | |
3514 | ext4_orphan_del(handle, inode); | |
3515 | if (ret > 0) { | |
3516 | loff_t end = offset + ret; | |
3517 | if (end > inode->i_size) { | |
3518 | ei->i_disksize = end; | |
3519 | i_size_write(inode, end); | |
3520 | /* | |
3521 | * We're going to return a positive `ret' | |
3522 | * here due to non-zero-length I/O, so there's | |
3523 | * no way of reporting error returns from | |
3524 | * ext4_mark_inode_dirty() to userspace. So | |
3525 | * ignore it. | |
3526 | */ | |
3527 | ext4_mark_inode_dirty(handle, inode); | |
3528 | } | |
3529 | } | |
3530 | err = ext4_journal_stop(handle); | |
3531 | if (ret == 0) | |
3532 | ret = err; | |
3533 | } | |
3534 | out: | |
3535 | return ret; | |
3536 | } | |
3537 | ||
0e01df10 | 3538 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
914f82a3 | 3539 | { |
16c54688 JK |
3540 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
3541 | struct inode *inode = mapping->host; | |
914f82a3 JK |
3542 | ssize_t ret; |
3543 | ||
16c54688 JK |
3544 | /* |
3545 | * Shared inode_lock is enough for us - it protects against concurrent | |
3546 | * writes & truncates and since we take care of writing back page cache, | |
3547 | * we are protected against page writeback as well. | |
3548 | */ | |
3549 | inode_lock_shared(inode); | |
914f82a3 | 3550 | if (IS_DAX(inode)) { |
16c54688 | 3551 | ret = dax_do_io(iocb, inode, iter, ext4_dio_get_block, NULL, 0); |
914f82a3 | 3552 | } else { |
16c54688 JK |
3553 | size_t count = iov_iter_count(iter); |
3554 | ||
3555 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, | |
3556 | iocb->ki_pos + count); | |
3557 | if (ret) | |
3558 | goto out_unlock; | |
914f82a3 | 3559 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, |
0e01df10 | 3560 | iter, ext4_dio_get_block, |
16c54688 | 3561 | NULL, NULL, 0); |
914f82a3 | 3562 | } |
16c54688 JK |
3563 | out_unlock: |
3564 | inode_unlock_shared(inode); | |
69c499d1 | 3565 | return ret; |
4c0425ff MC |
3566 | } |
3567 | ||
c8b8e32d | 3568 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3569 | { |
3570 | struct file *file = iocb->ki_filp; | |
3571 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3572 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3573 | loff_t offset = iocb->ki_pos; |
0562e0ba | 3574 | ssize_t ret; |
4c0425ff | 3575 | |
2058f83a MH |
3576 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3577 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3578 | return 0; | |
3579 | #endif | |
3580 | ||
84ebd795 TT |
3581 | /* |
3582 | * If we are doing data journalling we don't support O_DIRECT | |
3583 | */ | |
3584 | if (ext4_should_journal_data(inode)) | |
3585 | return 0; | |
3586 | ||
46c7f254 TM |
3587 | /* Let buffer I/O handle the inline data case. */ |
3588 | if (ext4_has_inline_data(inode)) | |
3589 | return 0; | |
3590 | ||
6f673763 | 3591 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
914f82a3 | 3592 | if (iov_iter_rw(iter) == READ) |
0e01df10 | 3593 | ret = ext4_direct_IO_read(iocb, iter); |
0562e0ba | 3594 | else |
0e01df10 | 3595 | ret = ext4_direct_IO_write(iocb, iter); |
6f673763 | 3596 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
0562e0ba | 3597 | return ret; |
4c0425ff MC |
3598 | } |
3599 | ||
ac27a0ec | 3600 | /* |
617ba13b | 3601 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3602 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3603 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3604 | * not necessarily locked. | |
3605 | * | |
3606 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3607 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3608 | * or jbddirty because all the journalling code will explode. | |
3609 | * | |
3610 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3611 | * is called, propagate that into the buffers appropriately. | |
3612 | */ | |
617ba13b | 3613 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3614 | { |
3615 | SetPageChecked(page); | |
3616 | return __set_page_dirty_nobuffers(page); | |
3617 | } | |
3618 | ||
74d553aa | 3619 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3620 | .readpage = ext4_readpage, |
3621 | .readpages = ext4_readpages, | |
43ce1d23 | 3622 | .writepage = ext4_writepage, |
20970ba6 | 3623 | .writepages = ext4_writepages, |
8ab22b9a | 3624 | .write_begin = ext4_write_begin, |
74d553aa | 3625 | .write_end = ext4_write_end, |
8ab22b9a HH |
3626 | .bmap = ext4_bmap, |
3627 | .invalidatepage = ext4_invalidatepage, | |
3628 | .releasepage = ext4_releasepage, | |
3629 | .direct_IO = ext4_direct_IO, | |
3630 | .migratepage = buffer_migrate_page, | |
3631 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3632 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3633 | }; |
3634 | ||
617ba13b | 3635 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3636 | .readpage = ext4_readpage, |
3637 | .readpages = ext4_readpages, | |
43ce1d23 | 3638 | .writepage = ext4_writepage, |
20970ba6 | 3639 | .writepages = ext4_writepages, |
8ab22b9a HH |
3640 | .write_begin = ext4_write_begin, |
3641 | .write_end = ext4_journalled_write_end, | |
3642 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3643 | .bmap = ext4_bmap, | |
4520fb3c | 3644 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3645 | .releasepage = ext4_releasepage, |
84ebd795 | 3646 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3647 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3648 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3649 | }; |
3650 | ||
64769240 | 3651 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3652 | .readpage = ext4_readpage, |
3653 | .readpages = ext4_readpages, | |
43ce1d23 | 3654 | .writepage = ext4_writepage, |
20970ba6 | 3655 | .writepages = ext4_writepages, |
8ab22b9a HH |
3656 | .write_begin = ext4_da_write_begin, |
3657 | .write_end = ext4_da_write_end, | |
3658 | .bmap = ext4_bmap, | |
3659 | .invalidatepage = ext4_da_invalidatepage, | |
3660 | .releasepage = ext4_releasepage, | |
3661 | .direct_IO = ext4_direct_IO, | |
3662 | .migratepage = buffer_migrate_page, | |
3663 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3664 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3665 | }; |
3666 | ||
617ba13b | 3667 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3668 | { |
3d2b1582 LC |
3669 | switch (ext4_inode_journal_mode(inode)) { |
3670 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3671 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3672 | break; |
3673 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3674 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3675 | return; |
3d2b1582 LC |
3676 | default: |
3677 | BUG(); | |
3678 | } | |
74d553aa TT |
3679 | if (test_opt(inode->i_sb, DELALLOC)) |
3680 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3681 | else | |
3682 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3683 | } |
3684 | ||
923ae0ff | 3685 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3686 | struct address_space *mapping, loff_t from, loff_t length) |
3687 | { | |
09cbfeaf KS |
3688 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3689 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3690 | unsigned blocksize, pos; |
d863dc36 LC |
3691 | ext4_lblk_t iblock; |
3692 | struct inode *inode = mapping->host; | |
3693 | struct buffer_head *bh; | |
3694 | struct page *page; | |
3695 | int err = 0; | |
3696 | ||
09cbfeaf | 3697 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3698 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3699 | if (!page) |
3700 | return -ENOMEM; | |
3701 | ||
3702 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3703 | |
09cbfeaf | 3704 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3705 | |
3706 | if (!page_has_buffers(page)) | |
3707 | create_empty_buffers(page, blocksize, 0); | |
3708 | ||
3709 | /* Find the buffer that contains "offset" */ | |
3710 | bh = page_buffers(page); | |
3711 | pos = blocksize; | |
3712 | while (offset >= pos) { | |
3713 | bh = bh->b_this_page; | |
3714 | iblock++; | |
3715 | pos += blocksize; | |
3716 | } | |
d863dc36 LC |
3717 | if (buffer_freed(bh)) { |
3718 | BUFFER_TRACE(bh, "freed: skip"); | |
3719 | goto unlock; | |
3720 | } | |
d863dc36 LC |
3721 | if (!buffer_mapped(bh)) { |
3722 | BUFFER_TRACE(bh, "unmapped"); | |
3723 | ext4_get_block(inode, iblock, bh, 0); | |
3724 | /* unmapped? It's a hole - nothing to do */ | |
3725 | if (!buffer_mapped(bh)) { | |
3726 | BUFFER_TRACE(bh, "still unmapped"); | |
3727 | goto unlock; | |
3728 | } | |
3729 | } | |
3730 | ||
3731 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3732 | if (PageUptodate(page)) | |
3733 | set_buffer_uptodate(bh); | |
3734 | ||
3735 | if (!buffer_uptodate(bh)) { | |
3736 | err = -EIO; | |
dfec8a14 | 3737 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
3738 | wait_on_buffer(bh); |
3739 | /* Uhhuh. Read error. Complain and punt. */ | |
3740 | if (!buffer_uptodate(bh)) | |
3741 | goto unlock; | |
c9c7429c MH |
3742 | if (S_ISREG(inode->i_mode) && |
3743 | ext4_encrypted_inode(inode)) { | |
3744 | /* We expect the key to be set. */ | |
a7550b30 | 3745 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
09cbfeaf | 3746 | BUG_ON(blocksize != PAGE_SIZE); |
b50f7b26 | 3747 | WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host, |
9c4bb8a3 | 3748 | page, PAGE_SIZE, 0, page->index)); |
c9c7429c | 3749 | } |
d863dc36 | 3750 | } |
d863dc36 LC |
3751 | if (ext4_should_journal_data(inode)) { |
3752 | BUFFER_TRACE(bh, "get write access"); | |
3753 | err = ext4_journal_get_write_access(handle, bh); | |
3754 | if (err) | |
3755 | goto unlock; | |
3756 | } | |
d863dc36 | 3757 | zero_user(page, offset, length); |
d863dc36 LC |
3758 | BUFFER_TRACE(bh, "zeroed end of block"); |
3759 | ||
d863dc36 LC |
3760 | if (ext4_should_journal_data(inode)) { |
3761 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3762 | } else { |
353eefd3 | 3763 | err = 0; |
d863dc36 | 3764 | mark_buffer_dirty(bh); |
3957ef53 | 3765 | if (ext4_should_order_data(inode)) |
ee0876bc | 3766 | err = ext4_jbd2_inode_add_write(handle, inode); |
0713ed0c | 3767 | } |
d863dc36 LC |
3768 | |
3769 | unlock: | |
3770 | unlock_page(page); | |
09cbfeaf | 3771 | put_page(page); |
d863dc36 LC |
3772 | return err; |
3773 | } | |
3774 | ||
923ae0ff RZ |
3775 | /* |
3776 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3777 | * starting from file offset 'from'. The range to be zero'd must | |
3778 | * be contained with in one block. If the specified range exceeds | |
3779 | * the end of the block it will be shortened to end of the block | |
3780 | * that cooresponds to 'from' | |
3781 | */ | |
3782 | static int ext4_block_zero_page_range(handle_t *handle, | |
3783 | struct address_space *mapping, loff_t from, loff_t length) | |
3784 | { | |
3785 | struct inode *inode = mapping->host; | |
09cbfeaf | 3786 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3787 | unsigned blocksize = inode->i_sb->s_blocksize; |
3788 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3789 | ||
3790 | /* | |
3791 | * correct length if it does not fall between | |
3792 | * 'from' and the end of the block | |
3793 | */ | |
3794 | if (length > max || length < 0) | |
3795 | length = max; | |
3796 | ||
3797 | if (IS_DAX(inode)) | |
3798 | return dax_zero_page_range(inode, from, length, ext4_get_block); | |
3799 | return __ext4_block_zero_page_range(handle, mapping, from, length); | |
3800 | } | |
3801 | ||
94350ab5 MW |
3802 | /* |
3803 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3804 | * up to the end of the block which corresponds to `from'. | |
3805 | * This required during truncate. We need to physically zero the tail end | |
3806 | * of that block so it doesn't yield old data if the file is later grown. | |
3807 | */ | |
c197855e | 3808 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3809 | struct address_space *mapping, loff_t from) |
3810 | { | |
09cbfeaf | 3811 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3812 | unsigned length; |
3813 | unsigned blocksize; | |
3814 | struct inode *inode = mapping->host; | |
3815 | ||
3816 | blocksize = inode->i_sb->s_blocksize; | |
3817 | length = blocksize - (offset & (blocksize - 1)); | |
3818 | ||
3819 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3820 | } | |
3821 | ||
a87dd18c LC |
3822 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3823 | loff_t lstart, loff_t length) | |
3824 | { | |
3825 | struct super_block *sb = inode->i_sb; | |
3826 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3827 | unsigned partial_start, partial_end; |
a87dd18c LC |
3828 | ext4_fsblk_t start, end; |
3829 | loff_t byte_end = (lstart + length - 1); | |
3830 | int err = 0; | |
3831 | ||
e1be3a92 LC |
3832 | partial_start = lstart & (sb->s_blocksize - 1); |
3833 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3834 | ||
a87dd18c LC |
3835 | start = lstart >> sb->s_blocksize_bits; |
3836 | end = byte_end >> sb->s_blocksize_bits; | |
3837 | ||
3838 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3839 | if (start == end && |
3840 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3841 | err = ext4_block_zero_page_range(handle, mapping, |
3842 | lstart, length); | |
3843 | return err; | |
3844 | } | |
3845 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3846 | if (partial_start) { |
a87dd18c LC |
3847 | err = ext4_block_zero_page_range(handle, mapping, |
3848 | lstart, sb->s_blocksize); | |
3849 | if (err) | |
3850 | return err; | |
3851 | } | |
3852 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3853 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3854 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3855 | byte_end - partial_end, |
3856 | partial_end + 1); | |
a87dd18c LC |
3857 | return err; |
3858 | } | |
3859 | ||
91ef4caf DG |
3860 | int ext4_can_truncate(struct inode *inode) |
3861 | { | |
91ef4caf DG |
3862 | if (S_ISREG(inode->i_mode)) |
3863 | return 1; | |
3864 | if (S_ISDIR(inode->i_mode)) | |
3865 | return 1; | |
3866 | if (S_ISLNK(inode->i_mode)) | |
3867 | return !ext4_inode_is_fast_symlink(inode); | |
3868 | return 0; | |
3869 | } | |
3870 | ||
01127848 JK |
3871 | /* |
3872 | * We have to make sure i_disksize gets properly updated before we truncate | |
3873 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3874 | * can get lost as it may have been postponed to submission of writeback but | |
3875 | * that will never happen after we truncate page cache. | |
3876 | */ | |
3877 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3878 | loff_t len) | |
3879 | { | |
3880 | handle_t *handle; | |
3881 | loff_t size = i_size_read(inode); | |
3882 | ||
5955102c | 3883 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3884 | if (offset > size || offset + len < size) |
3885 | return 0; | |
3886 | ||
3887 | if (EXT4_I(inode)->i_disksize >= size) | |
3888 | return 0; | |
3889 | ||
3890 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3891 | if (IS_ERR(handle)) | |
3892 | return PTR_ERR(handle); | |
3893 | ext4_update_i_disksize(inode, size); | |
3894 | ext4_mark_inode_dirty(handle, inode); | |
3895 | ext4_journal_stop(handle); | |
3896 | ||
3897 | return 0; | |
3898 | } | |
3899 | ||
a4bb6b64 | 3900 | /* |
cca32b7e | 3901 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3902 | * associated with the given offset and length |
3903 | * | |
3904 | * @inode: File inode | |
3905 | * @offset: The offset where the hole will begin | |
3906 | * @len: The length of the hole | |
3907 | * | |
4907cb7b | 3908 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3909 | */ |
3910 | ||
aeb2817a | 3911 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3912 | { |
26a4c0c6 TT |
3913 | struct super_block *sb = inode->i_sb; |
3914 | ext4_lblk_t first_block, stop_block; | |
3915 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3916 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3917 | handle_t *handle; |
3918 | unsigned int credits; | |
3919 | int ret = 0; | |
3920 | ||
a4bb6b64 | 3921 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3922 | return -EOPNOTSUPP; |
a4bb6b64 | 3923 | |
b8a86845 | 3924 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3925 | |
26a4c0c6 TT |
3926 | /* |
3927 | * Write out all dirty pages to avoid race conditions | |
3928 | * Then release them. | |
3929 | */ | |
cca32b7e | 3930 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
3931 | ret = filemap_write_and_wait_range(mapping, offset, |
3932 | offset + length - 1); | |
3933 | if (ret) | |
3934 | return ret; | |
3935 | } | |
3936 | ||
5955102c | 3937 | inode_lock(inode); |
9ef06cec | 3938 | |
26a4c0c6 TT |
3939 | /* No need to punch hole beyond i_size */ |
3940 | if (offset >= inode->i_size) | |
3941 | goto out_mutex; | |
3942 | ||
3943 | /* | |
3944 | * If the hole extends beyond i_size, set the hole | |
3945 | * to end after the page that contains i_size | |
3946 | */ | |
3947 | if (offset + length > inode->i_size) { | |
3948 | length = inode->i_size + | |
09cbfeaf | 3949 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
3950 | offset; |
3951 | } | |
3952 | ||
a361293f JK |
3953 | if (offset & (sb->s_blocksize - 1) || |
3954 | (offset + length) & (sb->s_blocksize - 1)) { | |
3955 | /* | |
3956 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3957 | * partial block | |
3958 | */ | |
3959 | ret = ext4_inode_attach_jinode(inode); | |
3960 | if (ret < 0) | |
3961 | goto out_mutex; | |
3962 | ||
3963 | } | |
3964 | ||
ea3d7209 JK |
3965 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
3966 | ext4_inode_block_unlocked_dio(inode); | |
3967 | inode_dio_wait(inode); | |
3968 | ||
3969 | /* | |
3970 | * Prevent page faults from reinstantiating pages we have released from | |
3971 | * page cache. | |
3972 | */ | |
3973 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
3974 | first_block_offset = round_up(offset, sb->s_blocksize); |
3975 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3976 | |
a87dd18c | 3977 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
3978 | if (last_block_offset > first_block_offset) { |
3979 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
3980 | if (ret) | |
3981 | goto out_dio; | |
a87dd18c LC |
3982 | truncate_pagecache_range(inode, first_block_offset, |
3983 | last_block_offset); | |
01127848 | 3984 | } |
26a4c0c6 TT |
3985 | |
3986 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3987 | credits = ext4_writepage_trans_blocks(inode); | |
3988 | else | |
3989 | credits = ext4_blocks_for_truncate(inode); | |
3990 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3991 | if (IS_ERR(handle)) { | |
3992 | ret = PTR_ERR(handle); | |
3993 | ext4_std_error(sb, ret); | |
3994 | goto out_dio; | |
3995 | } | |
3996 | ||
a87dd18c LC |
3997 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3998 | length); | |
3999 | if (ret) | |
4000 | goto out_stop; | |
26a4c0c6 TT |
4001 | |
4002 | first_block = (offset + sb->s_blocksize - 1) >> | |
4003 | EXT4_BLOCK_SIZE_BITS(sb); | |
4004 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4005 | ||
4006 | /* If there are no blocks to remove, return now */ | |
4007 | if (first_block >= stop_block) | |
4008 | goto out_stop; | |
4009 | ||
4010 | down_write(&EXT4_I(inode)->i_data_sem); | |
4011 | ext4_discard_preallocations(inode); | |
4012 | ||
4013 | ret = ext4_es_remove_extent(inode, first_block, | |
4014 | stop_block - first_block); | |
4015 | if (ret) { | |
4016 | up_write(&EXT4_I(inode)->i_data_sem); | |
4017 | goto out_stop; | |
4018 | } | |
4019 | ||
4020 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4021 | ret = ext4_ext_remove_space(inode, first_block, | |
4022 | stop_block - 1); | |
4023 | else | |
4f579ae7 | 4024 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
4025 | stop_block); |
4026 | ||
819c4920 | 4027 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
4028 | if (IS_SYNC(inode)) |
4029 | ext4_handle_sync(handle); | |
e251f9bc | 4030 | |
26a4c0c6 TT |
4031 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
4032 | ext4_mark_inode_dirty(handle, inode); | |
4033 | out_stop: | |
4034 | ext4_journal_stop(handle); | |
4035 | out_dio: | |
ea3d7209 | 4036 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 TT |
4037 | ext4_inode_resume_unlocked_dio(inode); |
4038 | out_mutex: | |
5955102c | 4039 | inode_unlock(inode); |
26a4c0c6 | 4040 | return ret; |
a4bb6b64 AH |
4041 | } |
4042 | ||
a361293f JK |
4043 | int ext4_inode_attach_jinode(struct inode *inode) |
4044 | { | |
4045 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4046 | struct jbd2_inode *jinode; | |
4047 | ||
4048 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4049 | return 0; | |
4050 | ||
4051 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4052 | spin_lock(&inode->i_lock); | |
4053 | if (!ei->jinode) { | |
4054 | if (!jinode) { | |
4055 | spin_unlock(&inode->i_lock); | |
4056 | return -ENOMEM; | |
4057 | } | |
4058 | ei->jinode = jinode; | |
4059 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4060 | jinode = NULL; | |
4061 | } | |
4062 | spin_unlock(&inode->i_lock); | |
4063 | if (unlikely(jinode != NULL)) | |
4064 | jbd2_free_inode(jinode); | |
4065 | return 0; | |
4066 | } | |
4067 | ||
ac27a0ec | 4068 | /* |
617ba13b | 4069 | * ext4_truncate() |
ac27a0ec | 4070 | * |
617ba13b MC |
4071 | * We block out ext4_get_block() block instantiations across the entire |
4072 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4073 | * simultaneously on behalf of the same inode. |
4074 | * | |
42b2aa86 | 4075 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4076 | * is one core, guiding principle: the file's tree must always be consistent on |
4077 | * disk. We must be able to restart the truncate after a crash. | |
4078 | * | |
4079 | * The file's tree may be transiently inconsistent in memory (although it | |
4080 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4081 | * the contents of (the filesystem + the journal) must be consistent and | |
4082 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4083 | * left-to-right works OK too). | |
4084 | * | |
4085 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4086 | * truncate against the orphan inode list. | |
4087 | * | |
4088 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4089 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4090 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4091 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4092 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4093 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4094 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4095 | */ |
617ba13b | 4096 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 4097 | { |
819c4920 TT |
4098 | struct ext4_inode_info *ei = EXT4_I(inode); |
4099 | unsigned int credits; | |
4100 | handle_t *handle; | |
4101 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4102 | |
19b5ef61 TT |
4103 | /* |
4104 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4105 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4106 | * have i_mutex locked because it's not necessary. |
4107 | */ | |
4108 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4109 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4110 | trace_ext4_truncate_enter(inode); |
4111 | ||
91ef4caf | 4112 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4113 | return; |
4114 | ||
12e9b892 | 4115 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4116 | |
5534fb5b | 4117 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4118 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4119 | |
aef1c851 TM |
4120 | if (ext4_has_inline_data(inode)) { |
4121 | int has_inline = 1; | |
4122 | ||
4123 | ext4_inline_data_truncate(inode, &has_inline); | |
4124 | if (has_inline) | |
4125 | return; | |
4126 | } | |
4127 | ||
a361293f JK |
4128 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4129 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4130 | if (ext4_inode_attach_jinode(inode) < 0) | |
4131 | return; | |
4132 | } | |
4133 | ||
819c4920 TT |
4134 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4135 | credits = ext4_writepage_trans_blocks(inode); | |
4136 | else | |
4137 | credits = ext4_blocks_for_truncate(inode); | |
4138 | ||
4139 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4140 | if (IS_ERR(handle)) { | |
4141 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
4142 | return; | |
4143 | } | |
4144 | ||
eb3544c6 LC |
4145 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4146 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4147 | |
4148 | /* | |
4149 | * We add the inode to the orphan list, so that if this | |
4150 | * truncate spans multiple transactions, and we crash, we will | |
4151 | * resume the truncate when the filesystem recovers. It also | |
4152 | * marks the inode dirty, to catch the new size. | |
4153 | * | |
4154 | * Implication: the file must always be in a sane, consistent | |
4155 | * truncatable state while each transaction commits. | |
4156 | */ | |
4157 | if (ext4_orphan_add(handle, inode)) | |
4158 | goto out_stop; | |
4159 | ||
4160 | down_write(&EXT4_I(inode)->i_data_sem); | |
4161 | ||
4162 | ext4_discard_preallocations(inode); | |
4163 | ||
ff9893dc | 4164 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 4165 | ext4_ext_truncate(handle, inode); |
ff9893dc | 4166 | else |
819c4920 TT |
4167 | ext4_ind_truncate(handle, inode); |
4168 | ||
4169 | up_write(&ei->i_data_sem); | |
4170 | ||
4171 | if (IS_SYNC(inode)) | |
4172 | ext4_handle_sync(handle); | |
4173 | ||
4174 | out_stop: | |
4175 | /* | |
4176 | * If this was a simple ftruncate() and the file will remain alive, | |
4177 | * then we need to clear up the orphan record which we created above. | |
4178 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4179 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4180 | * orphan info for us. |
4181 | */ | |
4182 | if (inode->i_nlink) | |
4183 | ext4_orphan_del(handle, inode); | |
4184 | ||
4185 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
4186 | ext4_mark_inode_dirty(handle, inode); | |
4187 | ext4_journal_stop(handle); | |
ac27a0ec | 4188 | |
0562e0ba | 4189 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
4190 | } |
4191 | ||
ac27a0ec | 4192 | /* |
617ba13b | 4193 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4194 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4195 | * data in memory that is needed to recreate the on-disk version of this | |
4196 | * inode. | |
4197 | */ | |
617ba13b MC |
4198 | static int __ext4_get_inode_loc(struct inode *inode, |
4199 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4200 | { |
240799cd TT |
4201 | struct ext4_group_desc *gdp; |
4202 | struct buffer_head *bh; | |
4203 | struct super_block *sb = inode->i_sb; | |
4204 | ext4_fsblk_t block; | |
4205 | int inodes_per_block, inode_offset; | |
4206 | ||
3a06d778 | 4207 | iloc->bh = NULL; |
240799cd | 4208 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 4209 | return -EFSCORRUPTED; |
ac27a0ec | 4210 | |
240799cd TT |
4211 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4212 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4213 | if (!gdp) | |
ac27a0ec DK |
4214 | return -EIO; |
4215 | ||
240799cd TT |
4216 | /* |
4217 | * Figure out the offset within the block group inode table | |
4218 | */ | |
00d09882 | 4219 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4220 | inode_offset = ((inode->i_ino - 1) % |
4221 | EXT4_INODES_PER_GROUP(sb)); | |
4222 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4223 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4224 | ||
4225 | bh = sb_getblk(sb, block); | |
aebf0243 | 4226 | if (unlikely(!bh)) |
860d21e2 | 4227 | return -ENOMEM; |
ac27a0ec DK |
4228 | if (!buffer_uptodate(bh)) { |
4229 | lock_buffer(bh); | |
9c83a923 HK |
4230 | |
4231 | /* | |
4232 | * If the buffer has the write error flag, we have failed | |
4233 | * to write out another inode in the same block. In this | |
4234 | * case, we don't have to read the block because we may | |
4235 | * read the old inode data successfully. | |
4236 | */ | |
4237 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4238 | set_buffer_uptodate(bh); | |
4239 | ||
ac27a0ec DK |
4240 | if (buffer_uptodate(bh)) { |
4241 | /* someone brought it uptodate while we waited */ | |
4242 | unlock_buffer(bh); | |
4243 | goto has_buffer; | |
4244 | } | |
4245 | ||
4246 | /* | |
4247 | * If we have all information of the inode in memory and this | |
4248 | * is the only valid inode in the block, we need not read the | |
4249 | * block. | |
4250 | */ | |
4251 | if (in_mem) { | |
4252 | struct buffer_head *bitmap_bh; | |
240799cd | 4253 | int i, start; |
ac27a0ec | 4254 | |
240799cd | 4255 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4256 | |
240799cd TT |
4257 | /* Is the inode bitmap in cache? */ |
4258 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4259 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4260 | goto make_io; |
4261 | ||
4262 | /* | |
4263 | * If the inode bitmap isn't in cache then the | |
4264 | * optimisation may end up performing two reads instead | |
4265 | * of one, so skip it. | |
4266 | */ | |
4267 | if (!buffer_uptodate(bitmap_bh)) { | |
4268 | brelse(bitmap_bh); | |
4269 | goto make_io; | |
4270 | } | |
240799cd | 4271 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4272 | if (i == inode_offset) |
4273 | continue; | |
617ba13b | 4274 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4275 | break; |
4276 | } | |
4277 | brelse(bitmap_bh); | |
240799cd | 4278 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4279 | /* all other inodes are free, so skip I/O */ |
4280 | memset(bh->b_data, 0, bh->b_size); | |
4281 | set_buffer_uptodate(bh); | |
4282 | unlock_buffer(bh); | |
4283 | goto has_buffer; | |
4284 | } | |
4285 | } | |
4286 | ||
4287 | make_io: | |
240799cd TT |
4288 | /* |
4289 | * If we need to do any I/O, try to pre-readahead extra | |
4290 | * blocks from the inode table. | |
4291 | */ | |
4292 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4293 | ext4_fsblk_t b, end, table; | |
4294 | unsigned num; | |
0d606e2c | 4295 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4296 | |
4297 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4298 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4299 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4300 | if (table > b) |
4301 | b = table; | |
0d606e2c | 4302 | end = b + ra_blks; |
240799cd | 4303 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4304 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4305 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4306 | table += num / inodes_per_block; |
4307 | if (end > table) | |
4308 | end = table; | |
4309 | while (b <= end) | |
4310 | sb_breadahead(sb, b++); | |
4311 | } | |
4312 | ||
ac27a0ec DK |
4313 | /* |
4314 | * There are other valid inodes in the buffer, this inode | |
4315 | * has in-inode xattrs, or we don't have this inode in memory. | |
4316 | * Read the block from disk. | |
4317 | */ | |
0562e0ba | 4318 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4319 | get_bh(bh); |
4320 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4321 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4322 | wait_on_buffer(bh); |
4323 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4324 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4325 | "unable to read itable block"); | |
ac27a0ec DK |
4326 | brelse(bh); |
4327 | return -EIO; | |
4328 | } | |
4329 | } | |
4330 | has_buffer: | |
4331 | iloc->bh = bh; | |
4332 | return 0; | |
4333 | } | |
4334 | ||
617ba13b | 4335 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4336 | { |
4337 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4338 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4339 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4340 | } |
4341 | ||
617ba13b | 4342 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4343 | { |
617ba13b | 4344 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4345 | unsigned int new_fl = 0; |
ac27a0ec | 4346 | |
617ba13b | 4347 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4348 | new_fl |= S_SYNC; |
617ba13b | 4349 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4350 | new_fl |= S_APPEND; |
617ba13b | 4351 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4352 | new_fl |= S_IMMUTABLE; |
617ba13b | 4353 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4354 | new_fl |= S_NOATIME; |
617ba13b | 4355 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4356 | new_fl |= S_DIRSYNC; |
0a6cf913 | 4357 | if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode)) |
923ae0ff | 4358 | new_fl |= S_DAX; |
5f16f322 | 4359 | inode_set_flags(inode, new_fl, |
923ae0ff | 4360 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
4361 | } |
4362 | ||
ff9ddf7e JK |
4363 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4364 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4365 | { | |
84a8dce2 DM |
4366 | unsigned int vfs_fl; |
4367 | unsigned long old_fl, new_fl; | |
4368 | ||
4369 | do { | |
4370 | vfs_fl = ei->vfs_inode.i_flags; | |
4371 | old_fl = ei->i_flags; | |
4372 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4373 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4374 | EXT4_DIRSYNC_FL); | |
4375 | if (vfs_fl & S_SYNC) | |
4376 | new_fl |= EXT4_SYNC_FL; | |
4377 | if (vfs_fl & S_APPEND) | |
4378 | new_fl |= EXT4_APPEND_FL; | |
4379 | if (vfs_fl & S_IMMUTABLE) | |
4380 | new_fl |= EXT4_IMMUTABLE_FL; | |
4381 | if (vfs_fl & S_NOATIME) | |
4382 | new_fl |= EXT4_NOATIME_FL; | |
4383 | if (vfs_fl & S_DIRSYNC) | |
4384 | new_fl |= EXT4_DIRSYNC_FL; | |
4385 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4386 | } |
de9a55b8 | 4387 | |
0fc1b451 | 4388 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4389 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4390 | { |
4391 | blkcnt_t i_blocks ; | |
8180a562 AK |
4392 | struct inode *inode = &(ei->vfs_inode); |
4393 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4394 | |
e2b911c5 | 4395 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4396 | /* we are using combined 48 bit field */ |
4397 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4398 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4399 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4400 | /* i_blocks represent file system block size */ |
4401 | return i_blocks << (inode->i_blkbits - 9); | |
4402 | } else { | |
4403 | return i_blocks; | |
4404 | } | |
0fc1b451 AK |
4405 | } else { |
4406 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4407 | } | |
4408 | } | |
ff9ddf7e | 4409 | |
152a7b0a TM |
4410 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4411 | struct ext4_inode *raw_inode, | |
4412 | struct ext4_inode_info *ei) | |
4413 | { | |
4414 | __le32 *magic = (void *)raw_inode + | |
4415 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4416 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4417 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4418 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4419 | } else |
4420 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4421 | } |
4422 | ||
040cb378 LX |
4423 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4424 | { | |
0b7b7779 | 4425 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4426 | return -EOPNOTSUPP; |
4427 | *projid = EXT4_I(inode)->i_projid; | |
4428 | return 0; | |
4429 | } | |
4430 | ||
1d1fe1ee | 4431 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4432 | { |
617ba13b MC |
4433 | struct ext4_iloc iloc; |
4434 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4435 | struct ext4_inode_info *ei; |
1d1fe1ee | 4436 | struct inode *inode; |
b436b9be | 4437 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4438 | long ret; |
ac27a0ec | 4439 | int block; |
08cefc7a EB |
4440 | uid_t i_uid; |
4441 | gid_t i_gid; | |
040cb378 | 4442 | projid_t i_projid; |
ac27a0ec | 4443 | |
1d1fe1ee DH |
4444 | inode = iget_locked(sb, ino); |
4445 | if (!inode) | |
4446 | return ERR_PTR(-ENOMEM); | |
4447 | if (!(inode->i_state & I_NEW)) | |
4448 | return inode; | |
4449 | ||
4450 | ei = EXT4_I(inode); | |
7dc57615 | 4451 | iloc.bh = NULL; |
ac27a0ec | 4452 | |
1d1fe1ee DH |
4453 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4454 | if (ret < 0) | |
ac27a0ec | 4455 | goto bad_inode; |
617ba13b | 4456 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4457 | |
4458 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4459 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4460 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4461 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4462 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4463 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4464 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4465 | ret = -EFSCORRUPTED; |
814525f4 DW |
4466 | goto bad_inode; |
4467 | } | |
4468 | } else | |
4469 | ei->i_extra_isize = 0; | |
4470 | ||
4471 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4472 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4473 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4474 | __u32 csum; | |
4475 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4476 | __le32 gen = raw_inode->i_generation; | |
4477 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4478 | sizeof(inum)); | |
4479 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4480 | sizeof(gen)); | |
4481 | } | |
4482 | ||
4483 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4484 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4485 | ret = -EFSBADCRC; |
814525f4 DW |
4486 | goto bad_inode; |
4487 | } | |
4488 | ||
ac27a0ec | 4489 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4490 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4491 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4492 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4493 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4494 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4495 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4496 | else | |
4497 | i_projid = EXT4_DEF_PROJID; | |
4498 | ||
af5bc92d | 4499 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4500 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4501 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4502 | } |
08cefc7a EB |
4503 | i_uid_write(inode, i_uid); |
4504 | i_gid_write(inode, i_gid); | |
040cb378 | 4505 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4506 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4507 | |
353eb83c | 4508 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4509 | ei->i_inline_off = 0; |
ac27a0ec DK |
4510 | ei->i_dir_start_lookup = 0; |
4511 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4512 | /* We now have enough fields to check if the inode was active or not. | |
4513 | * This is needed because nfsd might try to access dead inodes | |
4514 | * the test is that same one that e2fsck uses | |
4515 | * NeilBrown 1999oct15 | |
4516 | */ | |
4517 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4518 | if ((inode->i_mode == 0 || |
4519 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4520 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4521 | /* this inode is deleted */ |
1d1fe1ee | 4522 | ret = -ESTALE; |
ac27a0ec DK |
4523 | goto bad_inode; |
4524 | } | |
4525 | /* The only unlinked inodes we let through here have | |
4526 | * valid i_mode and are being read by the orphan | |
4527 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4528 | * the process of deleting those. |
4529 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4530 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4531 | } |
ac27a0ec | 4532 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4533 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4534 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4535 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4536 | ei->i_file_acl |= |
4537 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4538 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4539 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4540 | #ifdef CONFIG_QUOTA |
4541 | ei->i_reserved_quota = 0; | |
4542 | #endif | |
ac27a0ec DK |
4543 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4544 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4545 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4546 | /* |
4547 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4548 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4549 | */ | |
617ba13b | 4550 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4551 | ei->i_data[block] = raw_inode->i_block[block]; |
4552 | INIT_LIST_HEAD(&ei->i_orphan); | |
4553 | ||
b436b9be JK |
4554 | /* |
4555 | * Set transaction id's of transactions that have to be committed | |
4556 | * to finish f[data]sync. We set them to currently running transaction | |
4557 | * as we cannot be sure that the inode or some of its metadata isn't | |
4558 | * part of the transaction - the inode could have been reclaimed and | |
4559 | * now it is reread from disk. | |
4560 | */ | |
4561 | if (journal) { | |
4562 | transaction_t *transaction; | |
4563 | tid_t tid; | |
4564 | ||
a931da6a | 4565 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4566 | if (journal->j_running_transaction) |
4567 | transaction = journal->j_running_transaction; | |
4568 | else | |
4569 | transaction = journal->j_committing_transaction; | |
4570 | if (transaction) | |
4571 | tid = transaction->t_tid; | |
4572 | else | |
4573 | tid = journal->j_commit_sequence; | |
a931da6a | 4574 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4575 | ei->i_sync_tid = tid; |
4576 | ei->i_datasync_tid = tid; | |
4577 | } | |
4578 | ||
0040d987 | 4579 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4580 | if (ei->i_extra_isize == 0) { |
4581 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4582 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4583 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4584 | } else { |
152a7b0a | 4585 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4586 | } |
814525f4 | 4587 | } |
ac27a0ec | 4588 | |
ef7f3835 KS |
4589 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4590 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4591 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4592 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4593 | ||
ed3654eb | 4594 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4595 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4596 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4597 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4598 | inode->i_version |= | |
4599 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4600 | } | |
25ec56b5 JNC |
4601 | } |
4602 | ||
c4b5a614 | 4603 | ret = 0; |
485c26ec | 4604 | if (ei->i_file_acl && |
1032988c | 4605 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4606 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4607 | ei->i_file_acl); | |
6a797d27 | 4608 | ret = -EFSCORRUPTED; |
485c26ec | 4609 | goto bad_inode; |
f19d5870 TM |
4610 | } else if (!ext4_has_inline_data(inode)) { |
4611 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4612 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4613 | (S_ISLNK(inode->i_mode) && | |
4614 | !ext4_inode_is_fast_symlink(inode)))) | |
4615 | /* Validate extent which is part of inode */ | |
4616 | ret = ext4_ext_check_inode(inode); | |
4617 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4618 | (S_ISLNK(inode->i_mode) && | |
4619 | !ext4_inode_is_fast_symlink(inode))) { | |
4620 | /* Validate block references which are part of inode */ | |
4621 | ret = ext4_ind_check_inode(inode); | |
4622 | } | |
fe2c8191 | 4623 | } |
567f3e9a | 4624 | if (ret) |
de9a55b8 | 4625 | goto bad_inode; |
7a262f7c | 4626 | |
ac27a0ec | 4627 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4628 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4629 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4630 | ext4_set_aops(inode); |
ac27a0ec | 4631 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4632 | inode->i_op = &ext4_dir_inode_operations; |
4633 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4634 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4635 | if (ext4_encrypted_inode(inode)) { |
4636 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4637 | ext4_set_aops(inode); | |
4638 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4639 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4640 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4641 | nd_terminate_link(ei->i_data, inode->i_size, |
4642 | sizeof(ei->i_data) - 1); | |
4643 | } else { | |
617ba13b MC |
4644 | inode->i_op = &ext4_symlink_inode_operations; |
4645 | ext4_set_aops(inode); | |
ac27a0ec | 4646 | } |
21fc61c7 | 4647 | inode_nohighmem(inode); |
563bdd61 TT |
4648 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4649 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4650 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4651 | if (raw_inode->i_block[0]) |
4652 | init_special_inode(inode, inode->i_mode, | |
4653 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4654 | else | |
4655 | init_special_inode(inode, inode->i_mode, | |
4656 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4657 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4658 | make_bad_inode(inode); | |
563bdd61 | 4659 | } else { |
6a797d27 | 4660 | ret = -EFSCORRUPTED; |
24676da4 | 4661 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4662 | goto bad_inode; |
ac27a0ec | 4663 | } |
af5bc92d | 4664 | brelse(iloc.bh); |
617ba13b | 4665 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4666 | unlock_new_inode(inode); |
4667 | return inode; | |
ac27a0ec DK |
4668 | |
4669 | bad_inode: | |
567f3e9a | 4670 | brelse(iloc.bh); |
1d1fe1ee DH |
4671 | iget_failed(inode); |
4672 | return ERR_PTR(ret); | |
ac27a0ec DK |
4673 | } |
4674 | ||
f4bb2981 TT |
4675 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4676 | { | |
4677 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4678 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4679 | return ext4_iget(sb, ino); |
4680 | } | |
4681 | ||
0fc1b451 AK |
4682 | static int ext4_inode_blocks_set(handle_t *handle, |
4683 | struct ext4_inode *raw_inode, | |
4684 | struct ext4_inode_info *ei) | |
4685 | { | |
4686 | struct inode *inode = &(ei->vfs_inode); | |
4687 | u64 i_blocks = inode->i_blocks; | |
4688 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4689 | |
4690 | if (i_blocks <= ~0U) { | |
4691 | /* | |
4907cb7b | 4692 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4693 | * as multiple of 512 bytes |
4694 | */ | |
8180a562 | 4695 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4696 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4697 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4698 | return 0; |
4699 | } | |
e2b911c5 | 4700 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4701 | return -EFBIG; |
4702 | ||
4703 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4704 | /* |
4705 | * i_blocks can be represented in a 48 bit variable | |
4706 | * as multiple of 512 bytes | |
4707 | */ | |
8180a562 | 4708 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4709 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4710 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4711 | } else { |
84a8dce2 | 4712 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4713 | /* i_block is stored in file system block size */ |
4714 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4715 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4716 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4717 | } |
f287a1a5 | 4718 | return 0; |
0fc1b451 AK |
4719 | } |
4720 | ||
a26f4992 TT |
4721 | struct other_inode { |
4722 | unsigned long orig_ino; | |
4723 | struct ext4_inode *raw_inode; | |
4724 | }; | |
4725 | ||
4726 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4727 | void *data) | |
4728 | { | |
4729 | struct other_inode *oi = (struct other_inode *) data; | |
4730 | ||
4731 | if ((inode->i_ino != ino) || | |
4732 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4733 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4734 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4735 | return 0; | |
4736 | spin_lock(&inode->i_lock); | |
4737 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4738 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4739 | (inode->i_state & I_DIRTY_TIME)) { | |
4740 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4741 | ||
4742 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4743 | spin_unlock(&inode->i_lock); | |
4744 | ||
4745 | spin_lock(&ei->i_raw_lock); | |
4746 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4747 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4748 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4749 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4750 | spin_unlock(&ei->i_raw_lock); | |
4751 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4752 | return -1; | |
4753 | } | |
4754 | spin_unlock(&inode->i_lock); | |
4755 | return -1; | |
4756 | } | |
4757 | ||
4758 | /* | |
4759 | * Opportunistically update the other time fields for other inodes in | |
4760 | * the same inode table block. | |
4761 | */ | |
4762 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4763 | unsigned long orig_ino, char *buf) | |
4764 | { | |
4765 | struct other_inode oi; | |
4766 | unsigned long ino; | |
4767 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4768 | int inode_size = EXT4_INODE_SIZE(sb); | |
4769 | ||
4770 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
4771 | /* |
4772 | * Calculate the first inode in the inode table block. Inode | |
4773 | * numbers are one-based. That is, the first inode in a block | |
4774 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4775 | */ | |
4776 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
4777 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4778 | if (ino == orig_ino) | |
4779 | continue; | |
4780 | oi.raw_inode = (struct ext4_inode *) buf; | |
4781 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4782 | } | |
4783 | } | |
4784 | ||
ac27a0ec DK |
4785 | /* |
4786 | * Post the struct inode info into an on-disk inode location in the | |
4787 | * buffer-cache. This gobbles the caller's reference to the | |
4788 | * buffer_head in the inode location struct. | |
4789 | * | |
4790 | * The caller must have write access to iloc->bh. | |
4791 | */ | |
617ba13b | 4792 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4793 | struct inode *inode, |
830156c7 | 4794 | struct ext4_iloc *iloc) |
ac27a0ec | 4795 | { |
617ba13b MC |
4796 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4797 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4798 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4799 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4800 | int err = 0, rc, block; |
202ee5df | 4801 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4802 | uid_t i_uid; |
4803 | gid_t i_gid; | |
040cb378 | 4804 | projid_t i_projid; |
ac27a0ec | 4805 | |
202ee5df TT |
4806 | spin_lock(&ei->i_raw_lock); |
4807 | ||
4808 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4809 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4810 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4811 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4812 | |
ff9ddf7e | 4813 | ext4_get_inode_flags(ei); |
ac27a0ec | 4814 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4815 | i_uid = i_uid_read(inode); |
4816 | i_gid = i_gid_read(inode); | |
040cb378 | 4817 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 4818 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4819 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4820 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4821 | /* |
4822 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4823 | * re-used with the upper 16 bits of the uid/gid intact | |
4824 | */ | |
93e3b4e6 DJ |
4825 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
4826 | raw_inode->i_uid_high = 0; | |
4827 | raw_inode->i_gid_high = 0; | |
4828 | } else { | |
ac27a0ec | 4829 | raw_inode->i_uid_high = |
08cefc7a | 4830 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4831 | raw_inode->i_gid_high = |
08cefc7a | 4832 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4833 | } |
4834 | } else { | |
08cefc7a EB |
4835 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4836 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4837 | raw_inode->i_uid_high = 0; |
4838 | raw_inode->i_gid_high = 0; | |
4839 | } | |
4840 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4841 | |
4842 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4843 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4844 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4845 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4846 | ||
bce92d56 LX |
4847 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4848 | if (err) { | |
202ee5df | 4849 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4850 | goto out_brelse; |
202ee5df | 4851 | } |
ac27a0ec | 4852 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4853 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4854 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4855 | raw_inode->i_file_acl_high = |
4856 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4857 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4858 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4859 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4860 | need_datasync = 1; | |
4861 | } | |
a48380f7 | 4862 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 4863 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 4864 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
4865 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4866 | set_large_file = 1; | |
ac27a0ec DK |
4867 | } |
4868 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4869 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4870 | if (old_valid_dev(inode->i_rdev)) { | |
4871 | raw_inode->i_block[0] = | |
4872 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4873 | raw_inode->i_block[1] = 0; | |
4874 | } else { | |
4875 | raw_inode->i_block[0] = 0; | |
4876 | raw_inode->i_block[1] = | |
4877 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4878 | raw_inode->i_block[2] = 0; | |
4879 | } | |
f19d5870 | 4880 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4881 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4882 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4883 | } |
ac27a0ec | 4884 | |
ed3654eb | 4885 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4886 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4887 | if (ei->i_extra_isize) { | |
4888 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4889 | raw_inode->i_version_hi = | |
4890 | cpu_to_le32(inode->i_version >> 32); | |
4891 | raw_inode->i_extra_isize = | |
4892 | cpu_to_le16(ei->i_extra_isize); | |
4893 | } | |
25ec56b5 | 4894 | } |
040cb378 | 4895 | |
0b7b7779 | 4896 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
4897 | i_projid != EXT4_DEF_PROJID); |
4898 | ||
4899 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
4900 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4901 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
4902 | ||
814525f4 | 4903 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 4904 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
4905 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
4906 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
4907 | bh->b_data); | |
202ee5df | 4908 | |
830156c7 | 4909 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4910 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4911 | if (!err) |
4912 | err = rc; | |
19f5fb7a | 4913 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 4914 | if (set_large_file) { |
5d601255 | 4915 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4916 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4917 | if (err) | |
4918 | goto out_brelse; | |
4919 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 4920 | ext4_set_feature_large_file(sb); |
202ee5df TT |
4921 | ext4_handle_sync(handle); |
4922 | err = ext4_handle_dirty_super(handle, sb); | |
4923 | } | |
b71fc079 | 4924 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4925 | out_brelse: |
af5bc92d | 4926 | brelse(bh); |
617ba13b | 4927 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4928 | return err; |
4929 | } | |
4930 | ||
4931 | /* | |
617ba13b | 4932 | * ext4_write_inode() |
ac27a0ec DK |
4933 | * |
4934 | * We are called from a few places: | |
4935 | * | |
87f7e416 | 4936 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4937 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4938 | * transaction to commit. |
ac27a0ec | 4939 | * |
87f7e416 TT |
4940 | * - Within flush work (sys_sync(), kupdate and such). |
4941 | * We wait on commit, if told to. | |
ac27a0ec | 4942 | * |
87f7e416 TT |
4943 | * - Within iput_final() -> write_inode_now() |
4944 | * We wait on commit, if told to. | |
ac27a0ec DK |
4945 | * |
4946 | * In all cases it is actually safe for us to return without doing anything, | |
4947 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4948 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4949 | * writeback. | |
ac27a0ec DK |
4950 | * |
4951 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4952 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4953 | * which we are interested. | |
4954 | * | |
4955 | * It would be a bug for them to not do this. The code: | |
4956 | * | |
4957 | * mark_inode_dirty(inode) | |
4958 | * stuff(); | |
4959 | * inode->i_size = expr; | |
4960 | * | |
87f7e416 TT |
4961 | * is in error because write_inode() could occur while `stuff()' is running, |
4962 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4963 | * superblock's dirty inode list. | |
ac27a0ec | 4964 | */ |
a9185b41 | 4965 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4966 | { |
91ac6f43 FM |
4967 | int err; |
4968 | ||
87f7e416 | 4969 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4970 | return 0; |
4971 | ||
91ac6f43 FM |
4972 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4973 | if (ext4_journal_current_handle()) { | |
4974 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4975 | dump_stack(); | |
4976 | return -EIO; | |
4977 | } | |
ac27a0ec | 4978 | |
10542c22 JK |
4979 | /* |
4980 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4981 | * ext4_sync_fs() will force the commit after everything is | |
4982 | * written. | |
4983 | */ | |
4984 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4985 | return 0; |
4986 | ||
4987 | err = ext4_force_commit(inode->i_sb); | |
4988 | } else { | |
4989 | struct ext4_iloc iloc; | |
ac27a0ec | 4990 | |
8b472d73 | 4991 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4992 | if (err) |
4993 | return err; | |
10542c22 JK |
4994 | /* |
4995 | * sync(2) will flush the whole buffer cache. No need to do | |
4996 | * it here separately for each inode. | |
4997 | */ | |
4998 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4999 | sync_dirty_buffer(iloc.bh); |
5000 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5001 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5002 | "IO error syncing inode"); | |
830156c7 FM |
5003 | err = -EIO; |
5004 | } | |
fd2dd9fb | 5005 | brelse(iloc.bh); |
91ac6f43 FM |
5006 | } |
5007 | return err; | |
ac27a0ec DK |
5008 | } |
5009 | ||
53e87268 JK |
5010 | /* |
5011 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5012 | * buffers that are attached to a page stradding i_size and are undergoing | |
5013 | * commit. In that case we have to wait for commit to finish and try again. | |
5014 | */ | |
5015 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5016 | { | |
5017 | struct page *page; | |
5018 | unsigned offset; | |
5019 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5020 | tid_t commit_tid = 0; | |
5021 | int ret; | |
5022 | ||
09cbfeaf | 5023 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 JK |
5024 | /* |
5025 | * All buffers in the last page remain valid? Then there's nothing to | |
ea1754a0 | 5026 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
53e87268 JK |
5027 | * blocksize case |
5028 | */ | |
09cbfeaf | 5029 | if (offset > PAGE_SIZE - (1 << inode->i_blkbits)) |
53e87268 JK |
5030 | return; |
5031 | while (1) { | |
5032 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5033 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5034 | if (!page) |
5035 | return; | |
ca99fdd2 | 5036 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5037 | PAGE_SIZE - offset); |
53e87268 | 5038 | unlock_page(page); |
09cbfeaf | 5039 | put_page(page); |
53e87268 JK |
5040 | if (ret != -EBUSY) |
5041 | return; | |
5042 | commit_tid = 0; | |
5043 | read_lock(&journal->j_state_lock); | |
5044 | if (journal->j_committing_transaction) | |
5045 | commit_tid = journal->j_committing_transaction->t_tid; | |
5046 | read_unlock(&journal->j_state_lock); | |
5047 | if (commit_tid) | |
5048 | jbd2_log_wait_commit(journal, commit_tid); | |
5049 | } | |
5050 | } | |
5051 | ||
ac27a0ec | 5052 | /* |
617ba13b | 5053 | * ext4_setattr() |
ac27a0ec DK |
5054 | * |
5055 | * Called from notify_change. | |
5056 | * | |
5057 | * We want to trap VFS attempts to truncate the file as soon as | |
5058 | * possible. In particular, we want to make sure that when the VFS | |
5059 | * shrinks i_size, we put the inode on the orphan list and modify | |
5060 | * i_disksize immediately, so that during the subsequent flushing of | |
5061 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5062 | * commit will leave the blocks being flushed in an unused state on | |
5063 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5064 | * be freed, so we have a strong guarantee that no future commit will | |
5065 | * leave these blocks visible to the user.) | |
5066 | * | |
678aaf48 JK |
5067 | * Another thing we have to assure is that if we are in ordered mode |
5068 | * and inode is still attached to the committing transaction, we must | |
5069 | * we start writeout of all the dirty pages which are being truncated. | |
5070 | * This way we are sure that all the data written in the previous | |
5071 | * transaction are already on disk (truncate waits for pages under | |
5072 | * writeback). | |
5073 | * | |
5074 | * Called with inode->i_mutex down. | |
ac27a0ec | 5075 | */ |
617ba13b | 5076 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5077 | { |
2b0143b5 | 5078 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5079 | int error, rc = 0; |
3d287de3 | 5080 | int orphan = 0; |
ac27a0ec DK |
5081 | const unsigned int ia_valid = attr->ia_valid; |
5082 | ||
31051c85 | 5083 | error = setattr_prepare(dentry, attr); |
ac27a0ec DK |
5084 | if (error) |
5085 | return error; | |
5086 | ||
a7cdadee JK |
5087 | if (is_quota_modification(inode, attr)) { |
5088 | error = dquot_initialize(inode); | |
5089 | if (error) | |
5090 | return error; | |
5091 | } | |
08cefc7a EB |
5092 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5093 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5094 | handle_t *handle; |
5095 | ||
5096 | /* (user+group)*(old+new) structure, inode write (sb, | |
5097 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5098 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5099 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5100 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5101 | if (IS_ERR(handle)) { |
5102 | error = PTR_ERR(handle); | |
5103 | goto err_out; | |
5104 | } | |
b43fa828 | 5105 | error = dquot_transfer(inode, attr); |
ac27a0ec | 5106 | if (error) { |
617ba13b | 5107 | ext4_journal_stop(handle); |
ac27a0ec DK |
5108 | return error; |
5109 | } | |
5110 | /* Update corresponding info in inode so that everything is in | |
5111 | * one transaction */ | |
5112 | if (attr->ia_valid & ATTR_UID) | |
5113 | inode->i_uid = attr->ia_uid; | |
5114 | if (attr->ia_valid & ATTR_GID) | |
5115 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5116 | error = ext4_mark_inode_dirty(handle, inode); |
5117 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5118 | } |
5119 | ||
3da40c7b | 5120 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5121 | handle_t *handle; |
3da40c7b JB |
5122 | loff_t oldsize = inode->i_size; |
5123 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 5124 | |
12e9b892 | 5125 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5126 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5127 | ||
0c095c7f TT |
5128 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5129 | return -EFBIG; | |
e2b46574 | 5130 | } |
3da40c7b JB |
5131 | if (!S_ISREG(inode->i_mode)) |
5132 | return -EINVAL; | |
dff6efc3 CH |
5133 | |
5134 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5135 | inode_inc_iversion(inode); | |
5136 | ||
3da40c7b | 5137 | if (ext4_should_order_data(inode) && |
5208386c | 5138 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 5139 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 5140 | attr->ia_size); |
3da40c7b JB |
5141 | if (error) |
5142 | goto err_out; | |
5143 | } | |
5144 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
5145 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5146 | if (IS_ERR(handle)) { | |
5147 | error = PTR_ERR(handle); | |
5148 | goto err_out; | |
5149 | } | |
3da40c7b | 5150 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5151 | error = ext4_orphan_add(handle, inode); |
5152 | orphan = 1; | |
5153 | } | |
911af577 EG |
5154 | /* |
5155 | * Update c/mtime on truncate up, ext4_truncate() will | |
5156 | * update c/mtime in shrink case below | |
5157 | */ | |
5158 | if (!shrink) { | |
5159 | inode->i_mtime = ext4_current_time(inode); | |
5160 | inode->i_ctime = inode->i_mtime; | |
5161 | } | |
90e775b7 | 5162 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5163 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5164 | rc = ext4_mark_inode_dirty(handle, inode); | |
5165 | if (!error) | |
5166 | error = rc; | |
90e775b7 JK |
5167 | /* |
5168 | * We have to update i_size under i_data_sem together | |
5169 | * with i_disksize to avoid races with writeback code | |
5170 | * running ext4_wb_update_i_disksize(). | |
5171 | */ | |
5172 | if (!error) | |
5173 | i_size_write(inode, attr->ia_size); | |
5174 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
5175 | ext4_journal_stop(handle); |
5176 | if (error) { | |
3da40c7b JB |
5177 | if (orphan) |
5178 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
5179 | goto err_out; |
5180 | } | |
d6320cbf | 5181 | } |
3da40c7b JB |
5182 | if (!shrink) |
5183 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 5184 | |
5208386c JK |
5185 | /* |
5186 | * Blocks are going to be removed from the inode. Wait | |
5187 | * for dio in flight. Temporarily disable | |
5188 | * dioread_nolock to prevent livelock. | |
5189 | */ | |
5190 | if (orphan) { | |
5191 | if (!ext4_should_journal_data(inode)) { | |
5192 | ext4_inode_block_unlocked_dio(inode); | |
5193 | inode_dio_wait(inode); | |
5194 | ext4_inode_resume_unlocked_dio(inode); | |
5195 | } else | |
5196 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 5197 | } |
ea3d7209 | 5198 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
5199 | /* |
5200 | * Truncate pagecache after we've waited for commit | |
5201 | * in data=journal mode to make pages freeable. | |
5202 | */ | |
923ae0ff | 5203 | truncate_pagecache(inode, inode->i_size); |
3da40c7b JB |
5204 | if (shrink) |
5205 | ext4_truncate(inode); | |
ea3d7209 | 5206 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5207 | } |
ac27a0ec | 5208 | |
1025774c CH |
5209 | if (!rc) { |
5210 | setattr_copy(inode, attr); | |
5211 | mark_inode_dirty(inode); | |
5212 | } | |
5213 | ||
5214 | /* | |
5215 | * If the call to ext4_truncate failed to get a transaction handle at | |
5216 | * all, we need to clean up the in-core orphan list manually. | |
5217 | */ | |
3d287de3 | 5218 | if (orphan && inode->i_nlink) |
617ba13b | 5219 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5220 | |
5221 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 5222 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5223 | |
5224 | err_out: | |
617ba13b | 5225 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5226 | if (!error) |
5227 | error = rc; | |
5228 | return error; | |
5229 | } | |
5230 | ||
3e3398a0 MC |
5231 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5232 | struct kstat *stat) | |
5233 | { | |
5234 | struct inode *inode; | |
8af8eecc | 5235 | unsigned long long delalloc_blocks; |
3e3398a0 | 5236 | |
2b0143b5 | 5237 | inode = d_inode(dentry); |
3e3398a0 MC |
5238 | generic_fillattr(inode, stat); |
5239 | ||
9206c561 AD |
5240 | /* |
5241 | * If there is inline data in the inode, the inode will normally not | |
5242 | * have data blocks allocated (it may have an external xattr block). | |
5243 | * Report at least one sector for such files, so tools like tar, rsync, | |
5244 | * others doen't incorrectly think the file is completely sparse. | |
5245 | */ | |
5246 | if (unlikely(ext4_has_inline_data(inode))) | |
5247 | stat->blocks += (stat->size + 511) >> 9; | |
5248 | ||
3e3398a0 MC |
5249 | /* |
5250 | * We can't update i_blocks if the block allocation is delayed | |
5251 | * otherwise in the case of system crash before the real block | |
5252 | * allocation is done, we will have i_blocks inconsistent with | |
5253 | * on-disk file blocks. | |
5254 | * We always keep i_blocks updated together with real | |
5255 | * allocation. But to not confuse with user, stat | |
5256 | * will return the blocks that include the delayed allocation | |
5257 | * blocks for this file. | |
5258 | */ | |
96607551 | 5259 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5260 | EXT4_I(inode)->i_reserved_data_blocks); |
5261 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5262 | return 0; |
5263 | } | |
ac27a0ec | 5264 | |
fffb2739 JK |
5265 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5266 | int pextents) | |
a02908f1 | 5267 | { |
12e9b892 | 5268 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5269 | return ext4_ind_trans_blocks(inode, lblocks); |
5270 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5271 | } |
ac51d837 | 5272 | |
ac27a0ec | 5273 | /* |
a02908f1 MC |
5274 | * Account for index blocks, block groups bitmaps and block group |
5275 | * descriptor blocks if modify datablocks and index blocks | |
5276 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5277 | * |
a02908f1 | 5278 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5279 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5280 | * they could still across block group boundary. |
ac27a0ec | 5281 | * |
a02908f1 MC |
5282 | * Also account for superblock, inode, quota and xattr blocks |
5283 | */ | |
fffb2739 JK |
5284 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
5285 | int pextents) | |
a02908f1 | 5286 | { |
8df9675f TT |
5287 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5288 | int gdpblocks; | |
a02908f1 MC |
5289 | int idxblocks; |
5290 | int ret = 0; | |
5291 | ||
5292 | /* | |
fffb2739 JK |
5293 | * How many index blocks need to touch to map @lblocks logical blocks |
5294 | * to @pextents physical extents? | |
a02908f1 | 5295 | */ |
fffb2739 | 5296 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5297 | |
5298 | ret = idxblocks; | |
5299 | ||
5300 | /* | |
5301 | * Now let's see how many group bitmaps and group descriptors need | |
5302 | * to account | |
5303 | */ | |
fffb2739 | 5304 | groups = idxblocks + pextents; |
a02908f1 | 5305 | gdpblocks = groups; |
8df9675f TT |
5306 | if (groups > ngroups) |
5307 | groups = ngroups; | |
a02908f1 MC |
5308 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5309 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5310 | ||
5311 | /* bitmaps and block group descriptor blocks */ | |
5312 | ret += groups + gdpblocks; | |
5313 | ||
5314 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5315 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5316 | ||
5317 | return ret; | |
5318 | } | |
5319 | ||
5320 | /* | |
25985edc | 5321 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5322 | * the modification of a single pages into a single transaction, |
5323 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5324 | * |
525f4ed8 | 5325 | * This could be called via ext4_write_begin() |
ac27a0ec | 5326 | * |
525f4ed8 | 5327 | * We need to consider the worse case, when |
a02908f1 | 5328 | * one new block per extent. |
ac27a0ec | 5329 | */ |
a86c6181 | 5330 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5331 | { |
617ba13b | 5332 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5333 | int ret; |
5334 | ||
fffb2739 | 5335 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5336 | |
a02908f1 | 5337 | /* Account for data blocks for journalled mode */ |
617ba13b | 5338 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5339 | ret += bpp; |
ac27a0ec DK |
5340 | return ret; |
5341 | } | |
f3bd1f3f MC |
5342 | |
5343 | /* | |
5344 | * Calculate the journal credits for a chunk of data modification. | |
5345 | * | |
5346 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5347 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5348 | * |
5349 | * journal buffers for data blocks are not included here, as DIO | |
5350 | * and fallocate do no need to journal data buffers. | |
5351 | */ | |
5352 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5353 | { | |
5354 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5355 | } | |
5356 | ||
ac27a0ec | 5357 | /* |
617ba13b | 5358 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5359 | * Give this, we know that the caller already has write access to iloc->bh. |
5360 | */ | |
617ba13b | 5361 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5362 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5363 | { |
5364 | int err = 0; | |
5365 | ||
c64db50e | 5366 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5367 | inode_inc_iversion(inode); |
5368 | ||
ac27a0ec DK |
5369 | /* the do_update_inode consumes one bh->b_count */ |
5370 | get_bh(iloc->bh); | |
5371 | ||
dab291af | 5372 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5373 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5374 | put_bh(iloc->bh); |
5375 | return err; | |
5376 | } | |
5377 | ||
5378 | /* | |
5379 | * On success, We end up with an outstanding reference count against | |
5380 | * iloc->bh. This _must_ be cleaned up later. | |
5381 | */ | |
5382 | ||
5383 | int | |
617ba13b MC |
5384 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5385 | struct ext4_iloc *iloc) | |
ac27a0ec | 5386 | { |
0390131b FM |
5387 | int err; |
5388 | ||
5389 | err = ext4_get_inode_loc(inode, iloc); | |
5390 | if (!err) { | |
5391 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5392 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5393 | if (err) { | |
5394 | brelse(iloc->bh); | |
5395 | iloc->bh = NULL; | |
ac27a0ec DK |
5396 | } |
5397 | } | |
617ba13b | 5398 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5399 | return err; |
5400 | } | |
5401 | ||
6dd4ee7c KS |
5402 | /* |
5403 | * Expand an inode by new_extra_isize bytes. | |
5404 | * Returns 0 on success or negative error number on failure. | |
5405 | */ | |
1d03ec98 AK |
5406 | static int ext4_expand_extra_isize(struct inode *inode, |
5407 | unsigned int new_extra_isize, | |
5408 | struct ext4_iloc iloc, | |
5409 | handle_t *handle) | |
6dd4ee7c KS |
5410 | { |
5411 | struct ext4_inode *raw_inode; | |
5412 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5413 | |
5414 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5415 | return 0; | |
5416 | ||
5417 | raw_inode = ext4_raw_inode(&iloc); | |
5418 | ||
5419 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5420 | |
5421 | /* No extended attributes present */ | |
19f5fb7a TT |
5422 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5423 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5424 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5425 | new_extra_isize); | |
5426 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5427 | return 0; | |
5428 | } | |
5429 | ||
5430 | /* try to expand with EAs present */ | |
5431 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5432 | raw_inode, handle); | |
5433 | } | |
5434 | ||
ac27a0ec DK |
5435 | /* |
5436 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5437 | * dirtiness (it may still be data-dirty). | |
5438 | * This means that the in-core inode may be reaped by prune_icache | |
5439 | * without having to perform any I/O. This is a very good thing, | |
5440 | * because *any* task may call prune_icache - even ones which | |
5441 | * have a transaction open against a different journal. | |
5442 | * | |
5443 | * Is this cheating? Not really. Sure, we haven't written the | |
5444 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5445 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5446 | * we start and wait on commits. | |
ac27a0ec | 5447 | */ |
617ba13b | 5448 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5449 | { |
617ba13b | 5450 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5451 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5452 | static unsigned int mnt_count; | |
5453 | int err, ret; | |
ac27a0ec DK |
5454 | |
5455 | might_sleep(); | |
7ff9c073 | 5456 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5457 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
5458 | if (err) |
5459 | return err; | |
0390131b FM |
5460 | if (ext4_handle_valid(handle) && |
5461 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5462 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5463 | /* |
5464 | * We need extra buffer credits since we may write into EA block | |
5465 | * with this same handle. If journal_extend fails, then it will | |
5466 | * only result in a minor loss of functionality for that inode. | |
5467 | * If this is felt to be critical, then e2fsck should be run to | |
5468 | * force a large enough s_min_extra_isize. | |
5469 | */ | |
5470 | if ((jbd2_journal_extend(handle, | |
5471 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5472 | ret = ext4_expand_extra_isize(inode, | |
5473 | sbi->s_want_extra_isize, | |
5474 | iloc, handle); | |
5475 | if (ret) { | |
c1bddad9 AK |
5476 | if (mnt_count != |
5477 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5478 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5479 | "Unable to expand inode %lu. Delete" |
5480 | " some EAs or run e2fsck.", | |
5481 | inode->i_ino); | |
c1bddad9 AK |
5482 | mnt_count = |
5483 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5484 | } |
5485 | } | |
5486 | } | |
5487 | } | |
5e1021f2 | 5488 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5489 | } |
5490 | ||
5491 | /* | |
617ba13b | 5492 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5493 | * |
5494 | * We're really interested in the case where a file is being extended. | |
5495 | * i_size has been changed by generic_commit_write() and we thus need | |
5496 | * to include the updated inode in the current transaction. | |
5497 | * | |
5dd4056d | 5498 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5499 | * are allocated to the file. |
5500 | * | |
5501 | * If the inode is marked synchronous, we don't honour that here - doing | |
5502 | * so would cause a commit on atime updates, which we don't bother doing. | |
5503 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5504 | * |
5505 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5506 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5507 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5508 | */ |
aa385729 | 5509 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5510 | { |
ac27a0ec DK |
5511 | handle_t *handle; |
5512 | ||
0ae45f63 TT |
5513 | if (flags == I_DIRTY_TIME) |
5514 | return; | |
9924a92a | 5515 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5516 | if (IS_ERR(handle)) |
5517 | goto out; | |
f3dc272f | 5518 | |
f3dc272f CW |
5519 | ext4_mark_inode_dirty(handle, inode); |
5520 | ||
617ba13b | 5521 | ext4_journal_stop(handle); |
ac27a0ec DK |
5522 | out: |
5523 | return; | |
5524 | } | |
5525 | ||
5526 | #if 0 | |
5527 | /* | |
5528 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5529 | * it from being flushed to disk early. Unlike | |
617ba13b | 5530 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5531 | * returns no iloc structure, so the caller needs to repeat the iloc |
5532 | * lookup to mark the inode dirty later. | |
5533 | */ | |
617ba13b | 5534 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5535 | { |
617ba13b | 5536 | struct ext4_iloc iloc; |
ac27a0ec DK |
5537 | |
5538 | int err = 0; | |
5539 | if (handle) { | |
617ba13b | 5540 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5541 | if (!err) { |
5542 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5543 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5544 | if (!err) |
0390131b | 5545 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5546 | NULL, |
0390131b | 5547 | iloc.bh); |
ac27a0ec DK |
5548 | brelse(iloc.bh); |
5549 | } | |
5550 | } | |
617ba13b | 5551 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5552 | return err; |
5553 | } | |
5554 | #endif | |
5555 | ||
617ba13b | 5556 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5557 | { |
5558 | journal_t *journal; | |
5559 | handle_t *handle; | |
5560 | int err; | |
c8585c6f | 5561 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5562 | |
5563 | /* | |
5564 | * We have to be very careful here: changing a data block's | |
5565 | * journaling status dynamically is dangerous. If we write a | |
5566 | * data block to the journal, change the status and then delete | |
5567 | * that block, we risk forgetting to revoke the old log record | |
5568 | * from the journal and so a subsequent replay can corrupt data. | |
5569 | * So, first we make sure that the journal is empty and that | |
5570 | * nobody is changing anything. | |
5571 | */ | |
5572 | ||
617ba13b | 5573 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5574 | if (!journal) |
5575 | return 0; | |
d699594d | 5576 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5577 | return -EROFS; |
5578 | ||
17335dcc DM |
5579 | /* Wait for all existing dio workers */ |
5580 | ext4_inode_block_unlocked_dio(inode); | |
5581 | inode_dio_wait(inode); | |
5582 | ||
4c546592 DJ |
5583 | /* |
5584 | * Before flushing the journal and switching inode's aops, we have | |
5585 | * to flush all dirty data the inode has. There can be outstanding | |
5586 | * delayed allocations, there can be unwritten extents created by | |
5587 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5588 | * dirty data which can be converted only after flushing the dirty | |
5589 | * data (and journalled aops don't know how to handle these cases). | |
5590 | */ | |
5591 | if (val) { | |
5592 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5593 | err = filemap_write_and_wait(inode->i_mapping); | |
5594 | if (err < 0) { | |
5595 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
5596 | ext4_inode_resume_unlocked_dio(inode); | |
5597 | return err; | |
5598 | } | |
5599 | } | |
5600 | ||
c8585c6f | 5601 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
dab291af | 5602 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5603 | |
5604 | /* | |
5605 | * OK, there are no updates running now, and all cached data is | |
5606 | * synced to disk. We are now in a completely consistent state | |
5607 | * which doesn't have anything in the journal, and we know that | |
5608 | * no filesystem updates are running, so it is safe to modify | |
5609 | * the inode's in-core data-journaling state flag now. | |
5610 | */ | |
5611 | ||
5612 | if (val) | |
12e9b892 | 5613 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5614 | else { |
4f879ca6 JK |
5615 | err = jbd2_journal_flush(journal); |
5616 | if (err < 0) { | |
5617 | jbd2_journal_unlock_updates(journal); | |
c8585c6f | 5618 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
4f879ca6 JK |
5619 | ext4_inode_resume_unlocked_dio(inode); |
5620 | return err; | |
5621 | } | |
12e9b892 | 5622 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5623 | } |
617ba13b | 5624 | ext4_set_aops(inode); |
ac27a0ec | 5625 | |
dab291af | 5626 | jbd2_journal_unlock_updates(journal); |
c8585c6f DJ |
5627 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
5628 | ||
4c546592 DJ |
5629 | if (val) |
5630 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
17335dcc | 5631 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5632 | |
5633 | /* Finally we can mark the inode as dirty. */ | |
5634 | ||
9924a92a | 5635 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5636 | if (IS_ERR(handle)) |
5637 | return PTR_ERR(handle); | |
5638 | ||
617ba13b | 5639 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5640 | ext4_handle_sync(handle); |
617ba13b MC |
5641 | ext4_journal_stop(handle); |
5642 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5643 | |
5644 | return err; | |
5645 | } | |
2e9ee850 AK |
5646 | |
5647 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5648 | { | |
5649 | return !buffer_mapped(bh); | |
5650 | } | |
5651 | ||
c2ec175c | 5652 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5653 | { |
c2ec175c | 5654 | struct page *page = vmf->page; |
2e9ee850 AK |
5655 | loff_t size; |
5656 | unsigned long len; | |
9ea7df53 | 5657 | int ret; |
2e9ee850 | 5658 | struct file *file = vma->vm_file; |
496ad9aa | 5659 | struct inode *inode = file_inode(file); |
2e9ee850 | 5660 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5661 | handle_t *handle; |
5662 | get_block_t *get_block; | |
5663 | int retries = 0; | |
2e9ee850 | 5664 | |
8e8ad8a5 | 5665 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5666 | file_update_time(vma->vm_file); |
ea3d7209 JK |
5667 | |
5668 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
9ea7df53 JK |
5669 | /* Delalloc case is easy... */ |
5670 | if (test_opt(inode->i_sb, DELALLOC) && | |
5671 | !ext4_should_journal_data(inode) && | |
5672 | !ext4_nonda_switch(inode->i_sb)) { | |
5673 | do { | |
5c500029 | 5674 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
5675 | ext4_da_get_block_prep); |
5676 | } while (ret == -ENOSPC && | |
5677 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5678 | goto out_ret; | |
2e9ee850 | 5679 | } |
0e499890 DW |
5680 | |
5681 | lock_page(page); | |
9ea7df53 JK |
5682 | size = i_size_read(inode); |
5683 | /* Page got truncated from under us? */ | |
5684 | if (page->mapping != mapping || page_offset(page) > size) { | |
5685 | unlock_page(page); | |
5686 | ret = VM_FAULT_NOPAGE; | |
5687 | goto out; | |
0e499890 | 5688 | } |
2e9ee850 | 5689 | |
09cbfeaf KS |
5690 | if (page->index == size >> PAGE_SHIFT) |
5691 | len = size & ~PAGE_MASK; | |
2e9ee850 | 5692 | else |
09cbfeaf | 5693 | len = PAGE_SIZE; |
a827eaff | 5694 | /* |
9ea7df53 JK |
5695 | * Return if we have all the buffers mapped. This avoids the need to do |
5696 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5697 | */ |
2e9ee850 | 5698 | if (page_has_buffers(page)) { |
f19d5870 TM |
5699 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5700 | 0, len, NULL, | |
5701 | ext4_bh_unmapped)) { | |
9ea7df53 | 5702 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5703 | wait_for_stable_page(page); |
9ea7df53 JK |
5704 | ret = VM_FAULT_LOCKED; |
5705 | goto out; | |
a827eaff | 5706 | } |
2e9ee850 | 5707 | } |
a827eaff | 5708 | unlock_page(page); |
9ea7df53 JK |
5709 | /* OK, we need to fill the hole... */ |
5710 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 5711 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
5712 | else |
5713 | get_block = ext4_get_block; | |
5714 | retry_alloc: | |
9924a92a TT |
5715 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5716 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5717 | if (IS_ERR(handle)) { |
c2ec175c | 5718 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5719 | goto out; |
5720 | } | |
5c500029 | 5721 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 5722 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 5723 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 5724 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
5725 | unlock_page(page); |
5726 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5727 | ext4_journal_stop(handle); |
9ea7df53 JK |
5728 | goto out; |
5729 | } | |
5730 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5731 | } | |
5732 | ext4_journal_stop(handle); | |
5733 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5734 | goto retry_alloc; | |
5735 | out_ret: | |
5736 | ret = block_page_mkwrite_return(ret); | |
5737 | out: | |
ea3d7209 | 5738 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 5739 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5740 | return ret; |
5741 | } | |
ea3d7209 JK |
5742 | |
5743 | int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
5744 | { | |
5745 | struct inode *inode = file_inode(vma->vm_file); | |
5746 | int err; | |
5747 | ||
5748 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
5749 | err = filemap_fault(vma, vmf); | |
5750 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
5751 | ||
5752 | return err; | |
5753 | } | |
2d90c160 JK |
5754 | |
5755 | /* | |
5756 | * Find the first extent at or after @lblk in an inode that is not a hole. | |
5757 | * Search for @map_len blocks at most. The extent is returned in @result. | |
5758 | * | |
5759 | * The function returns 1 if we found an extent. The function returns 0 in | |
5760 | * case there is no extent at or after @lblk and in that case also sets | |
5761 | * @result->es_len to 0. In case of error, the error code is returned. | |
5762 | */ | |
5763 | int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk, | |
5764 | unsigned int map_len, struct extent_status *result) | |
5765 | { | |
5766 | struct ext4_map_blocks map; | |
5767 | struct extent_status es = {}; | |
5768 | int ret; | |
5769 | ||
5770 | map.m_lblk = lblk; | |
5771 | map.m_len = map_len; | |
5772 | ||
5773 | /* | |
5774 | * For non-extent based files this loop may iterate several times since | |
5775 | * we do not determine full hole size. | |
5776 | */ | |
5777 | while (map.m_len > 0) { | |
5778 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
5779 | if (ret < 0) | |
5780 | return ret; | |
5781 | /* There's extent covering m_lblk? Just return it. */ | |
5782 | if (ret > 0) { | |
5783 | int status; | |
5784 | ||
5785 | ext4_es_store_pblock(result, map.m_pblk); | |
5786 | result->es_lblk = map.m_lblk; | |
5787 | result->es_len = map.m_len; | |
5788 | if (map.m_flags & EXT4_MAP_UNWRITTEN) | |
5789 | status = EXTENT_STATUS_UNWRITTEN; | |
5790 | else | |
5791 | status = EXTENT_STATUS_WRITTEN; | |
5792 | ext4_es_store_status(result, status); | |
5793 | return 1; | |
5794 | } | |
5795 | ext4_es_find_delayed_extent_range(inode, map.m_lblk, | |
5796 | map.m_lblk + map.m_len - 1, | |
5797 | &es); | |
5798 | /* Is delalloc data before next block in extent tree? */ | |
5799 | if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) { | |
5800 | ext4_lblk_t offset = 0; | |
5801 | ||
5802 | if (es.es_lblk < lblk) | |
5803 | offset = lblk - es.es_lblk; | |
5804 | result->es_lblk = es.es_lblk + offset; | |
5805 | ext4_es_store_pblock(result, | |
5806 | ext4_es_pblock(&es) + offset); | |
5807 | result->es_len = es.es_len - offset; | |
5808 | ext4_es_store_status(result, ext4_es_status(&es)); | |
5809 | ||
5810 | return 1; | |
5811 | } | |
5812 | /* There's a hole at m_lblk, advance us after it */ | |
5813 | map.m_lblk += map.m_len; | |
5814 | map_len -= map.m_len; | |
5815 | map.m_len = map_len; | |
5816 | cond_resched(); | |
5817 | } | |
5818 | result->es_len = 0; | |
5819 | return 0; | |
5820 | } |