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> |
364443cb | 40 | #include <linux/iomap.h> |
9bffad1e | 41 | |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
ac27a0ec DK |
43 | #include "xattr.h" |
44 | #include "acl.h" | |
9f125d64 | 45 | #include "truncate.h" |
ac27a0ec | 46 | |
9bffad1e TT |
47 | #include <trace/events/ext4.h> |
48 | ||
a1d6cc56 AK |
49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
50 | ||
814525f4 DW |
51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
52 | struct ext4_inode_info *ei) | |
53 | { | |
54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
814525f4 | 55 | __u32 csum; |
b47820ed DJ |
56 | __u16 dummy_csum = 0; |
57 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
58 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 59 | |
b47820ed DJ |
60 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
61 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
62 | offset += csum_size; | |
63 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
64 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 65 | |
b47820ed DJ |
66 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
67 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
68 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
69 | EXT4_GOOD_OLD_INODE_SIZE, | |
70 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
71 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
72 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
73 | csum_size); | |
74 | offset += csum_size; | |
b47820ed | 75 | } |
05ac5aa1 DJ |
76 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
77 | EXT4_INODE_SIZE(inode->i_sb) - offset); | |
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 | } | |
2c98eb5e TT |
264 | if (inode->i_blocks) { |
265 | err = ext4_truncate(inode); | |
266 | if (err) { | |
267 | ext4_error(inode->i_sb, | |
268 | "couldn't truncate inode %lu (err %d)", | |
269 | inode->i_ino, err); | |
270 | goto stop_handle; | |
271 | } | |
272 | } | |
bc965ab3 TT |
273 | |
274 | /* | |
275 | * ext4_ext_truncate() doesn't reserve any slop when it | |
276 | * restarts journal transactions; therefore there may not be | |
277 | * enough credits left in the handle to remove the inode from | |
278 | * the orphan list and set the dtime field. | |
279 | */ | |
0390131b | 280 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
281 | err = ext4_journal_extend(handle, 3); |
282 | if (err > 0) | |
283 | err = ext4_journal_restart(handle, 3); | |
284 | if (err != 0) { | |
12062ddd | 285 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
286 | "couldn't extend journal (err %d)", err); |
287 | stop_handle: | |
288 | ext4_journal_stop(handle); | |
45388219 | 289 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 290 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
291 | goto no_delete; |
292 | } | |
293 | } | |
294 | ||
ac27a0ec | 295 | /* |
617ba13b | 296 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 297 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 298 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 299 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 300 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
301 | * (Well, we could do this if we need to, but heck - it works) |
302 | */ | |
617ba13b MC |
303 | ext4_orphan_del(handle, inode); |
304 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
305 | |
306 | /* | |
307 | * One subtle ordering requirement: if anything has gone wrong | |
308 | * (transaction abort, IO errors, whatever), then we can still | |
309 | * do these next steps (the fs will already have been marked as | |
310 | * having errors), but we can't free the inode if the mark_dirty | |
311 | * fails. | |
312 | */ | |
617ba13b | 313 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 314 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 315 | ext4_clear_inode(inode); |
ac27a0ec | 316 | else |
617ba13b MC |
317 | ext4_free_inode(handle, inode); |
318 | ext4_journal_stop(handle); | |
8e8ad8a5 | 319 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
320 | return; |
321 | no_delete: | |
0930fcc1 | 322 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
323 | } |
324 | ||
a9e7f447 DM |
325 | #ifdef CONFIG_QUOTA |
326 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 327 | { |
a9e7f447 | 328 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 329 | } |
a9e7f447 | 330 | #endif |
9d0be502 | 331 | |
0637c6f4 TT |
332 | /* |
333 | * Called with i_data_sem down, which is important since we can call | |
334 | * ext4_discard_preallocations() from here. | |
335 | */ | |
5f634d06 AK |
336 | void ext4_da_update_reserve_space(struct inode *inode, |
337 | int used, int quota_claim) | |
12219aea AK |
338 | { |
339 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 340 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
341 | |
342 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 343 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 344 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 345 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 346 | "with only %d reserved data blocks", |
0637c6f4 TT |
347 | __func__, inode->i_ino, used, |
348 | ei->i_reserved_data_blocks); | |
349 | WARN_ON(1); | |
350 | used = ei->i_reserved_data_blocks; | |
351 | } | |
12219aea | 352 | |
0637c6f4 TT |
353 | /* Update per-inode reservations */ |
354 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 355 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 356 | |
12219aea | 357 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 358 | |
72b8ab9d ES |
359 | /* Update quota subsystem for data blocks */ |
360 | if (quota_claim) | |
7b415bf6 | 361 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 362 | else { |
5f634d06 AK |
363 | /* |
364 | * We did fallocate with an offset that is already delayed | |
365 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 366 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 367 | */ |
7b415bf6 | 368 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 369 | } |
d6014301 AK |
370 | |
371 | /* | |
372 | * If we have done all the pending block allocations and if | |
373 | * there aren't any writers on the inode, we can discard the | |
374 | * inode's preallocations. | |
375 | */ | |
0637c6f4 TT |
376 | if ((ei->i_reserved_data_blocks == 0) && |
377 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 378 | ext4_discard_preallocations(inode); |
12219aea AK |
379 | } |
380 | ||
e29136f8 | 381 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
382 | unsigned int line, |
383 | struct ext4_map_blocks *map) | |
6fd058f7 | 384 | { |
24676da4 TT |
385 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
386 | map->m_len)) { | |
c398eda0 TT |
387 | ext4_error_inode(inode, func, line, map->m_pblk, |
388 | "lblock %lu mapped to illegal pblock " | |
389 | "(length %d)", (unsigned long) map->m_lblk, | |
390 | map->m_len); | |
6a797d27 | 391 | return -EFSCORRUPTED; |
6fd058f7 TT |
392 | } |
393 | return 0; | |
394 | } | |
395 | ||
53085fac JK |
396 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
397 | ext4_lblk_t len) | |
398 | { | |
399 | int ret; | |
400 | ||
401 | if (ext4_encrypted_inode(inode)) | |
a7550b30 | 402 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
403 | |
404 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
405 | if (ret > 0) | |
406 | ret = 0; | |
407 | ||
408 | return ret; | |
409 | } | |
410 | ||
e29136f8 | 411 | #define check_block_validity(inode, map) \ |
c398eda0 | 412 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 413 | |
921f266b DM |
414 | #ifdef ES_AGGRESSIVE_TEST |
415 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
416 | struct inode *inode, | |
417 | struct ext4_map_blocks *es_map, | |
418 | struct ext4_map_blocks *map, | |
419 | int flags) | |
420 | { | |
421 | int retval; | |
422 | ||
423 | map->m_flags = 0; | |
424 | /* | |
425 | * There is a race window that the result is not the same. | |
426 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
427 | * is that we lookup a block mapping in extent status tree with | |
428 | * out taking i_data_sem. So at the time the unwritten extent | |
429 | * could be converted. | |
430 | */ | |
2dcba478 | 431 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
432 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
433 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
434 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
435 | } else { | |
436 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
437 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
438 | } | |
2dcba478 | 439 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
440 | |
441 | /* | |
442 | * We don't check m_len because extent will be collpased in status | |
443 | * tree. So the m_len might not equal. | |
444 | */ | |
445 | if (es_map->m_lblk != map->m_lblk || | |
446 | es_map->m_flags != map->m_flags || | |
447 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 448 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
449 | "es_cached ex [%d/%d/%llu/%x] != " |
450 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
451 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
452 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
453 | map->m_len, map->m_pblk, map->m_flags, | |
454 | retval, flags); | |
455 | } | |
456 | } | |
457 | #endif /* ES_AGGRESSIVE_TEST */ | |
458 | ||
f5ab0d1f | 459 | /* |
e35fd660 | 460 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 461 | * and returns if the blocks are already mapped. |
f5ab0d1f | 462 | * |
f5ab0d1f MC |
463 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
464 | * and store the allocated blocks in the result buffer head and mark it | |
465 | * mapped. | |
466 | * | |
e35fd660 TT |
467 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
468 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
469 | * based files |
470 | * | |
facab4d9 JK |
471 | * On success, it returns the number of blocks being mapped or allocated. if |
472 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
473 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
474 | * |
475 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
476 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
477 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
478 | * |
479 | * It returns the error in case of allocation failure. | |
480 | */ | |
e35fd660 TT |
481 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
482 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 483 | { |
d100eef2 | 484 | struct extent_status es; |
0e855ac8 | 485 | int retval; |
b8a86845 | 486 | int ret = 0; |
921f266b DM |
487 | #ifdef ES_AGGRESSIVE_TEST |
488 | struct ext4_map_blocks orig_map; | |
489 | ||
490 | memcpy(&orig_map, map, sizeof(*map)); | |
491 | #endif | |
f5ab0d1f | 492 | |
e35fd660 TT |
493 | map->m_flags = 0; |
494 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
495 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
496 | (unsigned long) map->m_lblk); | |
d100eef2 | 497 | |
e861b5e9 TT |
498 | /* |
499 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
500 | */ | |
501 | if (unlikely(map->m_len > INT_MAX)) | |
502 | map->m_len = INT_MAX; | |
503 | ||
4adb6ab3 KM |
504 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
505 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 506 | return -EFSCORRUPTED; |
4adb6ab3 | 507 | |
d100eef2 ZL |
508 | /* Lookup extent status tree firstly */ |
509 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
510 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
511 | map->m_pblk = ext4_es_pblock(&es) + | |
512 | map->m_lblk - es.es_lblk; | |
513 | map->m_flags |= ext4_es_is_written(&es) ? | |
514 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
515 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
516 | if (retval > map->m_len) | |
517 | retval = map->m_len; | |
518 | map->m_len = retval; | |
519 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
520 | map->m_pblk = 0; |
521 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
522 | if (retval > map->m_len) | |
523 | retval = map->m_len; | |
524 | map->m_len = retval; | |
d100eef2 ZL |
525 | retval = 0; |
526 | } else { | |
527 | BUG_ON(1); | |
528 | } | |
921f266b DM |
529 | #ifdef ES_AGGRESSIVE_TEST |
530 | ext4_map_blocks_es_recheck(handle, inode, map, | |
531 | &orig_map, flags); | |
532 | #endif | |
d100eef2 ZL |
533 | goto found; |
534 | } | |
535 | ||
4df3d265 | 536 | /* |
b920c755 TT |
537 | * Try to see if we can get the block without requesting a new |
538 | * file system block. | |
4df3d265 | 539 | */ |
2dcba478 | 540 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 541 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
542 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
543 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 544 | } else { |
a4e5d88b DM |
545 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
546 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 547 | } |
f7fec032 | 548 | if (retval > 0) { |
3be78c73 | 549 | unsigned int status; |
f7fec032 | 550 | |
44fb851d ZL |
551 | if (unlikely(retval != map->m_len)) { |
552 | ext4_warning(inode->i_sb, | |
553 | "ES len assertion failed for inode " | |
554 | "%lu: retval %d != map->m_len %d", | |
555 | inode->i_ino, retval, map->m_len); | |
556 | WARN_ON(1); | |
921f266b | 557 | } |
921f266b | 558 | |
f7fec032 ZL |
559 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
560 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
561 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 562 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
563 | ext4_find_delalloc_range(inode, map->m_lblk, |
564 | map->m_lblk + map->m_len - 1)) | |
565 | status |= EXTENT_STATUS_DELAYED; | |
566 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
567 | map->m_len, map->m_pblk, status); | |
568 | if (ret < 0) | |
569 | retval = ret; | |
570 | } | |
2dcba478 | 571 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 572 | |
d100eef2 | 573 | found: |
e35fd660 | 574 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 575 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
576 | if (ret != 0) |
577 | return ret; | |
578 | } | |
579 | ||
f5ab0d1f | 580 | /* If it is only a block(s) look up */ |
c2177057 | 581 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
582 | return retval; |
583 | ||
584 | /* | |
585 | * Returns if the blocks have already allocated | |
586 | * | |
587 | * Note that if blocks have been preallocated | |
df3ab170 | 588 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
589 | * with buffer head unmapped. |
590 | */ | |
e35fd660 | 591 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
592 | /* |
593 | * If we need to convert extent to unwritten | |
594 | * we continue and do the actual work in | |
595 | * ext4_ext_map_blocks() | |
596 | */ | |
597 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
598 | return retval; | |
4df3d265 | 599 | |
2a8964d6 | 600 | /* |
a25a4e1a ZL |
601 | * Here we clear m_flags because after allocating an new extent, |
602 | * it will be set again. | |
2a8964d6 | 603 | */ |
a25a4e1a | 604 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 605 | |
4df3d265 | 606 | /* |
556615dc | 607 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 608 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 609 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 610 | * with create == 1 flag. |
4df3d265 | 611 | */ |
c8b459f4 | 612 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 613 | |
4df3d265 AK |
614 | /* |
615 | * We need to check for EXT4 here because migrate | |
616 | * could have changed the inode type in between | |
617 | */ | |
12e9b892 | 618 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 619 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 620 | } else { |
e35fd660 | 621 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 622 | |
e35fd660 | 623 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
624 | /* |
625 | * We allocated new blocks which will result in | |
626 | * i_data's format changing. Force the migrate | |
627 | * to fail by clearing migrate flags | |
628 | */ | |
19f5fb7a | 629 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 630 | } |
d2a17637 | 631 | |
5f634d06 AK |
632 | /* |
633 | * Update reserved blocks/metadata blocks after successful | |
634 | * block allocation which had been deferred till now. We don't | |
635 | * support fallocate for non extent files. So we can update | |
636 | * reserve space here. | |
637 | */ | |
638 | if ((retval > 0) && | |
1296cc85 | 639 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
640 | ext4_da_update_reserve_space(inode, retval, 1); |
641 | } | |
2ac3b6e0 | 642 | |
f7fec032 | 643 | if (retval > 0) { |
3be78c73 | 644 | unsigned int status; |
f7fec032 | 645 | |
44fb851d ZL |
646 | if (unlikely(retval != map->m_len)) { |
647 | ext4_warning(inode->i_sb, | |
648 | "ES len assertion failed for inode " | |
649 | "%lu: retval %d != map->m_len %d", | |
650 | inode->i_ino, retval, map->m_len); | |
651 | WARN_ON(1); | |
921f266b | 652 | } |
921f266b | 653 | |
c86d8db3 JK |
654 | /* |
655 | * We have to zeroout blocks before inserting them into extent | |
656 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
657 | * use them before they are really zeroed. We also have to |
658 | * unmap metadata before zeroing as otherwise writeback can | |
659 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
660 | */ |
661 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
662 | map->m_flags & EXT4_MAP_MAPPED && | |
663 | map->m_flags & EXT4_MAP_NEW) { | |
64e1c57f JK |
664 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
665 | map->m_len); | |
c86d8db3 JK |
666 | ret = ext4_issue_zeroout(inode, map->m_lblk, |
667 | map->m_pblk, map->m_len); | |
668 | if (ret) { | |
669 | retval = ret; | |
670 | goto out_sem; | |
671 | } | |
672 | } | |
673 | ||
adb23551 ZL |
674 | /* |
675 | * If the extent has been zeroed out, we don't need to update | |
676 | * extent status tree. | |
677 | */ | |
678 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
679 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
680 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 681 | goto out_sem; |
adb23551 | 682 | } |
f7fec032 ZL |
683 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
684 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
685 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 686 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
687 | ext4_find_delalloc_range(inode, map->m_lblk, |
688 | map->m_lblk + map->m_len - 1)) | |
689 | status |= EXTENT_STATUS_DELAYED; | |
690 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
691 | map->m_pblk, status); | |
c86d8db3 | 692 | if (ret < 0) { |
f7fec032 | 693 | retval = ret; |
c86d8db3 JK |
694 | goto out_sem; |
695 | } | |
5356f261 AK |
696 | } |
697 | ||
c86d8db3 | 698 | out_sem: |
4df3d265 | 699 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 700 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 701 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
702 | if (ret != 0) |
703 | return ret; | |
06bd3c36 JK |
704 | |
705 | /* | |
706 | * Inodes with freshly allocated blocks where contents will be | |
707 | * visible after transaction commit must be on transaction's | |
708 | * ordered data list. | |
709 | */ | |
710 | if (map->m_flags & EXT4_MAP_NEW && | |
711 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
712 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
713 | !IS_NOQUOTA(inode) && | |
714 | ext4_should_order_data(inode)) { | |
ee0876bc JK |
715 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
716 | ret = ext4_jbd2_inode_add_wait(handle, inode); | |
717 | else | |
718 | ret = ext4_jbd2_inode_add_write(handle, inode); | |
06bd3c36 JK |
719 | if (ret) |
720 | return ret; | |
721 | } | |
6fd058f7 | 722 | } |
0e855ac8 AK |
723 | return retval; |
724 | } | |
725 | ||
ed8ad838 JK |
726 | /* |
727 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
728 | * we have to be careful as someone else may be manipulating b_state as well. | |
729 | */ | |
730 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
731 | { | |
732 | unsigned long old_state; | |
733 | unsigned long new_state; | |
734 | ||
735 | flags &= EXT4_MAP_FLAGS; | |
736 | ||
737 | /* Dummy buffer_head? Set non-atomically. */ | |
738 | if (!bh->b_page) { | |
739 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
740 | return; | |
741 | } | |
742 | /* | |
743 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
744 | * once we get rid of using bh as a container for mapping information | |
745 | * to pass to / from get_block functions, this can go away. | |
746 | */ | |
747 | do { | |
748 | old_state = READ_ONCE(bh->b_state); | |
749 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
750 | } while (unlikely( | |
751 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
752 | } | |
753 | ||
2ed88685 TT |
754 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
755 | struct buffer_head *bh, int flags) | |
ac27a0ec | 756 | { |
2ed88685 | 757 | struct ext4_map_blocks map; |
efe70c29 | 758 | int ret = 0; |
ac27a0ec | 759 | |
46c7f254 TM |
760 | if (ext4_has_inline_data(inode)) |
761 | return -ERANGE; | |
762 | ||
2ed88685 TT |
763 | map.m_lblk = iblock; |
764 | map.m_len = bh->b_size >> inode->i_blkbits; | |
765 | ||
efe70c29 JK |
766 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
767 | flags); | |
7fb5409d | 768 | if (ret > 0) { |
2ed88685 | 769 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 770 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 771 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 772 | ret = 0; |
547edce3 RZ |
773 | } else if (ret == 0) { |
774 | /* hole case, need to fill in bh->b_size */ | |
775 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
776 | } |
777 | return ret; | |
778 | } | |
779 | ||
2ed88685 TT |
780 | int ext4_get_block(struct inode *inode, sector_t iblock, |
781 | struct buffer_head *bh, int create) | |
782 | { | |
783 | return _ext4_get_block(inode, iblock, bh, | |
784 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
785 | } | |
786 | ||
705965bd JK |
787 | /* |
788 | * Get block function used when preparing for buffered write if we require | |
789 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
790 | * will be converted to written after the IO is complete. | |
791 | */ | |
792 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
793 | struct buffer_head *bh_result, int create) | |
794 | { | |
795 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
796 | inode->i_ino, create); | |
797 | return _ext4_get_block(inode, iblock, bh_result, | |
798 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
799 | } | |
800 | ||
efe70c29 JK |
801 | /* Maximum number of blocks we map for direct IO at once. */ |
802 | #define DIO_MAX_BLOCKS 4096 | |
803 | ||
e84dfbe2 JK |
804 | /* |
805 | * Get blocks function for the cases that need to start a transaction - | |
806 | * generally difference cases of direct IO and DAX IO. It also handles retries | |
807 | * in case of ENOSPC. | |
808 | */ | |
809 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, | |
810 | struct buffer_head *bh_result, int flags) | |
efe70c29 JK |
811 | { |
812 | int dio_credits; | |
e84dfbe2 JK |
813 | handle_t *handle; |
814 | int retries = 0; | |
815 | int ret; | |
efe70c29 JK |
816 | |
817 | /* Trim mapping request to maximum we can map at once for DIO */ | |
818 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) | |
819 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; | |
820 | dio_credits = ext4_chunk_trans_blocks(inode, | |
821 | bh_result->b_size >> inode->i_blkbits); | |
e84dfbe2 JK |
822 | retry: |
823 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); | |
824 | if (IS_ERR(handle)) | |
825 | return PTR_ERR(handle); | |
826 | ||
827 | ret = _ext4_get_block(inode, iblock, bh_result, flags); | |
828 | ext4_journal_stop(handle); | |
829 | ||
830 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
831 | goto retry; | |
832 | return ret; | |
efe70c29 JK |
833 | } |
834 | ||
705965bd JK |
835 | /* Get block function for DIO reads and writes to inodes without extents */ |
836 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, | |
837 | struct buffer_head *bh, int create) | |
838 | { | |
efe70c29 JK |
839 | /* We don't expect handle for direct IO */ |
840 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
841 | ||
e84dfbe2 JK |
842 | if (!create) |
843 | return _ext4_get_block(inode, iblock, bh, 0); | |
844 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); | |
705965bd JK |
845 | } |
846 | ||
847 | /* | |
109811c2 | 848 | * Get block function for AIO DIO writes when we create unwritten extent if |
705965bd JK |
849 | * blocks are not allocated yet. The extent will be converted to written |
850 | * after IO is complete. | |
851 | */ | |
109811c2 JK |
852 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
853 | sector_t iblock, struct buffer_head *bh_result, int create) | |
705965bd | 854 | { |
efe70c29 JK |
855 | int ret; |
856 | ||
efe70c29 JK |
857 | /* We don't expect handle for direct IO */ |
858 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
859 | ||
e84dfbe2 JK |
860 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
861 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
efe70c29 | 862 | |
109811c2 JK |
863 | /* |
864 | * When doing DIO using unwritten extents, we need io_end to convert | |
865 | * unwritten extents to written on IO completion. We allocate io_end | |
866 | * once we spot unwritten extent and store it in b_private. Generic | |
867 | * DIO code keeps b_private set and furthermore passes the value to | |
868 | * our completion callback in 'private' argument. | |
869 | */ | |
870 | if (!ret && buffer_unwritten(bh_result)) { | |
871 | if (!bh_result->b_private) { | |
872 | ext4_io_end_t *io_end; | |
873 | ||
874 | io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
875 | if (!io_end) | |
876 | return -ENOMEM; | |
877 | bh_result->b_private = io_end; | |
878 | ext4_set_io_unwritten_flag(inode, io_end); | |
879 | } | |
efe70c29 | 880 | set_buffer_defer_completion(bh_result); |
efe70c29 JK |
881 | } |
882 | ||
883 | return ret; | |
705965bd JK |
884 | } |
885 | ||
109811c2 JK |
886 | /* |
887 | * Get block function for non-AIO DIO writes when we create unwritten extent if | |
888 | * blocks are not allocated yet. The extent will be converted to written | |
889 | * after IO is complete from ext4_ext_direct_IO() function. | |
890 | */ | |
891 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, | |
892 | sector_t iblock, struct buffer_head *bh_result, int create) | |
893 | { | |
109811c2 JK |
894 | int ret; |
895 | ||
896 | /* We don't expect handle for direct IO */ | |
897 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
898 | ||
e84dfbe2 JK |
899 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
900 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
109811c2 JK |
901 | |
902 | /* | |
903 | * Mark inode as having pending DIO writes to unwritten extents. | |
904 | * ext4_ext_direct_IO() checks this flag and converts extents to | |
905 | * written. | |
906 | */ | |
907 | if (!ret && buffer_unwritten(bh_result)) | |
908 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
909 | ||
910 | return ret; | |
911 | } | |
912 | ||
705965bd JK |
913 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
914 | struct buffer_head *bh_result, int create) | |
915 | { | |
916 | int ret; | |
917 | ||
918 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", | |
919 | inode->i_ino, create); | |
efe70c29 JK |
920 | /* We don't expect handle for direct IO */ |
921 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
922 | ||
705965bd JK |
923 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
924 | /* | |
925 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
926 | * that. | |
927 | */ | |
efe70c29 | 928 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
705965bd JK |
929 | |
930 | return ret; | |
931 | } | |
932 | ||
933 | ||
ac27a0ec DK |
934 | /* |
935 | * `handle' can be NULL if create is zero | |
936 | */ | |
617ba13b | 937 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 938 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 939 | { |
2ed88685 TT |
940 | struct ext4_map_blocks map; |
941 | struct buffer_head *bh; | |
c5e298ae | 942 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 943 | int err; |
ac27a0ec DK |
944 | |
945 | J_ASSERT(handle != NULL || create == 0); | |
946 | ||
2ed88685 TT |
947 | map.m_lblk = block; |
948 | map.m_len = 1; | |
c5e298ae | 949 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 950 | |
10560082 TT |
951 | if (err == 0) |
952 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 953 | if (err < 0) |
10560082 | 954 | return ERR_PTR(err); |
2ed88685 TT |
955 | |
956 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
957 | if (unlikely(!bh)) |
958 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
959 | if (map.m_flags & EXT4_MAP_NEW) { |
960 | J_ASSERT(create != 0); | |
961 | J_ASSERT(handle != NULL); | |
ac27a0ec | 962 | |
2ed88685 TT |
963 | /* |
964 | * Now that we do not always journal data, we should | |
965 | * keep in mind whether this should always journal the | |
966 | * new buffer as metadata. For now, regular file | |
967 | * writes use ext4_get_block instead, so it's not a | |
968 | * problem. | |
969 | */ | |
970 | lock_buffer(bh); | |
971 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
972 | err = ext4_journal_get_create_access(handle, bh); |
973 | if (unlikely(err)) { | |
974 | unlock_buffer(bh); | |
975 | goto errout; | |
976 | } | |
977 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
978 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
979 | set_buffer_uptodate(bh); | |
ac27a0ec | 980 | } |
2ed88685 TT |
981 | unlock_buffer(bh); |
982 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
983 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
984 | if (unlikely(err)) |
985 | goto errout; | |
986 | } else | |
2ed88685 | 987 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 988 | return bh; |
10560082 TT |
989 | errout: |
990 | brelse(bh); | |
991 | return ERR_PTR(err); | |
ac27a0ec DK |
992 | } |
993 | ||
617ba13b | 994 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 995 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 996 | { |
af5bc92d | 997 | struct buffer_head *bh; |
ac27a0ec | 998 | |
c5e298ae | 999 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 1000 | if (IS_ERR(bh)) |
ac27a0ec | 1001 | return bh; |
1c215028 | 1002 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 1003 | return bh; |
dfec8a14 | 1004 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
1005 | wait_on_buffer(bh); |
1006 | if (buffer_uptodate(bh)) | |
1007 | return bh; | |
1008 | put_bh(bh); | |
1c215028 | 1009 | return ERR_PTR(-EIO); |
ac27a0ec DK |
1010 | } |
1011 | ||
f19d5870 TM |
1012 | int ext4_walk_page_buffers(handle_t *handle, |
1013 | struct buffer_head *head, | |
1014 | unsigned from, | |
1015 | unsigned to, | |
1016 | int *partial, | |
1017 | int (*fn)(handle_t *handle, | |
1018 | struct buffer_head *bh)) | |
ac27a0ec DK |
1019 | { |
1020 | struct buffer_head *bh; | |
1021 | unsigned block_start, block_end; | |
1022 | unsigned blocksize = head->b_size; | |
1023 | int err, ret = 0; | |
1024 | struct buffer_head *next; | |
1025 | ||
af5bc92d TT |
1026 | for (bh = head, block_start = 0; |
1027 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1028 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1029 | next = bh->b_this_page; |
1030 | block_end = block_start + blocksize; | |
1031 | if (block_end <= from || block_start >= to) { | |
1032 | if (partial && !buffer_uptodate(bh)) | |
1033 | *partial = 1; | |
1034 | continue; | |
1035 | } | |
1036 | err = (*fn)(handle, bh); | |
1037 | if (!ret) | |
1038 | ret = err; | |
1039 | } | |
1040 | return ret; | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * To preserve ordering, it is essential that the hole instantiation and | |
1045 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1046 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1047 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1048 | * prepare_write() is the right place. |
1049 | * | |
36ade451 JK |
1050 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1051 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1052 | * whole page. So we won't block on the journal in that case, which is good, | |
1053 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1054 | * |
617ba13b | 1055 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1056 | * quota file writes. If we were to commit the transaction while thus |
1057 | * reentered, there can be a deadlock - we would be holding a quota | |
1058 | * lock, and the commit would never complete if another thread had a | |
1059 | * transaction open and was blocking on the quota lock - a ranking | |
1060 | * violation. | |
1061 | * | |
dab291af | 1062 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1063 | * will _not_ run commit under these circumstances because handle->h_ref |
1064 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1065 | * write. | |
1066 | */ | |
f19d5870 TM |
1067 | int do_journal_get_write_access(handle_t *handle, |
1068 | struct buffer_head *bh) | |
ac27a0ec | 1069 | { |
56d35a4c JK |
1070 | int dirty = buffer_dirty(bh); |
1071 | int ret; | |
1072 | ||
ac27a0ec DK |
1073 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1074 | return 0; | |
56d35a4c | 1075 | /* |
ebdec241 | 1076 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1077 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1078 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1079 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1080 | * the bit before releasing a page lock and thus writeback cannot |
1081 | * ever write the buffer. | |
1082 | */ | |
1083 | if (dirty) | |
1084 | clear_buffer_dirty(bh); | |
5d601255 | 1085 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1086 | ret = ext4_journal_get_write_access(handle, bh); |
1087 | if (!ret && dirty) | |
1088 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1089 | return ret; | |
ac27a0ec DK |
1090 | } |
1091 | ||
2058f83a MH |
1092 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1093 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
1094 | get_block_t *get_block) | |
1095 | { | |
09cbfeaf | 1096 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1097 | unsigned to = from + len; |
1098 | struct inode *inode = page->mapping->host; | |
1099 | unsigned block_start, block_end; | |
1100 | sector_t block; | |
1101 | int err = 0; | |
1102 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1103 | unsigned bbits; | |
1104 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
1105 | bool decrypt = false; | |
1106 | ||
1107 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1108 | BUG_ON(from > PAGE_SIZE); |
1109 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1110 | BUG_ON(from > to); |
1111 | ||
1112 | if (!page_has_buffers(page)) | |
1113 | create_empty_buffers(page, blocksize, 0); | |
1114 | head = page_buffers(page); | |
1115 | bbits = ilog2(blocksize); | |
09cbfeaf | 1116 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1117 | |
1118 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1119 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1120 | block_end = block_start + blocksize; | |
1121 | if (block_end <= from || block_start >= to) { | |
1122 | if (PageUptodate(page)) { | |
1123 | if (!buffer_uptodate(bh)) | |
1124 | set_buffer_uptodate(bh); | |
1125 | } | |
1126 | continue; | |
1127 | } | |
1128 | if (buffer_new(bh)) | |
1129 | clear_buffer_new(bh); | |
1130 | if (!buffer_mapped(bh)) { | |
1131 | WARN_ON(bh->b_size != blocksize); | |
1132 | err = get_block(inode, block, bh, 1); | |
1133 | if (err) | |
1134 | break; | |
1135 | if (buffer_new(bh)) { | |
e64855c6 | 1136 | clean_bdev_bh_alias(bh); |
2058f83a MH |
1137 | if (PageUptodate(page)) { |
1138 | clear_buffer_new(bh); | |
1139 | set_buffer_uptodate(bh); | |
1140 | mark_buffer_dirty(bh); | |
1141 | continue; | |
1142 | } | |
1143 | if (block_end > to || block_start < from) | |
1144 | zero_user_segments(page, to, block_end, | |
1145 | block_start, from); | |
1146 | continue; | |
1147 | } | |
1148 | } | |
1149 | if (PageUptodate(page)) { | |
1150 | if (!buffer_uptodate(bh)) | |
1151 | set_buffer_uptodate(bh); | |
1152 | continue; | |
1153 | } | |
1154 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1155 | !buffer_unwritten(bh) && | |
1156 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1157 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
2058f83a MH |
1158 | *wait_bh++ = bh; |
1159 | decrypt = ext4_encrypted_inode(inode) && | |
1160 | S_ISREG(inode->i_mode); | |
1161 | } | |
1162 | } | |
1163 | /* | |
1164 | * If we issued read requests, let them complete. | |
1165 | */ | |
1166 | while (wait_bh > wait) { | |
1167 | wait_on_buffer(*--wait_bh); | |
1168 | if (!buffer_uptodate(*wait_bh)) | |
1169 | err = -EIO; | |
1170 | } | |
1171 | if (unlikely(err)) | |
1172 | page_zero_new_buffers(page, from, to); | |
1173 | else if (decrypt) | |
7821d4dd | 1174 | err = fscrypt_decrypt_page(page->mapping->host, page, |
9c4bb8a3 | 1175 | PAGE_SIZE, 0, page->index); |
2058f83a MH |
1176 | return err; |
1177 | } | |
1178 | #endif | |
1179 | ||
bfc1af65 | 1180 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1181 | loff_t pos, unsigned len, unsigned flags, |
1182 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1183 | { |
af5bc92d | 1184 | struct inode *inode = mapping->host; |
1938a150 | 1185 | int ret, needed_blocks; |
ac27a0ec DK |
1186 | handle_t *handle; |
1187 | int retries = 0; | |
af5bc92d | 1188 | struct page *page; |
de9a55b8 | 1189 | pgoff_t index; |
af5bc92d | 1190 | unsigned from, to; |
bfc1af65 | 1191 | |
0db1ff22 TT |
1192 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1193 | return -EIO; | |
1194 | ||
9bffad1e | 1195 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1196 | /* |
1197 | * Reserve one block more for addition to orphan list in case | |
1198 | * we allocate blocks but write fails for some reason | |
1199 | */ | |
1200 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1201 | index = pos >> PAGE_SHIFT; |
1202 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1203 | to = from + len; |
ac27a0ec | 1204 | |
f19d5870 TM |
1205 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1206 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1207 | flags, pagep); | |
1208 | if (ret < 0) | |
47564bfb TT |
1209 | return ret; |
1210 | if (ret == 1) | |
1211 | return 0; | |
f19d5870 TM |
1212 | } |
1213 | ||
47564bfb TT |
1214 | /* |
1215 | * grab_cache_page_write_begin() can take a long time if the | |
1216 | * system is thrashing due to memory pressure, or if the page | |
1217 | * is being written back. So grab it first before we start | |
1218 | * the transaction handle. This also allows us to allocate | |
1219 | * the page (if needed) without using GFP_NOFS. | |
1220 | */ | |
1221 | retry_grab: | |
1222 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1223 | if (!page) | |
1224 | return -ENOMEM; | |
1225 | unlock_page(page); | |
1226 | ||
1227 | retry_journal: | |
9924a92a | 1228 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1229 | if (IS_ERR(handle)) { |
09cbfeaf | 1230 | put_page(page); |
47564bfb | 1231 | return PTR_ERR(handle); |
7479d2b9 | 1232 | } |
ac27a0ec | 1233 | |
47564bfb TT |
1234 | lock_page(page); |
1235 | if (page->mapping != mapping) { | |
1236 | /* The page got truncated from under us */ | |
1237 | unlock_page(page); | |
09cbfeaf | 1238 | put_page(page); |
cf108bca | 1239 | ext4_journal_stop(handle); |
47564bfb | 1240 | goto retry_grab; |
cf108bca | 1241 | } |
7afe5aa5 DM |
1242 | /* In case writeback began while the page was unlocked */ |
1243 | wait_for_stable_page(page); | |
cf108bca | 1244 | |
2058f83a MH |
1245 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1246 | if (ext4_should_dioread_nolock(inode)) | |
1247 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1248 | ext4_get_block_unwritten); |
2058f83a MH |
1249 | else |
1250 | ret = ext4_block_write_begin(page, pos, len, | |
1251 | ext4_get_block); | |
1252 | #else | |
744692dc | 1253 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1254 | ret = __block_write_begin(page, pos, len, |
1255 | ext4_get_block_unwritten); | |
744692dc | 1256 | else |
6e1db88d | 1257 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1258 | #endif |
bfc1af65 | 1259 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1260 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1261 | from, to, NULL, | |
1262 | do_journal_get_write_access); | |
ac27a0ec | 1263 | } |
bfc1af65 NP |
1264 | |
1265 | if (ret) { | |
af5bc92d | 1266 | unlock_page(page); |
ae4d5372 | 1267 | /* |
6e1db88d | 1268 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1269 | * outside i_size. Trim these off again. Don't need |
1270 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1271 | * |
1272 | * Add inode to orphan list in case we crash before | |
1273 | * truncate finishes | |
ae4d5372 | 1274 | */ |
ffacfa7a | 1275 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1276 | ext4_orphan_add(handle, inode); |
1277 | ||
1278 | ext4_journal_stop(handle); | |
1279 | if (pos + len > inode->i_size) { | |
b9a4207d | 1280 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1281 | /* |
ffacfa7a | 1282 | * If truncate failed early the inode might |
1938a150 AK |
1283 | * still be on the orphan list; we need to |
1284 | * make sure the inode is removed from the | |
1285 | * orphan list in that case. | |
1286 | */ | |
1287 | if (inode->i_nlink) | |
1288 | ext4_orphan_del(NULL, inode); | |
1289 | } | |
bfc1af65 | 1290 | |
47564bfb TT |
1291 | if (ret == -ENOSPC && |
1292 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1293 | goto retry_journal; | |
09cbfeaf | 1294 | put_page(page); |
47564bfb TT |
1295 | return ret; |
1296 | } | |
1297 | *pagep = page; | |
ac27a0ec DK |
1298 | return ret; |
1299 | } | |
1300 | ||
bfc1af65 NP |
1301 | /* For write_end() in data=journal mode */ |
1302 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1303 | { |
13fca323 | 1304 | int ret; |
ac27a0ec DK |
1305 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1306 | return 0; | |
1307 | set_buffer_uptodate(bh); | |
13fca323 TT |
1308 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1309 | clear_buffer_meta(bh); | |
1310 | clear_buffer_prio(bh); | |
1311 | return ret; | |
ac27a0ec DK |
1312 | } |
1313 | ||
eed4333f ZL |
1314 | /* |
1315 | * We need to pick up the new inode size which generic_commit_write gave us | |
1316 | * `file' can be NULL - eg, when called from page_symlink(). | |
1317 | * | |
1318 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1319 | * buffers are managed internally. | |
1320 | */ | |
1321 | static int ext4_write_end(struct file *file, | |
1322 | struct address_space *mapping, | |
1323 | loff_t pos, unsigned len, unsigned copied, | |
1324 | struct page *page, void *fsdata) | |
f8514083 | 1325 | { |
f8514083 | 1326 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1327 | struct inode *inode = mapping->host; |
0572639f | 1328 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1329 | int ret = 0, ret2; |
1330 | int i_size_changed = 0; | |
1331 | ||
1332 | trace_ext4_write_end(inode, pos, len, copied); | |
42c832de TT |
1333 | if (ext4_has_inline_data(inode)) { |
1334 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1335 | copied, page); | |
eb5efbcb TT |
1336 | if (ret < 0) { |
1337 | unlock_page(page); | |
1338 | put_page(page); | |
42c832de | 1339 | goto errout; |
eb5efbcb | 1340 | } |
42c832de TT |
1341 | copied = ret; |
1342 | } else | |
f19d5870 TM |
1343 | copied = block_write_end(file, mapping, pos, |
1344 | len, copied, page, fsdata); | |
f8514083 | 1345 | /* |
4631dbf6 | 1346 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1347 | * page writeout could otherwise come in and zero beyond i_size. |
1348 | */ | |
4631dbf6 | 1349 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 | 1350 | unlock_page(page); |
09cbfeaf | 1351 | put_page(page); |
f8514083 | 1352 | |
0572639f XW |
1353 | if (old_size < pos) |
1354 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1355 | /* |
1356 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1357 | * makes the holding time of page lock longer. Second, it forces lock | |
1358 | * ordering of page lock and transaction start for journaling | |
1359 | * filesystems. | |
1360 | */ | |
1361 | if (i_size_changed) | |
1362 | ext4_mark_inode_dirty(handle, inode); | |
1363 | ||
ffacfa7a | 1364 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1365 | /* if we have allocated more blocks and copied |
1366 | * less. We will have blocks allocated outside | |
1367 | * inode->i_size. So truncate them | |
1368 | */ | |
1369 | ext4_orphan_add(handle, inode); | |
74d553aa | 1370 | errout: |
617ba13b | 1371 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1372 | if (!ret) |
1373 | ret = ret2; | |
bfc1af65 | 1374 | |
f8514083 | 1375 | if (pos + len > inode->i_size) { |
b9a4207d | 1376 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1377 | /* |
ffacfa7a | 1378 | * If truncate failed early the inode might still be |
f8514083 AK |
1379 | * on the orphan list; we need to make sure the inode |
1380 | * is removed from the orphan list in that case. | |
1381 | */ | |
1382 | if (inode->i_nlink) | |
1383 | ext4_orphan_del(NULL, inode); | |
1384 | } | |
1385 | ||
bfc1af65 | 1386 | return ret ? ret : copied; |
ac27a0ec DK |
1387 | } |
1388 | ||
b90197b6 TT |
1389 | /* |
1390 | * This is a private version of page_zero_new_buffers() which doesn't | |
1391 | * set the buffer to be dirty, since in data=journalled mode we need | |
1392 | * to call ext4_handle_dirty_metadata() instead. | |
1393 | */ | |
3b136499 JK |
1394 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
1395 | struct page *page, | |
1396 | unsigned from, unsigned to) | |
b90197b6 TT |
1397 | { |
1398 | unsigned int block_start = 0, block_end; | |
1399 | struct buffer_head *head, *bh; | |
1400 | ||
1401 | bh = head = page_buffers(page); | |
1402 | do { | |
1403 | block_end = block_start + bh->b_size; | |
1404 | if (buffer_new(bh)) { | |
1405 | if (block_end > from && block_start < to) { | |
1406 | if (!PageUptodate(page)) { | |
1407 | unsigned start, size; | |
1408 | ||
1409 | start = max(from, block_start); | |
1410 | size = min(to, block_end) - start; | |
1411 | ||
1412 | zero_user(page, start, size); | |
3b136499 | 1413 | write_end_fn(handle, bh); |
b90197b6 TT |
1414 | } |
1415 | clear_buffer_new(bh); | |
1416 | } | |
1417 | } | |
1418 | block_start = block_end; | |
1419 | bh = bh->b_this_page; | |
1420 | } while (bh != head); | |
1421 | } | |
1422 | ||
bfc1af65 | 1423 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1424 | struct address_space *mapping, |
1425 | loff_t pos, unsigned len, unsigned copied, | |
1426 | struct page *page, void *fsdata) | |
ac27a0ec | 1427 | { |
617ba13b | 1428 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1429 | struct inode *inode = mapping->host; |
0572639f | 1430 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1431 | int ret = 0, ret2; |
1432 | int partial = 0; | |
bfc1af65 | 1433 | unsigned from, to; |
4631dbf6 | 1434 | int size_changed = 0; |
ac27a0ec | 1435 | |
9bffad1e | 1436 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1437 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1438 | to = from + len; |
1439 | ||
441c8508 CW |
1440 | BUG_ON(!ext4_handle_valid(handle)); |
1441 | ||
eb5efbcb TT |
1442 | if (ext4_has_inline_data(inode)) { |
1443 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1444 | copied, page); | |
1445 | if (ret < 0) { | |
1446 | unlock_page(page); | |
1447 | put_page(page); | |
1448 | goto errout; | |
1449 | } | |
1450 | copied = ret; | |
1451 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 JK |
1452 | copied = 0; |
1453 | ext4_journalled_zero_new_buffers(handle, page, from, to); | |
1454 | } else { | |
1455 | if (unlikely(copied < len)) | |
1456 | ext4_journalled_zero_new_buffers(handle, page, | |
1457 | from + copied, to); | |
3fdcfb66 | 1458 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
3b136499 JK |
1459 | from + copied, &partial, |
1460 | write_end_fn); | |
3fdcfb66 TM |
1461 | if (!partial) |
1462 | SetPageUptodate(page); | |
1463 | } | |
4631dbf6 | 1464 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1465 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1466 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1467 | unlock_page(page); |
09cbfeaf | 1468 | put_page(page); |
4631dbf6 | 1469 | |
0572639f XW |
1470 | if (old_size < pos) |
1471 | pagecache_isize_extended(inode, old_size, pos); | |
1472 | ||
4631dbf6 | 1473 | if (size_changed) { |
617ba13b | 1474 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1475 | if (!ret) |
1476 | ret = ret2; | |
1477 | } | |
bfc1af65 | 1478 | |
ffacfa7a | 1479 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1480 | /* if we have allocated more blocks and copied |
1481 | * less. We will have blocks allocated outside | |
1482 | * inode->i_size. So truncate them | |
1483 | */ | |
1484 | ext4_orphan_add(handle, inode); | |
1485 | ||
eb5efbcb | 1486 | errout: |
617ba13b | 1487 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1488 | if (!ret) |
1489 | ret = ret2; | |
f8514083 | 1490 | if (pos + len > inode->i_size) { |
b9a4207d | 1491 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1492 | /* |
ffacfa7a | 1493 | * If truncate failed early the inode might still be |
f8514083 AK |
1494 | * on the orphan list; we need to make sure the inode |
1495 | * is removed from the orphan list in that case. | |
1496 | */ | |
1497 | if (inode->i_nlink) | |
1498 | ext4_orphan_del(NULL, inode); | |
1499 | } | |
bfc1af65 NP |
1500 | |
1501 | return ret ? ret : copied; | |
ac27a0ec | 1502 | } |
d2a17637 | 1503 | |
9d0be502 | 1504 | /* |
c27e43a1 | 1505 | * Reserve space for a single cluster |
9d0be502 | 1506 | */ |
c27e43a1 | 1507 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1508 | { |
60e58e0f | 1509 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1510 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1511 | int ret; |
03179fe9 TT |
1512 | |
1513 | /* | |
1514 | * We will charge metadata quota at writeout time; this saves | |
1515 | * us from metadata over-estimation, though we may go over by | |
1516 | * a small amount in the end. Here we just reserve for data. | |
1517 | */ | |
1518 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1519 | if (ret) | |
1520 | return ret; | |
d2a17637 | 1521 | |
0637c6f4 | 1522 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1523 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1524 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1525 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1526 | return -ENOSPC; |
1527 | } | |
9d0be502 | 1528 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1529 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1530 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1531 | |
d2a17637 MC |
1532 | return 0; /* success */ |
1533 | } | |
1534 | ||
12219aea | 1535 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1536 | { |
1537 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1538 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1539 | |
cd213226 MC |
1540 | if (!to_free) |
1541 | return; /* Nothing to release, exit */ | |
1542 | ||
d2a17637 | 1543 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1544 | |
5a58ec87 | 1545 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1546 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1547 | /* |
0637c6f4 TT |
1548 | * if there aren't enough reserved blocks, then the |
1549 | * counter is messed up somewhere. Since this | |
1550 | * function is called from invalidate page, it's | |
1551 | * harmless to return without any action. | |
cd213226 | 1552 | */ |
8de5c325 | 1553 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1554 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1555 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1556 | ei->i_reserved_data_blocks); |
1557 | WARN_ON(1); | |
1558 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1559 | } |
0637c6f4 | 1560 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1561 | |
72b8ab9d | 1562 | /* update fs dirty data blocks counter */ |
57042651 | 1563 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1564 | |
d2a17637 | 1565 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1566 | |
7b415bf6 | 1567 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1568 | } |
1569 | ||
1570 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1571 | unsigned int offset, |
1572 | unsigned int length) | |
d2a17637 | 1573 | { |
9705acd6 | 1574 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1575 | struct buffer_head *head, *bh; |
1576 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1577 | struct inode *inode = page->mapping->host; |
1578 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1579 | unsigned int stop = offset + length; |
7b415bf6 | 1580 | int num_clusters; |
51865fda | 1581 | ext4_fsblk_t lblk; |
d2a17637 | 1582 | |
09cbfeaf | 1583 | BUG_ON(stop > PAGE_SIZE || stop < length); |
ca99fdd2 | 1584 | |
d2a17637 MC |
1585 | head = page_buffers(page); |
1586 | bh = head; | |
1587 | do { | |
1588 | unsigned int next_off = curr_off + bh->b_size; | |
1589 | ||
ca99fdd2 LC |
1590 | if (next_off > stop) |
1591 | break; | |
1592 | ||
d2a17637 MC |
1593 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1594 | to_release++; | |
9705acd6 | 1595 | contiguous_blks++; |
d2a17637 | 1596 | clear_buffer_delay(bh); |
9705acd6 LC |
1597 | } else if (contiguous_blks) { |
1598 | lblk = page->index << | |
09cbfeaf | 1599 | (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1600 | lblk += (curr_off >> inode->i_blkbits) - |
1601 | contiguous_blks; | |
1602 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1603 | contiguous_blks = 0; | |
d2a17637 MC |
1604 | } |
1605 | curr_off = next_off; | |
1606 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1607 | |
9705acd6 | 1608 | if (contiguous_blks) { |
09cbfeaf | 1609 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1610 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1611 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1612 | } |
1613 | ||
7b415bf6 AK |
1614 | /* If we have released all the blocks belonging to a cluster, then we |
1615 | * need to release the reserved space for that cluster. */ | |
1616 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1617 | while (num_clusters > 0) { | |
09cbfeaf | 1618 | lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
7b415bf6 AK |
1619 | ((num_clusters - 1) << sbi->s_cluster_bits); |
1620 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1621 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1622 | ext4_da_release_space(inode, 1); |
1623 | ||
1624 | num_clusters--; | |
1625 | } | |
d2a17637 | 1626 | } |
ac27a0ec | 1627 | |
64769240 AT |
1628 | /* |
1629 | * Delayed allocation stuff | |
1630 | */ | |
1631 | ||
4e7ea81d JK |
1632 | struct mpage_da_data { |
1633 | struct inode *inode; | |
1634 | struct writeback_control *wbc; | |
6b523df4 | 1635 | |
4e7ea81d JK |
1636 | pgoff_t first_page; /* The first page to write */ |
1637 | pgoff_t next_page; /* Current page to examine */ | |
1638 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1639 | /* |
4e7ea81d JK |
1640 | * Extent to map - this can be after first_page because that can be |
1641 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1642 | * is delalloc or unwritten. | |
791b7f08 | 1643 | */ |
4e7ea81d JK |
1644 | struct ext4_map_blocks map; |
1645 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1646 | unsigned int do_map:1; |
4e7ea81d | 1647 | }; |
64769240 | 1648 | |
4e7ea81d JK |
1649 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1650 | bool invalidate) | |
c4a0c46e AK |
1651 | { |
1652 | int nr_pages, i; | |
1653 | pgoff_t index, end; | |
1654 | struct pagevec pvec; | |
1655 | struct inode *inode = mpd->inode; | |
1656 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1657 | |
1658 | /* This is necessary when next_page == 0. */ | |
1659 | if (mpd->first_page >= mpd->next_page) | |
1660 | return; | |
c4a0c46e | 1661 | |
c7f5938a CW |
1662 | index = mpd->first_page; |
1663 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1664 | if (invalidate) { |
1665 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1666 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1667 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1668 | ext4_es_remove_extent(inode, start, last - start + 1); |
1669 | } | |
51865fda | 1670 | |
66bea92c | 1671 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1672 | while (index <= end) { |
1673 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1674 | if (nr_pages == 0) | |
1675 | break; | |
1676 | for (i = 0; i < nr_pages; i++) { | |
1677 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1678 | if (page->index > end) |
c4a0c46e | 1679 | break; |
c4a0c46e AK |
1680 | BUG_ON(!PageLocked(page)); |
1681 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1682 | if (invalidate) { |
4e800c03 | 1683 | if (page_mapped(page)) |
1684 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1685 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1686 | ClearPageUptodate(page); |
1687 | } | |
c4a0c46e AK |
1688 | unlock_page(page); |
1689 | } | |
9b1d0998 JK |
1690 | index = pvec.pages[nr_pages - 1]->index + 1; |
1691 | pagevec_release(&pvec); | |
c4a0c46e | 1692 | } |
c4a0c46e AK |
1693 | } |
1694 | ||
df22291f AK |
1695 | static void ext4_print_free_blocks(struct inode *inode) |
1696 | { | |
1697 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1698 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1699 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1700 | |
1701 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1702 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1703 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1704 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1705 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1706 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1707 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1708 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1709 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1710 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1711 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1712 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1713 | ei->i_reserved_data_blocks); |
df22291f AK |
1714 | return; |
1715 | } | |
1716 | ||
c364b22c | 1717 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1718 | { |
c364b22c | 1719 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1720 | } |
1721 | ||
5356f261 AK |
1722 | /* |
1723 | * This function is grabs code from the very beginning of | |
1724 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1725 | * time. This function looks up the requested blocks and sets the | |
1726 | * buffer delay bit under the protection of i_data_sem. | |
1727 | */ | |
1728 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1729 | struct ext4_map_blocks *map, | |
1730 | struct buffer_head *bh) | |
1731 | { | |
d100eef2 | 1732 | struct extent_status es; |
5356f261 AK |
1733 | int retval; |
1734 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1735 | #ifdef ES_AGGRESSIVE_TEST |
1736 | struct ext4_map_blocks orig_map; | |
1737 | ||
1738 | memcpy(&orig_map, map, sizeof(*map)); | |
1739 | #endif | |
5356f261 AK |
1740 | |
1741 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1742 | invalid_block = ~0; | |
1743 | ||
1744 | map->m_flags = 0; | |
1745 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1746 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1747 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1748 | |
1749 | /* Lookup extent status tree firstly */ | |
1750 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1751 | if (ext4_es_is_hole(&es)) { |
1752 | retval = 0; | |
c8b459f4 | 1753 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1754 | goto add_delayed; |
1755 | } | |
1756 | ||
1757 | /* | |
1758 | * Delayed extent could be allocated by fallocate. | |
1759 | * So we need to check it. | |
1760 | */ | |
1761 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1762 | map_bh(bh, inode->i_sb, invalid_block); | |
1763 | set_buffer_new(bh); | |
1764 | set_buffer_delay(bh); | |
1765 | return 0; | |
1766 | } | |
1767 | ||
1768 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1769 | retval = es.es_len - (iblock - es.es_lblk); | |
1770 | if (retval > map->m_len) | |
1771 | retval = map->m_len; | |
1772 | map->m_len = retval; | |
1773 | if (ext4_es_is_written(&es)) | |
1774 | map->m_flags |= EXT4_MAP_MAPPED; | |
1775 | else if (ext4_es_is_unwritten(&es)) | |
1776 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1777 | else | |
1778 | BUG_ON(1); | |
1779 | ||
921f266b DM |
1780 | #ifdef ES_AGGRESSIVE_TEST |
1781 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1782 | #endif | |
d100eef2 ZL |
1783 | return retval; |
1784 | } | |
1785 | ||
5356f261 AK |
1786 | /* |
1787 | * Try to see if we can get the block without requesting a new | |
1788 | * file system block. | |
1789 | */ | |
c8b459f4 | 1790 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1791 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1792 | retval = 0; |
cbd7584e | 1793 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1794 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1795 | else |
2f8e0a7c | 1796 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1797 | |
d100eef2 | 1798 | add_delayed: |
5356f261 | 1799 | if (retval == 0) { |
f7fec032 | 1800 | int ret; |
5356f261 AK |
1801 | /* |
1802 | * XXX: __block_prepare_write() unmaps passed block, | |
1803 | * is it OK? | |
1804 | */ | |
386ad67c LC |
1805 | /* |
1806 | * If the block was allocated from previously allocated cluster, | |
1807 | * then we don't need to reserve it again. However we still need | |
1808 | * to reserve metadata for every block we're going to write. | |
1809 | */ | |
c27e43a1 | 1810 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1811 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1812 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1813 | if (ret) { |
5356f261 | 1814 | /* not enough space to reserve */ |
f7fec032 | 1815 | retval = ret; |
5356f261 | 1816 | goto out_unlock; |
f7fec032 | 1817 | } |
5356f261 AK |
1818 | } |
1819 | ||
f7fec032 ZL |
1820 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1821 | ~0, EXTENT_STATUS_DELAYED); | |
1822 | if (ret) { | |
1823 | retval = ret; | |
51865fda | 1824 | goto out_unlock; |
f7fec032 | 1825 | } |
51865fda | 1826 | |
5356f261 AK |
1827 | map_bh(bh, inode->i_sb, invalid_block); |
1828 | set_buffer_new(bh); | |
1829 | set_buffer_delay(bh); | |
f7fec032 ZL |
1830 | } else if (retval > 0) { |
1831 | int ret; | |
3be78c73 | 1832 | unsigned int status; |
f7fec032 | 1833 | |
44fb851d ZL |
1834 | if (unlikely(retval != map->m_len)) { |
1835 | ext4_warning(inode->i_sb, | |
1836 | "ES len assertion failed for inode " | |
1837 | "%lu: retval %d != map->m_len %d", | |
1838 | inode->i_ino, retval, map->m_len); | |
1839 | WARN_ON(1); | |
921f266b | 1840 | } |
921f266b | 1841 | |
f7fec032 ZL |
1842 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1843 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1844 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1845 | map->m_pblk, status); | |
1846 | if (ret != 0) | |
1847 | retval = ret; | |
5356f261 AK |
1848 | } |
1849 | ||
1850 | out_unlock: | |
1851 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1852 | ||
1853 | return retval; | |
1854 | } | |
1855 | ||
64769240 | 1856 | /* |
d91bd2c1 | 1857 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1858 | * ext4_da_write_begin(). It will either return mapped block or |
1859 | * reserve space for a single block. | |
29fa89d0 AK |
1860 | * |
1861 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1862 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1863 | * | |
1864 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1865 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1866 | * initialized properly. | |
64769240 | 1867 | */ |
9c3569b5 TM |
1868 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1869 | struct buffer_head *bh, int create) | |
64769240 | 1870 | { |
2ed88685 | 1871 | struct ext4_map_blocks map; |
64769240 AT |
1872 | int ret = 0; |
1873 | ||
1874 | BUG_ON(create == 0); | |
2ed88685 TT |
1875 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1876 | ||
1877 | map.m_lblk = iblock; | |
1878 | map.m_len = 1; | |
64769240 AT |
1879 | |
1880 | /* | |
1881 | * first, we need to know whether the block is allocated already | |
1882 | * preallocated blocks are unmapped but should treated | |
1883 | * the same as allocated blocks. | |
1884 | */ | |
5356f261 AK |
1885 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1886 | if (ret <= 0) | |
2ed88685 | 1887 | return ret; |
64769240 | 1888 | |
2ed88685 | 1889 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1890 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1891 | |
1892 | if (buffer_unwritten(bh)) { | |
1893 | /* A delayed write to unwritten bh should be marked | |
1894 | * new and mapped. Mapped ensures that we don't do | |
1895 | * get_block multiple times when we write to the same | |
1896 | * offset and new ensures that we do proper zero out | |
1897 | * for partial write. | |
1898 | */ | |
1899 | set_buffer_new(bh); | |
c8205636 | 1900 | set_buffer_mapped(bh); |
2ed88685 TT |
1901 | } |
1902 | return 0; | |
64769240 | 1903 | } |
61628a3f | 1904 | |
62e086be AK |
1905 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1906 | { | |
1907 | get_bh(bh); | |
1908 | return 0; | |
1909 | } | |
1910 | ||
1911 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1912 | { | |
1913 | put_bh(bh); | |
1914 | return 0; | |
1915 | } | |
1916 | ||
1917 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1918 | unsigned int len) |
1919 | { | |
1920 | struct address_space *mapping = page->mapping; | |
1921 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1922 | struct buffer_head *page_bufs = NULL; |
62e086be | 1923 | handle_t *handle = NULL; |
3fdcfb66 TM |
1924 | int ret = 0, err = 0; |
1925 | int inline_data = ext4_has_inline_data(inode); | |
1926 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1927 | |
cb20d518 | 1928 | ClearPageChecked(page); |
3fdcfb66 TM |
1929 | |
1930 | if (inline_data) { | |
1931 | BUG_ON(page->index != 0); | |
1932 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1933 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1934 | if (inode_bh == NULL) | |
1935 | goto out; | |
1936 | } else { | |
1937 | page_bufs = page_buffers(page); | |
1938 | if (!page_bufs) { | |
1939 | BUG(); | |
1940 | goto out; | |
1941 | } | |
1942 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1943 | NULL, bget_one); | |
1944 | } | |
bdf96838 TT |
1945 | /* |
1946 | * We need to release the page lock before we start the | |
1947 | * journal, so grab a reference so the page won't disappear | |
1948 | * out from under us. | |
1949 | */ | |
1950 | get_page(page); | |
62e086be AK |
1951 | unlock_page(page); |
1952 | ||
9924a92a TT |
1953 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1954 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1955 | if (IS_ERR(handle)) { |
1956 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1957 | put_page(page); |
1958 | goto out_no_pagelock; | |
62e086be | 1959 | } |
441c8508 CW |
1960 | BUG_ON(!ext4_handle_valid(handle)); |
1961 | ||
bdf96838 TT |
1962 | lock_page(page); |
1963 | put_page(page); | |
1964 | if (page->mapping != mapping) { | |
1965 | /* The page got truncated from under us */ | |
1966 | ext4_journal_stop(handle); | |
1967 | ret = 0; | |
1968 | goto out; | |
1969 | } | |
1970 | ||
3fdcfb66 | 1971 | if (inline_data) { |
5d601255 | 1972 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1973 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1974 | |
3fdcfb66 TM |
1975 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1976 | ||
1977 | } else { | |
1978 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1979 | do_journal_get_write_access); | |
1980 | ||
1981 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1982 | write_end_fn); | |
1983 | } | |
62e086be AK |
1984 | if (ret == 0) |
1985 | ret = err; | |
2d859db3 | 1986 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1987 | err = ext4_journal_stop(handle); |
1988 | if (!ret) | |
1989 | ret = err; | |
1990 | ||
3fdcfb66 | 1991 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1992 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1993 | NULL, bput_one); |
19f5fb7a | 1994 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1995 | out: |
bdf96838 TT |
1996 | unlock_page(page); |
1997 | out_no_pagelock: | |
3fdcfb66 | 1998 | brelse(inode_bh); |
62e086be AK |
1999 | return ret; |
2000 | } | |
2001 | ||
61628a3f | 2002 | /* |
43ce1d23 AK |
2003 | * Note that we don't need to start a transaction unless we're journaling data |
2004 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2005 | * need to file the inode to the transaction's list in ordered mode because if | |
2006 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2007 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2008 | * transaction the data will hit the disk. In case we are journaling data, we |
2009 | * cannot start transaction directly because transaction start ranks above page | |
2010 | * lock so we have to do some magic. | |
2011 | * | |
b920c755 | 2012 | * This function can get called via... |
20970ba6 | 2013 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 2014 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 2015 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2016 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2017 | * |
2018 | * We don't do any block allocation in this function. If we have page with | |
2019 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2020 | * is important for mmaped based write. So if we do with blocksize 1K | |
2021 | * truncate(f, 1024); | |
2022 | * a = mmap(f, 0, 4096); | |
2023 | * a[0] = 'a'; | |
2024 | * truncate(f, 4096); | |
2025 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2026 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2027 | * do_wp_page). So writepage should write the first block. If we modify |
2028 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2029 | * page_mkwrite callback will do the block allocation and mark the | |
2030 | * buffer_heads mapped. | |
2031 | * | |
2032 | * We redirty the page if we have any buffer_heads that is either delay or | |
2033 | * unwritten in the page. | |
2034 | * | |
2035 | * We can get recursively called as show below. | |
2036 | * | |
2037 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2038 | * ext4_writepage() | |
2039 | * | |
2040 | * But since we don't do any block allocation we should not deadlock. | |
2041 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2042 | */ |
43ce1d23 | 2043 | static int ext4_writepage(struct page *page, |
62e086be | 2044 | struct writeback_control *wbc) |
64769240 | 2045 | { |
f8bec370 | 2046 | int ret = 0; |
61628a3f | 2047 | loff_t size; |
498e5f24 | 2048 | unsigned int len; |
744692dc | 2049 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2050 | struct inode *inode = page->mapping->host; |
36ade451 | 2051 | struct ext4_io_submit io_submit; |
1c8349a1 | 2052 | bool keep_towrite = false; |
61628a3f | 2053 | |
0db1ff22 TT |
2054 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
2055 | ext4_invalidatepage(page, 0, PAGE_SIZE); | |
2056 | unlock_page(page); | |
2057 | return -EIO; | |
2058 | } | |
2059 | ||
a9c667f8 | 2060 | trace_ext4_writepage(page); |
f0e6c985 | 2061 | size = i_size_read(inode); |
09cbfeaf KS |
2062 | if (page->index == size >> PAGE_SHIFT) |
2063 | len = size & ~PAGE_MASK; | |
f0e6c985 | 2064 | else |
09cbfeaf | 2065 | len = PAGE_SIZE; |
64769240 | 2066 | |
a42afc5f | 2067 | page_bufs = page_buffers(page); |
a42afc5f | 2068 | /* |
fe386132 JK |
2069 | * We cannot do block allocation or other extent handling in this |
2070 | * function. If there are buffers needing that, we have to redirty | |
2071 | * the page. But we may reach here when we do a journal commit via | |
2072 | * journal_submit_inode_data_buffers() and in that case we must write | |
2073 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2074 | * |
2075 | * Also, if there is only one buffer per page (the fs block | |
2076 | * size == the page size), if one buffer needs block | |
2077 | * allocation or needs to modify the extent tree to clear the | |
2078 | * unwritten flag, we know that the page can't be written at | |
2079 | * all, so we might as well refuse the write immediately. | |
2080 | * Unfortunately if the block size != page size, we can't as | |
2081 | * easily detect this case using ext4_walk_page_buffers(), but | |
2082 | * for the extremely common case, this is an optimization that | |
2083 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2084 | */ |
f19d5870 TM |
2085 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2086 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2087 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2088 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2089 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2090 | /* |
2091 | * For memory cleaning there's no point in writing only | |
2092 | * some buffers. So just bail out. Warn if we came here | |
2093 | * from direct reclaim. | |
2094 | */ | |
2095 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2096 | == PF_MEMALLOC); | |
f0e6c985 AK |
2097 | unlock_page(page); |
2098 | return 0; | |
2099 | } | |
1c8349a1 | 2100 | keep_towrite = true; |
a42afc5f | 2101 | } |
64769240 | 2102 | |
cb20d518 | 2103 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2104 | /* |
2105 | * It's mmapped pagecache. Add buffers and journal it. There | |
2106 | * doesn't seem much point in redirtying the page here. | |
2107 | */ | |
3f0ca309 | 2108 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2109 | |
97a851ed JK |
2110 | ext4_io_submit_init(&io_submit, wbc); |
2111 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2112 | if (!io_submit.io_end) { | |
2113 | redirty_page_for_writepage(wbc, page); | |
2114 | unlock_page(page); | |
2115 | return -ENOMEM; | |
2116 | } | |
1c8349a1 | 2117 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2118 | ext4_io_submit(&io_submit); |
97a851ed JK |
2119 | /* Drop io_end reference we got from init */ |
2120 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2121 | return ret; |
2122 | } | |
2123 | ||
5f1132b2 JK |
2124 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2125 | { | |
2126 | int len; | |
2127 | loff_t size = i_size_read(mpd->inode); | |
2128 | int err; | |
2129 | ||
2130 | BUG_ON(page->index != mpd->first_page); | |
09cbfeaf KS |
2131 | if (page->index == size >> PAGE_SHIFT) |
2132 | len = size & ~PAGE_MASK; | |
5f1132b2 | 2133 | else |
09cbfeaf | 2134 | len = PAGE_SIZE; |
5f1132b2 | 2135 | clear_page_dirty_for_io(page); |
1c8349a1 | 2136 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2137 | if (!err) |
2138 | mpd->wbc->nr_to_write--; | |
2139 | mpd->first_page++; | |
2140 | ||
2141 | return err; | |
2142 | } | |
2143 | ||
4e7ea81d JK |
2144 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2145 | ||
61628a3f | 2146 | /* |
fffb2739 JK |
2147 | * mballoc gives us at most this number of blocks... |
2148 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2149 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2150 | */ |
fffb2739 | 2151 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2152 | |
4e7ea81d JK |
2153 | /* |
2154 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2155 | * | |
2156 | * @mpd - extent of blocks | |
2157 | * @lblk - logical number of the block in the file | |
09930042 | 2158 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2159 | * |
09930042 JK |
2160 | * The function is used to collect contig. blocks in the same state. If the |
2161 | * buffer doesn't require mapping for writeback and we haven't started the | |
2162 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2163 | * caller can write the buffer right away. Otherwise the function returns true | |
2164 | * if the block has been added to the extent, false if the block couldn't be | |
2165 | * added. | |
4e7ea81d | 2166 | */ |
09930042 JK |
2167 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2168 | struct buffer_head *bh) | |
4e7ea81d JK |
2169 | { |
2170 | struct ext4_map_blocks *map = &mpd->map; | |
2171 | ||
09930042 JK |
2172 | /* Buffer that doesn't need mapping for writeback? */ |
2173 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2174 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2175 | /* So far no extent to map => we write the buffer right away */ | |
2176 | if (map->m_len == 0) | |
2177 | return true; | |
2178 | return false; | |
2179 | } | |
4e7ea81d JK |
2180 | |
2181 | /* First block in the extent? */ | |
2182 | if (map->m_len == 0) { | |
dddbd6ac JK |
2183 | /* We cannot map unless handle is started... */ |
2184 | if (!mpd->do_map) | |
2185 | return false; | |
4e7ea81d JK |
2186 | map->m_lblk = lblk; |
2187 | map->m_len = 1; | |
09930042 JK |
2188 | map->m_flags = bh->b_state & BH_FLAGS; |
2189 | return true; | |
4e7ea81d JK |
2190 | } |
2191 | ||
09930042 JK |
2192 | /* Don't go larger than mballoc is willing to allocate */ |
2193 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2194 | return false; | |
2195 | ||
4e7ea81d JK |
2196 | /* Can we merge the block to our big extent? */ |
2197 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2198 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2199 | map->m_len++; |
09930042 | 2200 | return true; |
4e7ea81d | 2201 | } |
09930042 | 2202 | return false; |
4e7ea81d JK |
2203 | } |
2204 | ||
5f1132b2 JK |
2205 | /* |
2206 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2207 | * | |
2208 | * @mpd - extent of blocks for mapping | |
2209 | * @head - the first buffer in the page | |
2210 | * @bh - buffer we should start processing from | |
2211 | * @lblk - logical number of the block in the file corresponding to @bh | |
2212 | * | |
2213 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2214 | * the page for IO if all buffers in this page were mapped and there's no | |
2215 | * accumulated extent of buffers to map or add buffers in the page to the | |
2216 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2217 | * by processing the next page, 0 if it should stop adding buffers to the | |
2218 | * extent to map because we cannot extend it anymore. It can also return value | |
2219 | * < 0 in case of error during IO submission. | |
2220 | */ | |
2221 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2222 | struct buffer_head *head, | |
2223 | struct buffer_head *bh, | |
2224 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2225 | { |
2226 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2227 | int err; |
93407472 | 2228 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2229 | >> inode->i_blkbits; |
2230 | ||
2231 | do { | |
2232 | BUG_ON(buffer_locked(bh)); | |
2233 | ||
09930042 | 2234 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2235 | /* Found extent to map? */ |
2236 | if (mpd->map.m_len) | |
5f1132b2 | 2237 | return 0; |
dddbd6ac JK |
2238 | /* Buffer needs mapping and handle is not started? */ |
2239 | if (!mpd->do_map) | |
2240 | return 0; | |
09930042 | 2241 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2242 | break; |
4e7ea81d | 2243 | } |
4e7ea81d | 2244 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2245 | /* So far everything mapped? Submit the page for IO. */ |
2246 | if (mpd->map.m_len == 0) { | |
2247 | err = mpage_submit_page(mpd, head->b_page); | |
2248 | if (err < 0) | |
2249 | return err; | |
2250 | } | |
2251 | return lblk < blocks; | |
4e7ea81d JK |
2252 | } |
2253 | ||
2254 | /* | |
2255 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2256 | * submit fully mapped pages for IO | |
2257 | * | |
2258 | * @mpd - description of extent to map, on return next extent to map | |
2259 | * | |
2260 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2261 | * to be already locked) and update buffer state according to new extent state. | |
2262 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2263 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2264 | * and do extent conversion after IO is finished. If the last page is not fully |
2265 | * mapped, we update @map to the next extent in the last page that needs | |
2266 | * mapping. Otherwise we submit the page for IO. | |
2267 | */ | |
2268 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2269 | { | |
2270 | struct pagevec pvec; | |
2271 | int nr_pages, i; | |
2272 | struct inode *inode = mpd->inode; | |
2273 | struct buffer_head *head, *bh; | |
09cbfeaf | 2274 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2275 | pgoff_t start, end; |
2276 | ext4_lblk_t lblk; | |
2277 | sector_t pblock; | |
2278 | int err; | |
2279 | ||
2280 | start = mpd->map.m_lblk >> bpp_bits; | |
2281 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2282 | lblk = start << bpp_bits; | |
2283 | pblock = mpd->map.m_pblk; | |
2284 | ||
2285 | pagevec_init(&pvec, 0); | |
2286 | while (start <= end) { | |
2287 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2288 | PAGEVEC_SIZE); | |
2289 | if (nr_pages == 0) | |
2290 | break; | |
2291 | for (i = 0; i < nr_pages; i++) { | |
2292 | struct page *page = pvec.pages[i]; | |
2293 | ||
2294 | if (page->index > end) | |
2295 | break; | |
70261f56 | 2296 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2297 | BUG_ON(page->index != start); |
2298 | bh = head = page_buffers(page); | |
2299 | do { | |
2300 | if (lblk < mpd->map.m_lblk) | |
2301 | continue; | |
2302 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2303 | /* | |
2304 | * Buffer after end of mapped extent. | |
2305 | * Find next buffer in the page to map. | |
2306 | */ | |
2307 | mpd->map.m_len = 0; | |
2308 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2309 | /* |
2310 | * FIXME: If dioread_nolock supports | |
2311 | * blocksize < pagesize, we need to make | |
2312 | * sure we add size mapped so far to | |
2313 | * io_end->size as the following call | |
2314 | * can submit the page for IO. | |
2315 | */ | |
2316 | err = mpage_process_page_bufs(mpd, head, | |
2317 | bh, lblk); | |
4e7ea81d | 2318 | pagevec_release(&pvec); |
5f1132b2 JK |
2319 | if (err > 0) |
2320 | err = 0; | |
2321 | return err; | |
4e7ea81d JK |
2322 | } |
2323 | if (buffer_delay(bh)) { | |
2324 | clear_buffer_delay(bh); | |
2325 | bh->b_blocknr = pblock++; | |
2326 | } | |
4e7ea81d | 2327 | clear_buffer_unwritten(bh); |
5f1132b2 | 2328 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2329 | |
2330 | /* | |
2331 | * FIXME: This is going to break if dioread_nolock | |
2332 | * supports blocksize < pagesize as we will try to | |
2333 | * convert potentially unmapped parts of inode. | |
2334 | */ | |
09cbfeaf | 2335 | mpd->io_submit.io_end->size += PAGE_SIZE; |
4e7ea81d JK |
2336 | /* Page fully mapped - let IO run! */ |
2337 | err = mpage_submit_page(mpd, page); | |
2338 | if (err < 0) { | |
2339 | pagevec_release(&pvec); | |
2340 | return err; | |
2341 | } | |
2342 | start++; | |
2343 | } | |
2344 | pagevec_release(&pvec); | |
2345 | } | |
2346 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2347 | mpd->map.m_len = 0; | |
2348 | mpd->map.m_flags = 0; | |
2349 | return 0; | |
2350 | } | |
2351 | ||
2352 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2353 | { | |
2354 | struct inode *inode = mpd->inode; | |
2355 | struct ext4_map_blocks *map = &mpd->map; | |
2356 | int get_blocks_flags; | |
090f32ee | 2357 | int err, dioread_nolock; |
4e7ea81d JK |
2358 | |
2359 | trace_ext4_da_write_pages_extent(inode, map); | |
2360 | /* | |
2361 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2362 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2363 | * where we have written into one or more preallocated blocks). It is |
2364 | * possible that we're going to need more metadata blocks than | |
2365 | * previously reserved. However we must not fail because we're in | |
2366 | * writeback and there is nothing we can do about it so it might result | |
2367 | * in data loss. So use reserved blocks to allocate metadata if | |
2368 | * possible. | |
2369 | * | |
754cfed6 TT |
2370 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2371 | * the blocks in question are delalloc blocks. This indicates | |
2372 | * that the blocks and quotas has already been checked when | |
2373 | * the data was copied into the page cache. | |
4e7ea81d JK |
2374 | */ |
2375 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2376 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2377 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2378 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2379 | if (dioread_nolock) | |
4e7ea81d JK |
2380 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2381 | if (map->m_flags & (1 << BH_Delay)) | |
2382 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2383 | ||
2384 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2385 | if (err < 0) | |
2386 | return err; | |
090f32ee | 2387 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2388 | if (!mpd->io_submit.io_end->handle && |
2389 | ext4_handle_valid(handle)) { | |
2390 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2391 | handle->h_rsv_handle = NULL; | |
2392 | } | |
3613d228 | 2393 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2394 | } |
4e7ea81d JK |
2395 | |
2396 | BUG_ON(map->m_len == 0); | |
2397 | if (map->m_flags & EXT4_MAP_NEW) { | |
64e1c57f JK |
2398 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
2399 | map->m_len); | |
4e7ea81d JK |
2400 | } |
2401 | return 0; | |
2402 | } | |
2403 | ||
2404 | /* | |
2405 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2406 | * mpd->len and submit pages underlying it for IO | |
2407 | * | |
2408 | * @handle - handle for journal operations | |
2409 | * @mpd - extent to map | |
7534e854 JK |
2410 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2411 | * is no hope of writing the data. The caller should discard | |
2412 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2413 | * |
2414 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2415 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2416 | * them to initialized or split the described range from larger unwritten | |
2417 | * extent. Note that we need not map all the described range since allocation | |
2418 | * can return less blocks or the range is covered by more unwritten extents. We | |
2419 | * cannot map more because we are limited by reserved transaction credits. On | |
2420 | * the other hand we always make sure that the last touched page is fully | |
2421 | * mapped so that it can be written out (and thus forward progress is | |
2422 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2423 | */ | |
2424 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2425 | struct mpage_da_data *mpd, |
2426 | bool *give_up_on_write) | |
4e7ea81d JK |
2427 | { |
2428 | struct inode *inode = mpd->inode; | |
2429 | struct ext4_map_blocks *map = &mpd->map; | |
2430 | int err; | |
2431 | loff_t disksize; | |
6603120e | 2432 | int progress = 0; |
4e7ea81d JK |
2433 | |
2434 | mpd->io_submit.io_end->offset = | |
2435 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2436 | do { |
4e7ea81d JK |
2437 | err = mpage_map_one_extent(handle, mpd); |
2438 | if (err < 0) { | |
2439 | struct super_block *sb = inode->i_sb; | |
2440 | ||
0db1ff22 TT |
2441 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
2442 | EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) | |
cb530541 | 2443 | goto invalidate_dirty_pages; |
4e7ea81d | 2444 | /* |
cb530541 TT |
2445 | * Let the uper layers retry transient errors. |
2446 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2447 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2448 | */ |
cb530541 | 2449 | if ((err == -ENOMEM) || |
6603120e DM |
2450 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2451 | if (progress) | |
2452 | goto update_disksize; | |
cb530541 | 2453 | return err; |
6603120e | 2454 | } |
cb530541 TT |
2455 | ext4_msg(sb, KERN_CRIT, |
2456 | "Delayed block allocation failed for " | |
2457 | "inode %lu at logical offset %llu with" | |
2458 | " max blocks %u with error %d", | |
2459 | inode->i_ino, | |
2460 | (unsigned long long)map->m_lblk, | |
2461 | (unsigned)map->m_len, -err); | |
2462 | ext4_msg(sb, KERN_CRIT, | |
2463 | "This should not happen!! Data will " | |
2464 | "be lost\n"); | |
2465 | if (err == -ENOSPC) | |
2466 | ext4_print_free_blocks(inode); | |
2467 | invalidate_dirty_pages: | |
2468 | *give_up_on_write = true; | |
4e7ea81d JK |
2469 | return err; |
2470 | } | |
6603120e | 2471 | progress = 1; |
4e7ea81d JK |
2472 | /* |
2473 | * Update buffer state, submit mapped pages, and get us new | |
2474 | * extent to map | |
2475 | */ | |
2476 | err = mpage_map_and_submit_buffers(mpd); | |
2477 | if (err < 0) | |
6603120e | 2478 | goto update_disksize; |
27d7c4ed | 2479 | } while (map->m_len); |
4e7ea81d | 2480 | |
6603120e | 2481 | update_disksize: |
622cad13 TT |
2482 | /* |
2483 | * Update on-disk size after IO is submitted. Races with | |
2484 | * truncate are avoided by checking i_size under i_data_sem. | |
2485 | */ | |
09cbfeaf | 2486 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
4e7ea81d JK |
2487 | if (disksize > EXT4_I(inode)->i_disksize) { |
2488 | int err2; | |
622cad13 TT |
2489 | loff_t i_size; |
2490 | ||
2491 | down_write(&EXT4_I(inode)->i_data_sem); | |
2492 | i_size = i_size_read(inode); | |
2493 | if (disksize > i_size) | |
2494 | disksize = i_size; | |
2495 | if (disksize > EXT4_I(inode)->i_disksize) | |
2496 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2497 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2498 | err2 = ext4_mark_inode_dirty(handle, inode); |
4e7ea81d JK |
2499 | if (err2) |
2500 | ext4_error(inode->i_sb, | |
2501 | "Failed to mark inode %lu dirty", | |
2502 | inode->i_ino); | |
2503 | if (!err) | |
2504 | err = err2; | |
2505 | } | |
2506 | return err; | |
2507 | } | |
2508 | ||
fffb2739 JK |
2509 | /* |
2510 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2511 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2512 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2513 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2514 | * bpp - 1 blocks in bpp different extents. | |
2515 | */ | |
525f4ed8 MC |
2516 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2517 | { | |
fffb2739 | 2518 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2519 | |
fffb2739 JK |
2520 | return ext4_meta_trans_blocks(inode, |
2521 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2522 | } |
61628a3f | 2523 | |
8e48dcfb | 2524 | /* |
4e7ea81d JK |
2525 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2526 | * and underlying extent to map | |
2527 | * | |
2528 | * @mpd - where to look for pages | |
2529 | * | |
2530 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2531 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2532 | * extent of buffers underlying these pages that needs mapping (formed by | |
2533 | * either delayed or unwritten buffers). We also lock the pages containing | |
2534 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2535 | * mpd->lblk with length mpd->len blocks). | |
2536 | * | |
2537 | * Note that this function can attach bios to one io_end structure which are | |
2538 | * neither logically nor physically contiguous. Although it may seem as an | |
2539 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2540 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2541 | */ |
4e7ea81d | 2542 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2543 | { |
4e7ea81d JK |
2544 | struct address_space *mapping = mpd->inode->i_mapping; |
2545 | struct pagevec pvec; | |
2546 | unsigned int nr_pages; | |
aeac589a | 2547 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2548 | pgoff_t index = mpd->first_page; |
2549 | pgoff_t end = mpd->last_page; | |
2550 | int tag; | |
2551 | int i, err = 0; | |
2552 | int blkbits = mpd->inode->i_blkbits; | |
2553 | ext4_lblk_t lblk; | |
2554 | struct buffer_head *head; | |
8e48dcfb | 2555 | |
4e7ea81d | 2556 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2557 | tag = PAGECACHE_TAG_TOWRITE; |
2558 | else | |
2559 | tag = PAGECACHE_TAG_DIRTY; | |
2560 | ||
4e7ea81d JK |
2561 | pagevec_init(&pvec, 0); |
2562 | mpd->map.m_len = 0; | |
2563 | mpd->next_page = index; | |
4f01b02c | 2564 | while (index <= end) { |
5b41d924 | 2565 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2566 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2567 | if (nr_pages == 0) | |
4e7ea81d | 2568 | goto out; |
8e48dcfb TT |
2569 | |
2570 | for (i = 0; i < nr_pages; i++) { | |
2571 | struct page *page = pvec.pages[i]; | |
2572 | ||
2573 | /* | |
2574 | * At this point, the page may be truncated or | |
2575 | * invalidated (changing page->mapping to NULL), or | |
2576 | * even swizzled back from swapper_space to tmpfs file | |
2577 | * mapping. However, page->index will not change | |
2578 | * because we have a reference on the page. | |
2579 | */ | |
4f01b02c TT |
2580 | if (page->index > end) |
2581 | goto out; | |
8e48dcfb | 2582 | |
aeac589a ML |
2583 | /* |
2584 | * Accumulated enough dirty pages? This doesn't apply | |
2585 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2586 | * keep going because someone may be concurrently | |
2587 | * dirtying pages, and we might have synced a lot of | |
2588 | * newly appeared dirty pages, but have not synced all | |
2589 | * of the old dirty pages. | |
2590 | */ | |
2591 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2592 | goto out; | |
2593 | ||
4e7ea81d JK |
2594 | /* If we can't merge this page, we are done. */ |
2595 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2596 | goto out; | |
78aaced3 | 2597 | |
8e48dcfb | 2598 | lock_page(page); |
8e48dcfb | 2599 | /* |
4e7ea81d JK |
2600 | * If the page is no longer dirty, or its mapping no |
2601 | * longer corresponds to inode we are writing (which | |
2602 | * means it has been truncated or invalidated), or the | |
2603 | * page is already under writeback and we are not doing | |
2604 | * a data integrity writeback, skip the page | |
8e48dcfb | 2605 | */ |
4f01b02c TT |
2606 | if (!PageDirty(page) || |
2607 | (PageWriteback(page) && | |
4e7ea81d | 2608 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2609 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2610 | unlock_page(page); |
2611 | continue; | |
2612 | } | |
2613 | ||
7cb1a535 | 2614 | wait_on_page_writeback(page); |
8e48dcfb | 2615 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2616 | |
4e7ea81d | 2617 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2618 | mpd->first_page = page->index; |
8eb9e5ce | 2619 | mpd->next_page = page->index + 1; |
f8bec370 | 2620 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2621 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2622 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2623 | head = page_buffers(page); |
5f1132b2 JK |
2624 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2625 | if (err <= 0) | |
4e7ea81d | 2626 | goto out; |
5f1132b2 | 2627 | err = 0; |
aeac589a | 2628 | left--; |
8e48dcfb TT |
2629 | } |
2630 | pagevec_release(&pvec); | |
2631 | cond_resched(); | |
2632 | } | |
4f01b02c | 2633 | return 0; |
8eb9e5ce TT |
2634 | out: |
2635 | pagevec_release(&pvec); | |
4e7ea81d | 2636 | return err; |
8e48dcfb TT |
2637 | } |
2638 | ||
20970ba6 TT |
2639 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2640 | void *data) | |
2641 | { | |
2642 | struct address_space *mapping = data; | |
2643 | int ret = ext4_writepage(page, wbc); | |
2644 | mapping_set_error(mapping, ret); | |
2645 | return ret; | |
2646 | } | |
2647 | ||
2648 | static int ext4_writepages(struct address_space *mapping, | |
2649 | struct writeback_control *wbc) | |
64769240 | 2650 | { |
4e7ea81d JK |
2651 | pgoff_t writeback_index = 0; |
2652 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2653 | int range_whole = 0; |
4e7ea81d | 2654 | int cycled = 1; |
61628a3f | 2655 | handle_t *handle = NULL; |
df22291f | 2656 | struct mpage_da_data mpd; |
5e745b04 | 2657 | struct inode *inode = mapping->host; |
6b523df4 | 2658 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2659 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2660 | bool done; |
1bce63d1 | 2661 | struct blk_plug plug; |
cb530541 | 2662 | bool give_up_on_write = false; |
61628a3f | 2663 | |
0db1ff22 TT |
2664 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2665 | return -EIO; | |
2666 | ||
c8585c6f | 2667 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
20970ba6 | 2668 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2669 | |
c8585c6f DJ |
2670 | if (dax_mapping(mapping)) { |
2671 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, | |
2672 | wbc); | |
2673 | goto out_writepages; | |
2674 | } | |
7f6d5b52 | 2675 | |
61628a3f MC |
2676 | /* |
2677 | * No pages to write? This is mainly a kludge to avoid starting | |
2678 | * a transaction for special inodes like journal inode on last iput() | |
2679 | * because that could violate lock ordering on umount | |
2680 | */ | |
a1d6cc56 | 2681 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2682 | goto out_writepages; |
2a21e37e | 2683 | |
20970ba6 TT |
2684 | if (ext4_should_journal_data(inode)) { |
2685 | struct blk_plug plug; | |
20970ba6 TT |
2686 | |
2687 | blk_start_plug(&plug); | |
2688 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2689 | blk_finish_plug(&plug); | |
bbf023c7 | 2690 | goto out_writepages; |
20970ba6 TT |
2691 | } |
2692 | ||
2a21e37e TT |
2693 | /* |
2694 | * If the filesystem has aborted, it is read-only, so return | |
2695 | * right away instead of dumping stack traces later on that | |
2696 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2697 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2698 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2699 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2700 | * *never* be called, so if that ever happens, we would want |
2701 | * the stack trace. | |
2702 | */ | |
0db1ff22 TT |
2703 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
2704 | sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { | |
bbf023c7 ML |
2705 | ret = -EROFS; |
2706 | goto out_writepages; | |
2707 | } | |
2a21e37e | 2708 | |
6b523df4 JK |
2709 | if (ext4_should_dioread_nolock(inode)) { |
2710 | /* | |
70261f56 | 2711 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2712 | * the page and we may dirty the inode. |
2713 | */ | |
09cbfeaf | 2714 | rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits); |
6b523df4 JK |
2715 | } |
2716 | ||
4e7ea81d JK |
2717 | /* |
2718 | * If we have inline data and arrive here, it means that | |
2719 | * we will soon create the block for the 1st page, so | |
2720 | * we'd better clear the inline data here. | |
2721 | */ | |
2722 | if (ext4_has_inline_data(inode)) { | |
2723 | /* Just inode will be modified... */ | |
2724 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2725 | if (IS_ERR(handle)) { | |
2726 | ret = PTR_ERR(handle); | |
2727 | goto out_writepages; | |
2728 | } | |
2729 | BUG_ON(ext4_test_inode_state(inode, | |
2730 | EXT4_STATE_MAY_INLINE_DATA)); | |
2731 | ext4_destroy_inline_data(handle, inode); | |
2732 | ext4_journal_stop(handle); | |
2733 | } | |
2734 | ||
22208ded AK |
2735 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2736 | range_whole = 1; | |
61628a3f | 2737 | |
2acf2c26 | 2738 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2739 | writeback_index = mapping->writeback_index; |
2740 | if (writeback_index) | |
2acf2c26 | 2741 | cycled = 0; |
4e7ea81d JK |
2742 | mpd.first_page = writeback_index; |
2743 | mpd.last_page = -1; | |
5b41d924 | 2744 | } else { |
09cbfeaf KS |
2745 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2746 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2747 | } |
a1d6cc56 | 2748 | |
4e7ea81d JK |
2749 | mpd.inode = inode; |
2750 | mpd.wbc = wbc; | |
2751 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2752 | retry: |
6e6938b6 | 2753 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2754 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2755 | done = false; | |
1bce63d1 | 2756 | blk_start_plug(&plug); |
dddbd6ac JK |
2757 | |
2758 | /* | |
2759 | * First writeback pages that don't need mapping - we can avoid | |
2760 | * starting a transaction unnecessarily and also avoid being blocked | |
2761 | * in the block layer on device congestion while having transaction | |
2762 | * started. | |
2763 | */ | |
2764 | mpd.do_map = 0; | |
2765 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2766 | if (!mpd.io_submit.io_end) { | |
2767 | ret = -ENOMEM; | |
2768 | goto unplug; | |
2769 | } | |
2770 | ret = mpage_prepare_extent_to_map(&mpd); | |
2771 | /* Submit prepared bio */ | |
2772 | ext4_io_submit(&mpd.io_submit); | |
2773 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2774 | mpd.io_submit.io_end = NULL; | |
2775 | /* Unlock pages we didn't use */ | |
2776 | mpage_release_unused_pages(&mpd, false); | |
2777 | if (ret < 0) | |
2778 | goto unplug; | |
2779 | ||
4e7ea81d JK |
2780 | while (!done && mpd.first_page <= mpd.last_page) { |
2781 | /* For each extent of pages we use new io_end */ | |
2782 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2783 | if (!mpd.io_submit.io_end) { | |
2784 | ret = -ENOMEM; | |
2785 | break; | |
2786 | } | |
a1d6cc56 AK |
2787 | |
2788 | /* | |
4e7ea81d JK |
2789 | * We have two constraints: We find one extent to map and we |
2790 | * must always write out whole page (makes a difference when | |
2791 | * blocksize < pagesize) so that we don't block on IO when we | |
2792 | * try to write out the rest of the page. Journalled mode is | |
2793 | * not supported by delalloc. | |
a1d6cc56 AK |
2794 | */ |
2795 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2796 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2797 | |
4e7ea81d | 2798 | /* start a new transaction */ |
6b523df4 JK |
2799 | handle = ext4_journal_start_with_reserve(inode, |
2800 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2801 | if (IS_ERR(handle)) { |
2802 | ret = PTR_ERR(handle); | |
1693918e | 2803 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2804 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2805 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2806 | /* Release allocated io_end */ |
2807 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2808 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2809 | break; |
61628a3f | 2810 | } |
dddbd6ac | 2811 | mpd.do_map = 1; |
f63e6005 | 2812 | |
4e7ea81d JK |
2813 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2814 | ret = mpage_prepare_extent_to_map(&mpd); | |
2815 | if (!ret) { | |
2816 | if (mpd.map.m_len) | |
cb530541 TT |
2817 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2818 | &give_up_on_write); | |
4e7ea81d JK |
2819 | else { |
2820 | /* | |
2821 | * We scanned the whole range (or exhausted | |
2822 | * nr_to_write), submitted what was mapped and | |
2823 | * didn't find anything needing mapping. We are | |
2824 | * done. | |
2825 | */ | |
2826 | done = true; | |
2827 | } | |
f63e6005 | 2828 | } |
646caa9c JK |
2829 | /* |
2830 | * Caution: If the handle is synchronous, | |
2831 | * ext4_journal_stop() can wait for transaction commit | |
2832 | * to finish which may depend on writeback of pages to | |
2833 | * complete or on page lock to be released. In that | |
2834 | * case, we have to wait until after after we have | |
2835 | * submitted all the IO, released page locks we hold, | |
2836 | * and dropped io_end reference (for extent conversion | |
2837 | * to be able to complete) before stopping the handle. | |
2838 | */ | |
2839 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2840 | ext4_journal_stop(handle); | |
2841 | handle = NULL; | |
dddbd6ac | 2842 | mpd.do_map = 0; |
646caa9c | 2843 | } |
4e7ea81d JK |
2844 | /* Submit prepared bio */ |
2845 | ext4_io_submit(&mpd.io_submit); | |
2846 | /* Unlock pages we didn't use */ | |
cb530541 | 2847 | mpage_release_unused_pages(&mpd, give_up_on_write); |
646caa9c JK |
2848 | /* |
2849 | * Drop our io_end reference we got from init. We have | |
2850 | * to be careful and use deferred io_end finishing if | |
2851 | * we are still holding the transaction as we can | |
2852 | * release the last reference to io_end which may end | |
2853 | * up doing unwritten extent conversion. | |
2854 | */ | |
2855 | if (handle) { | |
2856 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2857 | ext4_journal_stop(handle); | |
2858 | } else | |
2859 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2860 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2861 | |
2862 | if (ret == -ENOSPC && sbi->s_journal) { | |
2863 | /* | |
2864 | * Commit the transaction which would | |
22208ded AK |
2865 | * free blocks released in the transaction |
2866 | * and try again | |
2867 | */ | |
df22291f | 2868 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2869 | ret = 0; |
4e7ea81d JK |
2870 | continue; |
2871 | } | |
2872 | /* Fatal error - ENOMEM, EIO... */ | |
2873 | if (ret) | |
61628a3f | 2874 | break; |
a1d6cc56 | 2875 | } |
dddbd6ac | 2876 | unplug: |
1bce63d1 | 2877 | blk_finish_plug(&plug); |
9c12a831 | 2878 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2879 | cycled = 1; |
4e7ea81d JK |
2880 | mpd.last_page = writeback_index - 1; |
2881 | mpd.first_page = 0; | |
2acf2c26 AK |
2882 | goto retry; |
2883 | } | |
22208ded AK |
2884 | |
2885 | /* Update index */ | |
22208ded AK |
2886 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2887 | /* | |
4e7ea81d | 2888 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2889 | * mode will write it back later |
2890 | */ | |
4e7ea81d | 2891 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2892 | |
61628a3f | 2893 | out_writepages: |
20970ba6 TT |
2894 | trace_ext4_writepages_result(inode, wbc, ret, |
2895 | nr_to_write - wbc->nr_to_write); | |
c8585c6f | 2896 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
61628a3f | 2897 | return ret; |
64769240 AT |
2898 | } |
2899 | ||
79f0be8d AK |
2900 | static int ext4_nonda_switch(struct super_block *sb) |
2901 | { | |
5c1ff336 | 2902 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2903 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2904 | ||
2905 | /* | |
2906 | * switch to non delalloc mode if we are running low | |
2907 | * on free block. The free block accounting via percpu | |
179f7ebf | 2908 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2909 | * accumulated on each CPU without updating global counters |
2910 | * Delalloc need an accurate free block accounting. So switch | |
2911 | * to non delalloc when we are near to error range. | |
2912 | */ | |
5c1ff336 EW |
2913 | free_clusters = |
2914 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2915 | dirty_clusters = | |
2916 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2917 | /* |
2918 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2919 | */ | |
5c1ff336 | 2920 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2921 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2922 | |
5c1ff336 EW |
2923 | if (2 * free_clusters < 3 * dirty_clusters || |
2924 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2925 | /* |
c8afb446 ES |
2926 | * free block count is less than 150% of dirty blocks |
2927 | * or free blocks is less than watermark | |
79f0be8d AK |
2928 | */ |
2929 | return 1; | |
2930 | } | |
2931 | return 0; | |
2932 | } | |
2933 | ||
0ff8947f ES |
2934 | /* We always reserve for an inode update; the superblock could be there too */ |
2935 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2936 | { | |
e2b911c5 | 2937 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2938 | return 1; |
2939 | ||
2940 | if (pos + len <= 0x7fffffffULL) | |
2941 | return 1; | |
2942 | ||
2943 | /* We might need to update the superblock to set LARGE_FILE */ | |
2944 | return 2; | |
2945 | } | |
2946 | ||
64769240 | 2947 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2948 | loff_t pos, unsigned len, unsigned flags, |
2949 | struct page **pagep, void **fsdata) | |
64769240 | 2950 | { |
72b8ab9d | 2951 | int ret, retries = 0; |
64769240 AT |
2952 | struct page *page; |
2953 | pgoff_t index; | |
64769240 AT |
2954 | struct inode *inode = mapping->host; |
2955 | handle_t *handle; | |
2956 | ||
0db1ff22 TT |
2957 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2958 | return -EIO; | |
2959 | ||
09cbfeaf | 2960 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2961 | |
4db0d88e TT |
2962 | if (ext4_nonda_switch(inode->i_sb) || |
2963 | S_ISLNK(inode->i_mode)) { | |
79f0be8d AK |
2964 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2965 | return ext4_write_begin(file, mapping, pos, | |
2966 | len, flags, pagep, fsdata); | |
2967 | } | |
2968 | *fsdata = (void *)0; | |
9bffad1e | 2969 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2970 | |
2971 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2972 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2973 | pos, len, flags, | |
2974 | pagep, fsdata); | |
2975 | if (ret < 0) | |
47564bfb TT |
2976 | return ret; |
2977 | if (ret == 1) | |
2978 | return 0; | |
9c3569b5 TM |
2979 | } |
2980 | ||
47564bfb TT |
2981 | /* |
2982 | * grab_cache_page_write_begin() can take a long time if the | |
2983 | * system is thrashing due to memory pressure, or if the page | |
2984 | * is being written back. So grab it first before we start | |
2985 | * the transaction handle. This also allows us to allocate | |
2986 | * the page (if needed) without using GFP_NOFS. | |
2987 | */ | |
2988 | retry_grab: | |
2989 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2990 | if (!page) | |
2991 | return -ENOMEM; | |
2992 | unlock_page(page); | |
2993 | ||
64769240 AT |
2994 | /* |
2995 | * With delayed allocation, we don't log the i_disksize update | |
2996 | * if there is delayed block allocation. But we still need | |
2997 | * to journalling the i_disksize update if writes to the end | |
2998 | * of file which has an already mapped buffer. | |
2999 | */ | |
47564bfb | 3000 | retry_journal: |
0ff8947f ES |
3001 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
3002 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 3003 | if (IS_ERR(handle)) { |
09cbfeaf | 3004 | put_page(page); |
47564bfb | 3005 | return PTR_ERR(handle); |
64769240 AT |
3006 | } |
3007 | ||
47564bfb TT |
3008 | lock_page(page); |
3009 | if (page->mapping != mapping) { | |
3010 | /* The page got truncated from under us */ | |
3011 | unlock_page(page); | |
09cbfeaf | 3012 | put_page(page); |
d5a0d4f7 | 3013 | ext4_journal_stop(handle); |
47564bfb | 3014 | goto retry_grab; |
d5a0d4f7 | 3015 | } |
47564bfb | 3016 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3017 | wait_for_stable_page(page); |
64769240 | 3018 | |
2058f83a MH |
3019 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3020 | ret = ext4_block_write_begin(page, pos, len, | |
3021 | ext4_da_get_block_prep); | |
3022 | #else | |
6e1db88d | 3023 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3024 | #endif |
64769240 AT |
3025 | if (ret < 0) { |
3026 | unlock_page(page); | |
3027 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3028 | /* |
3029 | * block_write_begin may have instantiated a few blocks | |
3030 | * outside i_size. Trim these off again. Don't need | |
3031 | * i_size_read because we hold i_mutex. | |
3032 | */ | |
3033 | if (pos + len > inode->i_size) | |
b9a4207d | 3034 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3035 | |
3036 | if (ret == -ENOSPC && | |
3037 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3038 | goto retry_journal; | |
3039 | ||
09cbfeaf | 3040 | put_page(page); |
47564bfb | 3041 | return ret; |
64769240 AT |
3042 | } |
3043 | ||
47564bfb | 3044 | *pagep = page; |
64769240 AT |
3045 | return ret; |
3046 | } | |
3047 | ||
632eaeab MC |
3048 | /* |
3049 | * Check if we should update i_disksize | |
3050 | * when write to the end of file but not require block allocation | |
3051 | */ | |
3052 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3053 | unsigned long offset) |
632eaeab MC |
3054 | { |
3055 | struct buffer_head *bh; | |
3056 | struct inode *inode = page->mapping->host; | |
3057 | unsigned int idx; | |
3058 | int i; | |
3059 | ||
3060 | bh = page_buffers(page); | |
3061 | idx = offset >> inode->i_blkbits; | |
3062 | ||
af5bc92d | 3063 | for (i = 0; i < idx; i++) |
632eaeab MC |
3064 | bh = bh->b_this_page; |
3065 | ||
29fa89d0 | 3066 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3067 | return 0; |
3068 | return 1; | |
3069 | } | |
3070 | ||
64769240 | 3071 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3072 | struct address_space *mapping, |
3073 | loff_t pos, unsigned len, unsigned copied, | |
3074 | struct page *page, void *fsdata) | |
64769240 AT |
3075 | { |
3076 | struct inode *inode = mapping->host; | |
3077 | int ret = 0, ret2; | |
3078 | handle_t *handle = ext4_journal_current_handle(); | |
3079 | loff_t new_i_size; | |
632eaeab | 3080 | unsigned long start, end; |
79f0be8d AK |
3081 | int write_mode = (int)(unsigned long)fsdata; |
3082 | ||
74d553aa TT |
3083 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3084 | return ext4_write_end(file, mapping, pos, | |
3085 | len, copied, page, fsdata); | |
632eaeab | 3086 | |
9bffad1e | 3087 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3088 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3089 | end = start + copied - 1; |
64769240 AT |
3090 | |
3091 | /* | |
3092 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3093 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3094 | * into that. | |
3095 | */ | |
64769240 | 3096 | new_i_size = pos + copied; |
ea51d132 | 3097 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3098 | if (ext4_has_inline_data(inode) || |
3099 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3100 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3101 | /* We need to mark inode dirty even if |
3102 | * new_i_size is less that inode->i_size | |
3103 | * bu greater than i_disksize.(hint delalloc) | |
3104 | */ | |
3105 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3106 | } |
632eaeab | 3107 | } |
9c3569b5 TM |
3108 | |
3109 | if (write_mode != CONVERT_INLINE_DATA && | |
3110 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3111 | ext4_has_inline_data(inode)) | |
3112 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3113 | page); | |
3114 | else | |
3115 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3116 | page, fsdata); |
9c3569b5 | 3117 | |
64769240 AT |
3118 | copied = ret2; |
3119 | if (ret2 < 0) | |
3120 | ret = ret2; | |
3121 | ret2 = ext4_journal_stop(handle); | |
3122 | if (!ret) | |
3123 | ret = ret2; | |
3124 | ||
3125 | return ret ? ret : copied; | |
3126 | } | |
3127 | ||
d47992f8 LC |
3128 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
3129 | unsigned int length) | |
64769240 | 3130 | { |
64769240 AT |
3131 | /* |
3132 | * Drop reserved blocks | |
3133 | */ | |
3134 | BUG_ON(!PageLocked(page)); | |
3135 | if (!page_has_buffers(page)) | |
3136 | goto out; | |
3137 | ||
ca99fdd2 | 3138 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
3139 | |
3140 | out: | |
d47992f8 | 3141 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
3142 | |
3143 | return; | |
3144 | } | |
3145 | ||
ccd2506b TT |
3146 | /* |
3147 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3148 | */ | |
3149 | int ext4_alloc_da_blocks(struct inode *inode) | |
3150 | { | |
fb40ba0d TT |
3151 | trace_ext4_alloc_da_blocks(inode); |
3152 | ||
71d4f7d0 | 3153 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3154 | return 0; |
3155 | ||
3156 | /* | |
3157 | * We do something simple for now. The filemap_flush() will | |
3158 | * also start triggering a write of the data blocks, which is | |
3159 | * not strictly speaking necessary (and for users of | |
3160 | * laptop_mode, not even desirable). However, to do otherwise | |
3161 | * would require replicating code paths in: | |
de9a55b8 | 3162 | * |
20970ba6 | 3163 | * ext4_writepages() -> |
ccd2506b TT |
3164 | * write_cache_pages() ---> (via passed in callback function) |
3165 | * __mpage_da_writepage() --> | |
3166 | * mpage_add_bh_to_extent() | |
3167 | * mpage_da_map_blocks() | |
3168 | * | |
3169 | * The problem is that write_cache_pages(), located in | |
3170 | * mm/page-writeback.c, marks pages clean in preparation for | |
3171 | * doing I/O, which is not desirable if we're not planning on | |
3172 | * doing I/O at all. | |
3173 | * | |
3174 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3175 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3176 | * would be ugly in the extreme. So instead we would need to |
3177 | * replicate parts of the code in the above functions, | |
25985edc | 3178 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3179 | * write out the pages, but rather only collect contiguous |
3180 | * logical block extents, call the multi-block allocator, and | |
3181 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3182 | * |
ccd2506b TT |
3183 | * For now, though, we'll cheat by calling filemap_flush(), |
3184 | * which will map the blocks, and start the I/O, but not | |
3185 | * actually wait for the I/O to complete. | |
3186 | */ | |
3187 | return filemap_flush(inode->i_mapping); | |
3188 | } | |
64769240 | 3189 | |
ac27a0ec DK |
3190 | /* |
3191 | * bmap() is special. It gets used by applications such as lilo and by | |
3192 | * the swapper to find the on-disk block of a specific piece of data. | |
3193 | * | |
3194 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3195 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3196 | * filesystem and enables swap, then they may get a nasty shock when the |
3197 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3198 | * the original zero's written out previously to the journal and | |
3199 | * awaiting writeback in the kernel's buffer cache. | |
3200 | * | |
3201 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3202 | * take extra steps to flush any blocks which might be in the cache. | |
3203 | */ | |
617ba13b | 3204 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3205 | { |
3206 | struct inode *inode = mapping->host; | |
3207 | journal_t *journal; | |
3208 | int err; | |
3209 | ||
46c7f254 TM |
3210 | /* |
3211 | * We can get here for an inline file via the FIBMAP ioctl | |
3212 | */ | |
3213 | if (ext4_has_inline_data(inode)) | |
3214 | return 0; | |
3215 | ||
64769240 AT |
3216 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3217 | test_opt(inode->i_sb, DELALLOC)) { | |
3218 | /* | |
3219 | * With delalloc we want to sync the file | |
3220 | * so that we can make sure we allocate | |
3221 | * blocks for file | |
3222 | */ | |
3223 | filemap_write_and_wait(mapping); | |
3224 | } | |
3225 | ||
19f5fb7a TT |
3226 | if (EXT4_JOURNAL(inode) && |
3227 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3228 | /* |
3229 | * This is a REALLY heavyweight approach, but the use of | |
3230 | * bmap on dirty files is expected to be extremely rare: | |
3231 | * only if we run lilo or swapon on a freshly made file | |
3232 | * do we expect this to happen. | |
3233 | * | |
3234 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3235 | * represent an unprivileged user DOS attack --- we'd be | |
3236 | * in trouble if mortal users could trigger this path at | |
3237 | * will.) | |
3238 | * | |
617ba13b | 3239 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3240 | * regular files. If somebody wants to bmap a directory |
3241 | * or symlink and gets confused because the buffer | |
3242 | * hasn't yet been flushed to disk, they deserve | |
3243 | * everything they get. | |
3244 | */ | |
3245 | ||
19f5fb7a | 3246 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3247 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3248 | jbd2_journal_lock_updates(journal); |
3249 | err = jbd2_journal_flush(journal); | |
3250 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3251 | |
3252 | if (err) | |
3253 | return 0; | |
3254 | } | |
3255 | ||
af5bc92d | 3256 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3257 | } |
3258 | ||
617ba13b | 3259 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3260 | { |
46c7f254 TM |
3261 | int ret = -EAGAIN; |
3262 | struct inode *inode = page->mapping->host; | |
3263 | ||
0562e0ba | 3264 | trace_ext4_readpage(page); |
46c7f254 TM |
3265 | |
3266 | if (ext4_has_inline_data(inode)) | |
3267 | ret = ext4_readpage_inline(inode, page); | |
3268 | ||
3269 | if (ret == -EAGAIN) | |
f64e02fe | 3270 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3271 | |
3272 | return ret; | |
ac27a0ec DK |
3273 | } |
3274 | ||
3275 | static int | |
617ba13b | 3276 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3277 | struct list_head *pages, unsigned nr_pages) |
3278 | { | |
46c7f254 TM |
3279 | struct inode *inode = mapping->host; |
3280 | ||
3281 | /* If the file has inline data, no need to do readpages. */ | |
3282 | if (ext4_has_inline_data(inode)) | |
3283 | return 0; | |
3284 | ||
f64e02fe | 3285 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3286 | } |
3287 | ||
d47992f8 LC |
3288 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3289 | unsigned int length) | |
ac27a0ec | 3290 | { |
ca99fdd2 | 3291 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3292 | |
4520fb3c JK |
3293 | /* No journalling happens on data buffers when this function is used */ |
3294 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3295 | ||
ca99fdd2 | 3296 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3297 | } |
3298 | ||
53e87268 | 3299 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3300 | unsigned int offset, |
3301 | unsigned int length) | |
4520fb3c JK |
3302 | { |
3303 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3304 | ||
ca99fdd2 | 3305 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3306 | |
ac27a0ec DK |
3307 | /* |
3308 | * If it's a full truncate we just forget about the pending dirtying | |
3309 | */ | |
09cbfeaf | 3310 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3311 | ClearPageChecked(page); |
3312 | ||
ca99fdd2 | 3313 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3314 | } |
3315 | ||
3316 | /* Wrapper for aops... */ | |
3317 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3318 | unsigned int offset, |
3319 | unsigned int length) | |
53e87268 | 3320 | { |
ca99fdd2 | 3321 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3322 | } |
3323 | ||
617ba13b | 3324 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3325 | { |
617ba13b | 3326 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3327 | |
0562e0ba JZ |
3328 | trace_ext4_releasepage(page); |
3329 | ||
e1c36595 JK |
3330 | /* Page has dirty journalled data -> cannot release */ |
3331 | if (PageChecked(page)) | |
ac27a0ec | 3332 | return 0; |
0390131b FM |
3333 | if (journal) |
3334 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3335 | else | |
3336 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3337 | } |
3338 | ||
ba5843f5 | 3339 | #ifdef CONFIG_FS_DAX |
364443cb JK |
3340 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
3341 | unsigned flags, struct iomap *iomap) | |
3342 | { | |
fa5d932c | 3343 | struct block_device *bdev; |
364443cb JK |
3344 | unsigned int blkbits = inode->i_blkbits; |
3345 | unsigned long first_block = offset >> blkbits; | |
3346 | unsigned long last_block = (offset + length - 1) >> blkbits; | |
3347 | struct ext4_map_blocks map; | |
3348 | int ret; | |
3349 | ||
364443cb JK |
3350 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3351 | return -ERANGE; | |
3352 | ||
3353 | map.m_lblk = first_block; | |
3354 | map.m_len = last_block - first_block + 1; | |
3355 | ||
776722e8 JK |
3356 | if (!(flags & IOMAP_WRITE)) { |
3357 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3358 | } else { | |
3359 | int dio_credits; | |
3360 | handle_t *handle; | |
3361 | int retries = 0; | |
3362 | ||
3363 | /* Trim mapping request to maximum we can map at once for DIO */ | |
3364 | if (map.m_len > DIO_MAX_BLOCKS) | |
3365 | map.m_len = DIO_MAX_BLOCKS; | |
3366 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
3367 | retry: | |
3368 | /* | |
3369 | * Either we allocate blocks and then we don't get unwritten | |
3370 | * extent so we have reserved enough credits, or the blocks | |
3371 | * are already allocated and unwritten and in that case | |
3372 | * extent conversion fits in the credits as well. | |
3373 | */ | |
3374 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
3375 | dio_credits); | |
3376 | if (IS_ERR(handle)) | |
3377 | return PTR_ERR(handle); | |
3378 | ||
3379 | ret = ext4_map_blocks(handle, inode, &map, | |
776722e8 JK |
3380 | EXT4_GET_BLOCKS_CREATE_ZERO); |
3381 | if (ret < 0) { | |
3382 | ext4_journal_stop(handle); | |
3383 | if (ret == -ENOSPC && | |
3384 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3385 | goto retry; | |
3386 | return ret; | |
3387 | } | |
776722e8 JK |
3388 | |
3389 | /* | |
e2ae766c | 3390 | * If we added blocks beyond i_size, we need to make sure they |
776722e8 | 3391 | * will get truncated if we crash before updating i_size in |
e2ae766c JK |
3392 | * ext4_iomap_end(). For faults we don't need to do that (and |
3393 | * even cannot because for orphan list operations inode_lock is | |
3394 | * required) - if we happen to instantiate block beyond i_size, | |
3395 | * it is because we race with truncate which has already added | |
3396 | * the inode to the orphan list. | |
776722e8 | 3397 | */ |
e2ae766c JK |
3398 | if (!(flags & IOMAP_FAULT) && first_block + map.m_len > |
3399 | (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) { | |
776722e8 JK |
3400 | int err; |
3401 | ||
3402 | err = ext4_orphan_add(handle, inode); | |
3403 | if (err < 0) { | |
3404 | ext4_journal_stop(handle); | |
3405 | return err; | |
3406 | } | |
3407 | } | |
3408 | ext4_journal_stop(handle); | |
3409 | } | |
364443cb JK |
3410 | |
3411 | iomap->flags = 0; | |
fa5d932c DW |
3412 | bdev = inode->i_sb->s_bdev; |
3413 | iomap->bdev = bdev; | |
3414 | if (blk_queue_dax(bdev->bd_queue)) | |
3415 | iomap->dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); | |
3416 | else | |
3417 | iomap->dax_dev = NULL; | |
364443cb JK |
3418 | iomap->offset = first_block << blkbits; |
3419 | ||
3420 | if (ret == 0) { | |
3421 | iomap->type = IOMAP_HOLE; | |
3422 | iomap->blkno = IOMAP_NULL_BLOCK; | |
3423 | iomap->length = (u64)map.m_len << blkbits; | |
3424 | } else { | |
3425 | if (map.m_flags & EXT4_MAP_MAPPED) { | |
3426 | iomap->type = IOMAP_MAPPED; | |
3427 | } else if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
3428 | iomap->type = IOMAP_UNWRITTEN; | |
3429 | } else { | |
3430 | WARN_ON_ONCE(1); | |
3431 | return -EIO; | |
3432 | } | |
3433 | iomap->blkno = (sector_t)map.m_pblk << (blkbits - 9); | |
3434 | iomap->length = (u64)map.m_len << blkbits; | |
3435 | } | |
3436 | ||
3437 | if (map.m_flags & EXT4_MAP_NEW) | |
3438 | iomap->flags |= IOMAP_F_NEW; | |
3439 | return 0; | |
3440 | } | |
3441 | ||
776722e8 JK |
3442 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3443 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3444 | { | |
3445 | int ret = 0; | |
3446 | handle_t *handle; | |
3447 | int blkbits = inode->i_blkbits; | |
3448 | bool truncate = false; | |
3449 | ||
fa5d932c | 3450 | put_dax(iomap->dax_dev); |
e2ae766c | 3451 | if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT)) |
776722e8 JK |
3452 | return 0; |
3453 | ||
3454 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3455 | if (IS_ERR(handle)) { | |
3456 | ret = PTR_ERR(handle); | |
3457 | goto orphan_del; | |
3458 | } | |
3459 | if (ext4_update_inode_size(inode, offset + written)) | |
3460 | ext4_mark_inode_dirty(handle, inode); | |
3461 | /* | |
3462 | * We may need to truncate allocated but not written blocks beyond EOF. | |
3463 | */ | |
3464 | if (iomap->offset + iomap->length > | |
3465 | ALIGN(inode->i_size, 1 << blkbits)) { | |
3466 | ext4_lblk_t written_blk, end_blk; | |
3467 | ||
3468 | written_blk = (offset + written) >> blkbits; | |
3469 | end_blk = (offset + length) >> blkbits; | |
3470 | if (written_blk < end_blk && ext4_can_truncate(inode)) | |
3471 | truncate = true; | |
3472 | } | |
3473 | /* | |
3474 | * Remove inode from orphan list if we were extending a inode and | |
3475 | * everything went fine. | |
3476 | */ | |
3477 | if (!truncate && inode->i_nlink && | |
3478 | !list_empty(&EXT4_I(inode)->i_orphan)) | |
3479 | ext4_orphan_del(handle, inode); | |
3480 | ext4_journal_stop(handle); | |
3481 | if (truncate) { | |
3482 | ext4_truncate_failed_write(inode); | |
3483 | orphan_del: | |
3484 | /* | |
3485 | * If truncate failed early the inode might still be on the | |
3486 | * orphan list; we need to make sure the inode is removed from | |
3487 | * the orphan list in that case. | |
3488 | */ | |
3489 | if (inode->i_nlink) | |
3490 | ext4_orphan_del(NULL, inode); | |
3491 | } | |
3492 | return ret; | |
3493 | } | |
3494 | ||
8ff6daa1 | 3495 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3496 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3497 | .iomap_end = ext4_iomap_end, |
364443cb JK |
3498 | }; |
3499 | ||
ba5843f5 | 3500 | #endif |
ed923b57 | 3501 | |
187372a3 | 3502 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3503 | ssize_t size, void *private) |
4c0425ff | 3504 | { |
109811c2 | 3505 | ext4_io_end_t *io_end = private; |
4c0425ff | 3506 | |
97a851ed | 3507 | /* if not async direct IO just return */ |
7b7a8665 | 3508 | if (!io_end) |
187372a3 | 3509 | return 0; |
4b70df18 | 3510 | |
88635ca2 | 3511 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3512 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
109811c2 | 3513 | io_end, io_end->inode->i_ino, iocb, offset, size); |
8d5d02e6 | 3514 | |
74c66bcb JK |
3515 | /* |
3516 | * Error during AIO DIO. We cannot convert unwritten extents as the | |
3517 | * data was not written. Just clear the unwritten flag and drop io_end. | |
3518 | */ | |
3519 | if (size <= 0) { | |
3520 | ext4_clear_io_unwritten_flag(io_end); | |
3521 | size = 0; | |
3522 | } | |
4c0425ff MC |
3523 | io_end->offset = offset; |
3524 | io_end->size = size; | |
7b7a8665 | 3525 | ext4_put_io_end(io_end); |
187372a3 CH |
3526 | |
3527 | return 0; | |
4c0425ff | 3528 | } |
c7064ef1 | 3529 | |
4c0425ff | 3530 | /* |
914f82a3 JK |
3531 | * Handling of direct IO writes. |
3532 | * | |
3533 | * For ext4 extent files, ext4 will do direct-io write even to holes, | |
4c0425ff MC |
3534 | * preallocated extents, and those write extend the file, no need to |
3535 | * fall back to buffered IO. | |
3536 | * | |
556615dc | 3537 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3538 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3539 | * still keep the range to write as unwritten. |
4c0425ff | 3540 | * |
69c499d1 | 3541 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3542 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3543 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3544 | * when async direct IO completed. |
4c0425ff MC |
3545 | * |
3546 | * If the O_DIRECT write will extend the file then add this inode to the | |
3547 | * orphan list. So recovery will truncate it back to the original size | |
3548 | * if the machine crashes during the write. | |
3549 | * | |
3550 | */ | |
0e01df10 | 3551 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3552 | { |
3553 | struct file *file = iocb->ki_filp; | |
3554 | struct inode *inode = file->f_mapping->host; | |
914f82a3 | 3555 | struct ext4_inode_info *ei = EXT4_I(inode); |
4c0425ff | 3556 | ssize_t ret; |
c8b8e32d | 3557 | loff_t offset = iocb->ki_pos; |
a6cbcd4a | 3558 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3559 | int overwrite = 0; |
3560 | get_block_t *get_block_func = NULL; | |
3561 | int dio_flags = 0; | |
4c0425ff | 3562 | loff_t final_size = offset + count; |
914f82a3 JK |
3563 | int orphan = 0; |
3564 | handle_t *handle; | |
729f52c6 | 3565 | |
914f82a3 JK |
3566 | if (final_size > inode->i_size) { |
3567 | /* Credits for sb + inode write */ | |
3568 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3569 | if (IS_ERR(handle)) { | |
3570 | ret = PTR_ERR(handle); | |
3571 | goto out; | |
3572 | } | |
3573 | ret = ext4_orphan_add(handle, inode); | |
3574 | if (ret) { | |
3575 | ext4_journal_stop(handle); | |
3576 | goto out; | |
3577 | } | |
3578 | orphan = 1; | |
3579 | ei->i_disksize = inode->i_size; | |
3580 | ext4_journal_stop(handle); | |
3581 | } | |
4bd809db | 3582 | |
69c499d1 | 3583 | BUG_ON(iocb->private == NULL); |
4bd809db | 3584 | |
e8340395 JK |
3585 | /* |
3586 | * Make all waiters for direct IO properly wait also for extent | |
3587 | * conversion. This also disallows race between truncate() and | |
3588 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3589 | */ | |
914f82a3 | 3590 | inode_dio_begin(inode); |
e8340395 | 3591 | |
69c499d1 TT |
3592 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3593 | overwrite = *((int *)iocb->private); | |
4bd809db | 3594 | |
2dcba478 | 3595 | if (overwrite) |
5955102c | 3596 | inode_unlock(inode); |
8d5d02e6 | 3597 | |
69c499d1 | 3598 | /* |
914f82a3 | 3599 | * For extent mapped files we could direct write to holes and fallocate. |
69c499d1 | 3600 | * |
109811c2 JK |
3601 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
3602 | * parallel buffered read to expose the stale data before DIO complete | |
3603 | * the data IO. | |
69c499d1 | 3604 | * |
109811c2 JK |
3605 | * As to previously fallocated extents, ext4 get_block will just simply |
3606 | * mark the buffer mapped but still keep the extents unwritten. | |
69c499d1 | 3607 | * |
109811c2 JK |
3608 | * For non AIO case, we will convert those unwritten extents to written |
3609 | * after return back from blockdev_direct_IO. That way we save us from | |
3610 | * allocating io_end structure and also the overhead of offloading | |
3611 | * the extent convertion to a workqueue. | |
69c499d1 TT |
3612 | * |
3613 | * For async DIO, the conversion needs to be deferred when the | |
3614 | * IO is completed. The ext4 end_io callback function will be | |
3615 | * called to take care of the conversion work. Here for async | |
3616 | * case, we allocate an io_end structure to hook to the iocb. | |
3617 | */ | |
3618 | iocb->private = NULL; | |
109811c2 | 3619 | if (overwrite) |
705965bd | 3620 | get_block_func = ext4_dio_get_block_overwrite; |
0bd2d5ec | 3621 | else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || |
93407472 | 3622 | round_down(offset, i_blocksize(inode)) >= inode->i_size) { |
914f82a3 JK |
3623 | get_block_func = ext4_dio_get_block; |
3624 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
3625 | } else if (is_sync_kiocb(iocb)) { | |
109811c2 JK |
3626 | get_block_func = ext4_dio_get_block_unwritten_sync; |
3627 | dio_flags = DIO_LOCKING; | |
69c499d1 | 3628 | } else { |
109811c2 | 3629 | get_block_func = ext4_dio_get_block_unwritten_async; |
69c499d1 TT |
3630 | dio_flags = DIO_LOCKING; |
3631 | } | |
2058f83a MH |
3632 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3633 | BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); | |
3634 | #endif | |
0bd2d5ec JK |
3635 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
3636 | get_block_func, ext4_end_io_dio, NULL, | |
3637 | dio_flags); | |
69c499d1 | 3638 | |
97a851ed | 3639 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
69c499d1 TT |
3640 | EXT4_STATE_DIO_UNWRITTEN)) { |
3641 | int err; | |
3642 | /* | |
3643 | * for non AIO case, since the IO is already | |
3644 | * completed, we could do the conversion right here | |
3645 | */ | |
6b523df4 | 3646 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3647 | offset, ret); |
3648 | if (err < 0) | |
3649 | ret = err; | |
3650 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3651 | } | |
4bd809db | 3652 | |
914f82a3 | 3653 | inode_dio_end(inode); |
69c499d1 | 3654 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3655 | if (overwrite) |
5955102c | 3656 | inode_lock(inode); |
8d5d02e6 | 3657 | |
914f82a3 JK |
3658 | if (ret < 0 && final_size > inode->i_size) |
3659 | ext4_truncate_failed_write(inode); | |
3660 | ||
3661 | /* Handle extending of i_size after direct IO write */ | |
3662 | if (orphan) { | |
3663 | int err; | |
3664 | ||
3665 | /* Credits for sb + inode write */ | |
3666 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3667 | if (IS_ERR(handle)) { | |
3668 | /* This is really bad luck. We've written the data | |
3669 | * but cannot extend i_size. Bail out and pretend | |
3670 | * the write failed... */ | |
3671 | ret = PTR_ERR(handle); | |
3672 | if (inode->i_nlink) | |
3673 | ext4_orphan_del(NULL, inode); | |
3674 | ||
3675 | goto out; | |
3676 | } | |
3677 | if (inode->i_nlink) | |
3678 | ext4_orphan_del(handle, inode); | |
3679 | if (ret > 0) { | |
3680 | loff_t end = offset + ret; | |
3681 | if (end > inode->i_size) { | |
3682 | ei->i_disksize = end; | |
3683 | i_size_write(inode, end); | |
3684 | /* | |
3685 | * We're going to return a positive `ret' | |
3686 | * here due to non-zero-length I/O, so there's | |
3687 | * no way of reporting error returns from | |
3688 | * ext4_mark_inode_dirty() to userspace. So | |
3689 | * ignore it. | |
3690 | */ | |
3691 | ext4_mark_inode_dirty(handle, inode); | |
3692 | } | |
3693 | } | |
3694 | err = ext4_journal_stop(handle); | |
3695 | if (ret == 0) | |
3696 | ret = err; | |
3697 | } | |
3698 | out: | |
3699 | return ret; | |
3700 | } | |
3701 | ||
0e01df10 | 3702 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
914f82a3 | 3703 | { |
16c54688 JK |
3704 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
3705 | struct inode *inode = mapping->host; | |
0bd2d5ec | 3706 | size_t count = iov_iter_count(iter); |
914f82a3 JK |
3707 | ssize_t ret; |
3708 | ||
16c54688 JK |
3709 | /* |
3710 | * Shared inode_lock is enough for us - it protects against concurrent | |
3711 | * writes & truncates and since we take care of writing back page cache, | |
3712 | * we are protected against page writeback as well. | |
3713 | */ | |
3714 | inode_lock_shared(inode); | |
0bd2d5ec JK |
3715 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
3716 | iocb->ki_pos + count); | |
3717 | if (ret) | |
3718 | goto out_unlock; | |
3719 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, | |
3720 | iter, ext4_dio_get_block, NULL, NULL, 0); | |
16c54688 JK |
3721 | out_unlock: |
3722 | inode_unlock_shared(inode); | |
69c499d1 | 3723 | return ret; |
4c0425ff MC |
3724 | } |
3725 | ||
c8b8e32d | 3726 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3727 | { |
3728 | struct file *file = iocb->ki_filp; | |
3729 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3730 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3731 | loff_t offset = iocb->ki_pos; |
0562e0ba | 3732 | ssize_t ret; |
4c0425ff | 3733 | |
2058f83a MH |
3734 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3735 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3736 | return 0; | |
3737 | #endif | |
3738 | ||
84ebd795 TT |
3739 | /* |
3740 | * If we are doing data journalling we don't support O_DIRECT | |
3741 | */ | |
3742 | if (ext4_should_journal_data(inode)) | |
3743 | return 0; | |
3744 | ||
46c7f254 TM |
3745 | /* Let buffer I/O handle the inline data case. */ |
3746 | if (ext4_has_inline_data(inode)) | |
3747 | return 0; | |
3748 | ||
0bd2d5ec JK |
3749 | /* DAX uses iomap path now */ |
3750 | if (WARN_ON_ONCE(IS_DAX(inode))) | |
3751 | return 0; | |
3752 | ||
6f673763 | 3753 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
914f82a3 | 3754 | if (iov_iter_rw(iter) == READ) |
0e01df10 | 3755 | ret = ext4_direct_IO_read(iocb, iter); |
0562e0ba | 3756 | else |
0e01df10 | 3757 | ret = ext4_direct_IO_write(iocb, iter); |
6f673763 | 3758 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
0562e0ba | 3759 | return ret; |
4c0425ff MC |
3760 | } |
3761 | ||
ac27a0ec | 3762 | /* |
617ba13b | 3763 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3764 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3765 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3766 | * not necessarily locked. | |
3767 | * | |
3768 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3769 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3770 | * or jbddirty because all the journalling code will explode. | |
3771 | * | |
3772 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3773 | * is called, propagate that into the buffers appropriately. | |
3774 | */ | |
617ba13b | 3775 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3776 | { |
3777 | SetPageChecked(page); | |
3778 | return __set_page_dirty_nobuffers(page); | |
3779 | } | |
3780 | ||
6dcc693b JK |
3781 | static int ext4_set_page_dirty(struct page *page) |
3782 | { | |
3783 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3784 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3785 | return __set_page_dirty_buffers(page); | |
3786 | } | |
3787 | ||
74d553aa | 3788 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3789 | .readpage = ext4_readpage, |
3790 | .readpages = ext4_readpages, | |
43ce1d23 | 3791 | .writepage = ext4_writepage, |
20970ba6 | 3792 | .writepages = ext4_writepages, |
8ab22b9a | 3793 | .write_begin = ext4_write_begin, |
74d553aa | 3794 | .write_end = ext4_write_end, |
6dcc693b | 3795 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3796 | .bmap = ext4_bmap, |
3797 | .invalidatepage = ext4_invalidatepage, | |
3798 | .releasepage = ext4_releasepage, | |
3799 | .direct_IO = ext4_direct_IO, | |
3800 | .migratepage = buffer_migrate_page, | |
3801 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3802 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3803 | }; |
3804 | ||
617ba13b | 3805 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3806 | .readpage = ext4_readpage, |
3807 | .readpages = ext4_readpages, | |
43ce1d23 | 3808 | .writepage = ext4_writepage, |
20970ba6 | 3809 | .writepages = ext4_writepages, |
8ab22b9a HH |
3810 | .write_begin = ext4_write_begin, |
3811 | .write_end = ext4_journalled_write_end, | |
3812 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3813 | .bmap = ext4_bmap, | |
4520fb3c | 3814 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3815 | .releasepage = ext4_releasepage, |
84ebd795 | 3816 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3817 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3818 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3819 | }; |
3820 | ||
64769240 | 3821 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3822 | .readpage = ext4_readpage, |
3823 | .readpages = ext4_readpages, | |
43ce1d23 | 3824 | .writepage = ext4_writepage, |
20970ba6 | 3825 | .writepages = ext4_writepages, |
8ab22b9a HH |
3826 | .write_begin = ext4_da_write_begin, |
3827 | .write_end = ext4_da_write_end, | |
6dcc693b | 3828 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3829 | .bmap = ext4_bmap, |
3830 | .invalidatepage = ext4_da_invalidatepage, | |
3831 | .releasepage = ext4_releasepage, | |
3832 | .direct_IO = ext4_direct_IO, | |
3833 | .migratepage = buffer_migrate_page, | |
3834 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3835 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3836 | }; |
3837 | ||
617ba13b | 3838 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3839 | { |
3d2b1582 LC |
3840 | switch (ext4_inode_journal_mode(inode)) { |
3841 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3842 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3843 | break; |
3844 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3845 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3846 | return; |
3d2b1582 LC |
3847 | default: |
3848 | BUG(); | |
3849 | } | |
74d553aa TT |
3850 | if (test_opt(inode->i_sb, DELALLOC)) |
3851 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3852 | else | |
3853 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3854 | } |
3855 | ||
923ae0ff | 3856 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3857 | struct address_space *mapping, loff_t from, loff_t length) |
3858 | { | |
09cbfeaf KS |
3859 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3860 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3861 | unsigned blocksize, pos; |
d863dc36 LC |
3862 | ext4_lblk_t iblock; |
3863 | struct inode *inode = mapping->host; | |
3864 | struct buffer_head *bh; | |
3865 | struct page *page; | |
3866 | int err = 0; | |
3867 | ||
09cbfeaf | 3868 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3869 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3870 | if (!page) |
3871 | return -ENOMEM; | |
3872 | ||
3873 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3874 | |
09cbfeaf | 3875 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3876 | |
3877 | if (!page_has_buffers(page)) | |
3878 | create_empty_buffers(page, blocksize, 0); | |
3879 | ||
3880 | /* Find the buffer that contains "offset" */ | |
3881 | bh = page_buffers(page); | |
3882 | pos = blocksize; | |
3883 | while (offset >= pos) { | |
3884 | bh = bh->b_this_page; | |
3885 | iblock++; | |
3886 | pos += blocksize; | |
3887 | } | |
d863dc36 LC |
3888 | if (buffer_freed(bh)) { |
3889 | BUFFER_TRACE(bh, "freed: skip"); | |
3890 | goto unlock; | |
3891 | } | |
d863dc36 LC |
3892 | if (!buffer_mapped(bh)) { |
3893 | BUFFER_TRACE(bh, "unmapped"); | |
3894 | ext4_get_block(inode, iblock, bh, 0); | |
3895 | /* unmapped? It's a hole - nothing to do */ | |
3896 | if (!buffer_mapped(bh)) { | |
3897 | BUFFER_TRACE(bh, "still unmapped"); | |
3898 | goto unlock; | |
3899 | } | |
3900 | } | |
3901 | ||
3902 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3903 | if (PageUptodate(page)) | |
3904 | set_buffer_uptodate(bh); | |
3905 | ||
3906 | if (!buffer_uptodate(bh)) { | |
3907 | err = -EIO; | |
dfec8a14 | 3908 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
3909 | wait_on_buffer(bh); |
3910 | /* Uhhuh. Read error. Complain and punt. */ | |
3911 | if (!buffer_uptodate(bh)) | |
3912 | goto unlock; | |
c9c7429c MH |
3913 | if (S_ISREG(inode->i_mode) && |
3914 | ext4_encrypted_inode(inode)) { | |
3915 | /* We expect the key to be set. */ | |
a7550b30 | 3916 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
09cbfeaf | 3917 | BUG_ON(blocksize != PAGE_SIZE); |
b50f7b26 | 3918 | WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host, |
9c4bb8a3 | 3919 | page, PAGE_SIZE, 0, page->index)); |
c9c7429c | 3920 | } |
d863dc36 | 3921 | } |
d863dc36 LC |
3922 | if (ext4_should_journal_data(inode)) { |
3923 | BUFFER_TRACE(bh, "get write access"); | |
3924 | err = ext4_journal_get_write_access(handle, bh); | |
3925 | if (err) | |
3926 | goto unlock; | |
3927 | } | |
d863dc36 | 3928 | zero_user(page, offset, length); |
d863dc36 LC |
3929 | BUFFER_TRACE(bh, "zeroed end of block"); |
3930 | ||
d863dc36 LC |
3931 | if (ext4_should_journal_data(inode)) { |
3932 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3933 | } else { |
353eefd3 | 3934 | err = 0; |
d863dc36 | 3935 | mark_buffer_dirty(bh); |
3957ef53 | 3936 | if (ext4_should_order_data(inode)) |
ee0876bc | 3937 | err = ext4_jbd2_inode_add_write(handle, inode); |
0713ed0c | 3938 | } |
d863dc36 LC |
3939 | |
3940 | unlock: | |
3941 | unlock_page(page); | |
09cbfeaf | 3942 | put_page(page); |
d863dc36 LC |
3943 | return err; |
3944 | } | |
3945 | ||
923ae0ff RZ |
3946 | /* |
3947 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3948 | * starting from file offset 'from'. The range to be zero'd must | |
3949 | * be contained with in one block. If the specified range exceeds | |
3950 | * the end of the block it will be shortened to end of the block | |
3951 | * that cooresponds to 'from' | |
3952 | */ | |
3953 | static int ext4_block_zero_page_range(handle_t *handle, | |
3954 | struct address_space *mapping, loff_t from, loff_t length) | |
3955 | { | |
3956 | struct inode *inode = mapping->host; | |
09cbfeaf | 3957 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3958 | unsigned blocksize = inode->i_sb->s_blocksize; |
3959 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3960 | ||
3961 | /* | |
3962 | * correct length if it does not fall between | |
3963 | * 'from' and the end of the block | |
3964 | */ | |
3965 | if (length > max || length < 0) | |
3966 | length = max; | |
3967 | ||
47e69351 JK |
3968 | if (IS_DAX(inode)) { |
3969 | return iomap_zero_range(inode, from, length, NULL, | |
3970 | &ext4_iomap_ops); | |
3971 | } | |
923ae0ff RZ |
3972 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3973 | } | |
3974 | ||
94350ab5 MW |
3975 | /* |
3976 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3977 | * up to the end of the block which corresponds to `from'. | |
3978 | * This required during truncate. We need to physically zero the tail end | |
3979 | * of that block so it doesn't yield old data if the file is later grown. | |
3980 | */ | |
c197855e | 3981 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3982 | struct address_space *mapping, loff_t from) |
3983 | { | |
09cbfeaf | 3984 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3985 | unsigned length; |
3986 | unsigned blocksize; | |
3987 | struct inode *inode = mapping->host; | |
3988 | ||
0d06863f TT |
3989 | /* If we are processing an encrypted inode during orphan list handling */ |
3990 | if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode)) | |
3991 | return 0; | |
3992 | ||
94350ab5 MW |
3993 | blocksize = inode->i_sb->s_blocksize; |
3994 | length = blocksize - (offset & (blocksize - 1)); | |
3995 | ||
3996 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3997 | } | |
3998 | ||
a87dd18c LC |
3999 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
4000 | loff_t lstart, loff_t length) | |
4001 | { | |
4002 | struct super_block *sb = inode->i_sb; | |
4003 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 4004 | unsigned partial_start, partial_end; |
a87dd18c LC |
4005 | ext4_fsblk_t start, end; |
4006 | loff_t byte_end = (lstart + length - 1); | |
4007 | int err = 0; | |
4008 | ||
e1be3a92 LC |
4009 | partial_start = lstart & (sb->s_blocksize - 1); |
4010 | partial_end = byte_end & (sb->s_blocksize - 1); | |
4011 | ||
a87dd18c LC |
4012 | start = lstart >> sb->s_blocksize_bits; |
4013 | end = byte_end >> sb->s_blocksize_bits; | |
4014 | ||
4015 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
4016 | if (start == end && |
4017 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
4018 | err = ext4_block_zero_page_range(handle, mapping, |
4019 | lstart, length); | |
4020 | return err; | |
4021 | } | |
4022 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 4023 | if (partial_start) { |
a87dd18c LC |
4024 | err = ext4_block_zero_page_range(handle, mapping, |
4025 | lstart, sb->s_blocksize); | |
4026 | if (err) | |
4027 | return err; | |
4028 | } | |
4029 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 4030 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 4031 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
4032 | byte_end - partial_end, |
4033 | partial_end + 1); | |
a87dd18c LC |
4034 | return err; |
4035 | } | |
4036 | ||
91ef4caf DG |
4037 | int ext4_can_truncate(struct inode *inode) |
4038 | { | |
91ef4caf DG |
4039 | if (S_ISREG(inode->i_mode)) |
4040 | return 1; | |
4041 | if (S_ISDIR(inode->i_mode)) | |
4042 | return 1; | |
4043 | if (S_ISLNK(inode->i_mode)) | |
4044 | return !ext4_inode_is_fast_symlink(inode); | |
4045 | return 0; | |
4046 | } | |
4047 | ||
01127848 JK |
4048 | /* |
4049 | * We have to make sure i_disksize gets properly updated before we truncate | |
4050 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
4051 | * can get lost as it may have been postponed to submission of writeback but | |
4052 | * that will never happen after we truncate page cache. | |
4053 | */ | |
4054 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
4055 | loff_t len) | |
4056 | { | |
4057 | handle_t *handle; | |
4058 | loff_t size = i_size_read(inode); | |
4059 | ||
5955102c | 4060 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
4061 | if (offset > size || offset + len < size) |
4062 | return 0; | |
4063 | ||
4064 | if (EXT4_I(inode)->i_disksize >= size) | |
4065 | return 0; | |
4066 | ||
4067 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
4068 | if (IS_ERR(handle)) | |
4069 | return PTR_ERR(handle); | |
4070 | ext4_update_i_disksize(inode, size); | |
4071 | ext4_mark_inode_dirty(handle, inode); | |
4072 | ext4_journal_stop(handle); | |
4073 | ||
4074 | return 0; | |
4075 | } | |
4076 | ||
a4bb6b64 | 4077 | /* |
cca32b7e | 4078 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
4079 | * associated with the given offset and length |
4080 | * | |
4081 | * @inode: File inode | |
4082 | * @offset: The offset where the hole will begin | |
4083 | * @len: The length of the hole | |
4084 | * | |
4907cb7b | 4085 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
4086 | */ |
4087 | ||
aeb2817a | 4088 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 4089 | { |
26a4c0c6 TT |
4090 | struct super_block *sb = inode->i_sb; |
4091 | ext4_lblk_t first_block, stop_block; | |
4092 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 4093 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
4094 | handle_t *handle; |
4095 | unsigned int credits; | |
4096 | int ret = 0; | |
4097 | ||
a4bb6b64 | 4098 | if (!S_ISREG(inode->i_mode)) |
73355192 | 4099 | return -EOPNOTSUPP; |
a4bb6b64 | 4100 | |
b8a86845 | 4101 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4102 | |
26a4c0c6 TT |
4103 | /* |
4104 | * Write out all dirty pages to avoid race conditions | |
4105 | * Then release them. | |
4106 | */ | |
cca32b7e | 4107 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4108 | ret = filemap_write_and_wait_range(mapping, offset, |
4109 | offset + length - 1); | |
4110 | if (ret) | |
4111 | return ret; | |
4112 | } | |
4113 | ||
5955102c | 4114 | inode_lock(inode); |
9ef06cec | 4115 | |
26a4c0c6 TT |
4116 | /* No need to punch hole beyond i_size */ |
4117 | if (offset >= inode->i_size) | |
4118 | goto out_mutex; | |
4119 | ||
4120 | /* | |
4121 | * If the hole extends beyond i_size, set the hole | |
4122 | * to end after the page that contains i_size | |
4123 | */ | |
4124 | if (offset + length > inode->i_size) { | |
4125 | length = inode->i_size + | |
09cbfeaf | 4126 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4127 | offset; |
4128 | } | |
4129 | ||
a361293f JK |
4130 | if (offset & (sb->s_blocksize - 1) || |
4131 | (offset + length) & (sb->s_blocksize - 1)) { | |
4132 | /* | |
4133 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4134 | * partial block | |
4135 | */ | |
4136 | ret = ext4_inode_attach_jinode(inode); | |
4137 | if (ret < 0) | |
4138 | goto out_mutex; | |
4139 | ||
4140 | } | |
4141 | ||
ea3d7209 JK |
4142 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
4143 | ext4_inode_block_unlocked_dio(inode); | |
4144 | inode_dio_wait(inode); | |
4145 | ||
4146 | /* | |
4147 | * Prevent page faults from reinstantiating pages we have released from | |
4148 | * page cache. | |
4149 | */ | |
4150 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
4151 | first_block_offset = round_up(offset, sb->s_blocksize); |
4152 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4153 | |
a87dd18c | 4154 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4155 | if (last_block_offset > first_block_offset) { |
4156 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4157 | if (ret) | |
4158 | goto out_dio; | |
a87dd18c LC |
4159 | truncate_pagecache_range(inode, first_block_offset, |
4160 | last_block_offset); | |
01127848 | 4161 | } |
26a4c0c6 TT |
4162 | |
4163 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4164 | credits = ext4_writepage_trans_blocks(inode); | |
4165 | else | |
4166 | credits = ext4_blocks_for_truncate(inode); | |
4167 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4168 | if (IS_ERR(handle)) { | |
4169 | ret = PTR_ERR(handle); | |
4170 | ext4_std_error(sb, ret); | |
4171 | goto out_dio; | |
4172 | } | |
4173 | ||
a87dd18c LC |
4174 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4175 | length); | |
4176 | if (ret) | |
4177 | goto out_stop; | |
26a4c0c6 TT |
4178 | |
4179 | first_block = (offset + sb->s_blocksize - 1) >> | |
4180 | EXT4_BLOCK_SIZE_BITS(sb); | |
4181 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4182 | ||
4183 | /* If there are no blocks to remove, return now */ | |
4184 | if (first_block >= stop_block) | |
4185 | goto out_stop; | |
4186 | ||
4187 | down_write(&EXT4_I(inode)->i_data_sem); | |
4188 | ext4_discard_preallocations(inode); | |
4189 | ||
4190 | ret = ext4_es_remove_extent(inode, first_block, | |
4191 | stop_block - first_block); | |
4192 | if (ret) { | |
4193 | up_write(&EXT4_I(inode)->i_data_sem); | |
4194 | goto out_stop; | |
4195 | } | |
4196 | ||
4197 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4198 | ret = ext4_ext_remove_space(inode, first_block, | |
4199 | stop_block - 1); | |
4200 | else | |
4f579ae7 | 4201 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
4202 | stop_block); |
4203 | ||
819c4920 | 4204 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
4205 | if (IS_SYNC(inode)) |
4206 | ext4_handle_sync(handle); | |
e251f9bc | 4207 | |
eeca7ea1 | 4208 | inode->i_mtime = inode->i_ctime = current_time(inode); |
26a4c0c6 TT |
4209 | ext4_mark_inode_dirty(handle, inode); |
4210 | out_stop: | |
4211 | ext4_journal_stop(handle); | |
4212 | out_dio: | |
ea3d7209 | 4213 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 TT |
4214 | ext4_inode_resume_unlocked_dio(inode); |
4215 | out_mutex: | |
5955102c | 4216 | inode_unlock(inode); |
26a4c0c6 | 4217 | return ret; |
a4bb6b64 AH |
4218 | } |
4219 | ||
a361293f JK |
4220 | int ext4_inode_attach_jinode(struct inode *inode) |
4221 | { | |
4222 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4223 | struct jbd2_inode *jinode; | |
4224 | ||
4225 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4226 | return 0; | |
4227 | ||
4228 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4229 | spin_lock(&inode->i_lock); | |
4230 | if (!ei->jinode) { | |
4231 | if (!jinode) { | |
4232 | spin_unlock(&inode->i_lock); | |
4233 | return -ENOMEM; | |
4234 | } | |
4235 | ei->jinode = jinode; | |
4236 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4237 | jinode = NULL; | |
4238 | } | |
4239 | spin_unlock(&inode->i_lock); | |
4240 | if (unlikely(jinode != NULL)) | |
4241 | jbd2_free_inode(jinode); | |
4242 | return 0; | |
4243 | } | |
4244 | ||
ac27a0ec | 4245 | /* |
617ba13b | 4246 | * ext4_truncate() |
ac27a0ec | 4247 | * |
617ba13b MC |
4248 | * We block out ext4_get_block() block instantiations across the entire |
4249 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4250 | * simultaneously on behalf of the same inode. |
4251 | * | |
42b2aa86 | 4252 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4253 | * is one core, guiding principle: the file's tree must always be consistent on |
4254 | * disk. We must be able to restart the truncate after a crash. | |
4255 | * | |
4256 | * The file's tree may be transiently inconsistent in memory (although it | |
4257 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4258 | * the contents of (the filesystem + the journal) must be consistent and | |
4259 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4260 | * left-to-right works OK too). | |
4261 | * | |
4262 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4263 | * truncate against the orphan inode list. | |
4264 | * | |
4265 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4266 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4267 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4268 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4269 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4270 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4271 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4272 | */ |
2c98eb5e | 4273 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4274 | { |
819c4920 TT |
4275 | struct ext4_inode_info *ei = EXT4_I(inode); |
4276 | unsigned int credits; | |
2c98eb5e | 4277 | int err = 0; |
819c4920 TT |
4278 | handle_t *handle; |
4279 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4280 | |
19b5ef61 TT |
4281 | /* |
4282 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4283 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4284 | * have i_mutex locked because it's not necessary. |
4285 | */ | |
4286 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4287 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4288 | trace_ext4_truncate_enter(inode); |
4289 | ||
91ef4caf | 4290 | if (!ext4_can_truncate(inode)) |
2c98eb5e | 4291 | return 0; |
ac27a0ec | 4292 | |
12e9b892 | 4293 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4294 | |
5534fb5b | 4295 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4296 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4297 | |
aef1c851 TM |
4298 | if (ext4_has_inline_data(inode)) { |
4299 | int has_inline = 1; | |
4300 | ||
01daf945 TT |
4301 | err = ext4_inline_data_truncate(inode, &has_inline); |
4302 | if (err) | |
4303 | return err; | |
aef1c851 | 4304 | if (has_inline) |
2c98eb5e | 4305 | return 0; |
aef1c851 TM |
4306 | } |
4307 | ||
a361293f JK |
4308 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4309 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4310 | if (ext4_inode_attach_jinode(inode) < 0) | |
2c98eb5e | 4311 | return 0; |
a361293f JK |
4312 | } |
4313 | ||
819c4920 TT |
4314 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4315 | credits = ext4_writepage_trans_blocks(inode); | |
4316 | else | |
4317 | credits = ext4_blocks_for_truncate(inode); | |
4318 | ||
4319 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
2c98eb5e TT |
4320 | if (IS_ERR(handle)) |
4321 | return PTR_ERR(handle); | |
819c4920 | 4322 | |
eb3544c6 LC |
4323 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4324 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4325 | |
4326 | /* | |
4327 | * We add the inode to the orphan list, so that if this | |
4328 | * truncate spans multiple transactions, and we crash, we will | |
4329 | * resume the truncate when the filesystem recovers. It also | |
4330 | * marks the inode dirty, to catch the new size. | |
4331 | * | |
4332 | * Implication: the file must always be in a sane, consistent | |
4333 | * truncatable state while each transaction commits. | |
4334 | */ | |
2c98eb5e TT |
4335 | err = ext4_orphan_add(handle, inode); |
4336 | if (err) | |
819c4920 TT |
4337 | goto out_stop; |
4338 | ||
4339 | down_write(&EXT4_I(inode)->i_data_sem); | |
4340 | ||
4341 | ext4_discard_preallocations(inode); | |
4342 | ||
ff9893dc | 4343 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4344 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4345 | else |
819c4920 TT |
4346 | ext4_ind_truncate(handle, inode); |
4347 | ||
4348 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4349 | if (err) |
4350 | goto out_stop; | |
819c4920 TT |
4351 | |
4352 | if (IS_SYNC(inode)) | |
4353 | ext4_handle_sync(handle); | |
4354 | ||
4355 | out_stop: | |
4356 | /* | |
4357 | * If this was a simple ftruncate() and the file will remain alive, | |
4358 | * then we need to clear up the orphan record which we created above. | |
4359 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4360 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4361 | * orphan info for us. |
4362 | */ | |
4363 | if (inode->i_nlink) | |
4364 | ext4_orphan_del(handle, inode); | |
4365 | ||
eeca7ea1 | 4366 | inode->i_mtime = inode->i_ctime = current_time(inode); |
819c4920 TT |
4367 | ext4_mark_inode_dirty(handle, inode); |
4368 | ext4_journal_stop(handle); | |
ac27a0ec | 4369 | |
0562e0ba | 4370 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4371 | return err; |
ac27a0ec DK |
4372 | } |
4373 | ||
ac27a0ec | 4374 | /* |
617ba13b | 4375 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4376 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4377 | * data in memory that is needed to recreate the on-disk version of this | |
4378 | * inode. | |
4379 | */ | |
617ba13b MC |
4380 | static int __ext4_get_inode_loc(struct inode *inode, |
4381 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4382 | { |
240799cd TT |
4383 | struct ext4_group_desc *gdp; |
4384 | struct buffer_head *bh; | |
4385 | struct super_block *sb = inode->i_sb; | |
4386 | ext4_fsblk_t block; | |
4387 | int inodes_per_block, inode_offset; | |
4388 | ||
3a06d778 | 4389 | iloc->bh = NULL; |
240799cd | 4390 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 4391 | return -EFSCORRUPTED; |
ac27a0ec | 4392 | |
240799cd TT |
4393 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4394 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4395 | if (!gdp) | |
ac27a0ec DK |
4396 | return -EIO; |
4397 | ||
240799cd TT |
4398 | /* |
4399 | * Figure out the offset within the block group inode table | |
4400 | */ | |
00d09882 | 4401 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4402 | inode_offset = ((inode->i_ino - 1) % |
4403 | EXT4_INODES_PER_GROUP(sb)); | |
4404 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4405 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4406 | ||
4407 | bh = sb_getblk(sb, block); | |
aebf0243 | 4408 | if (unlikely(!bh)) |
860d21e2 | 4409 | return -ENOMEM; |
ac27a0ec DK |
4410 | if (!buffer_uptodate(bh)) { |
4411 | lock_buffer(bh); | |
9c83a923 HK |
4412 | |
4413 | /* | |
4414 | * If the buffer has the write error flag, we have failed | |
4415 | * to write out another inode in the same block. In this | |
4416 | * case, we don't have to read the block because we may | |
4417 | * read the old inode data successfully. | |
4418 | */ | |
4419 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4420 | set_buffer_uptodate(bh); | |
4421 | ||
ac27a0ec DK |
4422 | if (buffer_uptodate(bh)) { |
4423 | /* someone brought it uptodate while we waited */ | |
4424 | unlock_buffer(bh); | |
4425 | goto has_buffer; | |
4426 | } | |
4427 | ||
4428 | /* | |
4429 | * If we have all information of the inode in memory and this | |
4430 | * is the only valid inode in the block, we need not read the | |
4431 | * block. | |
4432 | */ | |
4433 | if (in_mem) { | |
4434 | struct buffer_head *bitmap_bh; | |
240799cd | 4435 | int i, start; |
ac27a0ec | 4436 | |
240799cd | 4437 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4438 | |
240799cd TT |
4439 | /* Is the inode bitmap in cache? */ |
4440 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4441 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4442 | goto make_io; |
4443 | ||
4444 | /* | |
4445 | * If the inode bitmap isn't in cache then the | |
4446 | * optimisation may end up performing two reads instead | |
4447 | * of one, so skip it. | |
4448 | */ | |
4449 | if (!buffer_uptodate(bitmap_bh)) { | |
4450 | brelse(bitmap_bh); | |
4451 | goto make_io; | |
4452 | } | |
240799cd | 4453 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4454 | if (i == inode_offset) |
4455 | continue; | |
617ba13b | 4456 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4457 | break; |
4458 | } | |
4459 | brelse(bitmap_bh); | |
240799cd | 4460 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4461 | /* all other inodes are free, so skip I/O */ |
4462 | memset(bh->b_data, 0, bh->b_size); | |
4463 | set_buffer_uptodate(bh); | |
4464 | unlock_buffer(bh); | |
4465 | goto has_buffer; | |
4466 | } | |
4467 | } | |
4468 | ||
4469 | make_io: | |
240799cd TT |
4470 | /* |
4471 | * If we need to do any I/O, try to pre-readahead extra | |
4472 | * blocks from the inode table. | |
4473 | */ | |
4474 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4475 | ext4_fsblk_t b, end, table; | |
4476 | unsigned num; | |
0d606e2c | 4477 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4478 | |
4479 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4480 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4481 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4482 | if (table > b) |
4483 | b = table; | |
0d606e2c | 4484 | end = b + ra_blks; |
240799cd | 4485 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4486 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4487 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4488 | table += num / inodes_per_block; |
4489 | if (end > table) | |
4490 | end = table; | |
4491 | while (b <= end) | |
4492 | sb_breadahead(sb, b++); | |
4493 | } | |
4494 | ||
ac27a0ec DK |
4495 | /* |
4496 | * There are other valid inodes in the buffer, this inode | |
4497 | * has in-inode xattrs, or we don't have this inode in memory. | |
4498 | * Read the block from disk. | |
4499 | */ | |
0562e0ba | 4500 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4501 | get_bh(bh); |
4502 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4503 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4504 | wait_on_buffer(bh); |
4505 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4506 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4507 | "unable to read itable block"); | |
ac27a0ec DK |
4508 | brelse(bh); |
4509 | return -EIO; | |
4510 | } | |
4511 | } | |
4512 | has_buffer: | |
4513 | iloc->bh = bh; | |
4514 | return 0; | |
4515 | } | |
4516 | ||
617ba13b | 4517 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4518 | { |
4519 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4520 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4521 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4522 | } |
4523 | ||
617ba13b | 4524 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4525 | { |
617ba13b | 4526 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4527 | unsigned int new_fl = 0; |
ac27a0ec | 4528 | |
617ba13b | 4529 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4530 | new_fl |= S_SYNC; |
617ba13b | 4531 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4532 | new_fl |= S_APPEND; |
617ba13b | 4533 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4534 | new_fl |= S_IMMUTABLE; |
617ba13b | 4535 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4536 | new_fl |= S_NOATIME; |
617ba13b | 4537 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4538 | new_fl |= S_DIRSYNC; |
a3caa24b JK |
4539 | if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) && |
4540 | !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) && | |
4541 | !ext4_encrypted_inode(inode)) | |
923ae0ff | 4542 | new_fl |= S_DAX; |
5f16f322 | 4543 | inode_set_flags(inode, new_fl, |
923ae0ff | 4544 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
4545 | } |
4546 | ||
0fc1b451 | 4547 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4548 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4549 | { |
4550 | blkcnt_t i_blocks ; | |
8180a562 AK |
4551 | struct inode *inode = &(ei->vfs_inode); |
4552 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4553 | |
e2b911c5 | 4554 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4555 | /* we are using combined 48 bit field */ |
4556 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4557 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4558 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4559 | /* i_blocks represent file system block size */ |
4560 | return i_blocks << (inode->i_blkbits - 9); | |
4561 | } else { | |
4562 | return i_blocks; | |
4563 | } | |
0fc1b451 AK |
4564 | } else { |
4565 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4566 | } | |
4567 | } | |
ff9ddf7e | 4568 | |
152a7b0a TM |
4569 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4570 | struct ext4_inode *raw_inode, | |
4571 | struct ext4_inode_info *ei) | |
4572 | { | |
4573 | __le32 *magic = (void *)raw_inode + | |
4574 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
290ab230 EB |
4575 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4576 | EXT4_INODE_SIZE(inode->i_sb) && | |
4577 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4578 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4579 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4580 | } else |
4581 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4582 | } |
4583 | ||
040cb378 LX |
4584 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4585 | { | |
0b7b7779 | 4586 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4587 | return -EOPNOTSUPP; |
4588 | *projid = EXT4_I(inode)->i_projid; | |
4589 | return 0; | |
4590 | } | |
4591 | ||
1d1fe1ee | 4592 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4593 | { |
617ba13b MC |
4594 | struct ext4_iloc iloc; |
4595 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4596 | struct ext4_inode_info *ei; |
1d1fe1ee | 4597 | struct inode *inode; |
b436b9be | 4598 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4599 | long ret; |
7e6e1ef4 | 4600 | loff_t size; |
ac27a0ec | 4601 | int block; |
08cefc7a EB |
4602 | uid_t i_uid; |
4603 | gid_t i_gid; | |
040cb378 | 4604 | projid_t i_projid; |
ac27a0ec | 4605 | |
1d1fe1ee DH |
4606 | inode = iget_locked(sb, ino); |
4607 | if (!inode) | |
4608 | return ERR_PTR(-ENOMEM); | |
4609 | if (!(inode->i_state & I_NEW)) | |
4610 | return inode; | |
4611 | ||
4612 | ei = EXT4_I(inode); | |
7dc57615 | 4613 | iloc.bh = NULL; |
ac27a0ec | 4614 | |
1d1fe1ee DH |
4615 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4616 | if (ret < 0) | |
ac27a0ec | 4617 | goto bad_inode; |
617ba13b | 4618 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4619 | |
4620 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4621 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4622 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4623 | EXT4_INODE_SIZE(inode->i_sb) || |
4624 | (ei->i_extra_isize & 3)) { | |
4625 | EXT4_ERROR_INODE(inode, | |
4626 | "bad extra_isize %u (inode size %u)", | |
4627 | ei->i_extra_isize, | |
4628 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4629 | ret = -EFSCORRUPTED; |
814525f4 DW |
4630 | goto bad_inode; |
4631 | } | |
4632 | } else | |
4633 | ei->i_extra_isize = 0; | |
4634 | ||
4635 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4636 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4637 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4638 | __u32 csum; | |
4639 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4640 | __le32 gen = raw_inode->i_generation; | |
4641 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4642 | sizeof(inum)); | |
4643 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4644 | sizeof(gen)); | |
4645 | } | |
4646 | ||
4647 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4648 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4649 | ret = -EFSBADCRC; |
814525f4 DW |
4650 | goto bad_inode; |
4651 | } | |
4652 | ||
ac27a0ec | 4653 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4654 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4655 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4656 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4657 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4658 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4659 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4660 | else | |
4661 | i_projid = EXT4_DEF_PROJID; | |
4662 | ||
af5bc92d | 4663 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4664 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4665 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4666 | } |
08cefc7a EB |
4667 | i_uid_write(inode, i_uid); |
4668 | i_gid_write(inode, i_gid); | |
040cb378 | 4669 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4670 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4671 | |
353eb83c | 4672 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4673 | ei->i_inline_off = 0; |
ac27a0ec DK |
4674 | ei->i_dir_start_lookup = 0; |
4675 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4676 | /* We now have enough fields to check if the inode was active or not. | |
4677 | * This is needed because nfsd might try to access dead inodes | |
4678 | * the test is that same one that e2fsck uses | |
4679 | * NeilBrown 1999oct15 | |
4680 | */ | |
4681 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4682 | if ((inode->i_mode == 0 || |
4683 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4684 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4685 | /* this inode is deleted */ |
1d1fe1ee | 4686 | ret = -ESTALE; |
ac27a0ec DK |
4687 | goto bad_inode; |
4688 | } | |
4689 | /* The only unlinked inodes we let through here have | |
4690 | * valid i_mode and are being read by the orphan | |
4691 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4692 | * the process of deleting those. |
4693 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4694 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4695 | } |
ac27a0ec | 4696 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4697 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4698 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4699 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4700 | ei->i_file_acl |= |
4701 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4702 | inode->i_size = ext4_isize(raw_inode); |
7e6e1ef4 DW |
4703 | if ((size = i_size_read(inode)) < 0) { |
4704 | EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size); | |
4705 | ret = -EFSCORRUPTED; | |
4706 | goto bad_inode; | |
4707 | } | |
ac27a0ec | 4708 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4709 | #ifdef CONFIG_QUOTA |
4710 | ei->i_reserved_quota = 0; | |
4711 | #endif | |
ac27a0ec DK |
4712 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4713 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4714 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4715 | /* |
4716 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4717 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4718 | */ | |
617ba13b | 4719 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4720 | ei->i_data[block] = raw_inode->i_block[block]; |
4721 | INIT_LIST_HEAD(&ei->i_orphan); | |
4722 | ||
b436b9be JK |
4723 | /* |
4724 | * Set transaction id's of transactions that have to be committed | |
4725 | * to finish f[data]sync. We set them to currently running transaction | |
4726 | * as we cannot be sure that the inode or some of its metadata isn't | |
4727 | * part of the transaction - the inode could have been reclaimed and | |
4728 | * now it is reread from disk. | |
4729 | */ | |
4730 | if (journal) { | |
4731 | transaction_t *transaction; | |
4732 | tid_t tid; | |
4733 | ||
a931da6a | 4734 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4735 | if (journal->j_running_transaction) |
4736 | transaction = journal->j_running_transaction; | |
4737 | else | |
4738 | transaction = journal->j_committing_transaction; | |
4739 | if (transaction) | |
4740 | tid = transaction->t_tid; | |
4741 | else | |
4742 | tid = journal->j_commit_sequence; | |
a931da6a | 4743 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4744 | ei->i_sync_tid = tid; |
4745 | ei->i_datasync_tid = tid; | |
4746 | } | |
4747 | ||
0040d987 | 4748 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4749 | if (ei->i_extra_isize == 0) { |
4750 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4751 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4752 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4753 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4754 | } else { |
152a7b0a | 4755 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4756 | } |
814525f4 | 4757 | } |
ac27a0ec | 4758 | |
ef7f3835 KS |
4759 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4760 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4761 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4762 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4763 | ||
ed3654eb | 4764 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4765 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4766 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4767 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4768 | inode->i_version |= | |
4769 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4770 | } | |
25ec56b5 JNC |
4771 | } |
4772 | ||
c4b5a614 | 4773 | ret = 0; |
485c26ec | 4774 | if (ei->i_file_acl && |
1032988c | 4775 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4776 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4777 | ei->i_file_acl); | |
6a797d27 | 4778 | ret = -EFSCORRUPTED; |
485c26ec | 4779 | goto bad_inode; |
f19d5870 TM |
4780 | } else if (!ext4_has_inline_data(inode)) { |
4781 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4782 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4783 | (S_ISLNK(inode->i_mode) && | |
4784 | !ext4_inode_is_fast_symlink(inode)))) | |
4785 | /* Validate extent which is part of inode */ | |
4786 | ret = ext4_ext_check_inode(inode); | |
4787 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4788 | (S_ISLNK(inode->i_mode) && | |
4789 | !ext4_inode_is_fast_symlink(inode))) { | |
4790 | /* Validate block references which are part of inode */ | |
4791 | ret = ext4_ind_check_inode(inode); | |
4792 | } | |
fe2c8191 | 4793 | } |
567f3e9a | 4794 | if (ret) |
de9a55b8 | 4795 | goto bad_inode; |
7a262f7c | 4796 | |
ac27a0ec | 4797 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4798 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4799 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4800 | ext4_set_aops(inode); |
ac27a0ec | 4801 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4802 | inode->i_op = &ext4_dir_inode_operations; |
4803 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4804 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4805 | if (ext4_encrypted_inode(inode)) { |
4806 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4807 | ext4_set_aops(inode); | |
4808 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4809 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4810 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4811 | nd_terminate_link(ei->i_data, inode->i_size, |
4812 | sizeof(ei->i_data) - 1); | |
4813 | } else { | |
617ba13b MC |
4814 | inode->i_op = &ext4_symlink_inode_operations; |
4815 | ext4_set_aops(inode); | |
ac27a0ec | 4816 | } |
21fc61c7 | 4817 | inode_nohighmem(inode); |
563bdd61 TT |
4818 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4819 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4820 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4821 | if (raw_inode->i_block[0]) |
4822 | init_special_inode(inode, inode->i_mode, | |
4823 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4824 | else | |
4825 | init_special_inode(inode, inode->i_mode, | |
4826 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4827 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4828 | make_bad_inode(inode); | |
563bdd61 | 4829 | } else { |
6a797d27 | 4830 | ret = -EFSCORRUPTED; |
24676da4 | 4831 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4832 | goto bad_inode; |
ac27a0ec | 4833 | } |
af5bc92d | 4834 | brelse(iloc.bh); |
617ba13b | 4835 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4836 | unlock_new_inode(inode); |
4837 | return inode; | |
ac27a0ec DK |
4838 | |
4839 | bad_inode: | |
567f3e9a | 4840 | brelse(iloc.bh); |
1d1fe1ee DH |
4841 | iget_failed(inode); |
4842 | return ERR_PTR(ret); | |
ac27a0ec DK |
4843 | } |
4844 | ||
f4bb2981 TT |
4845 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4846 | { | |
4847 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4848 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4849 | return ext4_iget(sb, ino); |
4850 | } | |
4851 | ||
0fc1b451 AK |
4852 | static int ext4_inode_blocks_set(handle_t *handle, |
4853 | struct ext4_inode *raw_inode, | |
4854 | struct ext4_inode_info *ei) | |
4855 | { | |
4856 | struct inode *inode = &(ei->vfs_inode); | |
4857 | u64 i_blocks = inode->i_blocks; | |
4858 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4859 | |
4860 | if (i_blocks <= ~0U) { | |
4861 | /* | |
4907cb7b | 4862 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4863 | * as multiple of 512 bytes |
4864 | */ | |
8180a562 | 4865 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4866 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4867 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4868 | return 0; |
4869 | } | |
e2b911c5 | 4870 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4871 | return -EFBIG; |
4872 | ||
4873 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4874 | /* |
4875 | * i_blocks can be represented in a 48 bit variable | |
4876 | * as multiple of 512 bytes | |
4877 | */ | |
8180a562 | 4878 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4879 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4880 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4881 | } else { |
84a8dce2 | 4882 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4883 | /* i_block is stored in file system block size */ |
4884 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4885 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4886 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4887 | } |
f287a1a5 | 4888 | return 0; |
0fc1b451 AK |
4889 | } |
4890 | ||
a26f4992 TT |
4891 | struct other_inode { |
4892 | unsigned long orig_ino; | |
4893 | struct ext4_inode *raw_inode; | |
4894 | }; | |
4895 | ||
4896 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4897 | void *data) | |
4898 | { | |
4899 | struct other_inode *oi = (struct other_inode *) data; | |
4900 | ||
4901 | if ((inode->i_ino != ino) || | |
4902 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4903 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4904 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4905 | return 0; | |
4906 | spin_lock(&inode->i_lock); | |
4907 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4908 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4909 | (inode->i_state & I_DIRTY_TIME)) { | |
4910 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4911 | ||
4912 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4913 | spin_unlock(&inode->i_lock); | |
4914 | ||
4915 | spin_lock(&ei->i_raw_lock); | |
4916 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4917 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4918 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4919 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4920 | spin_unlock(&ei->i_raw_lock); | |
4921 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4922 | return -1; | |
4923 | } | |
4924 | spin_unlock(&inode->i_lock); | |
4925 | return -1; | |
4926 | } | |
4927 | ||
4928 | /* | |
4929 | * Opportunistically update the other time fields for other inodes in | |
4930 | * the same inode table block. | |
4931 | */ | |
4932 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4933 | unsigned long orig_ino, char *buf) | |
4934 | { | |
4935 | struct other_inode oi; | |
4936 | unsigned long ino; | |
4937 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4938 | int inode_size = EXT4_INODE_SIZE(sb); | |
4939 | ||
4940 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
4941 | /* |
4942 | * Calculate the first inode in the inode table block. Inode | |
4943 | * numbers are one-based. That is, the first inode in a block | |
4944 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4945 | */ | |
4946 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
4947 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4948 | if (ino == orig_ino) | |
4949 | continue; | |
4950 | oi.raw_inode = (struct ext4_inode *) buf; | |
4951 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4952 | } | |
4953 | } | |
4954 | ||
ac27a0ec DK |
4955 | /* |
4956 | * Post the struct inode info into an on-disk inode location in the | |
4957 | * buffer-cache. This gobbles the caller's reference to the | |
4958 | * buffer_head in the inode location struct. | |
4959 | * | |
4960 | * The caller must have write access to iloc->bh. | |
4961 | */ | |
617ba13b | 4962 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4963 | struct inode *inode, |
830156c7 | 4964 | struct ext4_iloc *iloc) |
ac27a0ec | 4965 | { |
617ba13b MC |
4966 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4967 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4968 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4969 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4970 | int err = 0, rc, block; |
202ee5df | 4971 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4972 | uid_t i_uid; |
4973 | gid_t i_gid; | |
040cb378 | 4974 | projid_t i_projid; |
ac27a0ec | 4975 | |
202ee5df TT |
4976 | spin_lock(&ei->i_raw_lock); |
4977 | ||
4978 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4979 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4980 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4981 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec DK |
4982 | |
4983 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
08cefc7a EB |
4984 | i_uid = i_uid_read(inode); |
4985 | i_gid = i_gid_read(inode); | |
040cb378 | 4986 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 4987 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4988 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4989 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4990 | /* |
4991 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4992 | * re-used with the upper 16 bits of the uid/gid intact | |
4993 | */ | |
93e3b4e6 DJ |
4994 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
4995 | raw_inode->i_uid_high = 0; | |
4996 | raw_inode->i_gid_high = 0; | |
4997 | } else { | |
ac27a0ec | 4998 | raw_inode->i_uid_high = |
08cefc7a | 4999 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5000 | raw_inode->i_gid_high = |
08cefc7a | 5001 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5002 | } |
5003 | } else { | |
08cefc7a EB |
5004 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5005 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5006 | raw_inode->i_uid_high = 0; |
5007 | raw_inode->i_gid_high = 0; | |
5008 | } | |
5009 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5010 | |
5011 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5012 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5013 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5014 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5015 | ||
bce92d56 LX |
5016 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
5017 | if (err) { | |
202ee5df | 5018 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 5019 | goto out_brelse; |
202ee5df | 5020 | } |
ac27a0ec | 5021 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5022 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5023 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5024 | raw_inode->i_file_acl_high = |
5025 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5026 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
5027 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
5028 | ext4_isize_set(raw_inode, ei->i_disksize); | |
5029 | need_datasync = 1; | |
5030 | } | |
a48380f7 | 5031 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5032 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5033 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5034 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5035 | set_large_file = 1; | |
ac27a0ec DK |
5036 | } |
5037 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5038 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5039 | if (old_valid_dev(inode->i_rdev)) { | |
5040 | raw_inode->i_block[0] = | |
5041 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5042 | raw_inode->i_block[1] = 0; | |
5043 | } else { | |
5044 | raw_inode->i_block[0] = 0; | |
5045 | raw_inode->i_block[1] = | |
5046 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5047 | raw_inode->i_block[2] = 0; | |
5048 | } | |
f19d5870 | 5049 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5050 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5051 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5052 | } |
ac27a0ec | 5053 | |
ed3654eb | 5054 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
5055 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
5056 | if (ei->i_extra_isize) { | |
5057 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5058 | raw_inode->i_version_hi = | |
5059 | cpu_to_le32(inode->i_version >> 32); | |
5060 | raw_inode->i_extra_isize = | |
5061 | cpu_to_le16(ei->i_extra_isize); | |
5062 | } | |
25ec56b5 | 5063 | } |
040cb378 | 5064 | |
0b7b7779 | 5065 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
5066 | i_projid != EXT4_DEF_PROJID); |
5067 | ||
5068 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5069 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5070 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5071 | ||
814525f4 | 5072 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5073 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
5074 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
5075 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
5076 | bh->b_data); | |
202ee5df | 5077 | |
830156c7 | 5078 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5079 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5080 | if (!err) |
5081 | err = rc; | |
19f5fb7a | 5082 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5083 | if (set_large_file) { |
5d601255 | 5084 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
5085 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
5086 | if (err) | |
5087 | goto out_brelse; | |
5088 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 5089 | ext4_set_feature_large_file(sb); |
202ee5df TT |
5090 | ext4_handle_sync(handle); |
5091 | err = ext4_handle_dirty_super(handle, sb); | |
5092 | } | |
b71fc079 | 5093 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5094 | out_brelse: |
af5bc92d | 5095 | brelse(bh); |
617ba13b | 5096 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5097 | return err; |
5098 | } | |
5099 | ||
5100 | /* | |
617ba13b | 5101 | * ext4_write_inode() |
ac27a0ec DK |
5102 | * |
5103 | * We are called from a few places: | |
5104 | * | |
87f7e416 | 5105 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5106 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5107 | * transaction to commit. |
ac27a0ec | 5108 | * |
87f7e416 TT |
5109 | * - Within flush work (sys_sync(), kupdate and such). |
5110 | * We wait on commit, if told to. | |
ac27a0ec | 5111 | * |
87f7e416 TT |
5112 | * - Within iput_final() -> write_inode_now() |
5113 | * We wait on commit, if told to. | |
ac27a0ec DK |
5114 | * |
5115 | * In all cases it is actually safe for us to return without doing anything, | |
5116 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5117 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5118 | * writeback. | |
ac27a0ec DK |
5119 | * |
5120 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5121 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5122 | * which we are interested. | |
5123 | * | |
5124 | * It would be a bug for them to not do this. The code: | |
5125 | * | |
5126 | * mark_inode_dirty(inode) | |
5127 | * stuff(); | |
5128 | * inode->i_size = expr; | |
5129 | * | |
87f7e416 TT |
5130 | * is in error because write_inode() could occur while `stuff()' is running, |
5131 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5132 | * superblock's dirty inode list. | |
ac27a0ec | 5133 | */ |
a9185b41 | 5134 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5135 | { |
91ac6f43 FM |
5136 | int err; |
5137 | ||
87f7e416 | 5138 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
5139 | return 0; |
5140 | ||
91ac6f43 FM |
5141 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5142 | if (ext4_journal_current_handle()) { | |
5143 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5144 | dump_stack(); | |
5145 | return -EIO; | |
5146 | } | |
ac27a0ec | 5147 | |
10542c22 JK |
5148 | /* |
5149 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5150 | * ext4_sync_fs() will force the commit after everything is | |
5151 | * written. | |
5152 | */ | |
5153 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5154 | return 0; |
5155 | ||
5156 | err = ext4_force_commit(inode->i_sb); | |
5157 | } else { | |
5158 | struct ext4_iloc iloc; | |
ac27a0ec | 5159 | |
8b472d73 | 5160 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5161 | if (err) |
5162 | return err; | |
10542c22 JK |
5163 | /* |
5164 | * sync(2) will flush the whole buffer cache. No need to do | |
5165 | * it here separately for each inode. | |
5166 | */ | |
5167 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5168 | sync_dirty_buffer(iloc.bh); |
5169 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5170 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5171 | "IO error syncing inode"); | |
830156c7 FM |
5172 | err = -EIO; |
5173 | } | |
fd2dd9fb | 5174 | brelse(iloc.bh); |
91ac6f43 FM |
5175 | } |
5176 | return err; | |
ac27a0ec DK |
5177 | } |
5178 | ||
53e87268 JK |
5179 | /* |
5180 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5181 | * buffers that are attached to a page stradding i_size and are undergoing | |
5182 | * commit. In that case we have to wait for commit to finish and try again. | |
5183 | */ | |
5184 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5185 | { | |
5186 | struct page *page; | |
5187 | unsigned offset; | |
5188 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5189 | tid_t commit_tid = 0; | |
5190 | int ret; | |
5191 | ||
09cbfeaf | 5192 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 JK |
5193 | /* |
5194 | * All buffers in the last page remain valid? Then there's nothing to | |
ea1754a0 | 5195 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
53e87268 JK |
5196 | * blocksize case |
5197 | */ | |
93407472 | 5198 | if (offset > PAGE_SIZE - i_blocksize(inode)) |
53e87268 JK |
5199 | return; |
5200 | while (1) { | |
5201 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5202 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5203 | if (!page) |
5204 | return; | |
ca99fdd2 | 5205 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5206 | PAGE_SIZE - offset); |
53e87268 | 5207 | unlock_page(page); |
09cbfeaf | 5208 | put_page(page); |
53e87268 JK |
5209 | if (ret != -EBUSY) |
5210 | return; | |
5211 | commit_tid = 0; | |
5212 | read_lock(&journal->j_state_lock); | |
5213 | if (journal->j_committing_transaction) | |
5214 | commit_tid = journal->j_committing_transaction->t_tid; | |
5215 | read_unlock(&journal->j_state_lock); | |
5216 | if (commit_tid) | |
5217 | jbd2_log_wait_commit(journal, commit_tid); | |
5218 | } | |
5219 | } | |
5220 | ||
ac27a0ec | 5221 | /* |
617ba13b | 5222 | * ext4_setattr() |
ac27a0ec DK |
5223 | * |
5224 | * Called from notify_change. | |
5225 | * | |
5226 | * We want to trap VFS attempts to truncate the file as soon as | |
5227 | * possible. In particular, we want to make sure that when the VFS | |
5228 | * shrinks i_size, we put the inode on the orphan list and modify | |
5229 | * i_disksize immediately, so that during the subsequent flushing of | |
5230 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5231 | * commit will leave the blocks being flushed in an unused state on | |
5232 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5233 | * be freed, so we have a strong guarantee that no future commit will | |
5234 | * leave these blocks visible to the user.) | |
5235 | * | |
678aaf48 JK |
5236 | * Another thing we have to assure is that if we are in ordered mode |
5237 | * and inode is still attached to the committing transaction, we must | |
5238 | * we start writeout of all the dirty pages which are being truncated. | |
5239 | * This way we are sure that all the data written in the previous | |
5240 | * transaction are already on disk (truncate waits for pages under | |
5241 | * writeback). | |
5242 | * | |
5243 | * Called with inode->i_mutex down. | |
ac27a0ec | 5244 | */ |
617ba13b | 5245 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5246 | { |
2b0143b5 | 5247 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5248 | int error, rc = 0; |
3d287de3 | 5249 | int orphan = 0; |
ac27a0ec DK |
5250 | const unsigned int ia_valid = attr->ia_valid; |
5251 | ||
0db1ff22 TT |
5252 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5253 | return -EIO; | |
5254 | ||
31051c85 | 5255 | error = setattr_prepare(dentry, attr); |
ac27a0ec DK |
5256 | if (error) |
5257 | return error; | |
5258 | ||
a7cdadee JK |
5259 | if (is_quota_modification(inode, attr)) { |
5260 | error = dquot_initialize(inode); | |
5261 | if (error) | |
5262 | return error; | |
5263 | } | |
08cefc7a EB |
5264 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5265 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5266 | handle_t *handle; |
5267 | ||
5268 | /* (user+group)*(old+new) structure, inode write (sb, | |
5269 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5270 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5271 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5272 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5273 | if (IS_ERR(handle)) { |
5274 | error = PTR_ERR(handle); | |
5275 | goto err_out; | |
5276 | } | |
b43fa828 | 5277 | error = dquot_transfer(inode, attr); |
ac27a0ec | 5278 | if (error) { |
617ba13b | 5279 | ext4_journal_stop(handle); |
ac27a0ec DK |
5280 | return error; |
5281 | } | |
5282 | /* Update corresponding info in inode so that everything is in | |
5283 | * one transaction */ | |
5284 | if (attr->ia_valid & ATTR_UID) | |
5285 | inode->i_uid = attr->ia_uid; | |
5286 | if (attr->ia_valid & ATTR_GID) | |
5287 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5288 | error = ext4_mark_inode_dirty(handle, inode); |
5289 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5290 | } |
5291 | ||
3da40c7b | 5292 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5293 | handle_t *handle; |
3da40c7b JB |
5294 | loff_t oldsize = inode->i_size; |
5295 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 5296 | |
12e9b892 | 5297 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5298 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5299 | ||
0c095c7f TT |
5300 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5301 | return -EFBIG; | |
e2b46574 | 5302 | } |
3da40c7b JB |
5303 | if (!S_ISREG(inode->i_mode)) |
5304 | return -EINVAL; | |
dff6efc3 CH |
5305 | |
5306 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5307 | inode_inc_iversion(inode); | |
5308 | ||
3da40c7b | 5309 | if (ext4_should_order_data(inode) && |
5208386c | 5310 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 5311 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 5312 | attr->ia_size); |
3da40c7b JB |
5313 | if (error) |
5314 | goto err_out; | |
5315 | } | |
5316 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
5317 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5318 | if (IS_ERR(handle)) { | |
5319 | error = PTR_ERR(handle); | |
5320 | goto err_out; | |
5321 | } | |
3da40c7b | 5322 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5323 | error = ext4_orphan_add(handle, inode); |
5324 | orphan = 1; | |
5325 | } | |
911af577 EG |
5326 | /* |
5327 | * Update c/mtime on truncate up, ext4_truncate() will | |
5328 | * update c/mtime in shrink case below | |
5329 | */ | |
5330 | if (!shrink) { | |
eeca7ea1 | 5331 | inode->i_mtime = current_time(inode); |
911af577 EG |
5332 | inode->i_ctime = inode->i_mtime; |
5333 | } | |
90e775b7 | 5334 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5335 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5336 | rc = ext4_mark_inode_dirty(handle, inode); | |
5337 | if (!error) | |
5338 | error = rc; | |
90e775b7 JK |
5339 | /* |
5340 | * We have to update i_size under i_data_sem together | |
5341 | * with i_disksize to avoid races with writeback code | |
5342 | * running ext4_wb_update_i_disksize(). | |
5343 | */ | |
5344 | if (!error) | |
5345 | i_size_write(inode, attr->ia_size); | |
5346 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
5347 | ext4_journal_stop(handle); |
5348 | if (error) { | |
3da40c7b JB |
5349 | if (orphan) |
5350 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
5351 | goto err_out; |
5352 | } | |
d6320cbf | 5353 | } |
3da40c7b JB |
5354 | if (!shrink) |
5355 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 5356 | |
5208386c JK |
5357 | /* |
5358 | * Blocks are going to be removed from the inode. Wait | |
5359 | * for dio in flight. Temporarily disable | |
5360 | * dioread_nolock to prevent livelock. | |
5361 | */ | |
5362 | if (orphan) { | |
5363 | if (!ext4_should_journal_data(inode)) { | |
5364 | ext4_inode_block_unlocked_dio(inode); | |
5365 | inode_dio_wait(inode); | |
5366 | ext4_inode_resume_unlocked_dio(inode); | |
5367 | } else | |
5368 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 5369 | } |
ea3d7209 | 5370 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
5371 | /* |
5372 | * Truncate pagecache after we've waited for commit | |
5373 | * in data=journal mode to make pages freeable. | |
5374 | */ | |
923ae0ff | 5375 | truncate_pagecache(inode, inode->i_size); |
2c98eb5e TT |
5376 | if (shrink) { |
5377 | rc = ext4_truncate(inode); | |
5378 | if (rc) | |
5379 | error = rc; | |
5380 | } | |
ea3d7209 | 5381 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5382 | } |
ac27a0ec | 5383 | |
2c98eb5e | 5384 | if (!error) { |
1025774c CH |
5385 | setattr_copy(inode, attr); |
5386 | mark_inode_dirty(inode); | |
5387 | } | |
5388 | ||
5389 | /* | |
5390 | * If the call to ext4_truncate failed to get a transaction handle at | |
5391 | * all, we need to clean up the in-core orphan list manually. | |
5392 | */ | |
3d287de3 | 5393 | if (orphan && inode->i_nlink) |
617ba13b | 5394 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5395 | |
2c98eb5e | 5396 | if (!error && (ia_valid & ATTR_MODE)) |
64e178a7 | 5397 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5398 | |
5399 | err_out: | |
617ba13b | 5400 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5401 | if (!error) |
5402 | error = rc; | |
5403 | return error; | |
5404 | } | |
5405 | ||
a528d35e DH |
5406 | int ext4_getattr(const struct path *path, struct kstat *stat, |
5407 | u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5408 | { |
99652ea5 DH |
5409 | struct inode *inode = d_inode(path->dentry); |
5410 | struct ext4_inode *raw_inode; | |
5411 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5412 | unsigned int flags; | |
5413 | ||
5414 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
5415 | stat->result_mask |= STATX_BTIME; | |
5416 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5417 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5418 | } | |
5419 | ||
5420 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5421 | if (flags & EXT4_APPEND_FL) | |
5422 | stat->attributes |= STATX_ATTR_APPEND; | |
5423 | if (flags & EXT4_COMPR_FL) | |
5424 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5425 | if (flags & EXT4_ENCRYPT_FL) | |
5426 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5427 | if (flags & EXT4_IMMUTABLE_FL) | |
5428 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5429 | if (flags & EXT4_NODUMP_FL) | |
5430 | stat->attributes |= STATX_ATTR_NODUMP; | |
3e3398a0 | 5431 | |
3209f68b DH |
5432 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5433 | STATX_ATTR_COMPRESSED | | |
5434 | STATX_ATTR_ENCRYPTED | | |
5435 | STATX_ATTR_IMMUTABLE | | |
5436 | STATX_ATTR_NODUMP); | |
5437 | ||
3e3398a0 | 5438 | generic_fillattr(inode, stat); |
99652ea5 DH |
5439 | return 0; |
5440 | } | |
5441 | ||
5442 | int ext4_file_getattr(const struct path *path, struct kstat *stat, | |
5443 | u32 request_mask, unsigned int query_flags) | |
5444 | { | |
5445 | struct inode *inode = d_inode(path->dentry); | |
5446 | u64 delalloc_blocks; | |
5447 | ||
5448 | ext4_getattr(path, stat, request_mask, query_flags); | |
3e3398a0 | 5449 | |
9206c561 AD |
5450 | /* |
5451 | * If there is inline data in the inode, the inode will normally not | |
5452 | * have data blocks allocated (it may have an external xattr block). | |
5453 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5454 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5455 | */ |
5456 | if (unlikely(ext4_has_inline_data(inode))) | |
5457 | stat->blocks += (stat->size + 511) >> 9; | |
5458 | ||
3e3398a0 MC |
5459 | /* |
5460 | * We can't update i_blocks if the block allocation is delayed | |
5461 | * otherwise in the case of system crash before the real block | |
5462 | * allocation is done, we will have i_blocks inconsistent with | |
5463 | * on-disk file blocks. | |
5464 | * We always keep i_blocks updated together with real | |
5465 | * allocation. But to not confuse with user, stat | |
5466 | * will return the blocks that include the delayed allocation | |
5467 | * blocks for this file. | |
5468 | */ | |
96607551 | 5469 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5470 | EXT4_I(inode)->i_reserved_data_blocks); |
5471 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5472 | return 0; |
5473 | } | |
ac27a0ec | 5474 | |
fffb2739 JK |
5475 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5476 | int pextents) | |
a02908f1 | 5477 | { |
12e9b892 | 5478 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5479 | return ext4_ind_trans_blocks(inode, lblocks); |
5480 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5481 | } |
ac51d837 | 5482 | |
ac27a0ec | 5483 | /* |
a02908f1 MC |
5484 | * Account for index blocks, block groups bitmaps and block group |
5485 | * descriptor blocks if modify datablocks and index blocks | |
5486 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5487 | * |
a02908f1 | 5488 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5489 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5490 | * they could still across block group boundary. |
ac27a0ec | 5491 | * |
a02908f1 MC |
5492 | * Also account for superblock, inode, quota and xattr blocks |
5493 | */ | |
fffb2739 JK |
5494 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
5495 | int pextents) | |
a02908f1 | 5496 | { |
8df9675f TT |
5497 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5498 | int gdpblocks; | |
a02908f1 MC |
5499 | int idxblocks; |
5500 | int ret = 0; | |
5501 | ||
5502 | /* | |
fffb2739 JK |
5503 | * How many index blocks need to touch to map @lblocks logical blocks |
5504 | * to @pextents physical extents? | |
a02908f1 | 5505 | */ |
fffb2739 | 5506 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5507 | |
5508 | ret = idxblocks; | |
5509 | ||
5510 | /* | |
5511 | * Now let's see how many group bitmaps and group descriptors need | |
5512 | * to account | |
5513 | */ | |
fffb2739 | 5514 | groups = idxblocks + pextents; |
a02908f1 | 5515 | gdpblocks = groups; |
8df9675f TT |
5516 | if (groups > ngroups) |
5517 | groups = ngroups; | |
a02908f1 MC |
5518 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5519 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5520 | ||
5521 | /* bitmaps and block group descriptor blocks */ | |
5522 | ret += groups + gdpblocks; | |
5523 | ||
5524 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5525 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5526 | ||
5527 | return ret; | |
5528 | } | |
5529 | ||
5530 | /* | |
25985edc | 5531 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5532 | * the modification of a single pages into a single transaction, |
5533 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5534 | * |
525f4ed8 | 5535 | * This could be called via ext4_write_begin() |
ac27a0ec | 5536 | * |
525f4ed8 | 5537 | * We need to consider the worse case, when |
a02908f1 | 5538 | * one new block per extent. |
ac27a0ec | 5539 | */ |
a86c6181 | 5540 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5541 | { |
617ba13b | 5542 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5543 | int ret; |
5544 | ||
fffb2739 | 5545 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5546 | |
a02908f1 | 5547 | /* Account for data blocks for journalled mode */ |
617ba13b | 5548 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5549 | ret += bpp; |
ac27a0ec DK |
5550 | return ret; |
5551 | } | |
f3bd1f3f MC |
5552 | |
5553 | /* | |
5554 | * Calculate the journal credits for a chunk of data modification. | |
5555 | * | |
5556 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5557 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5558 | * |
5559 | * journal buffers for data blocks are not included here, as DIO | |
5560 | * and fallocate do no need to journal data buffers. | |
5561 | */ | |
5562 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5563 | { | |
5564 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5565 | } | |
5566 | ||
ac27a0ec | 5567 | /* |
617ba13b | 5568 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5569 | * Give this, we know that the caller already has write access to iloc->bh. |
5570 | */ | |
617ba13b | 5571 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5572 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5573 | { |
5574 | int err = 0; | |
5575 | ||
0db1ff22 TT |
5576 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5577 | return -EIO; | |
5578 | ||
c64db50e | 5579 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5580 | inode_inc_iversion(inode); |
5581 | ||
ac27a0ec DK |
5582 | /* the do_update_inode consumes one bh->b_count */ |
5583 | get_bh(iloc->bh); | |
5584 | ||
dab291af | 5585 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5586 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5587 | put_bh(iloc->bh); |
5588 | return err; | |
5589 | } | |
5590 | ||
5591 | /* | |
5592 | * On success, We end up with an outstanding reference count against | |
5593 | * iloc->bh. This _must_ be cleaned up later. | |
5594 | */ | |
5595 | ||
5596 | int | |
617ba13b MC |
5597 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5598 | struct ext4_iloc *iloc) | |
ac27a0ec | 5599 | { |
0390131b FM |
5600 | int err; |
5601 | ||
0db1ff22 TT |
5602 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5603 | return -EIO; | |
5604 | ||
0390131b FM |
5605 | err = ext4_get_inode_loc(inode, iloc); |
5606 | if (!err) { | |
5607 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5608 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5609 | if (err) { | |
5610 | brelse(iloc->bh); | |
5611 | iloc->bh = NULL; | |
ac27a0ec DK |
5612 | } |
5613 | } | |
617ba13b | 5614 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5615 | return err; |
5616 | } | |
5617 | ||
6dd4ee7c KS |
5618 | /* |
5619 | * Expand an inode by new_extra_isize bytes. | |
5620 | * Returns 0 on success or negative error number on failure. | |
5621 | */ | |
1d03ec98 AK |
5622 | static int ext4_expand_extra_isize(struct inode *inode, |
5623 | unsigned int new_extra_isize, | |
5624 | struct ext4_iloc iloc, | |
5625 | handle_t *handle) | |
6dd4ee7c KS |
5626 | { |
5627 | struct ext4_inode *raw_inode; | |
5628 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5629 | |
5630 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5631 | return 0; | |
5632 | ||
5633 | raw_inode = ext4_raw_inode(&iloc); | |
5634 | ||
5635 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5636 | |
5637 | /* No extended attributes present */ | |
19f5fb7a TT |
5638 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5639 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5640 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5641 | new_extra_isize); | |
5642 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5643 | return 0; | |
5644 | } | |
5645 | ||
5646 | /* try to expand with EAs present */ | |
5647 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5648 | raw_inode, handle); | |
5649 | } | |
5650 | ||
ac27a0ec DK |
5651 | /* |
5652 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5653 | * dirtiness (it may still be data-dirty). | |
5654 | * This means that the in-core inode may be reaped by prune_icache | |
5655 | * without having to perform any I/O. This is a very good thing, | |
5656 | * because *any* task may call prune_icache - even ones which | |
5657 | * have a transaction open against a different journal. | |
5658 | * | |
5659 | * Is this cheating? Not really. Sure, we haven't written the | |
5660 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5661 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5662 | * we start and wait on commits. | |
ac27a0ec | 5663 | */ |
617ba13b | 5664 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5665 | { |
617ba13b | 5666 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5667 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5668 | static unsigned int mnt_count; | |
5669 | int err, ret; | |
ac27a0ec DK |
5670 | |
5671 | might_sleep(); | |
7ff9c073 | 5672 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5673 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
5674 | if (err) |
5675 | return err; | |
88e03877 | 5676 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
19f5fb7a | 5677 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c | 5678 | /* |
88e03877 EW |
5679 | * In nojournal mode, we can immediately attempt to expand |
5680 | * the inode. When journaled, we first need to obtain extra | |
5681 | * buffer credits since we may write into the EA block | |
6dd4ee7c KS |
5682 | * with this same handle. If journal_extend fails, then it will |
5683 | * only result in a minor loss of functionality for that inode. | |
5684 | * If this is felt to be critical, then e2fsck should be run to | |
5685 | * force a large enough s_min_extra_isize. | |
5686 | */ | |
88e03877 EW |
5687 | if (!ext4_handle_valid(handle) || |
5688 | jbd2_journal_extend(handle, | |
5689 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) == 0) { | |
6dd4ee7c KS |
5690 | ret = ext4_expand_extra_isize(inode, |
5691 | sbi->s_want_extra_isize, | |
5692 | iloc, handle); | |
5693 | if (ret) { | |
c1bddad9 AK |
5694 | if (mnt_count != |
5695 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5696 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5697 | "Unable to expand inode %lu. Delete" |
5698 | " some EAs or run e2fsck.", | |
5699 | inode->i_ino); | |
c1bddad9 AK |
5700 | mnt_count = |
5701 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5702 | } |
5703 | } | |
5704 | } | |
5705 | } | |
5e1021f2 | 5706 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5707 | } |
5708 | ||
5709 | /* | |
617ba13b | 5710 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5711 | * |
5712 | * We're really interested in the case where a file is being extended. | |
5713 | * i_size has been changed by generic_commit_write() and we thus need | |
5714 | * to include the updated inode in the current transaction. | |
5715 | * | |
5dd4056d | 5716 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5717 | * are allocated to the file. |
5718 | * | |
5719 | * If the inode is marked synchronous, we don't honour that here - doing | |
5720 | * so would cause a commit on atime updates, which we don't bother doing. | |
5721 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5722 | * |
5723 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5724 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5725 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5726 | */ |
aa385729 | 5727 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5728 | { |
ac27a0ec DK |
5729 | handle_t *handle; |
5730 | ||
0ae45f63 TT |
5731 | if (flags == I_DIRTY_TIME) |
5732 | return; | |
9924a92a | 5733 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5734 | if (IS_ERR(handle)) |
5735 | goto out; | |
f3dc272f | 5736 | |
f3dc272f CW |
5737 | ext4_mark_inode_dirty(handle, inode); |
5738 | ||
617ba13b | 5739 | ext4_journal_stop(handle); |
ac27a0ec DK |
5740 | out: |
5741 | return; | |
5742 | } | |
5743 | ||
5744 | #if 0 | |
5745 | /* | |
5746 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5747 | * it from being flushed to disk early. Unlike | |
617ba13b | 5748 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5749 | * returns no iloc structure, so the caller needs to repeat the iloc |
5750 | * lookup to mark the inode dirty later. | |
5751 | */ | |
617ba13b | 5752 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5753 | { |
617ba13b | 5754 | struct ext4_iloc iloc; |
ac27a0ec DK |
5755 | |
5756 | int err = 0; | |
5757 | if (handle) { | |
617ba13b | 5758 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5759 | if (!err) { |
5760 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5761 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5762 | if (!err) |
0390131b | 5763 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5764 | NULL, |
0390131b | 5765 | iloc.bh); |
ac27a0ec DK |
5766 | brelse(iloc.bh); |
5767 | } | |
5768 | } | |
617ba13b | 5769 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5770 | return err; |
5771 | } | |
5772 | #endif | |
5773 | ||
617ba13b | 5774 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5775 | { |
5776 | journal_t *journal; | |
5777 | handle_t *handle; | |
5778 | int err; | |
c8585c6f | 5779 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5780 | |
5781 | /* | |
5782 | * We have to be very careful here: changing a data block's | |
5783 | * journaling status dynamically is dangerous. If we write a | |
5784 | * data block to the journal, change the status and then delete | |
5785 | * that block, we risk forgetting to revoke the old log record | |
5786 | * from the journal and so a subsequent replay can corrupt data. | |
5787 | * So, first we make sure that the journal is empty and that | |
5788 | * nobody is changing anything. | |
5789 | */ | |
5790 | ||
617ba13b | 5791 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5792 | if (!journal) |
5793 | return 0; | |
d699594d | 5794 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5795 | return -EROFS; |
5796 | ||
17335dcc DM |
5797 | /* Wait for all existing dio workers */ |
5798 | ext4_inode_block_unlocked_dio(inode); | |
5799 | inode_dio_wait(inode); | |
5800 | ||
4c546592 DJ |
5801 | /* |
5802 | * Before flushing the journal and switching inode's aops, we have | |
5803 | * to flush all dirty data the inode has. There can be outstanding | |
5804 | * delayed allocations, there can be unwritten extents created by | |
5805 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5806 | * dirty data which can be converted only after flushing the dirty | |
5807 | * data (and journalled aops don't know how to handle these cases). | |
5808 | */ | |
5809 | if (val) { | |
5810 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5811 | err = filemap_write_and_wait(inode->i_mapping); | |
5812 | if (err < 0) { | |
5813 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
5814 | ext4_inode_resume_unlocked_dio(inode); | |
5815 | return err; | |
5816 | } | |
5817 | } | |
5818 | ||
c8585c6f | 5819 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
dab291af | 5820 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5821 | |
5822 | /* | |
5823 | * OK, there are no updates running now, and all cached data is | |
5824 | * synced to disk. We are now in a completely consistent state | |
5825 | * which doesn't have anything in the journal, and we know that | |
5826 | * no filesystem updates are running, so it is safe to modify | |
5827 | * the inode's in-core data-journaling state flag now. | |
5828 | */ | |
5829 | ||
5830 | if (val) | |
12e9b892 | 5831 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5832 | else { |
4f879ca6 JK |
5833 | err = jbd2_journal_flush(journal); |
5834 | if (err < 0) { | |
5835 | jbd2_journal_unlock_updates(journal); | |
c8585c6f | 5836 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
4f879ca6 JK |
5837 | ext4_inode_resume_unlocked_dio(inode); |
5838 | return err; | |
5839 | } | |
12e9b892 | 5840 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5841 | } |
617ba13b | 5842 | ext4_set_aops(inode); |
a3caa24b JK |
5843 | /* |
5844 | * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated. | |
5845 | * E.g. S_DAX may get cleared / set. | |
5846 | */ | |
5847 | ext4_set_inode_flags(inode); | |
ac27a0ec | 5848 | |
dab291af | 5849 | jbd2_journal_unlock_updates(journal); |
c8585c6f DJ |
5850 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
5851 | ||
4c546592 DJ |
5852 | if (val) |
5853 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
17335dcc | 5854 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5855 | |
5856 | /* Finally we can mark the inode as dirty. */ | |
5857 | ||
9924a92a | 5858 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5859 | if (IS_ERR(handle)) |
5860 | return PTR_ERR(handle); | |
5861 | ||
617ba13b | 5862 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5863 | ext4_handle_sync(handle); |
617ba13b MC |
5864 | ext4_journal_stop(handle); |
5865 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5866 | |
5867 | return err; | |
5868 | } | |
2e9ee850 AK |
5869 | |
5870 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5871 | { | |
5872 | return !buffer_mapped(bh); | |
5873 | } | |
5874 | ||
11bac800 | 5875 | int ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 5876 | { |
11bac800 | 5877 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 5878 | struct page *page = vmf->page; |
2e9ee850 AK |
5879 | loff_t size; |
5880 | unsigned long len; | |
9ea7df53 | 5881 | int ret; |
2e9ee850 | 5882 | struct file *file = vma->vm_file; |
496ad9aa | 5883 | struct inode *inode = file_inode(file); |
2e9ee850 | 5884 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5885 | handle_t *handle; |
5886 | get_block_t *get_block; | |
5887 | int retries = 0; | |
2e9ee850 | 5888 | |
8e8ad8a5 | 5889 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5890 | file_update_time(vma->vm_file); |
ea3d7209 JK |
5891 | |
5892 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 EB |
5893 | |
5894 | ret = ext4_convert_inline_data(inode); | |
5895 | if (ret) | |
5896 | goto out_ret; | |
5897 | ||
9ea7df53 JK |
5898 | /* Delalloc case is easy... */ |
5899 | if (test_opt(inode->i_sb, DELALLOC) && | |
5900 | !ext4_should_journal_data(inode) && | |
5901 | !ext4_nonda_switch(inode->i_sb)) { | |
5902 | do { | |
5c500029 | 5903 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
5904 | ext4_da_get_block_prep); |
5905 | } while (ret == -ENOSPC && | |
5906 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5907 | goto out_ret; | |
2e9ee850 | 5908 | } |
0e499890 DW |
5909 | |
5910 | lock_page(page); | |
9ea7df53 JK |
5911 | size = i_size_read(inode); |
5912 | /* Page got truncated from under us? */ | |
5913 | if (page->mapping != mapping || page_offset(page) > size) { | |
5914 | unlock_page(page); | |
5915 | ret = VM_FAULT_NOPAGE; | |
5916 | goto out; | |
0e499890 | 5917 | } |
2e9ee850 | 5918 | |
09cbfeaf KS |
5919 | if (page->index == size >> PAGE_SHIFT) |
5920 | len = size & ~PAGE_MASK; | |
2e9ee850 | 5921 | else |
09cbfeaf | 5922 | len = PAGE_SIZE; |
a827eaff | 5923 | /* |
9ea7df53 JK |
5924 | * Return if we have all the buffers mapped. This avoids the need to do |
5925 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5926 | */ |
2e9ee850 | 5927 | if (page_has_buffers(page)) { |
f19d5870 TM |
5928 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5929 | 0, len, NULL, | |
5930 | ext4_bh_unmapped)) { | |
9ea7df53 | 5931 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5932 | wait_for_stable_page(page); |
9ea7df53 JK |
5933 | ret = VM_FAULT_LOCKED; |
5934 | goto out; | |
a827eaff | 5935 | } |
2e9ee850 | 5936 | } |
a827eaff | 5937 | unlock_page(page); |
9ea7df53 JK |
5938 | /* OK, we need to fill the hole... */ |
5939 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 5940 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
5941 | else |
5942 | get_block = ext4_get_block; | |
5943 | retry_alloc: | |
9924a92a TT |
5944 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5945 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5946 | if (IS_ERR(handle)) { |
c2ec175c | 5947 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5948 | goto out; |
5949 | } | |
5c500029 | 5950 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 5951 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 5952 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 5953 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
5954 | unlock_page(page); |
5955 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5956 | ext4_journal_stop(handle); |
9ea7df53 JK |
5957 | goto out; |
5958 | } | |
5959 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5960 | } | |
5961 | ext4_journal_stop(handle); | |
5962 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5963 | goto retry_alloc; | |
5964 | out_ret: | |
5965 | ret = block_page_mkwrite_return(ret); | |
5966 | out: | |
ea3d7209 | 5967 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 5968 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5969 | return ret; |
5970 | } | |
ea3d7209 | 5971 | |
11bac800 | 5972 | int ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 5973 | { |
11bac800 | 5974 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ea3d7209 JK |
5975 | int err; |
5976 | ||
5977 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
11bac800 | 5978 | err = filemap_fault(vmf); |
ea3d7209 JK |
5979 | up_read(&EXT4_I(inode)->i_mmap_sem); |
5980 | ||
5981 | return err; | |
5982 | } | |
2d90c160 JK |
5983 | |
5984 | /* | |
5985 | * Find the first extent at or after @lblk in an inode that is not a hole. | |
5986 | * Search for @map_len blocks at most. The extent is returned in @result. | |
5987 | * | |
5988 | * The function returns 1 if we found an extent. The function returns 0 in | |
5989 | * case there is no extent at or after @lblk and in that case also sets | |
5990 | * @result->es_len to 0. In case of error, the error code is returned. | |
5991 | */ | |
5992 | int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk, | |
5993 | unsigned int map_len, struct extent_status *result) | |
5994 | { | |
5995 | struct ext4_map_blocks map; | |
5996 | struct extent_status es = {}; | |
5997 | int ret; | |
5998 | ||
5999 | map.m_lblk = lblk; | |
6000 | map.m_len = map_len; | |
6001 | ||
6002 | /* | |
6003 | * For non-extent based files this loop may iterate several times since | |
6004 | * we do not determine full hole size. | |
6005 | */ | |
6006 | while (map.m_len > 0) { | |
6007 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
6008 | if (ret < 0) | |
6009 | return ret; | |
6010 | /* There's extent covering m_lblk? Just return it. */ | |
6011 | if (ret > 0) { | |
6012 | int status; | |
6013 | ||
6014 | ext4_es_store_pblock(result, map.m_pblk); | |
6015 | result->es_lblk = map.m_lblk; | |
6016 | result->es_len = map.m_len; | |
6017 | if (map.m_flags & EXT4_MAP_UNWRITTEN) | |
6018 | status = EXTENT_STATUS_UNWRITTEN; | |
6019 | else | |
6020 | status = EXTENT_STATUS_WRITTEN; | |
6021 | ext4_es_store_status(result, status); | |
6022 | return 1; | |
6023 | } | |
6024 | ext4_es_find_delayed_extent_range(inode, map.m_lblk, | |
6025 | map.m_lblk + map.m_len - 1, | |
6026 | &es); | |
6027 | /* Is delalloc data before next block in extent tree? */ | |
6028 | if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) { | |
6029 | ext4_lblk_t offset = 0; | |
6030 | ||
6031 | if (es.es_lblk < lblk) | |
6032 | offset = lblk - es.es_lblk; | |
6033 | result->es_lblk = es.es_lblk + offset; | |
6034 | ext4_es_store_pblock(result, | |
6035 | ext4_es_pblock(&es) + offset); | |
6036 | result->es_len = es.es_len - offset; | |
6037 | ext4_es_store_status(result, ext4_es_status(&es)); | |
6038 | ||
6039 | return 1; | |
6040 | } | |
6041 | /* There's a hole at m_lblk, advance us after it */ | |
6042 | map.m_lblk += map.m_len; | |
6043 | map_len -= map.m_len; | |
6044 | map.m_len = map_len; | |
6045 | cond_resched(); | |
6046 | } | |
6047 | result->es_len = 0; | |
6048 | return 0; | |
6049 | } |