1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/inode.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
18 *
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20 */
21
22 #include <linux/fs.h>
23 #include <linux/mount.h>
24 #include <linux/time.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/string.h>
30 #include <linux/buffer_head.h>
31 #include <linux/writeback.h>
32 #include <linux/pagevec.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include <linux/bio.h>
37 #include <linux/workqueue.h>
38 #include <linux/kernel.h>
39 #include <linux/printk.h>
40 #include <linux/slab.h>
41 #include <linux/bitops.h>
42 #include <linux/iomap.h>
43 #include <linux/iversion.h>
44
45 #include "ext4_jbd2.h"
46 #include "xattr.h"
47 #include "acl.h"
48 #include "truncate.h"
49
50 #include <trace/events/ext4.h>
51
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
54 {
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 __u32 csum;
57 __u16 dummy_csum = 0;
58 int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 unsigned int csum_size = sizeof(dummy_csum);
60
61 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 offset += csum_size;
64 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 EXT4_GOOD_OLD_INODE_SIZE - offset);
66
67 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 offset = offsetof(struct ext4_inode, i_checksum_hi);
69 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 EXT4_GOOD_OLD_INODE_SIZE,
71 offset - EXT4_GOOD_OLD_INODE_SIZE);
72 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
74 csum_size);
75 offset += csum_size;
76 }
77 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 EXT4_INODE_SIZE(inode->i_sb) - offset);
79 }
80
81 return csum;
82 }
83
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
86 {
87 __u32 provided, calculated;
88
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
92 return 1;
93
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
99 else
100 calculated &= 0xFFFF;
101
102 return provided == calculated;
103 }
104
105 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
107 {
108 __u32 csum;
109
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
113 return;
114
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
120 }
121
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
123 loff_t new_size)
124 {
125 trace_ext4_begin_ordered_truncate(inode, new_size);
126 /*
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
131 */
132 if (!EXT4_I(inode)->jinode)
133 return 0;
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
136 new_size);
137 }
138
139 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 int pextents);
141
142 /*
143 * Test whether an inode is a fast symlink.
144 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
145 */
146 int ext4_inode_is_fast_symlink(struct inode *inode)
147 {
148 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
149 int ea_blocks = EXT4_I(inode)->i_file_acl ?
150 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
151
152 if (ext4_has_inline_data(inode))
153 return 0;
154
155 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
156 }
157 return S_ISLNK(inode->i_mode) && inode->i_size &&
158 (inode->i_size < EXT4_N_BLOCKS * 4);
159 }
160
161 /*
162 * Called at the last iput() if i_nlink is zero.
163 */
164 void ext4_evict_inode(struct inode *inode)
165 {
166 handle_t *handle;
167 int err;
168 /*
169 * Credits for final inode cleanup and freeing:
170 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
171 * (xattr block freeing), bitmap, group descriptor (inode freeing)
172 */
173 int extra_credits = 6;
174 struct ext4_xattr_inode_array *ea_inode_array = NULL;
175 bool freeze_protected = false;
176
177 trace_ext4_evict_inode(inode);
178
179 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
180 ext4_evict_ea_inode(inode);
181 if (inode->i_nlink) {
182 truncate_inode_pages_final(&inode->i_data);
183
184 goto no_delete;
185 }
186
187 if (is_bad_inode(inode))
188 goto no_delete;
189 dquot_initialize(inode);
190
191 if (ext4_should_order_data(inode))
192 ext4_begin_ordered_truncate(inode, 0);
193 truncate_inode_pages_final(&inode->i_data);
194
195 /*
196 * For inodes with journalled data, transaction commit could have
197 * dirtied the inode. And for inodes with dioread_nolock, unwritten
198 * extents converting worker could merge extents and also have dirtied
199 * the inode. Flush worker is ignoring it because of I_FREEING flag but
200 * we still need to remove the inode from the writeback lists.
201 */
202 if (!list_empty_careful(&inode->i_io_list))
203 inode_io_list_del(inode);
204
205 /*
206 * Protect us against freezing - iput() caller didn't have to have any
207 * protection against it. When we are in a running transaction though,
208 * we are already protected against freezing and we cannot grab further
209 * protection due to lock ordering constraints.
210 */
211 if (!ext4_journal_current_handle()) {
212 sb_start_intwrite(inode->i_sb);
213 freeze_protected = true;
214 }
215
216 if (!IS_NOQUOTA(inode))
217 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
218
219 /*
220 * Block bitmap, group descriptor, and inode are accounted in both
221 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
222 */
223 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
224 ext4_blocks_for_truncate(inode) + extra_credits - 3);
225 if (IS_ERR(handle)) {
226 ext4_std_error(inode->i_sb, PTR_ERR(handle));
227 /*
228 * If we're going to skip the normal cleanup, we still need to
229 * make sure that the in-core orphan linked list is properly
230 * cleaned up.
231 */
232 ext4_orphan_del(NULL, inode);
233 if (freeze_protected)
234 sb_end_intwrite(inode->i_sb);
235 goto no_delete;
236 }
237
238 if (IS_SYNC(inode))
239 ext4_handle_sync(handle);
240
241 /*
242 * Set inode->i_size to 0 before calling ext4_truncate(). We need
243 * special handling of symlinks here because i_size is used to
244 * determine whether ext4_inode_info->i_data contains symlink data or
245 * block mappings. Setting i_size to 0 will remove its fast symlink
246 * status. Erase i_data so that it becomes a valid empty block map.
247 */
248 if (ext4_inode_is_fast_symlink(inode))
249 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
250 inode->i_size = 0;
251 err = ext4_mark_inode_dirty(handle, inode);
252 if (err) {
253 ext4_warning(inode->i_sb,
254 "couldn't mark inode dirty (err %d)", err);
255 goto stop_handle;
256 }
257 if (inode->i_blocks) {
258 err = ext4_truncate(inode);
259 if (err) {
260 ext4_error_err(inode->i_sb, -err,
261 "couldn't truncate inode %lu (err %d)",
262 inode->i_ino, err);
263 goto stop_handle;
264 }
265 }
266
267 /* Remove xattr references. */
268 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
269 extra_credits);
270 if (err) {
271 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
272 stop_handle:
273 ext4_journal_stop(handle);
274 ext4_orphan_del(NULL, inode);
275 if (freeze_protected)
276 sb_end_intwrite(inode->i_sb);
277 ext4_xattr_inode_array_free(ea_inode_array);
278 goto no_delete;
279 }
280
281 /*
282 * Kill off the orphan record which ext4_truncate created.
283 * AKPM: I think this can be inside the above `if'.
284 * Note that ext4_orphan_del() has to be able to cope with the
285 * deletion of a non-existent orphan - this is because we don't
286 * know if ext4_truncate() actually created an orphan record.
287 * (Well, we could do this if we need to, but heck - it works)
288 */
289 ext4_orphan_del(handle, inode);
290 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
291
292 /*
293 * One subtle ordering requirement: if anything has gone wrong
294 * (transaction abort, IO errors, whatever), then we can still
295 * do these next steps (the fs will already have been marked as
296 * having errors), but we can't free the inode if the mark_dirty
297 * fails.
298 */
299 if (ext4_mark_inode_dirty(handle, inode))
300 /* If that failed, just do the required in-core inode clear. */
301 ext4_clear_inode(inode);
302 else
303 ext4_free_inode(handle, inode);
304 ext4_journal_stop(handle);
305 if (freeze_protected)
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
308 return;
309 no_delete:
310 /*
311 * Check out some where else accidentally dirty the evicting inode,
312 * which may probably cause inode use-after-free issues later.
313 */
314 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
315
316 if (!list_empty(&EXT4_I(inode)->i_fc_list))
317 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
318 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
319 }
320
321 #ifdef CONFIG_QUOTA
322 qsize_t *ext4_get_reserved_space(struct inode *inode)
323 {
324 return &EXT4_I(inode)->i_reserved_quota;
325 }
326 #endif
327
328 /*
329 * Called with i_data_sem down, which is important since we can call
330 * ext4_discard_preallocations() from here.
331 */
332 void ext4_da_update_reserve_space(struct inode *inode,
333 int used, int quota_claim)
334 {
335 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
336 struct ext4_inode_info *ei = EXT4_I(inode);
337
338 spin_lock(&ei->i_block_reservation_lock);
339 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
340 if (unlikely(used > ei->i_reserved_data_blocks)) {
341 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
342 "with only %d reserved data blocks",
343 __func__, inode->i_ino, used,
344 ei->i_reserved_data_blocks);
345 WARN_ON(1);
346 used = ei->i_reserved_data_blocks;
347 }
348
349 /* Update per-inode reservations */
350 ei->i_reserved_data_blocks -= used;
351 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
352
353 spin_unlock(&ei->i_block_reservation_lock);
354
355 /* Update quota subsystem for data blocks */
356 if (quota_claim)
357 dquot_claim_block(inode, EXT4_C2B(sbi, used));
358 else {
359 /*
360 * We did fallocate with an offset that is already delayed
361 * allocated. So on delayed allocated writeback we should
362 * not re-claim the quota for fallocated blocks.
363 */
364 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
365 }
366
367 /*
368 * If we have done all the pending block allocations and if
369 * there aren't any writers on the inode, we can discard the
370 * inode's preallocations.
371 */
372 if ((ei->i_reserved_data_blocks == 0) &&
373 !inode_is_open_for_write(inode))
374 ext4_discard_preallocations(inode);
375 }
376
377 static int __check_block_validity(struct inode *inode, const char *func,
378 unsigned int line,
379 struct ext4_map_blocks *map)
380 {
381 if (ext4_has_feature_journal(inode->i_sb) &&
382 (inode->i_ino ==
383 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
384 return 0;
385 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
386 ext4_error_inode(inode, func, line, map->m_pblk,
387 "lblock %lu mapped to illegal pblock %llu "
388 "(length %d)", (unsigned long) map->m_lblk,
389 map->m_pblk, map->m_len);
390 return -EFSCORRUPTED;
391 }
392 return 0;
393 }
394
395 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
396 ext4_lblk_t len)
397 {
398 int ret;
399
400 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
401 return fscrypt_zeroout_range(inode, lblk, pblk, len);
402
403 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
404 if (ret > 0)
405 ret = 0;
406
407 return ret;
408 }
409
410 #define check_block_validity(inode, map) \
411 __check_block_validity((inode), __func__, __LINE__, (map))
412
413 #ifdef ES_AGGRESSIVE_TEST
414 static void ext4_map_blocks_es_recheck(handle_t *handle,
415 struct inode *inode,
416 struct ext4_map_blocks *es_map,
417 struct ext4_map_blocks *map,
418 int flags)
419 {
420 int retval;
421
422 map->m_flags = 0;
423 /*
424 * There is a race window that the result is not the same.
425 * e.g. xfstests #223 when dioread_nolock enables. The reason
426 * is that we lookup a block mapping in extent status tree with
427 * out taking i_data_sem. So at the time the unwritten extent
428 * could be converted.
429 */
430 down_read(&EXT4_I(inode)->i_data_sem);
431 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
432 retval = ext4_ext_map_blocks(handle, inode, map, 0);
433 } else {
434 retval = ext4_ind_map_blocks(handle, inode, map, 0);
435 }
436 up_read((&EXT4_I(inode)->i_data_sem));
437
438 /*
439 * We don't check m_len because extent will be collpased in status
440 * tree. So the m_len might not equal.
441 */
442 if (es_map->m_lblk != map->m_lblk ||
443 es_map->m_flags != map->m_flags ||
444 es_map->m_pblk != map->m_pblk) {
445 printk("ES cache assertion failed for inode: %lu "
446 "es_cached ex [%d/%d/%llu/%x] != "
447 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
448 inode->i_ino, es_map->m_lblk, es_map->m_len,
449 es_map->m_pblk, es_map->m_flags, map->m_lblk,
450 map->m_len, map->m_pblk, map->m_flags,
451 retval, flags);
452 }
453 }
454 #endif /* ES_AGGRESSIVE_TEST */
455
456 static int ext4_map_query_blocks(handle_t *handle, struct inode *inode,
457 struct ext4_map_blocks *map)
458 {
459 unsigned int status;
460 int retval;
461
462 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
463 retval = ext4_ext_map_blocks(handle, inode, map, 0);
464 else
465 retval = ext4_ind_map_blocks(handle, inode, map, 0);
466
467 if (retval <= 0)
468 return retval;
469
470 if (unlikely(retval != map->m_len)) {
471 ext4_warning(inode->i_sb,
472 "ES len assertion failed for inode "
473 "%lu: retval %d != map->m_len %d",
474 inode->i_ino, retval, map->m_len);
475 WARN_ON(1);
476 }
477
478 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
479 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
480 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
481 map->m_pblk, status);
482 return retval;
483 }
484
485 /*
486 * The ext4_map_blocks() function tries to look up the requested blocks,
487 * and returns if the blocks are already mapped.
488 *
489 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
490 * and store the allocated blocks in the result buffer head and mark it
491 * mapped.
492 *
493 * If file type is extents based, it will call ext4_ext_map_blocks(),
494 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
495 * based files
496 *
497 * On success, it returns the number of blocks being mapped or allocated.
498 * If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
499 * pre-allocated and unwritten, the resulting @map is marked as unwritten.
500 * If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
501 *
502 * It returns 0 if plain look up failed (blocks have not been allocated), in
503 * that case, @map is returned as unmapped but we still do fill map->m_len to
504 * indicate the length of a hole starting at map->m_lblk.
505 *
506 * It returns the error in case of allocation failure.
507 */
508 int ext4_map_blocks(handle_t *handle, struct inode *inode,
509 struct ext4_map_blocks *map, int flags)
510 {
511 struct extent_status es;
512 int retval;
513 int ret = 0;
514 #ifdef ES_AGGRESSIVE_TEST
515 struct ext4_map_blocks orig_map;
516
517 memcpy(&orig_map, map, sizeof(*map));
518 #endif
519
520 map->m_flags = 0;
521 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
522 flags, map->m_len, (unsigned long) map->m_lblk);
523
524 /*
525 * ext4_map_blocks returns an int, and m_len is an unsigned int
526 */
527 if (unlikely(map->m_len > INT_MAX))
528 map->m_len = INT_MAX;
529
530 /* We can handle the block number less than EXT_MAX_BLOCKS */
531 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
532 return -EFSCORRUPTED;
533
534 /* Lookup extent status tree firstly */
535 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
536 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
537 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
538 map->m_pblk = ext4_es_pblock(&es) +
539 map->m_lblk - es.es_lblk;
540 map->m_flags |= ext4_es_is_written(&es) ?
541 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
542 retval = es.es_len - (map->m_lblk - es.es_lblk);
543 if (retval > map->m_len)
544 retval = map->m_len;
545 map->m_len = retval;
546 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
547 map->m_pblk = 0;
548 map->m_flags |= ext4_es_is_delayed(&es) ?
549 EXT4_MAP_DELAYED : 0;
550 retval = es.es_len - (map->m_lblk - es.es_lblk);
551 if (retval > map->m_len)
552 retval = map->m_len;
553 map->m_len = retval;
554 retval = 0;
555 } else {
556 BUG();
557 }
558
559 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
560 return retval;
561 #ifdef ES_AGGRESSIVE_TEST
562 ext4_map_blocks_es_recheck(handle, inode, map,
563 &orig_map, flags);
564 #endif
565 goto found;
566 }
567 /*
568 * In the query cache no-wait mode, nothing we can do more if we
569 * cannot find extent in the cache.
570 */
571 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
572 return 0;
573
574 /*
575 * Try to see if we can get the block without requesting a new
576 * file system block.
577 */
578 down_read(&EXT4_I(inode)->i_data_sem);
579 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
580 retval = ext4_ext_map_blocks(handle, inode, map, 0);
581 } else {
582 retval = ext4_ind_map_blocks(handle, inode, map, 0);
583 }
584 if (retval > 0) {
585 unsigned int status;
586
587 if (unlikely(retval != map->m_len)) {
588 ext4_warning(inode->i_sb,
589 "ES len assertion failed for inode "
590 "%lu: retval %d != map->m_len %d",
591 inode->i_ino, retval, map->m_len);
592 WARN_ON(1);
593 }
594
595 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
596 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
597 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
598 !(status & EXTENT_STATUS_WRITTEN) &&
599 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
600 map->m_lblk + map->m_len - 1))
601 status |= EXTENT_STATUS_DELAYED;
602 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
603 map->m_pblk, status);
604 }
605 up_read((&EXT4_I(inode)->i_data_sem));
606
607 found:
608 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
609 ret = check_block_validity(inode, map);
610 if (ret != 0)
611 return ret;
612 }
613
614 /* If it is only a block(s) look up */
615 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
616 return retval;
617
618 /*
619 * Returns if the blocks have already allocated
620 *
621 * Note that if blocks have been preallocated
622 * ext4_ext_map_blocks() returns with buffer head unmapped
623 */
624 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
625 /*
626 * If we need to convert extent to unwritten
627 * we continue and do the actual work in
628 * ext4_ext_map_blocks()
629 */
630 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
631 return retval;
632
633 /*
634 * Here we clear m_flags because after allocating an new extent,
635 * it will be set again.
636 */
637 map->m_flags &= ~EXT4_MAP_FLAGS;
638
639 /*
640 * New blocks allocate and/or writing to unwritten extent
641 * will possibly result in updating i_data, so we take
642 * the write lock of i_data_sem, and call get_block()
643 * with create == 1 flag.
644 */
645 down_write(&EXT4_I(inode)->i_data_sem);
646
647 /*
648 * We need to check for EXT4 here because migrate
649 * could have changed the inode type in between
650 */
651 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
652 retval = ext4_ext_map_blocks(handle, inode, map, flags);
653 } else {
654 retval = ext4_ind_map_blocks(handle, inode, map, flags);
655
656 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
657 /*
658 * We allocated new blocks which will result in
659 * i_data's format changing. Force the migrate
660 * to fail by clearing migrate flags
661 */
662 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
663 }
664 }
665
666 if (retval > 0) {
667 unsigned int status;
668
669 if (unlikely(retval != map->m_len)) {
670 ext4_warning(inode->i_sb,
671 "ES len assertion failed for inode "
672 "%lu: retval %d != map->m_len %d",
673 inode->i_ino, retval, map->m_len);
674 WARN_ON(1);
675 }
676
677 /*
678 * We have to zeroout blocks before inserting them into extent
679 * status tree. Otherwise someone could look them up there and
680 * use them before they are really zeroed. We also have to
681 * unmap metadata before zeroing as otherwise writeback can
682 * overwrite zeros with stale data from block device.
683 */
684 if (flags & EXT4_GET_BLOCKS_ZERO &&
685 map->m_flags & EXT4_MAP_MAPPED &&
686 map->m_flags & EXT4_MAP_NEW) {
687 ret = ext4_issue_zeroout(inode, map->m_lblk,
688 map->m_pblk, map->m_len);
689 if (ret) {
690 retval = ret;
691 goto out_sem;
692 }
693 }
694
695 /*
696 * If the extent has been zeroed out, we don't need to update
697 * extent status tree.
698 */
699 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
700 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
701 if (ext4_es_is_written(&es))
702 goto out_sem;
703 }
704 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
705 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
706 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
707 !(status & EXTENT_STATUS_WRITTEN) &&
708 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
709 map->m_lblk + map->m_len - 1))
710 status |= EXTENT_STATUS_DELAYED;
711 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
712 map->m_pblk, status);
713 }
714
715 out_sem:
716 up_write((&EXT4_I(inode)->i_data_sem));
717 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718 ret = check_block_validity(inode, map);
719 if (ret != 0)
720 return ret;
721
722 /*
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
725 * ordered data list.
726 */
727 if (map->m_flags & EXT4_MAP_NEW &&
728 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729 !(flags & EXT4_GET_BLOCKS_ZERO) &&
730 !ext4_is_quota_file(inode) &&
731 ext4_should_order_data(inode)) {
732 loff_t start_byte =
733 (loff_t)map->m_lblk << inode->i_blkbits;
734 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
735
736 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
737 ret = ext4_jbd2_inode_add_wait(handle, inode,
738 start_byte, length);
739 else
740 ret = ext4_jbd2_inode_add_write(handle, inode,
741 start_byte, length);
742 if (ret)
743 return ret;
744 }
745 }
746 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
747 map->m_flags & EXT4_MAP_MAPPED))
748 ext4_fc_track_range(handle, inode, map->m_lblk,
749 map->m_lblk + map->m_len - 1);
750 if (retval < 0)
751 ext_debug(inode, "failed with err %d\n", retval);
752 return retval;
753 }
754
755 /*
756 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
757 * we have to be careful as someone else may be manipulating b_state as well.
758 */
759 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
760 {
761 unsigned long old_state;
762 unsigned long new_state;
763
764 flags &= EXT4_MAP_FLAGS;
765
766 /* Dummy buffer_head? Set non-atomically. */
767 if (!bh->b_page) {
768 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
769 return;
770 }
771 /*
772 * Someone else may be modifying b_state. Be careful! This is ugly but
773 * once we get rid of using bh as a container for mapping information
774 * to pass to / from get_block functions, this can go away.
775 */
776 old_state = READ_ONCE(bh->b_state);
777 do {
778 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
779 } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
780 }
781
782 static int _ext4_get_block(struct inode *inode, sector_t iblock,
783 struct buffer_head *bh, int flags)
784 {
785 struct ext4_map_blocks map;
786 int ret = 0;
787
788 if (ext4_has_inline_data(inode))
789 return -ERANGE;
790
791 map.m_lblk = iblock;
792 map.m_len = bh->b_size >> inode->i_blkbits;
793
794 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
795 flags);
796 if (ret > 0) {
797 map_bh(bh, inode->i_sb, map.m_pblk);
798 ext4_update_bh_state(bh, map.m_flags);
799 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
800 ret = 0;
801 } else if (ret == 0) {
802 /* hole case, need to fill in bh->b_size */
803 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
804 }
805 return ret;
806 }
807
808 int ext4_get_block(struct inode *inode, sector_t iblock,
809 struct buffer_head *bh, int create)
810 {
811 return _ext4_get_block(inode, iblock, bh,
812 create ? EXT4_GET_BLOCKS_CREATE : 0);
813 }
814
815 /*
816 * Get block function used when preparing for buffered write if we require
817 * creating an unwritten extent if blocks haven't been allocated. The extent
818 * will be converted to written after the IO is complete.
819 */
820 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
821 struct buffer_head *bh_result, int create)
822 {
823 int ret = 0;
824
825 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
826 inode->i_ino, create);
827 ret = _ext4_get_block(inode, iblock, bh_result,
828 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
829
830 /*
831 * If the buffer is marked unwritten, mark it as new to make sure it is
832 * zeroed out correctly in case of partial writes. Otherwise, there is
833 * a chance of stale data getting exposed.
834 */
835 if (ret == 0 && buffer_unwritten(bh_result))
836 set_buffer_new(bh_result);
837
838 return ret;
839 }
840
841 /* Maximum number of blocks we map for direct IO at once. */
842 #define DIO_MAX_BLOCKS 4096
843
844 /*
845 * `handle' can be NULL if create is zero
846 */
847 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
848 ext4_lblk_t block, int map_flags)
849 {
850 struct ext4_map_blocks map;
851 struct buffer_head *bh;
852 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
853 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
854 int err;
855
856 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
857 || handle != NULL || create == 0);
858 ASSERT(create == 0 || !nowait);
859
860 map.m_lblk = block;
861 map.m_len = 1;
862 err = ext4_map_blocks(handle, inode, &map, map_flags);
863
864 if (err == 0)
865 return create ? ERR_PTR(-ENOSPC) : NULL;
866 if (err < 0)
867 return ERR_PTR(err);
868
869 if (nowait)
870 return sb_find_get_block(inode->i_sb, map.m_pblk);
871
872 bh = sb_getblk(inode->i_sb, map.m_pblk);
873 if (unlikely(!bh))
874 return ERR_PTR(-ENOMEM);
875 if (map.m_flags & EXT4_MAP_NEW) {
876 ASSERT(create != 0);
877 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
878 || (handle != NULL));
879
880 /*
881 * Now that we do not always journal data, we should
882 * keep in mind whether this should always journal the
883 * new buffer as metadata. For now, regular file
884 * writes use ext4_get_block instead, so it's not a
885 * problem.
886 */
887 lock_buffer(bh);
888 BUFFER_TRACE(bh, "call get_create_access");
889 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
890 EXT4_JTR_NONE);
891 if (unlikely(err)) {
892 unlock_buffer(bh);
893 goto errout;
894 }
895 if (!buffer_uptodate(bh)) {
896 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
897 set_buffer_uptodate(bh);
898 }
899 unlock_buffer(bh);
900 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
901 err = ext4_handle_dirty_metadata(handle, inode, bh);
902 if (unlikely(err))
903 goto errout;
904 } else
905 BUFFER_TRACE(bh, "not a new buffer");
906 return bh;
907 errout:
908 brelse(bh);
909 return ERR_PTR(err);
910 }
911
912 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
913 ext4_lblk_t block, int map_flags)
914 {
915 struct buffer_head *bh;
916 int ret;
917
918 bh = ext4_getblk(handle, inode, block, map_flags);
919 if (IS_ERR(bh))
920 return bh;
921 if (!bh || ext4_buffer_uptodate(bh))
922 return bh;
923
924 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
925 if (ret) {
926 put_bh(bh);
927 return ERR_PTR(ret);
928 }
929 return bh;
930 }
931
932 /* Read a contiguous batch of blocks. */
933 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
934 bool wait, struct buffer_head **bhs)
935 {
936 int i, err;
937
938 for (i = 0; i < bh_count; i++) {
939 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
940 if (IS_ERR(bhs[i])) {
941 err = PTR_ERR(bhs[i]);
942 bh_count = i;
943 goto out_brelse;
944 }
945 }
946
947 for (i = 0; i < bh_count; i++)
948 /* Note that NULL bhs[i] is valid because of holes. */
949 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
950 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
951
952 if (!wait)
953 return 0;
954
955 for (i = 0; i < bh_count; i++)
956 if (bhs[i])
957 wait_on_buffer(bhs[i]);
958
959 for (i = 0; i < bh_count; i++) {
960 if (bhs[i] && !buffer_uptodate(bhs[i])) {
961 err = -EIO;
962 goto out_brelse;
963 }
964 }
965 return 0;
966
967 out_brelse:
968 for (i = 0; i < bh_count; i++) {
969 brelse(bhs[i]);
970 bhs[i] = NULL;
971 }
972 return err;
973 }
974
975 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
976 struct buffer_head *head,
977 unsigned from,
978 unsigned to,
979 int *partial,
980 int (*fn)(handle_t *handle, struct inode *inode,
981 struct buffer_head *bh))
982 {
983 struct buffer_head *bh;
984 unsigned block_start, block_end;
985 unsigned blocksize = head->b_size;
986 int err, ret = 0;
987 struct buffer_head *next;
988
989 for (bh = head, block_start = 0;
990 ret == 0 && (bh != head || !block_start);
991 block_start = block_end, bh = next) {
992 next = bh->b_this_page;
993 block_end = block_start + blocksize;
994 if (block_end <= from || block_start >= to) {
995 if (partial && !buffer_uptodate(bh))
996 *partial = 1;
997 continue;
998 }
999 err = (*fn)(handle, inode, bh);
1000 if (!ret)
1001 ret = err;
1002 }
1003 return ret;
1004 }
1005
1006 /*
1007 * Helper for handling dirtying of journalled data. We also mark the folio as
1008 * dirty so that writeback code knows about this page (and inode) contains
1009 * dirty data. ext4_writepages() then commits appropriate transaction to
1010 * make data stable.
1011 */
1012 static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
1013 {
1014 folio_mark_dirty(bh->b_folio);
1015 return ext4_handle_dirty_metadata(handle, NULL, bh);
1016 }
1017
1018 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1019 struct buffer_head *bh)
1020 {
1021 int dirty = buffer_dirty(bh);
1022 int ret;
1023
1024 if (!buffer_mapped(bh) || buffer_freed(bh))
1025 return 0;
1026 /*
1027 * __block_write_begin() could have dirtied some buffers. Clean
1028 * the dirty bit as jbd2_journal_get_write_access() could complain
1029 * otherwise about fs integrity issues. Setting of the dirty bit
1030 * by __block_write_begin() isn't a real problem here as we clear
1031 * the bit before releasing a page lock and thus writeback cannot
1032 * ever write the buffer.
1033 */
1034 if (dirty)
1035 clear_buffer_dirty(bh);
1036 BUFFER_TRACE(bh, "get write access");
1037 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1038 EXT4_JTR_NONE);
1039 if (!ret && dirty)
1040 ret = ext4_dirty_journalled_data(handle, bh);
1041 return ret;
1042 }
1043
1044 #ifdef CONFIG_FS_ENCRYPTION
1045 static int ext4_block_write_begin(struct folio *folio, loff_t pos, unsigned len,
1046 get_block_t *get_block)
1047 {
1048 unsigned from = pos & (PAGE_SIZE - 1);
1049 unsigned to = from + len;
1050 struct inode *inode = folio->mapping->host;
1051 unsigned block_start, block_end;
1052 sector_t block;
1053 int err = 0;
1054 unsigned blocksize = inode->i_sb->s_blocksize;
1055 unsigned bbits;
1056 struct buffer_head *bh, *head, *wait[2];
1057 int nr_wait = 0;
1058 int i;
1059
1060 BUG_ON(!folio_test_locked(folio));
1061 BUG_ON(from > PAGE_SIZE);
1062 BUG_ON(to > PAGE_SIZE);
1063 BUG_ON(from > to);
1064
1065 head = folio_buffers(folio);
1066 if (!head)
1067 head = create_empty_buffers(folio, blocksize, 0);
1068 bbits = ilog2(blocksize);
1069 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1070
1071 for (bh = head, block_start = 0; bh != head || !block_start;
1072 block++, block_start = block_end, bh = bh->b_this_page) {
1073 block_end = block_start + blocksize;
1074 if (block_end <= from || block_start >= to) {
1075 if (folio_test_uptodate(folio)) {
1076 set_buffer_uptodate(bh);
1077 }
1078 continue;
1079 }
1080 if (buffer_new(bh))
1081 clear_buffer_new(bh);
1082 if (!buffer_mapped(bh)) {
1083 WARN_ON(bh->b_size != blocksize);
1084 err = get_block(inode, block, bh, 1);
1085 if (err)
1086 break;
1087 if (buffer_new(bh)) {
1088 if (folio_test_uptodate(folio)) {
1089 clear_buffer_new(bh);
1090 set_buffer_uptodate(bh);
1091 mark_buffer_dirty(bh);
1092 continue;
1093 }
1094 if (block_end > to || block_start < from)
1095 folio_zero_segments(folio, to,
1096 block_end,
1097 block_start, from);
1098 continue;
1099 }
1100 }
1101 if (folio_test_uptodate(folio)) {
1102 set_buffer_uptodate(bh);
1103 continue;
1104 }
1105 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1106 !buffer_unwritten(bh) &&
1107 (block_start < from || block_end > to)) {
1108 ext4_read_bh_lock(bh, 0, false);
1109 wait[nr_wait++] = bh;
1110 }
1111 }
1112 /*
1113 * If we issued read requests, let them complete.
1114 */
1115 for (i = 0; i < nr_wait; i++) {
1116 wait_on_buffer(wait[i]);
1117 if (!buffer_uptodate(wait[i]))
1118 err = -EIO;
1119 }
1120 if (unlikely(err)) {
1121 folio_zero_new_buffers(folio, from, to);
1122 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1123 for (i = 0; i < nr_wait; i++) {
1124 int err2;
1125
1126 err2 = fscrypt_decrypt_pagecache_blocks(folio,
1127 blocksize, bh_offset(wait[i]));
1128 if (err2) {
1129 clear_buffer_uptodate(wait[i]);
1130 err = err2;
1131 }
1132 }
1133 }
1134
1135 return err;
1136 }
1137 #endif
1138
1139 /*
1140 * To preserve ordering, it is essential that the hole instantiation and
1141 * the data write be encapsulated in a single transaction. We cannot
1142 * close off a transaction and start a new one between the ext4_get_block()
1143 * and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1144 * ext4_write_begin() is the right place.
1145 */
1146 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1147 loff_t pos, unsigned len,
1148 struct page **pagep, void **fsdata)
1149 {
1150 struct inode *inode = mapping->host;
1151 int ret, needed_blocks;
1152 handle_t *handle;
1153 int retries = 0;
1154 struct folio *folio;
1155 pgoff_t index;
1156 unsigned from, to;
1157
1158 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
1159 return -EIO;
1160
1161 trace_ext4_write_begin(inode, pos, len);
1162 /*
1163 * Reserve one block more for addition to orphan list in case
1164 * we allocate blocks but write fails for some reason
1165 */
1166 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1167 index = pos >> PAGE_SHIFT;
1168 from = pos & (PAGE_SIZE - 1);
1169 to = from + len;
1170
1171 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1172 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1173 pagep);
1174 if (ret < 0)
1175 return ret;
1176 if (ret == 1)
1177 return 0;
1178 }
1179
1180 /*
1181 * __filemap_get_folio() can take a long time if the
1182 * system is thrashing due to memory pressure, or if the folio
1183 * is being written back. So grab it first before we start
1184 * the transaction handle. This also allows us to allocate
1185 * the folio (if needed) without using GFP_NOFS.
1186 */
1187 retry_grab:
1188 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1189 mapping_gfp_mask(mapping));
1190 if (IS_ERR(folio))
1191 return PTR_ERR(folio);
1192 /*
1193 * The same as page allocation, we prealloc buffer heads before
1194 * starting the handle.
1195 */
1196 if (!folio_buffers(folio))
1197 create_empty_buffers(folio, inode->i_sb->s_blocksize, 0);
1198
1199 folio_unlock(folio);
1200
1201 retry_journal:
1202 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1203 if (IS_ERR(handle)) {
1204 folio_put(folio);
1205 return PTR_ERR(handle);
1206 }
1207
1208 folio_lock(folio);
1209 if (folio->mapping != mapping) {
1210 /* The folio got truncated from under us */
1211 folio_unlock(folio);
1212 folio_put(folio);
1213 ext4_journal_stop(handle);
1214 goto retry_grab;
1215 }
1216 /* In case writeback began while the folio was unlocked */
1217 folio_wait_stable(folio);
1218
1219 #ifdef CONFIG_FS_ENCRYPTION
1220 if (ext4_should_dioread_nolock(inode))
1221 ret = ext4_block_write_begin(folio, pos, len,
1222 ext4_get_block_unwritten);
1223 else
1224 ret = ext4_block_write_begin(folio, pos, len, ext4_get_block);
1225 #else
1226 if (ext4_should_dioread_nolock(inode))
1227 ret = __block_write_begin(&folio->page, pos, len,
1228 ext4_get_block_unwritten);
1229 else
1230 ret = __block_write_begin(&folio->page, pos, len, ext4_get_block);
1231 #endif
1232 if (!ret && ext4_should_journal_data(inode)) {
1233 ret = ext4_walk_page_buffers(handle, inode,
1234 folio_buffers(folio), from, to,
1235 NULL, do_journal_get_write_access);
1236 }
1237
1238 if (ret) {
1239 bool extended = (pos + len > inode->i_size) &&
1240 !ext4_verity_in_progress(inode);
1241
1242 folio_unlock(folio);
1243 /*
1244 * __block_write_begin may have instantiated a few blocks
1245 * outside i_size. Trim these off again. Don't need
1246 * i_size_read because we hold i_rwsem.
1247 *
1248 * Add inode to orphan list in case we crash before
1249 * truncate finishes
1250 */
1251 if (extended && ext4_can_truncate(inode))
1252 ext4_orphan_add(handle, inode);
1253
1254 ext4_journal_stop(handle);
1255 if (extended) {
1256 ext4_truncate_failed_write(inode);
1257 /*
1258 * If truncate failed early the inode might
1259 * still be on the orphan list; we need to
1260 * make sure the inode is removed from the
1261 * orphan list in that case.
1262 */
1263 if (inode->i_nlink)
1264 ext4_orphan_del(NULL, inode);
1265 }
1266
1267 if (ret == -ENOSPC &&
1268 ext4_should_retry_alloc(inode->i_sb, &retries))
1269 goto retry_journal;
1270 folio_put(folio);
1271 return ret;
1272 }
1273 *pagep = &folio->page;
1274 return ret;
1275 }
1276
1277 /* For write_end() in data=journal mode */
1278 static int write_end_fn(handle_t *handle, struct inode *inode,
1279 struct buffer_head *bh)
1280 {
1281 int ret;
1282 if (!buffer_mapped(bh) || buffer_freed(bh))
1283 return 0;
1284 set_buffer_uptodate(bh);
1285 ret = ext4_dirty_journalled_data(handle, bh);
1286 clear_buffer_meta(bh);
1287 clear_buffer_prio(bh);
1288 return ret;
1289 }
1290
1291 /*
1292 * We need to pick up the new inode size which generic_commit_write gave us
1293 * `file' can be NULL - eg, when called from page_symlink().
1294 *
1295 * ext4 never places buffers on inode->i_mapping->i_private_list. metadata
1296 * buffers are managed internally.
1297 */
1298 static int ext4_write_end(struct file *file,
1299 struct address_space *mapping,
1300 loff_t pos, unsigned len, unsigned copied,
1301 struct page *page, void *fsdata)
1302 {
1303 struct folio *folio = page_folio(page);
1304 handle_t *handle = ext4_journal_current_handle();
1305 struct inode *inode = mapping->host;
1306 loff_t old_size = inode->i_size;
1307 int ret = 0, ret2;
1308 int i_size_changed = 0;
1309 bool verity = ext4_verity_in_progress(inode);
1310
1311 trace_ext4_write_end(inode, pos, len, copied);
1312
1313 if (ext4_has_inline_data(inode) &&
1314 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1315 return ext4_write_inline_data_end(inode, pos, len, copied,
1316 folio);
1317
1318 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1319 /*
1320 * it's important to update i_size while still holding folio lock:
1321 * page writeout could otherwise come in and zero beyond i_size.
1322 *
1323 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1324 * blocks are being written past EOF, so skip the i_size update.
1325 */
1326 if (!verity)
1327 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1328 folio_unlock(folio);
1329 folio_put(folio);
1330
1331 if (old_size < pos && !verity)
1332 pagecache_isize_extended(inode, old_size, pos);
1333 /*
1334 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1335 * makes the holding time of folio lock longer. Second, it forces lock
1336 * ordering of folio lock and transaction start for journaling
1337 * filesystems.
1338 */
1339 if (i_size_changed)
1340 ret = ext4_mark_inode_dirty(handle, inode);
1341
1342 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1343 /* if we have allocated more blocks and copied
1344 * less. We will have blocks allocated outside
1345 * inode->i_size. So truncate them
1346 */
1347 ext4_orphan_add(handle, inode);
1348
1349 ret2 = ext4_journal_stop(handle);
1350 if (!ret)
1351 ret = ret2;
1352
1353 if (pos + len > inode->i_size && !verity) {
1354 ext4_truncate_failed_write(inode);
1355 /*
1356 * If truncate failed early the inode might still be
1357 * on the orphan list; we need to make sure the inode
1358 * is removed from the orphan list in that case.
1359 */
1360 if (inode->i_nlink)
1361 ext4_orphan_del(NULL, inode);
1362 }
1363
1364 return ret ? ret : copied;
1365 }
1366
1367 /*
1368 * This is a private version of folio_zero_new_buffers() which doesn't
1369 * set the buffer to be dirty, since in data=journalled mode we need
1370 * to call ext4_dirty_journalled_data() instead.
1371 */
1372 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1373 struct inode *inode,
1374 struct folio *folio,
1375 unsigned from, unsigned to)
1376 {
1377 unsigned int block_start = 0, block_end;
1378 struct buffer_head *head, *bh;
1379
1380 bh = head = folio_buffers(folio);
1381 do {
1382 block_end = block_start + bh->b_size;
1383 if (buffer_new(bh)) {
1384 if (block_end > from && block_start < to) {
1385 if (!folio_test_uptodate(folio)) {
1386 unsigned start, size;
1387
1388 start = max(from, block_start);
1389 size = min(to, block_end) - start;
1390
1391 folio_zero_range(folio, start, size);
1392 write_end_fn(handle, inode, bh);
1393 }
1394 clear_buffer_new(bh);
1395 }
1396 }
1397 block_start = block_end;
1398 bh = bh->b_this_page;
1399 } while (bh != head);
1400 }
1401
1402 static int ext4_journalled_write_end(struct file *file,
1403 struct address_space *mapping,
1404 loff_t pos, unsigned len, unsigned copied,
1405 struct page *page, void *fsdata)
1406 {
1407 struct folio *folio = page_folio(page);
1408 handle_t *handle = ext4_journal_current_handle();
1409 struct inode *inode = mapping->host;
1410 loff_t old_size = inode->i_size;
1411 int ret = 0, ret2;
1412 int partial = 0;
1413 unsigned from, to;
1414 int size_changed = 0;
1415 bool verity = ext4_verity_in_progress(inode);
1416
1417 trace_ext4_journalled_write_end(inode, pos, len, copied);
1418 from = pos & (PAGE_SIZE - 1);
1419 to = from + len;
1420
1421 BUG_ON(!ext4_handle_valid(handle));
1422
1423 if (ext4_has_inline_data(inode))
1424 return ext4_write_inline_data_end(inode, pos, len, copied,
1425 folio);
1426
1427 if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1428 copied = 0;
1429 ext4_journalled_zero_new_buffers(handle, inode, folio,
1430 from, to);
1431 } else {
1432 if (unlikely(copied < len))
1433 ext4_journalled_zero_new_buffers(handle, inode, folio,
1434 from + copied, to);
1435 ret = ext4_walk_page_buffers(handle, inode,
1436 folio_buffers(folio),
1437 from, from + copied, &partial,
1438 write_end_fn);
1439 if (!partial)
1440 folio_mark_uptodate(folio);
1441 }
1442 if (!verity)
1443 size_changed = ext4_update_inode_size(inode, pos + copied);
1444 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1445 folio_unlock(folio);
1446 folio_put(folio);
1447
1448 if (old_size < pos && !verity)
1449 pagecache_isize_extended(inode, old_size, pos);
1450
1451 if (size_changed) {
1452 ret2 = ext4_mark_inode_dirty(handle, inode);
1453 if (!ret)
1454 ret = ret2;
1455 }
1456
1457 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1458 /* if we have allocated more blocks and copied
1459 * less. We will have blocks allocated outside
1460 * inode->i_size. So truncate them
1461 */
1462 ext4_orphan_add(handle, inode);
1463
1464 ret2 = ext4_journal_stop(handle);
1465 if (!ret)
1466 ret = ret2;
1467 if (pos + len > inode->i_size && !verity) {
1468 ext4_truncate_failed_write(inode);
1469 /*
1470 * If truncate failed early the inode might still be
1471 * on the orphan list; we need to make sure the inode
1472 * is removed from the orphan list in that case.
1473 */
1474 if (inode->i_nlink)
1475 ext4_orphan_del(NULL, inode);
1476 }
1477
1478 return ret ? ret : copied;
1479 }
1480
1481 /*
1482 * Reserve space for 'nr_resv' clusters
1483 */
1484 static int ext4_da_reserve_space(struct inode *inode, int nr_resv)
1485 {
1486 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1487 struct ext4_inode_info *ei = EXT4_I(inode);
1488 int ret;
1489
1490 /*
1491 * We will charge metadata quota at writeout time; this saves
1492 * us from metadata over-estimation, though we may go over by
1493 * a small amount in the end. Here we just reserve for data.
1494 */
1495 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv));
1496 if (ret)
1497 return ret;
1498
1499 spin_lock(&ei->i_block_reservation_lock);
1500 if (ext4_claim_free_clusters(sbi, nr_resv, 0)) {
1501 spin_unlock(&ei->i_block_reservation_lock);
1502 dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv));
1503 return -ENOSPC;
1504 }
1505 ei->i_reserved_data_blocks += nr_resv;
1506 trace_ext4_da_reserve_space(inode, nr_resv);
1507 spin_unlock(&ei->i_block_reservation_lock);
1508
1509 return 0; /* success */
1510 }
1511
1512 void ext4_da_release_space(struct inode *inode, int to_free)
1513 {
1514 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1515 struct ext4_inode_info *ei = EXT4_I(inode);
1516
1517 if (!to_free)
1518 return; /* Nothing to release, exit */
1519
1520 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1521
1522 trace_ext4_da_release_space(inode, to_free);
1523 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1524 /*
1525 * if there aren't enough reserved blocks, then the
1526 * counter is messed up somewhere. Since this
1527 * function is called from invalidate page, it's
1528 * harmless to return without any action.
1529 */
1530 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1531 "ino %lu, to_free %d with only %d reserved "
1532 "data blocks", inode->i_ino, to_free,
1533 ei->i_reserved_data_blocks);
1534 WARN_ON(1);
1535 to_free = ei->i_reserved_data_blocks;
1536 }
1537 ei->i_reserved_data_blocks -= to_free;
1538
1539 /* update fs dirty data blocks counter */
1540 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1541
1542 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1543
1544 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1545 }
1546
1547 /*
1548 * Delayed allocation stuff
1549 */
1550
1551 struct mpage_da_data {
1552 /* These are input fields for ext4_do_writepages() */
1553 struct inode *inode;
1554 struct writeback_control *wbc;
1555 unsigned int can_map:1; /* Can writepages call map blocks? */
1556
1557 /* These are internal state of ext4_do_writepages() */
1558 pgoff_t first_page; /* The first page to write */
1559 pgoff_t next_page; /* Current page to examine */
1560 pgoff_t last_page; /* Last page to examine */
1561 /*
1562 * Extent to map - this can be after first_page because that can be
1563 * fully mapped. We somewhat abuse m_flags to store whether the extent
1564 * is delalloc or unwritten.
1565 */
1566 struct ext4_map_blocks map;
1567 struct ext4_io_submit io_submit; /* IO submission data */
1568 unsigned int do_map:1;
1569 unsigned int scanned_until_end:1;
1570 unsigned int journalled_more_data:1;
1571 };
1572
1573 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1574 bool invalidate)
1575 {
1576 unsigned nr, i;
1577 pgoff_t index, end;
1578 struct folio_batch fbatch;
1579 struct inode *inode = mpd->inode;
1580 struct address_space *mapping = inode->i_mapping;
1581
1582 /* This is necessary when next_page == 0. */
1583 if (mpd->first_page >= mpd->next_page)
1584 return;
1585
1586 mpd->scanned_until_end = 0;
1587 index = mpd->first_page;
1588 end = mpd->next_page - 1;
1589 if (invalidate) {
1590 ext4_lblk_t start, last;
1591 start = index << (PAGE_SHIFT - inode->i_blkbits);
1592 last = end << (PAGE_SHIFT - inode->i_blkbits);
1593
1594 /*
1595 * avoid racing with extent status tree scans made by
1596 * ext4_insert_delayed_block()
1597 */
1598 down_write(&EXT4_I(inode)->i_data_sem);
1599 ext4_es_remove_extent(inode, start, last - start + 1);
1600 up_write(&EXT4_I(inode)->i_data_sem);
1601 }
1602
1603 folio_batch_init(&fbatch);
1604 while (index <= end) {
1605 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1606 if (nr == 0)
1607 break;
1608 for (i = 0; i < nr; i++) {
1609 struct folio *folio = fbatch.folios[i];
1610
1611 if (folio->index < mpd->first_page)
1612 continue;
1613 if (folio_next_index(folio) - 1 > end)
1614 continue;
1615 BUG_ON(!folio_test_locked(folio));
1616 BUG_ON(folio_test_writeback(folio));
1617 if (invalidate) {
1618 if (folio_mapped(folio))
1619 folio_clear_dirty_for_io(folio);
1620 block_invalidate_folio(folio, 0,
1621 folio_size(folio));
1622 folio_clear_uptodate(folio);
1623 }
1624 folio_unlock(folio);
1625 }
1626 folio_batch_release(&fbatch);
1627 }
1628 }
1629
1630 static void ext4_print_free_blocks(struct inode *inode)
1631 {
1632 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1633 struct super_block *sb = inode->i_sb;
1634 struct ext4_inode_info *ei = EXT4_I(inode);
1635
1636 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1637 EXT4_C2B(EXT4_SB(inode->i_sb),
1638 ext4_count_free_clusters(sb)));
1639 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1640 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1641 (long long) EXT4_C2B(EXT4_SB(sb),
1642 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1643 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1644 (long long) EXT4_C2B(EXT4_SB(sb),
1645 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1646 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1647 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1648 ei->i_reserved_data_blocks);
1649 return;
1650 }
1651
1652 /*
1653 * Check whether the cluster containing lblk has been allocated or has
1654 * delalloc reservation.
1655 *
1656 * Returns 0 if the cluster doesn't have either, 1 if it has delalloc
1657 * reservation, 2 if it's already been allocated, negative error code on
1658 * failure.
1659 */
1660 static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk)
1661 {
1662 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1663 int ret;
1664
1665 /* Has delalloc reservation? */
1666 if (ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk))
1667 return 1;
1668
1669 /* Already been allocated? */
1670 if (ext4_es_scan_clu(inode, &ext4_es_is_mapped, lblk))
1671 return 2;
1672 ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk));
1673 if (ret < 0)
1674 return ret;
1675 if (ret > 0)
1676 return 2;
1677
1678 return 0;
1679 }
1680
1681 /*
1682 * ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents
1683 * status tree, incrementing the reserved
1684 * cluster/block count or making pending
1685 * reservations where needed
1686 *
1687 * @inode - file containing the newly added block
1688 * @lblk - start logical block to be added
1689 * @len - length of blocks to be added
1690 *
1691 * Returns 0 on success, negative error code on failure.
1692 */
1693 static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk,
1694 ext4_lblk_t len)
1695 {
1696 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1697 int ret;
1698 bool lclu_allocated = false;
1699 bool end_allocated = false;
1700 ext4_lblk_t resv_clu;
1701 ext4_lblk_t end = lblk + len - 1;
1702
1703 /*
1704 * If the cluster containing lblk or end is shared with a delayed,
1705 * written, or unwritten extent in a bigalloc file system, it's
1706 * already been accounted for and does not need to be reserved.
1707 * A pending reservation must be made for the cluster if it's
1708 * shared with a written or unwritten extent and doesn't already
1709 * have one. Written and unwritten extents can be purged from the
1710 * extents status tree if the system is under memory pressure, so
1711 * it's necessary to examine the extent tree if a search of the
1712 * extents status tree doesn't get a match.
1713 */
1714 if (sbi->s_cluster_ratio == 1) {
1715 ret = ext4_da_reserve_space(inode, len);
1716 if (ret != 0) /* ENOSPC */
1717 return ret;
1718 } else { /* bigalloc */
1719 resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1;
1720
1721 ret = ext4_clu_alloc_state(inode, lblk);
1722 if (ret < 0)
1723 return ret;
1724 if (ret > 0) {
1725 resv_clu--;
1726 lclu_allocated = (ret == 2);
1727 }
1728
1729 if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) {
1730 ret = ext4_clu_alloc_state(inode, end);
1731 if (ret < 0)
1732 return ret;
1733 if (ret > 0) {
1734 resv_clu--;
1735 end_allocated = (ret == 2);
1736 }
1737 }
1738
1739 if (resv_clu) {
1740 ret = ext4_da_reserve_space(inode, resv_clu);
1741 if (ret != 0) /* ENOSPC */
1742 return ret;
1743 }
1744 }
1745
1746 ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated,
1747 end_allocated);
1748 return 0;
1749 }
1750
1751 /*
1752 * Looks up the requested blocks and sets the delalloc extent map.
1753 * First try to look up for the extent entry that contains the requested
1754 * blocks in the extent status tree without i_data_sem, then try to look
1755 * up for the ondisk extent mapping with i_data_sem in read mode,
1756 * finally hold i_data_sem in write mode, looks up again and add a
1757 * delalloc extent entry if it still couldn't find any extent. Pass out
1758 * the mapped extent through @map and return 0 on success.
1759 */
1760 static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map)
1761 {
1762 struct extent_status es;
1763 int retval;
1764 #ifdef ES_AGGRESSIVE_TEST
1765 struct ext4_map_blocks orig_map;
1766
1767 memcpy(&orig_map, map, sizeof(*map));
1768 #endif
1769
1770 map->m_flags = 0;
1771 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1772 (unsigned long) map->m_lblk);
1773
1774 /* Lookup extent status tree firstly */
1775 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
1776 map->m_len = min_t(unsigned int, map->m_len,
1777 es.es_len - (map->m_lblk - es.es_lblk));
1778
1779 if (ext4_es_is_hole(&es))
1780 goto add_delayed;
1781
1782 found:
1783 /*
1784 * Delayed extent could be allocated by fallocate.
1785 * So we need to check it.
1786 */
1787 if (ext4_es_is_delonly(&es)) {
1788 map->m_flags |= EXT4_MAP_DELAYED;
1789 return 0;
1790 }
1791
1792 map->m_pblk = ext4_es_pblock(&es) + map->m_lblk - es.es_lblk;
1793 if (ext4_es_is_written(&es))
1794 map->m_flags |= EXT4_MAP_MAPPED;
1795 else if (ext4_es_is_unwritten(&es))
1796 map->m_flags |= EXT4_MAP_UNWRITTEN;
1797 else
1798 BUG();
1799
1800 #ifdef ES_AGGRESSIVE_TEST
1801 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1802 #endif
1803 return 0;
1804 }
1805
1806 /*
1807 * Try to see if we can get the block without requesting a new
1808 * file system block.
1809 */
1810 down_read(&EXT4_I(inode)->i_data_sem);
1811 if (ext4_has_inline_data(inode))
1812 retval = 0;
1813 else
1814 retval = ext4_map_query_blocks(NULL, inode, map);
1815 up_read(&EXT4_I(inode)->i_data_sem);
1816 if (retval)
1817 return retval < 0 ? retval : 0;
1818
1819 add_delayed:
1820 down_write(&EXT4_I(inode)->i_data_sem);
1821 /*
1822 * Page fault path (ext4_page_mkwrite does not take i_rwsem)
1823 * and fallocate path (no folio lock) can race. Make sure we
1824 * lookup the extent status tree here again while i_data_sem
1825 * is held in write mode, before inserting a new da entry in
1826 * the extent status tree.
1827 */
1828 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
1829 map->m_len = min_t(unsigned int, map->m_len,
1830 es.es_len - (map->m_lblk - es.es_lblk));
1831
1832 if (!ext4_es_is_hole(&es)) {
1833 up_write(&EXT4_I(inode)->i_data_sem);
1834 goto found;
1835 }
1836 } else if (!ext4_has_inline_data(inode)) {
1837 retval = ext4_map_query_blocks(NULL, inode, map);
1838 if (retval) {
1839 up_write(&EXT4_I(inode)->i_data_sem);
1840 return retval < 0 ? retval : 0;
1841 }
1842 }
1843
1844 map->m_flags |= EXT4_MAP_DELAYED;
1845 retval = ext4_insert_delayed_blocks(inode, map->m_lblk, map->m_len);
1846 up_write(&EXT4_I(inode)->i_data_sem);
1847
1848 return retval;
1849 }
1850
1851 /*
1852 * This is a special get_block_t callback which is used by
1853 * ext4_da_write_begin(). It will either return mapped block or
1854 * reserve space for a single block.
1855 *
1856 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1857 * We also have b_blocknr = -1 and b_bdev initialized properly
1858 *
1859 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1860 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1861 * initialized properly.
1862 */
1863 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1864 struct buffer_head *bh, int create)
1865 {
1866 struct ext4_map_blocks map;
1867 sector_t invalid_block = ~((sector_t) 0xffff);
1868 int ret = 0;
1869
1870 BUG_ON(create == 0);
1871 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1872
1873 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1874 invalid_block = ~0;
1875
1876 map.m_lblk = iblock;
1877 map.m_len = 1;
1878
1879 /*
1880 * first, we need to know whether the block is allocated already
1881 * preallocated blocks are unmapped but should treated
1882 * the same as allocated blocks.
1883 */
1884 ret = ext4_da_map_blocks(inode, &map);
1885 if (ret < 0)
1886 return ret;
1887
1888 if (map.m_flags & EXT4_MAP_DELAYED) {
1889 map_bh(bh, inode->i_sb, invalid_block);
1890 set_buffer_new(bh);
1891 set_buffer_delay(bh);
1892 return 0;
1893 }
1894
1895 map_bh(bh, inode->i_sb, map.m_pblk);
1896 ext4_update_bh_state(bh, map.m_flags);
1897
1898 if (buffer_unwritten(bh)) {
1899 /* A delayed write to unwritten bh should be marked
1900 * new and mapped. Mapped ensures that we don't do
1901 * get_block multiple times when we write to the same
1902 * offset and new ensures that we do proper zero out
1903 * for partial write.
1904 */
1905 set_buffer_new(bh);
1906 set_buffer_mapped(bh);
1907 }
1908 return 0;
1909 }
1910
1911 static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
1912 {
1913 mpd->first_page += folio_nr_pages(folio);
1914 folio_unlock(folio);
1915 }
1916
1917 static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1918 {
1919 size_t len;
1920 loff_t size;
1921 int err;
1922
1923 BUG_ON(folio->index != mpd->first_page);
1924 folio_clear_dirty_for_io(folio);
1925 /*
1926 * We have to be very careful here! Nothing protects writeback path
1927 * against i_size changes and the page can be writeably mapped into
1928 * page tables. So an application can be growing i_size and writing
1929 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1930 * write-protects our page in page tables and the page cannot get
1931 * written to again until we release folio lock. So only after
1932 * folio_clear_dirty_for_io() we are safe to sample i_size for
1933 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1934 * on the barrier provided by folio_test_clear_dirty() in
1935 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1936 * after page tables are updated.
1937 */
1938 size = i_size_read(mpd->inode);
1939 len = folio_size(folio);
1940 if (folio_pos(folio) + len > size &&
1941 !ext4_verity_in_progress(mpd->inode))
1942 len = size & (len - 1);
1943 err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
1944 if (!err)
1945 mpd->wbc->nr_to_write--;
1946
1947 return err;
1948 }
1949
1950 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1951
1952 /*
1953 * mballoc gives us at most this number of blocks...
1954 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1955 * The rest of mballoc seems to handle chunks up to full group size.
1956 */
1957 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1958
1959 /*
1960 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1961 *
1962 * @mpd - extent of blocks
1963 * @lblk - logical number of the block in the file
1964 * @bh - buffer head we want to add to the extent
1965 *
1966 * The function is used to collect contig. blocks in the same state. If the
1967 * buffer doesn't require mapping for writeback and we haven't started the
1968 * extent of buffers to map yet, the function returns 'true' immediately - the
1969 * caller can write the buffer right away. Otherwise the function returns true
1970 * if the block has been added to the extent, false if the block couldn't be
1971 * added.
1972 */
1973 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1974 struct buffer_head *bh)
1975 {
1976 struct ext4_map_blocks *map = &mpd->map;
1977
1978 /* Buffer that doesn't need mapping for writeback? */
1979 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1980 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1981 /* So far no extent to map => we write the buffer right away */
1982 if (map->m_len == 0)
1983 return true;
1984 return false;
1985 }
1986
1987 /* First block in the extent? */
1988 if (map->m_len == 0) {
1989 /* We cannot map unless handle is started... */
1990 if (!mpd->do_map)
1991 return false;
1992 map->m_lblk = lblk;
1993 map->m_len = 1;
1994 map->m_flags = bh->b_state & BH_FLAGS;
1995 return true;
1996 }
1997
1998 /* Don't go larger than mballoc is willing to allocate */
1999 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2000 return false;
2001
2002 /* Can we merge the block to our big extent? */
2003 if (lblk == map->m_lblk + map->m_len &&
2004 (bh->b_state & BH_FLAGS) == map->m_flags) {
2005 map->m_len++;
2006 return true;
2007 }
2008 return false;
2009 }
2010
2011 /*
2012 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2013 *
2014 * @mpd - extent of blocks for mapping
2015 * @head - the first buffer in the page
2016 * @bh - buffer we should start processing from
2017 * @lblk - logical number of the block in the file corresponding to @bh
2018 *
2019 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2020 * the page for IO if all buffers in this page were mapped and there's no
2021 * accumulated extent of buffers to map or add buffers in the page to the
2022 * extent of buffers to map. The function returns 1 if the caller can continue
2023 * by processing the next page, 0 if it should stop adding buffers to the
2024 * extent to map because we cannot extend it anymore. It can also return value
2025 * < 0 in case of error during IO submission.
2026 */
2027 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2028 struct buffer_head *head,
2029 struct buffer_head *bh,
2030 ext4_lblk_t lblk)
2031 {
2032 struct inode *inode = mpd->inode;
2033 int err;
2034 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2035 >> inode->i_blkbits;
2036
2037 if (ext4_verity_in_progress(inode))
2038 blocks = EXT_MAX_BLOCKS;
2039
2040 do {
2041 BUG_ON(buffer_locked(bh));
2042
2043 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2044 /* Found extent to map? */
2045 if (mpd->map.m_len)
2046 return 0;
2047 /* Buffer needs mapping and handle is not started? */
2048 if (!mpd->do_map)
2049 return 0;
2050 /* Everything mapped so far and we hit EOF */
2051 break;
2052 }
2053 } while (lblk++, (bh = bh->b_this_page) != head);
2054 /* So far everything mapped? Submit the page for IO. */
2055 if (mpd->map.m_len == 0) {
2056 err = mpage_submit_folio(mpd, head->b_folio);
2057 if (err < 0)
2058 return err;
2059 mpage_folio_done(mpd, head->b_folio);
2060 }
2061 if (lblk >= blocks) {
2062 mpd->scanned_until_end = 1;
2063 return 0;
2064 }
2065 return 1;
2066 }
2067
2068 /*
2069 * mpage_process_folio - update folio buffers corresponding to changed extent
2070 * and may submit fully mapped page for IO
2071 * @mpd: description of extent to map, on return next extent to map
2072 * @folio: Contains these buffers.
2073 * @m_lblk: logical block mapping.
2074 * @m_pblk: corresponding physical mapping.
2075 * @map_bh: determines on return whether this page requires any further
2076 * mapping or not.
2077 *
2078 * Scan given folio buffers corresponding to changed extent and update buffer
2079 * state according to new extent state.
2080 * We map delalloc buffers to their physical location, clear unwritten bits.
2081 * If the given folio is not fully mapped, we update @mpd to the next extent in
2082 * the given folio that needs mapping & return @map_bh as true.
2083 */
2084 static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2085 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2086 bool *map_bh)
2087 {
2088 struct buffer_head *head, *bh;
2089 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2090 ext4_lblk_t lblk = *m_lblk;
2091 ext4_fsblk_t pblock = *m_pblk;
2092 int err = 0;
2093 int blkbits = mpd->inode->i_blkbits;
2094 ssize_t io_end_size = 0;
2095 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2096
2097 bh = head = folio_buffers(folio);
2098 do {
2099 if (lblk < mpd->map.m_lblk)
2100 continue;
2101 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2102 /*
2103 * Buffer after end of mapped extent.
2104 * Find next buffer in the folio to map.
2105 */
2106 mpd->map.m_len = 0;
2107 mpd->map.m_flags = 0;
2108 io_end_vec->size += io_end_size;
2109
2110 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2111 if (err > 0)
2112 err = 0;
2113 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2114 io_end_vec = ext4_alloc_io_end_vec(io_end);
2115 if (IS_ERR(io_end_vec)) {
2116 err = PTR_ERR(io_end_vec);
2117 goto out;
2118 }
2119 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2120 }
2121 *map_bh = true;
2122 goto out;
2123 }
2124 if (buffer_delay(bh)) {
2125 clear_buffer_delay(bh);
2126 bh->b_blocknr = pblock++;
2127 }
2128 clear_buffer_unwritten(bh);
2129 io_end_size += (1 << blkbits);
2130 } while (lblk++, (bh = bh->b_this_page) != head);
2131
2132 io_end_vec->size += io_end_size;
2133 *map_bh = false;
2134 out:
2135 *m_lblk = lblk;
2136 *m_pblk = pblock;
2137 return err;
2138 }
2139
2140 /*
2141 * mpage_map_buffers - update buffers corresponding to changed extent and
2142 * submit fully mapped pages for IO
2143 *
2144 * @mpd - description of extent to map, on return next extent to map
2145 *
2146 * Scan buffers corresponding to changed extent (we expect corresponding pages
2147 * to be already locked) and update buffer state according to new extent state.
2148 * We map delalloc buffers to their physical location, clear unwritten bits,
2149 * and mark buffers as uninit when we perform writes to unwritten extents
2150 * and do extent conversion after IO is finished. If the last page is not fully
2151 * mapped, we update @map to the next extent in the last page that needs
2152 * mapping. Otherwise we submit the page for IO.
2153 */
2154 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2155 {
2156 struct folio_batch fbatch;
2157 unsigned nr, i;
2158 struct inode *inode = mpd->inode;
2159 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2160 pgoff_t start, end;
2161 ext4_lblk_t lblk;
2162 ext4_fsblk_t pblock;
2163 int err;
2164 bool map_bh = false;
2165
2166 start = mpd->map.m_lblk >> bpp_bits;
2167 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2168 lblk = start << bpp_bits;
2169 pblock = mpd->map.m_pblk;
2170
2171 folio_batch_init(&fbatch);
2172 while (start <= end) {
2173 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2174 if (nr == 0)
2175 break;
2176 for (i = 0; i < nr; i++) {
2177 struct folio *folio = fbatch.folios[i];
2178
2179 err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2180 &map_bh);
2181 /*
2182 * If map_bh is true, means page may require further bh
2183 * mapping, or maybe the page was submitted for IO.
2184 * So we return to call further extent mapping.
2185 */
2186 if (err < 0 || map_bh)
2187 goto out;
2188 /* Page fully mapped - let IO run! */
2189 err = mpage_submit_folio(mpd, folio);
2190 if (err < 0)
2191 goto out;
2192 mpage_folio_done(mpd, folio);
2193 }
2194 folio_batch_release(&fbatch);
2195 }
2196 /* Extent fully mapped and matches with page boundary. We are done. */
2197 mpd->map.m_len = 0;
2198 mpd->map.m_flags = 0;
2199 return 0;
2200 out:
2201 folio_batch_release(&fbatch);
2202 return err;
2203 }
2204
2205 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2206 {
2207 struct inode *inode = mpd->inode;
2208 struct ext4_map_blocks *map = &mpd->map;
2209 int get_blocks_flags;
2210 int err, dioread_nolock;
2211
2212 trace_ext4_da_write_pages_extent(inode, map);
2213 /*
2214 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2215 * to convert an unwritten extent to be initialized (in the case
2216 * where we have written into one or more preallocated blocks). It is
2217 * possible that we're going to need more metadata blocks than
2218 * previously reserved. However we must not fail because we're in
2219 * writeback and there is nothing we can do about it so it might result
2220 * in data loss. So use reserved blocks to allocate metadata if
2221 * possible.
2222 *
2223 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2224 * the blocks in question are delalloc blocks. This indicates
2225 * that the blocks and quotas has already been checked when
2226 * the data was copied into the page cache.
2227 */
2228 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2229 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2230 EXT4_GET_BLOCKS_IO_SUBMIT;
2231 dioread_nolock = ext4_should_dioread_nolock(inode);
2232 if (dioread_nolock)
2233 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2234 if (map->m_flags & BIT(BH_Delay))
2235 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2236
2237 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2238 if (err < 0)
2239 return err;
2240 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2241 if (!mpd->io_submit.io_end->handle &&
2242 ext4_handle_valid(handle)) {
2243 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2244 handle->h_rsv_handle = NULL;
2245 }
2246 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2247 }
2248
2249 BUG_ON(map->m_len == 0);
2250 return 0;
2251 }
2252
2253 /*
2254 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2255 * mpd->len and submit pages underlying it for IO
2256 *
2257 * @handle - handle for journal operations
2258 * @mpd - extent to map
2259 * @give_up_on_write - we set this to true iff there is a fatal error and there
2260 * is no hope of writing the data. The caller should discard
2261 * dirty pages to avoid infinite loops.
2262 *
2263 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2264 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2265 * them to initialized or split the described range from larger unwritten
2266 * extent. Note that we need not map all the described range since allocation
2267 * can return less blocks or the range is covered by more unwritten extents. We
2268 * cannot map more because we are limited by reserved transaction credits. On
2269 * the other hand we always make sure that the last touched page is fully
2270 * mapped so that it can be written out (and thus forward progress is
2271 * guaranteed). After mapping we submit all mapped pages for IO.
2272 */
2273 static int mpage_map_and_submit_extent(handle_t *handle,
2274 struct mpage_da_data *mpd,
2275 bool *give_up_on_write)
2276 {
2277 struct inode *inode = mpd->inode;
2278 struct ext4_map_blocks *map = &mpd->map;
2279 int err;
2280 loff_t disksize;
2281 int progress = 0;
2282 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2283 struct ext4_io_end_vec *io_end_vec;
2284
2285 io_end_vec = ext4_alloc_io_end_vec(io_end);
2286 if (IS_ERR(io_end_vec))
2287 return PTR_ERR(io_end_vec);
2288 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2289 do {
2290 err = mpage_map_one_extent(handle, mpd);
2291 if (err < 0) {
2292 struct super_block *sb = inode->i_sb;
2293
2294 if (ext4_forced_shutdown(sb))
2295 goto invalidate_dirty_pages;
2296 /*
2297 * Let the uper layers retry transient errors.
2298 * In the case of ENOSPC, if ext4_count_free_blocks()
2299 * is non-zero, a commit should free up blocks.
2300 */
2301 if ((err == -ENOMEM) ||
2302 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2303 if (progress)
2304 goto update_disksize;
2305 return err;
2306 }
2307 ext4_msg(sb, KERN_CRIT,
2308 "Delayed block allocation failed for "
2309 "inode %lu at logical offset %llu with"
2310 " max blocks %u with error %d",
2311 inode->i_ino,
2312 (unsigned long long)map->m_lblk,
2313 (unsigned)map->m_len, -err);
2314 ext4_msg(sb, KERN_CRIT,
2315 "This should not happen!! Data will "
2316 "be lost\n");
2317 if (err == -ENOSPC)
2318 ext4_print_free_blocks(inode);
2319 invalidate_dirty_pages:
2320 *give_up_on_write = true;
2321 return err;
2322 }
2323 progress = 1;
2324 /*
2325 * Update buffer state, submit mapped pages, and get us new
2326 * extent to map
2327 */
2328 err = mpage_map_and_submit_buffers(mpd);
2329 if (err < 0)
2330 goto update_disksize;
2331 } while (map->m_len);
2332
2333 update_disksize:
2334 /*
2335 * Update on-disk size after IO is submitted. Races with
2336 * truncate are avoided by checking i_size under i_data_sem.
2337 */
2338 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2339 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2340 int err2;
2341 loff_t i_size;
2342
2343 down_write(&EXT4_I(inode)->i_data_sem);
2344 i_size = i_size_read(inode);
2345 if (disksize > i_size)
2346 disksize = i_size;
2347 if (disksize > EXT4_I(inode)->i_disksize)
2348 EXT4_I(inode)->i_disksize = disksize;
2349 up_write(&EXT4_I(inode)->i_data_sem);
2350 err2 = ext4_mark_inode_dirty(handle, inode);
2351 if (err2) {
2352 ext4_error_err(inode->i_sb, -err2,
2353 "Failed to mark inode %lu dirty",
2354 inode->i_ino);
2355 }
2356 if (!err)
2357 err = err2;
2358 }
2359 return err;
2360 }
2361
2362 /*
2363 * Calculate the total number of credits to reserve for one writepages
2364 * iteration. This is called from ext4_writepages(). We map an extent of
2365 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2366 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2367 * bpp - 1 blocks in bpp different extents.
2368 */
2369 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2370 {
2371 int bpp = ext4_journal_blocks_per_page(inode);
2372
2373 return ext4_meta_trans_blocks(inode,
2374 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2375 }
2376
2377 static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2378 size_t len)
2379 {
2380 struct buffer_head *page_bufs = folio_buffers(folio);
2381 struct inode *inode = folio->mapping->host;
2382 int ret, err;
2383
2384 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2385 NULL, do_journal_get_write_access);
2386 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2387 NULL, write_end_fn);
2388 if (ret == 0)
2389 ret = err;
2390 err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len);
2391 if (ret == 0)
2392 ret = err;
2393 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2394
2395 return ret;
2396 }
2397
2398 static int mpage_journal_page_buffers(handle_t *handle,
2399 struct mpage_da_data *mpd,
2400 struct folio *folio)
2401 {
2402 struct inode *inode = mpd->inode;
2403 loff_t size = i_size_read(inode);
2404 size_t len = folio_size(folio);
2405
2406 folio_clear_checked(folio);
2407 mpd->wbc->nr_to_write--;
2408
2409 if (folio_pos(folio) + len > size &&
2410 !ext4_verity_in_progress(inode))
2411 len = size & (len - 1);
2412
2413 return ext4_journal_folio_buffers(handle, folio, len);
2414 }
2415
2416 /*
2417 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2418 * needing mapping, submit mapped pages
2419 *
2420 * @mpd - where to look for pages
2421 *
2422 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2423 * IO immediately. If we cannot map blocks, we submit just already mapped
2424 * buffers in the page for IO and keep page dirty. When we can map blocks and
2425 * we find a page which isn't mapped we start accumulating extent of buffers
2426 * underlying these pages that needs mapping (formed by either delayed or
2427 * unwritten buffers). We also lock the pages containing these buffers. The
2428 * extent found is returned in @mpd structure (starting at mpd->lblk with
2429 * length mpd->len blocks).
2430 *
2431 * Note that this function can attach bios to one io_end structure which are
2432 * neither logically nor physically contiguous. Although it may seem as an
2433 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2434 * case as we need to track IO to all buffers underlying a page in one io_end.
2435 */
2436 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2437 {
2438 struct address_space *mapping = mpd->inode->i_mapping;
2439 struct folio_batch fbatch;
2440 unsigned int nr_folios;
2441 pgoff_t index = mpd->first_page;
2442 pgoff_t end = mpd->last_page;
2443 xa_mark_t tag;
2444 int i, err = 0;
2445 int blkbits = mpd->inode->i_blkbits;
2446 ext4_lblk_t lblk;
2447 struct buffer_head *head;
2448 handle_t *handle = NULL;
2449 int bpp = ext4_journal_blocks_per_page(mpd->inode);
2450
2451 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2452 tag = PAGECACHE_TAG_TOWRITE;
2453 else
2454 tag = PAGECACHE_TAG_DIRTY;
2455
2456 mpd->map.m_len = 0;
2457 mpd->next_page = index;
2458 if (ext4_should_journal_data(mpd->inode)) {
2459 handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2460 bpp);
2461 if (IS_ERR(handle))
2462 return PTR_ERR(handle);
2463 }
2464 folio_batch_init(&fbatch);
2465 while (index <= end) {
2466 nr_folios = filemap_get_folios_tag(mapping, &index, end,
2467 tag, &fbatch);
2468 if (nr_folios == 0)
2469 break;
2470
2471 for (i = 0; i < nr_folios; i++) {
2472 struct folio *folio = fbatch.folios[i];
2473
2474 /*
2475 * Accumulated enough dirty pages? This doesn't apply
2476 * to WB_SYNC_ALL mode. For integrity sync we have to
2477 * keep going because someone may be concurrently
2478 * dirtying pages, and we might have synced a lot of
2479 * newly appeared dirty pages, but have not synced all
2480 * of the old dirty pages.
2481 */
2482 if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2483 mpd->wbc->nr_to_write <=
2484 mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2485 goto out;
2486
2487 /* If we can't merge this page, we are done. */
2488 if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2489 goto out;
2490
2491 if (handle) {
2492 err = ext4_journal_ensure_credits(handle, bpp,
2493 0);
2494 if (err < 0)
2495 goto out;
2496 }
2497
2498 folio_lock(folio);
2499 /*
2500 * If the page is no longer dirty, or its mapping no
2501 * longer corresponds to inode we are writing (which
2502 * means it has been truncated or invalidated), or the
2503 * page is already under writeback and we are not doing
2504 * a data integrity writeback, skip the page
2505 */
2506 if (!folio_test_dirty(folio) ||
2507 (folio_test_writeback(folio) &&
2508 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2509 unlikely(folio->mapping != mapping)) {
2510 folio_unlock(folio);
2511 continue;
2512 }
2513
2514 folio_wait_writeback(folio);
2515 BUG_ON(folio_test_writeback(folio));
2516
2517 /*
2518 * Should never happen but for buggy code in
2519 * other subsystems that call
2520 * set_page_dirty() without properly warning
2521 * the file system first. See [1] for more
2522 * information.
2523 *
2524 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2525 */
2526 if (!folio_buffers(folio)) {
2527 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2528 folio_clear_dirty(folio);
2529 folio_unlock(folio);
2530 continue;
2531 }
2532
2533 if (mpd->map.m_len == 0)
2534 mpd->first_page = folio->index;
2535 mpd->next_page = folio_next_index(folio);
2536 /*
2537 * Writeout when we cannot modify metadata is simple.
2538 * Just submit the page. For data=journal mode we
2539 * first handle writeout of the page for checkpoint and
2540 * only after that handle delayed page dirtying. This
2541 * makes sure current data is checkpointed to the final
2542 * location before possibly journalling it again which
2543 * is desirable when the page is frequently dirtied
2544 * through a pin.
2545 */
2546 if (!mpd->can_map) {
2547 err = mpage_submit_folio(mpd, folio);
2548 if (err < 0)
2549 goto out;
2550 /* Pending dirtying of journalled data? */
2551 if (folio_test_checked(folio)) {
2552 err = mpage_journal_page_buffers(handle,
2553 mpd, folio);
2554 if (err < 0)
2555 goto out;
2556 mpd->journalled_more_data = 1;
2557 }
2558 mpage_folio_done(mpd, folio);
2559 } else {
2560 /* Add all dirty buffers to mpd */
2561 lblk = ((ext4_lblk_t)folio->index) <<
2562 (PAGE_SHIFT - blkbits);
2563 head = folio_buffers(folio);
2564 err = mpage_process_page_bufs(mpd, head, head,
2565 lblk);
2566 if (err <= 0)
2567 goto out;
2568 err = 0;
2569 }
2570 }
2571 folio_batch_release(&fbatch);
2572 cond_resched();
2573 }
2574 mpd->scanned_until_end = 1;
2575 if (handle)
2576 ext4_journal_stop(handle);
2577 return 0;
2578 out:
2579 folio_batch_release(&fbatch);
2580 if (handle)
2581 ext4_journal_stop(handle);
2582 return err;
2583 }
2584
2585 static int ext4_do_writepages(struct mpage_da_data *mpd)
2586 {
2587 struct writeback_control *wbc = mpd->wbc;
2588 pgoff_t writeback_index = 0;
2589 long nr_to_write = wbc->nr_to_write;
2590 int range_whole = 0;
2591 int cycled = 1;
2592 handle_t *handle = NULL;
2593 struct inode *inode = mpd->inode;
2594 struct address_space *mapping = inode->i_mapping;
2595 int needed_blocks, rsv_blocks = 0, ret = 0;
2596 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2597 struct blk_plug plug;
2598 bool give_up_on_write = false;
2599
2600 trace_ext4_writepages(inode, wbc);
2601
2602 /*
2603 * No pages to write? This is mainly a kludge to avoid starting
2604 * a transaction for special inodes like journal inode on last iput()
2605 * because that could violate lock ordering on umount
2606 */
2607 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2608 goto out_writepages;
2609
2610 /*
2611 * If the filesystem has aborted, it is read-only, so return
2612 * right away instead of dumping stack traces later on that
2613 * will obscure the real source of the problem. We test
2614 * fs shutdown state instead of sb->s_flag's SB_RDONLY because
2615 * the latter could be true if the filesystem is mounted
2616 * read-only, and in that case, ext4_writepages should
2617 * *never* be called, so if that ever happens, we would want
2618 * the stack trace.
2619 */
2620 if (unlikely(ext4_forced_shutdown(mapping->host->i_sb))) {
2621 ret = -EROFS;
2622 goto out_writepages;
2623 }
2624
2625 /*
2626 * If we have inline data and arrive here, it means that
2627 * we will soon create the block for the 1st page, so
2628 * we'd better clear the inline data here.
2629 */
2630 if (ext4_has_inline_data(inode)) {
2631 /* Just inode will be modified... */
2632 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2633 if (IS_ERR(handle)) {
2634 ret = PTR_ERR(handle);
2635 goto out_writepages;
2636 }
2637 BUG_ON(ext4_test_inode_state(inode,
2638 EXT4_STATE_MAY_INLINE_DATA));
2639 ext4_destroy_inline_data(handle, inode);
2640 ext4_journal_stop(handle);
2641 }
2642
2643 /*
2644 * data=journal mode does not do delalloc so we just need to writeout /
2645 * journal already mapped buffers. On the other hand we need to commit
2646 * transaction to make data stable. We expect all the data to be
2647 * already in the journal (the only exception are DMA pinned pages
2648 * dirtied behind our back) so we commit transaction here and run the
2649 * writeback loop to checkpoint them. The checkpointing is not actually
2650 * necessary to make data persistent *but* quite a few places (extent
2651 * shifting operations, fsverity, ...) depend on being able to drop
2652 * pagecache pages after calling filemap_write_and_wait() and for that
2653 * checkpointing needs to happen.
2654 */
2655 if (ext4_should_journal_data(inode)) {
2656 mpd->can_map = 0;
2657 if (wbc->sync_mode == WB_SYNC_ALL)
2658 ext4_fc_commit(sbi->s_journal,
2659 EXT4_I(inode)->i_datasync_tid);
2660 }
2661 mpd->journalled_more_data = 0;
2662
2663 if (ext4_should_dioread_nolock(inode)) {
2664 /*
2665 * We may need to convert up to one extent per block in
2666 * the page and we may dirty the inode.
2667 */
2668 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2669 PAGE_SIZE >> inode->i_blkbits);
2670 }
2671
2672 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2673 range_whole = 1;
2674
2675 if (wbc->range_cyclic) {
2676 writeback_index = mapping->writeback_index;
2677 if (writeback_index)
2678 cycled = 0;
2679 mpd->first_page = writeback_index;
2680 mpd->last_page = -1;
2681 } else {
2682 mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2683 mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2684 }
2685
2686 ext4_io_submit_init(&mpd->io_submit, wbc);
2687 retry:
2688 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2689 tag_pages_for_writeback(mapping, mpd->first_page,
2690 mpd->last_page);
2691 blk_start_plug(&plug);
2692
2693 /*
2694 * First writeback pages that don't need mapping - we can avoid
2695 * starting a transaction unnecessarily and also avoid being blocked
2696 * in the block layer on device congestion while having transaction
2697 * started.
2698 */
2699 mpd->do_map = 0;
2700 mpd->scanned_until_end = 0;
2701 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2702 if (!mpd->io_submit.io_end) {
2703 ret = -ENOMEM;
2704 goto unplug;
2705 }
2706 ret = mpage_prepare_extent_to_map(mpd);
2707 /* Unlock pages we didn't use */
2708 mpage_release_unused_pages(mpd, false);
2709 /* Submit prepared bio */
2710 ext4_io_submit(&mpd->io_submit);
2711 ext4_put_io_end_defer(mpd->io_submit.io_end);
2712 mpd->io_submit.io_end = NULL;
2713 if (ret < 0)
2714 goto unplug;
2715
2716 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2717 /* For each extent of pages we use new io_end */
2718 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2719 if (!mpd->io_submit.io_end) {
2720 ret = -ENOMEM;
2721 break;
2722 }
2723
2724 WARN_ON_ONCE(!mpd->can_map);
2725 /*
2726 * We have two constraints: We find one extent to map and we
2727 * must always write out whole page (makes a difference when
2728 * blocksize < pagesize) so that we don't block on IO when we
2729 * try to write out the rest of the page. Journalled mode is
2730 * not supported by delalloc.
2731 */
2732 BUG_ON(ext4_should_journal_data(inode));
2733 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2734
2735 /* start a new transaction */
2736 handle = ext4_journal_start_with_reserve(inode,
2737 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2738 if (IS_ERR(handle)) {
2739 ret = PTR_ERR(handle);
2740 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2741 "%ld pages, ino %lu; err %d", __func__,
2742 wbc->nr_to_write, inode->i_ino, ret);
2743 /* Release allocated io_end */
2744 ext4_put_io_end(mpd->io_submit.io_end);
2745 mpd->io_submit.io_end = NULL;
2746 break;
2747 }
2748 mpd->do_map = 1;
2749
2750 trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2751 ret = mpage_prepare_extent_to_map(mpd);
2752 if (!ret && mpd->map.m_len)
2753 ret = mpage_map_and_submit_extent(handle, mpd,
2754 &give_up_on_write);
2755 /*
2756 * Caution: If the handle is synchronous,
2757 * ext4_journal_stop() can wait for transaction commit
2758 * to finish which may depend on writeback of pages to
2759 * complete or on page lock to be released. In that
2760 * case, we have to wait until after we have
2761 * submitted all the IO, released page locks we hold,
2762 * and dropped io_end reference (for extent conversion
2763 * to be able to complete) before stopping the handle.
2764 */
2765 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2766 ext4_journal_stop(handle);
2767 handle = NULL;
2768 mpd->do_map = 0;
2769 }
2770 /* Unlock pages we didn't use */
2771 mpage_release_unused_pages(mpd, give_up_on_write);
2772 /* Submit prepared bio */
2773 ext4_io_submit(&mpd->io_submit);
2774
2775 /*
2776 * Drop our io_end reference we got from init. We have
2777 * to be careful and use deferred io_end finishing if
2778 * we are still holding the transaction as we can
2779 * release the last reference to io_end which may end
2780 * up doing unwritten extent conversion.
2781 */
2782 if (handle) {
2783 ext4_put_io_end_defer(mpd->io_submit.io_end);
2784 ext4_journal_stop(handle);
2785 } else
2786 ext4_put_io_end(mpd->io_submit.io_end);
2787 mpd->io_submit.io_end = NULL;
2788
2789 if (ret == -ENOSPC && sbi->s_journal) {
2790 /*
2791 * Commit the transaction which would
2792 * free blocks released in the transaction
2793 * and try again
2794 */
2795 jbd2_journal_force_commit_nested(sbi->s_journal);
2796 ret = 0;
2797 continue;
2798 }
2799 /* Fatal error - ENOMEM, EIO... */
2800 if (ret)
2801 break;
2802 }
2803 unplug:
2804 blk_finish_plug(&plug);
2805 if (!ret && !cycled && wbc->nr_to_write > 0) {
2806 cycled = 1;
2807 mpd->last_page = writeback_index - 1;
2808 mpd->first_page = 0;
2809 goto retry;
2810 }
2811
2812 /* Update index */
2813 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2814 /*
2815 * Set the writeback_index so that range_cyclic
2816 * mode will write it back later
2817 */
2818 mapping->writeback_index = mpd->first_page;
2819
2820 out_writepages:
2821 trace_ext4_writepages_result(inode, wbc, ret,
2822 nr_to_write - wbc->nr_to_write);
2823 return ret;
2824 }
2825
2826 static int ext4_writepages(struct address_space *mapping,
2827 struct writeback_control *wbc)
2828 {
2829 struct super_block *sb = mapping->host->i_sb;
2830 struct mpage_da_data mpd = {
2831 .inode = mapping->host,
2832 .wbc = wbc,
2833 .can_map = 1,
2834 };
2835 int ret;
2836 int alloc_ctx;
2837
2838 if (unlikely(ext4_forced_shutdown(sb)))
2839 return -EIO;
2840
2841 alloc_ctx = ext4_writepages_down_read(sb);
2842 ret = ext4_do_writepages(&mpd);
2843 /*
2844 * For data=journal writeback we could have come across pages marked
2845 * for delayed dirtying (PageChecked) which were just added to the
2846 * running transaction. Try once more to get them to stable storage.
2847 */
2848 if (!ret && mpd.journalled_more_data)
2849 ret = ext4_do_writepages(&mpd);
2850 ext4_writepages_up_read(sb, alloc_ctx);
2851
2852 return ret;
2853 }
2854
2855 int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2856 {
2857 struct writeback_control wbc = {
2858 .sync_mode = WB_SYNC_ALL,
2859 .nr_to_write = LONG_MAX,
2860 .range_start = jinode->i_dirty_start,
2861 .range_end = jinode->i_dirty_end,
2862 };
2863 struct mpage_da_data mpd = {
2864 .inode = jinode->i_vfs_inode,
2865 .wbc = &wbc,
2866 .can_map = 0,
2867 };
2868 return ext4_do_writepages(&mpd);
2869 }
2870
2871 static int ext4_dax_writepages(struct address_space *mapping,
2872 struct writeback_control *wbc)
2873 {
2874 int ret;
2875 long nr_to_write = wbc->nr_to_write;
2876 struct inode *inode = mapping->host;
2877 int alloc_ctx;
2878
2879 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2880 return -EIO;
2881
2882 alloc_ctx = ext4_writepages_down_read(inode->i_sb);
2883 trace_ext4_writepages(inode, wbc);
2884
2885 ret = dax_writeback_mapping_range(mapping,
2886 EXT4_SB(inode->i_sb)->s_daxdev, wbc);
2887 trace_ext4_writepages_result(inode, wbc, ret,
2888 nr_to_write - wbc->nr_to_write);
2889 ext4_writepages_up_read(inode->i_sb, alloc_ctx);
2890 return ret;
2891 }
2892
2893 static int ext4_nonda_switch(struct super_block *sb)
2894 {
2895 s64 free_clusters, dirty_clusters;
2896 struct ext4_sb_info *sbi = EXT4_SB(sb);
2897
2898 /*
2899 * switch to non delalloc mode if we are running low
2900 * on free block. The free block accounting via percpu
2901 * counters can get slightly wrong with percpu_counter_batch getting
2902 * accumulated on each CPU without updating global counters
2903 * Delalloc need an accurate free block accounting. So switch
2904 * to non delalloc when we are near to error range.
2905 */
2906 free_clusters =
2907 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2908 dirty_clusters =
2909 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2910 /*
2911 * Start pushing delalloc when 1/2 of free blocks are dirty.
2912 */
2913 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2914 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2915
2916 if (2 * free_clusters < 3 * dirty_clusters ||
2917 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2918 /*
2919 * free block count is less than 150% of dirty blocks
2920 * or free blocks is less than watermark
2921 */
2922 return 1;
2923 }
2924 return 0;
2925 }
2926
2927 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2928 loff_t pos, unsigned len,
2929 struct page **pagep, void **fsdata)
2930 {
2931 int ret, retries = 0;
2932 struct folio *folio;
2933 pgoff_t index;
2934 struct inode *inode = mapping->host;
2935
2936 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2937 return -EIO;
2938
2939 index = pos >> PAGE_SHIFT;
2940
2941 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2942 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2943 return ext4_write_begin(file, mapping, pos,
2944 len, pagep, fsdata);
2945 }
2946 *fsdata = (void *)0;
2947 trace_ext4_da_write_begin(inode, pos, len);
2948
2949 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2950 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2951 pagep, fsdata);
2952 if (ret < 0)
2953 return ret;
2954 if (ret == 1)
2955 return 0;
2956 }
2957
2958 retry:
2959 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2960 mapping_gfp_mask(mapping));
2961 if (IS_ERR(folio))
2962 return PTR_ERR(folio);
2963
2964 #ifdef CONFIG_FS_ENCRYPTION
2965 ret = ext4_block_write_begin(folio, pos, len, ext4_da_get_block_prep);
2966 #else
2967 ret = __block_write_begin(&folio->page, pos, len, ext4_da_get_block_prep);
2968 #endif
2969 if (ret < 0) {
2970 folio_unlock(folio);
2971 folio_put(folio);
2972 /*
2973 * block_write_begin may have instantiated a few blocks
2974 * outside i_size. Trim these off again. Don't need
2975 * i_size_read because we hold inode lock.
2976 */
2977 if (pos + len > inode->i_size)
2978 ext4_truncate_failed_write(inode);
2979
2980 if (ret == -ENOSPC &&
2981 ext4_should_retry_alloc(inode->i_sb, &retries))
2982 goto retry;
2983 return ret;
2984 }
2985
2986 *pagep = &folio->page;
2987 return ret;
2988 }
2989
2990 /*
2991 * Check if we should update i_disksize
2992 * when write to the end of file but not require block allocation
2993 */
2994 static int ext4_da_should_update_i_disksize(struct folio *folio,
2995 unsigned long offset)
2996 {
2997 struct buffer_head *bh;
2998 struct inode *inode = folio->mapping->host;
2999 unsigned int idx;
3000 int i;
3001
3002 bh = folio_buffers(folio);
3003 idx = offset >> inode->i_blkbits;
3004
3005 for (i = 0; i < idx; i++)
3006 bh = bh->b_this_page;
3007
3008 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3009 return 0;
3010 return 1;
3011 }
3012
3013 static int ext4_da_do_write_end(struct address_space *mapping,
3014 loff_t pos, unsigned len, unsigned copied,
3015 struct folio *folio)
3016 {
3017 struct inode *inode = mapping->host;
3018 loff_t old_size = inode->i_size;
3019 bool disksize_changed = false;
3020 loff_t new_i_size;
3021
3022 if (unlikely(!folio_buffers(folio))) {
3023 folio_unlock(folio);
3024 folio_put(folio);
3025 return -EIO;
3026 }
3027 /*
3028 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
3029 * flag, which all that's needed to trigger page writeback.
3030 */
3031 copied = block_write_end(NULL, mapping, pos, len, copied,
3032 &folio->page, NULL);
3033 new_i_size = pos + copied;
3034
3035 /*
3036 * It's important to update i_size while still holding folio lock,
3037 * because folio writeout could otherwise come in and zero beyond
3038 * i_size.
3039 *
3040 * Since we are holding inode lock, we are sure i_disksize <=
3041 * i_size. We also know that if i_disksize < i_size, there are
3042 * delalloc writes pending in the range up to i_size. If the end of
3043 * the current write is <= i_size, there's no need to touch
3044 * i_disksize since writeback will push i_disksize up to i_size
3045 * eventually. If the end of the current write is > i_size and
3046 * inside an allocated block which ext4_da_should_update_i_disksize()
3047 * checked, we need to update i_disksize here as certain
3048 * ext4_writepages() paths not allocating blocks and update i_disksize.
3049 */
3050 if (new_i_size > inode->i_size) {
3051 unsigned long end;
3052
3053 i_size_write(inode, new_i_size);
3054 end = (new_i_size - 1) & (PAGE_SIZE - 1);
3055 if (copied && ext4_da_should_update_i_disksize(folio, end)) {
3056 ext4_update_i_disksize(inode, new_i_size);
3057 disksize_changed = true;
3058 }
3059 }
3060
3061 folio_unlock(folio);
3062 folio_put(folio);
3063
3064 if (old_size < pos)
3065 pagecache_isize_extended(inode, old_size, pos);
3066
3067 if (disksize_changed) {
3068 handle_t *handle;
3069
3070 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3071 if (IS_ERR(handle))
3072 return PTR_ERR(handle);
3073 ext4_mark_inode_dirty(handle, inode);
3074 ext4_journal_stop(handle);
3075 }
3076
3077 return copied;
3078 }
3079
3080 static int ext4_da_write_end(struct file *file,
3081 struct address_space *mapping,
3082 loff_t pos, unsigned len, unsigned copied,
3083 struct page *page, void *fsdata)
3084 {
3085 struct inode *inode = mapping->host;
3086 int write_mode = (int)(unsigned long)fsdata;
3087 struct folio *folio = page_folio(page);
3088
3089 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3090 return ext4_write_end(file, mapping, pos,
3091 len, copied, &folio->page, fsdata);
3092
3093 trace_ext4_da_write_end(inode, pos, len, copied);
3094
3095 if (write_mode != CONVERT_INLINE_DATA &&
3096 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3097 ext4_has_inline_data(inode))
3098 return ext4_write_inline_data_end(inode, pos, len, copied,
3099 folio);
3100
3101 if (unlikely(copied < len) && !folio_test_uptodate(folio))
3102 copied = 0;
3103
3104 return ext4_da_do_write_end(mapping, pos, len, copied, folio);
3105 }
3106
3107 /*
3108 * Force all delayed allocation blocks to be allocated for a given inode.
3109 */
3110 int ext4_alloc_da_blocks(struct inode *inode)
3111 {
3112 trace_ext4_alloc_da_blocks(inode);
3113
3114 if (!EXT4_I(inode)->i_reserved_data_blocks)
3115 return 0;
3116
3117 /*
3118 * We do something simple for now. The filemap_flush() will
3119 * also start triggering a write of the data blocks, which is
3120 * not strictly speaking necessary (and for users of
3121 * laptop_mode, not even desirable). However, to do otherwise
3122 * would require replicating code paths in:
3123 *
3124 * ext4_writepages() ->
3125 * write_cache_pages() ---> (via passed in callback function)
3126 * __mpage_da_writepage() -->
3127 * mpage_add_bh_to_extent()
3128 * mpage_da_map_blocks()
3129 *
3130 * The problem is that write_cache_pages(), located in
3131 * mm/page-writeback.c, marks pages clean in preparation for
3132 * doing I/O, which is not desirable if we're not planning on
3133 * doing I/O at all.
3134 *
3135 * We could call write_cache_pages(), and then redirty all of
3136 * the pages by calling redirty_page_for_writepage() but that
3137 * would be ugly in the extreme. So instead we would need to
3138 * replicate parts of the code in the above functions,
3139 * simplifying them because we wouldn't actually intend to
3140 * write out the pages, but rather only collect contiguous
3141 * logical block extents, call the multi-block allocator, and
3142 * then update the buffer heads with the block allocations.
3143 *
3144 * For now, though, we'll cheat by calling filemap_flush(),
3145 * which will map the blocks, and start the I/O, but not
3146 * actually wait for the I/O to complete.
3147 */
3148 return filemap_flush(inode->i_mapping);
3149 }
3150
3151 /*
3152 * bmap() is special. It gets used by applications such as lilo and by
3153 * the swapper to find the on-disk block of a specific piece of data.
3154 *
3155 * Naturally, this is dangerous if the block concerned is still in the
3156 * journal. If somebody makes a swapfile on an ext4 data-journaling
3157 * filesystem and enables swap, then they may get a nasty shock when the
3158 * data getting swapped to that swapfile suddenly gets overwritten by
3159 * the original zero's written out previously to the journal and
3160 * awaiting writeback in the kernel's buffer cache.
3161 *
3162 * So, if we see any bmap calls here on a modified, data-journaled file,
3163 * take extra steps to flush any blocks which might be in the cache.
3164 */
3165 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3166 {
3167 struct inode *inode = mapping->host;
3168 sector_t ret = 0;
3169
3170 inode_lock_shared(inode);
3171 /*
3172 * We can get here for an inline file via the FIBMAP ioctl
3173 */
3174 if (ext4_has_inline_data(inode))
3175 goto out;
3176
3177 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3178 (test_opt(inode->i_sb, DELALLOC) ||
3179 ext4_should_journal_data(inode))) {
3180 /*
3181 * With delalloc or journalled data we want to sync the file so
3182 * that we can make sure we allocate blocks for file and data
3183 * is in place for the user to see it
3184 */
3185 filemap_write_and_wait(mapping);
3186 }
3187
3188 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3189
3190 out:
3191 inode_unlock_shared(inode);
3192 return ret;
3193 }
3194
3195 static int ext4_read_folio(struct file *file, struct folio *folio)
3196 {
3197 int ret = -EAGAIN;
3198 struct inode *inode = folio->mapping->host;
3199
3200 trace_ext4_read_folio(inode, folio);
3201
3202 if (ext4_has_inline_data(inode))
3203 ret = ext4_readpage_inline(inode, folio);
3204
3205 if (ret == -EAGAIN)
3206 return ext4_mpage_readpages(inode, NULL, folio);
3207
3208 return ret;
3209 }
3210
3211 static void ext4_readahead(struct readahead_control *rac)
3212 {
3213 struct inode *inode = rac->mapping->host;
3214
3215 /* If the file has inline data, no need to do readahead. */
3216 if (ext4_has_inline_data(inode))
3217 return;
3218
3219 ext4_mpage_readpages(inode, rac, NULL);
3220 }
3221
3222 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3223 size_t length)
3224 {
3225 trace_ext4_invalidate_folio(folio, offset, length);
3226
3227 /* No journalling happens on data buffers when this function is used */
3228 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3229
3230 block_invalidate_folio(folio, offset, length);
3231 }
3232
3233 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3234 size_t offset, size_t length)
3235 {
3236 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3237
3238 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3239
3240 /*
3241 * If it's a full truncate we just forget about the pending dirtying
3242 */
3243 if (offset == 0 && length == folio_size(folio))
3244 folio_clear_checked(folio);
3245
3246 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3247 }
3248
3249 /* Wrapper for aops... */
3250 static void ext4_journalled_invalidate_folio(struct folio *folio,
3251 size_t offset,
3252 size_t length)
3253 {
3254 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3255 }
3256
3257 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3258 {
3259 struct inode *inode = folio->mapping->host;
3260 journal_t *journal = EXT4_JOURNAL(inode);
3261
3262 trace_ext4_release_folio(inode, folio);
3263
3264 /* Page has dirty journalled data -> cannot release */
3265 if (folio_test_checked(folio))
3266 return false;
3267 if (journal)
3268 return jbd2_journal_try_to_free_buffers(journal, folio);
3269 else
3270 return try_to_free_buffers(folio);
3271 }
3272
3273 static bool ext4_inode_datasync_dirty(struct inode *inode)
3274 {
3275 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3276
3277 if (journal) {
3278 if (jbd2_transaction_committed(journal,
3279 EXT4_I(inode)->i_datasync_tid))
3280 return false;
3281 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3282 return !list_empty(&EXT4_I(inode)->i_fc_list);
3283 return true;
3284 }
3285
3286 /* Any metadata buffers to write? */
3287 if (!list_empty(&inode->i_mapping->i_private_list))
3288 return true;
3289 return inode->i_state & I_DIRTY_DATASYNC;
3290 }
3291
3292 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3293 struct ext4_map_blocks *map, loff_t offset,
3294 loff_t length, unsigned int flags)
3295 {
3296 u8 blkbits = inode->i_blkbits;
3297
3298 /*
3299 * Writes that span EOF might trigger an I/O size update on completion,
3300 * so consider them to be dirty for the purpose of O_DSYNC, even if
3301 * there is no other metadata changes being made or are pending.
3302 */
3303 iomap->flags = 0;
3304 if (ext4_inode_datasync_dirty(inode) ||
3305 offset + length > i_size_read(inode))
3306 iomap->flags |= IOMAP_F_DIRTY;
3307
3308 if (map->m_flags & EXT4_MAP_NEW)
3309 iomap->flags |= IOMAP_F_NEW;
3310
3311 if (flags & IOMAP_DAX)
3312 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3313 else
3314 iomap->bdev = inode->i_sb->s_bdev;
3315 iomap->offset = (u64) map->m_lblk << blkbits;
3316 iomap->length = (u64) map->m_len << blkbits;
3317
3318 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3319 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3320 iomap->flags |= IOMAP_F_MERGED;
3321
3322 /*
3323 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3324 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3325 * set. In order for any allocated unwritten extents to be converted
3326 * into written extents correctly within the ->end_io() handler, we
3327 * need to ensure that the iomap->type is set appropriately. Hence, the
3328 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3329 * been set first.
3330 */
3331 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3332 iomap->type = IOMAP_UNWRITTEN;
3333 iomap->addr = (u64) map->m_pblk << blkbits;
3334 if (flags & IOMAP_DAX)
3335 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3336 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3337 iomap->type = IOMAP_MAPPED;
3338 iomap->addr = (u64) map->m_pblk << blkbits;
3339 if (flags & IOMAP_DAX)
3340 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3341 } else if (map->m_flags & EXT4_MAP_DELAYED) {
3342 iomap->type = IOMAP_DELALLOC;
3343 iomap->addr = IOMAP_NULL_ADDR;
3344 } else {
3345 iomap->type = IOMAP_HOLE;
3346 iomap->addr = IOMAP_NULL_ADDR;
3347 }
3348 }
3349
3350 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3351 unsigned int flags)
3352 {
3353 handle_t *handle;
3354 u8 blkbits = inode->i_blkbits;
3355 int ret, dio_credits, m_flags = 0, retries = 0;
3356
3357 /*
3358 * Trim the mapping request to the maximum value that we can map at
3359 * once for direct I/O.
3360 */
3361 if (map->m_len > DIO_MAX_BLOCKS)
3362 map->m_len = DIO_MAX_BLOCKS;
3363 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3364
3365 retry:
3366 /*
3367 * Either we allocate blocks and then don't get an unwritten extent, so
3368 * in that case we have reserved enough credits. Or, the blocks are
3369 * already allocated and unwritten. In that case, the extent conversion
3370 * fits into the credits as well.
3371 */
3372 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3373 if (IS_ERR(handle))
3374 return PTR_ERR(handle);
3375
3376 /*
3377 * DAX and direct I/O are the only two operations that are currently
3378 * supported with IOMAP_WRITE.
3379 */
3380 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3381 if (flags & IOMAP_DAX)
3382 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3383 /*
3384 * We use i_size instead of i_disksize here because delalloc writeback
3385 * can complete at any point during the I/O and subsequently push the
3386 * i_disksize out to i_size. This could be beyond where direct I/O is
3387 * happening and thus expose allocated blocks to direct I/O reads.
3388 */
3389 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3390 m_flags = EXT4_GET_BLOCKS_CREATE;
3391 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3392 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3393
3394 ret = ext4_map_blocks(handle, inode, map, m_flags);
3395
3396 /*
3397 * We cannot fill holes in indirect tree based inodes as that could
3398 * expose stale data in the case of a crash. Use the magic error code
3399 * to fallback to buffered I/O.
3400 */
3401 if (!m_flags && !ret)
3402 ret = -ENOTBLK;
3403
3404 ext4_journal_stop(handle);
3405 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3406 goto retry;
3407
3408 return ret;
3409 }
3410
3411
3412 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3413 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3414 {
3415 int ret;
3416 struct ext4_map_blocks map;
3417 u8 blkbits = inode->i_blkbits;
3418
3419 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3420 return -EINVAL;
3421
3422 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3423 return -ERANGE;
3424
3425 /*
3426 * Calculate the first and last logical blocks respectively.
3427 */
3428 map.m_lblk = offset >> blkbits;
3429 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3430 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3431
3432 if (flags & IOMAP_WRITE) {
3433 /*
3434 * We check here if the blocks are already allocated, then we
3435 * don't need to start a journal txn and we can directly return
3436 * the mapping information. This could boost performance
3437 * especially in multi-threaded overwrite requests.
3438 */
3439 if (offset + length <= i_size_read(inode)) {
3440 ret = ext4_map_blocks(NULL, inode, &map, 0);
3441 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3442 goto out;
3443 }
3444 ret = ext4_iomap_alloc(inode, &map, flags);
3445 } else {
3446 ret = ext4_map_blocks(NULL, inode, &map, 0);
3447 }
3448
3449 if (ret < 0)
3450 return ret;
3451 out:
3452 /*
3453 * When inline encryption is enabled, sometimes I/O to an encrypted file
3454 * has to be broken up to guarantee DUN contiguity. Handle this by
3455 * limiting the length of the mapping returned.
3456 */
3457 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3458
3459 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3460
3461 return 0;
3462 }
3463
3464 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3465 loff_t length, unsigned flags, struct iomap *iomap,
3466 struct iomap *srcmap)
3467 {
3468 int ret;
3469
3470 /*
3471 * Even for writes we don't need to allocate blocks, so just pretend
3472 * we are reading to save overhead of starting a transaction.
3473 */
3474 flags &= ~IOMAP_WRITE;
3475 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3476 WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3477 return ret;
3478 }
3479
3480 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3481 ssize_t written, unsigned flags, struct iomap *iomap)
3482 {
3483 /*
3484 * Check to see whether an error occurred while writing out the data to
3485 * the allocated blocks. If so, return the magic error code so that we
3486 * fallback to buffered I/O and attempt to complete the remainder of
3487 * the I/O. Any blocks that may have been allocated in preparation for
3488 * the direct I/O will be reused during buffered I/O.
3489 */
3490 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3491 return -ENOTBLK;
3492
3493 return 0;
3494 }
3495
3496 const struct iomap_ops ext4_iomap_ops = {
3497 .iomap_begin = ext4_iomap_begin,
3498 .iomap_end = ext4_iomap_end,
3499 };
3500
3501 const struct iomap_ops ext4_iomap_overwrite_ops = {
3502 .iomap_begin = ext4_iomap_overwrite_begin,
3503 .iomap_end = ext4_iomap_end,
3504 };
3505
3506 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3507 loff_t length, unsigned int flags,
3508 struct iomap *iomap, struct iomap *srcmap)
3509 {
3510 int ret;
3511 struct ext4_map_blocks map;
3512 u8 blkbits = inode->i_blkbits;
3513
3514 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3515 return -EINVAL;
3516
3517 if (ext4_has_inline_data(inode)) {
3518 ret = ext4_inline_data_iomap(inode, iomap);
3519 if (ret != -EAGAIN) {
3520 if (ret == 0 && offset >= iomap->length)
3521 ret = -ENOENT;
3522 return ret;
3523 }
3524 }
3525
3526 /*
3527 * Calculate the first and last logical block respectively.
3528 */
3529 map.m_lblk = offset >> blkbits;
3530 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3531 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3532
3533 /*
3534 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3535 * So handle it here itself instead of querying ext4_map_blocks().
3536 * Since ext4_map_blocks() will warn about it and will return
3537 * -EIO error.
3538 */
3539 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3540 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3541
3542 if (offset >= sbi->s_bitmap_maxbytes) {
3543 map.m_flags = 0;
3544 goto set_iomap;
3545 }
3546 }
3547
3548 ret = ext4_map_blocks(NULL, inode, &map, 0);
3549 if (ret < 0)
3550 return ret;
3551 set_iomap:
3552 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3553
3554 return 0;
3555 }
3556
3557 const struct iomap_ops ext4_iomap_report_ops = {
3558 .iomap_begin = ext4_iomap_begin_report,
3559 };
3560
3561 /*
3562 * For data=journal mode, folio should be marked dirty only when it was
3563 * writeably mapped. When that happens, it was already attached to the
3564 * transaction and marked as jbddirty (we take care of this in
3565 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3566 * so we should have nothing to do here, except for the case when someone
3567 * had the page pinned and dirtied the page through this pin (e.g. by doing
3568 * direct IO to it). In that case we'd need to attach buffers here to the
3569 * transaction but we cannot due to lock ordering. We cannot just dirty the
3570 * folio and leave attached buffers clean, because the buffers' dirty state is
3571 * "definitive". We cannot just set the buffers dirty or jbddirty because all
3572 * the journalling code will explode. So what we do is to mark the folio
3573 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3574 * to the transaction appropriately.
3575 */
3576 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3577 struct folio *folio)
3578 {
3579 WARN_ON_ONCE(!folio_buffers(folio));
3580 if (folio_maybe_dma_pinned(folio))
3581 folio_set_checked(folio);
3582 return filemap_dirty_folio(mapping, folio);
3583 }
3584
3585 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3586 {
3587 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3588 WARN_ON_ONCE(!folio_buffers(folio));
3589 return block_dirty_folio(mapping, folio);
3590 }
3591
3592 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3593 struct file *file, sector_t *span)
3594 {
3595 return iomap_swapfile_activate(sis, file, span,
3596 &ext4_iomap_report_ops);
3597 }
3598
3599 static const struct address_space_operations ext4_aops = {
3600 .read_folio = ext4_read_folio,
3601 .readahead = ext4_readahead,
3602 .writepages = ext4_writepages,
3603 .write_begin = ext4_write_begin,
3604 .write_end = ext4_write_end,
3605 .dirty_folio = ext4_dirty_folio,
3606 .bmap = ext4_bmap,
3607 .invalidate_folio = ext4_invalidate_folio,
3608 .release_folio = ext4_release_folio,
3609 .migrate_folio = buffer_migrate_folio,
3610 .is_partially_uptodate = block_is_partially_uptodate,
3611 .error_remove_folio = generic_error_remove_folio,
3612 .swap_activate = ext4_iomap_swap_activate,
3613 };
3614
3615 static const struct address_space_operations ext4_journalled_aops = {
3616 .read_folio = ext4_read_folio,
3617 .readahead = ext4_readahead,
3618 .writepages = ext4_writepages,
3619 .write_begin = ext4_write_begin,
3620 .write_end = ext4_journalled_write_end,
3621 .dirty_folio = ext4_journalled_dirty_folio,
3622 .bmap = ext4_bmap,
3623 .invalidate_folio = ext4_journalled_invalidate_folio,
3624 .release_folio = ext4_release_folio,
3625 .migrate_folio = buffer_migrate_folio_norefs,
3626 .is_partially_uptodate = block_is_partially_uptodate,
3627 .error_remove_folio = generic_error_remove_folio,
3628 .swap_activate = ext4_iomap_swap_activate,
3629 };
3630
3631 static const struct address_space_operations ext4_da_aops = {
3632 .read_folio = ext4_read_folio,
3633 .readahead = ext4_readahead,
3634 .writepages = ext4_writepages,
3635 .write_begin = ext4_da_write_begin,
3636 .write_end = ext4_da_write_end,
3637 .dirty_folio = ext4_dirty_folio,
3638 .bmap = ext4_bmap,
3639 .invalidate_folio = ext4_invalidate_folio,
3640 .release_folio = ext4_release_folio,
3641 .migrate_folio = buffer_migrate_folio,
3642 .is_partially_uptodate = block_is_partially_uptodate,
3643 .error_remove_folio = generic_error_remove_folio,
3644 .swap_activate = ext4_iomap_swap_activate,
3645 };
3646
3647 static const struct address_space_operations ext4_dax_aops = {
3648 .writepages = ext4_dax_writepages,
3649 .dirty_folio = noop_dirty_folio,
3650 .bmap = ext4_bmap,
3651 .swap_activate = ext4_iomap_swap_activate,
3652 };
3653
3654 void ext4_set_aops(struct inode *inode)
3655 {
3656 switch (ext4_inode_journal_mode(inode)) {
3657 case EXT4_INODE_ORDERED_DATA_MODE:
3658 case EXT4_INODE_WRITEBACK_DATA_MODE:
3659 break;
3660 case EXT4_INODE_JOURNAL_DATA_MODE:
3661 inode->i_mapping->a_ops = &ext4_journalled_aops;
3662 return;
3663 default:
3664 BUG();
3665 }
3666 if (IS_DAX(inode))
3667 inode->i_mapping->a_ops = &ext4_dax_aops;
3668 else if (test_opt(inode->i_sb, DELALLOC))
3669 inode->i_mapping->a_ops = &ext4_da_aops;
3670 else
3671 inode->i_mapping->a_ops = &ext4_aops;
3672 }
3673
3674 /*
3675 * Here we can't skip an unwritten buffer even though it usually reads zero
3676 * because it might have data in pagecache (eg, if called from ext4_zero_range,
3677 * ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
3678 * racing writeback can come later and flush the stale pagecache to disk.
3679 */
3680 static int __ext4_block_zero_page_range(handle_t *handle,
3681 struct address_space *mapping, loff_t from, loff_t length)
3682 {
3683 ext4_fsblk_t index = from >> PAGE_SHIFT;
3684 unsigned offset = from & (PAGE_SIZE-1);
3685 unsigned blocksize, pos;
3686 ext4_lblk_t iblock;
3687 struct inode *inode = mapping->host;
3688 struct buffer_head *bh;
3689 struct folio *folio;
3690 int err = 0;
3691
3692 folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3693 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3694 mapping_gfp_constraint(mapping, ~__GFP_FS));
3695 if (IS_ERR(folio))
3696 return PTR_ERR(folio);
3697
3698 blocksize = inode->i_sb->s_blocksize;
3699
3700 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3701
3702 bh = folio_buffers(folio);
3703 if (!bh)
3704 bh = create_empty_buffers(folio, blocksize, 0);
3705
3706 /* Find the buffer that contains "offset" */
3707 pos = blocksize;
3708 while (offset >= pos) {
3709 bh = bh->b_this_page;
3710 iblock++;
3711 pos += blocksize;
3712 }
3713 if (buffer_freed(bh)) {
3714 BUFFER_TRACE(bh, "freed: skip");
3715 goto unlock;
3716 }
3717 if (!buffer_mapped(bh)) {
3718 BUFFER_TRACE(bh, "unmapped");
3719 ext4_get_block(inode, iblock, bh, 0);
3720 /* unmapped? It's a hole - nothing to do */
3721 if (!buffer_mapped(bh)) {
3722 BUFFER_TRACE(bh, "still unmapped");
3723 goto unlock;
3724 }
3725 }
3726
3727 /* Ok, it's mapped. Make sure it's up-to-date */
3728 if (folio_test_uptodate(folio))
3729 set_buffer_uptodate(bh);
3730
3731 if (!buffer_uptodate(bh)) {
3732 err = ext4_read_bh_lock(bh, 0, true);
3733 if (err)
3734 goto unlock;
3735 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3736 /* We expect the key to be set. */
3737 BUG_ON(!fscrypt_has_encryption_key(inode));
3738 err = fscrypt_decrypt_pagecache_blocks(folio,
3739 blocksize,
3740 bh_offset(bh));
3741 if (err) {
3742 clear_buffer_uptodate(bh);
3743 goto unlock;
3744 }
3745 }
3746 }
3747 if (ext4_should_journal_data(inode)) {
3748 BUFFER_TRACE(bh, "get write access");
3749 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3750 EXT4_JTR_NONE);
3751 if (err)
3752 goto unlock;
3753 }
3754 folio_zero_range(folio, offset, length);
3755 BUFFER_TRACE(bh, "zeroed end of block");
3756
3757 if (ext4_should_journal_data(inode)) {
3758 err = ext4_dirty_journalled_data(handle, bh);
3759 } else {
3760 err = 0;
3761 mark_buffer_dirty(bh);
3762 if (ext4_should_order_data(inode))
3763 err = ext4_jbd2_inode_add_write(handle, inode, from,
3764 length);
3765 }
3766
3767 unlock:
3768 folio_unlock(folio);
3769 folio_put(folio);
3770 return err;
3771 }
3772
3773 /*
3774 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3775 * starting from file offset 'from'. The range to be zero'd must
3776 * be contained with in one block. If the specified range exceeds
3777 * the end of the block it will be shortened to end of the block
3778 * that corresponds to 'from'
3779 */
3780 static int ext4_block_zero_page_range(handle_t *handle,
3781 struct address_space *mapping, loff_t from, loff_t length)
3782 {
3783 struct inode *inode = mapping->host;
3784 unsigned offset = from & (PAGE_SIZE-1);
3785 unsigned blocksize = inode->i_sb->s_blocksize;
3786 unsigned max = blocksize - (offset & (blocksize - 1));
3787
3788 /*
3789 * correct length if it does not fall between
3790 * 'from' and the end of the block
3791 */
3792 if (length > max || length < 0)
3793 length = max;
3794
3795 if (IS_DAX(inode)) {
3796 return dax_zero_range(inode, from, length, NULL,
3797 &ext4_iomap_ops);
3798 }
3799 return __ext4_block_zero_page_range(handle, mapping, from, length);
3800 }
3801
3802 /*
3803 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3804 * up to the end of the block which corresponds to `from'.
3805 * This required during truncate. We need to physically zero the tail end
3806 * of that block so it doesn't yield old data if the file is later grown.
3807 */
3808 static int ext4_block_truncate_page(handle_t *handle,
3809 struct address_space *mapping, loff_t from)
3810 {
3811 unsigned offset = from & (PAGE_SIZE-1);
3812 unsigned length;
3813 unsigned blocksize;
3814 struct inode *inode = mapping->host;
3815
3816 /* If we are processing an encrypted inode during orphan list handling */
3817 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3818 return 0;
3819
3820 blocksize = inode->i_sb->s_blocksize;
3821 length = blocksize - (offset & (blocksize - 1));
3822
3823 return ext4_block_zero_page_range(handle, mapping, from, length);
3824 }
3825
3826 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3827 loff_t lstart, loff_t length)
3828 {
3829 struct super_block *sb = inode->i_sb;
3830 struct address_space *mapping = inode->i_mapping;
3831 unsigned partial_start, partial_end;
3832 ext4_fsblk_t start, end;
3833 loff_t byte_end = (lstart + length - 1);
3834 int err = 0;
3835
3836 partial_start = lstart & (sb->s_blocksize - 1);
3837 partial_end = byte_end & (sb->s_blocksize - 1);
3838
3839 start = lstart >> sb->s_blocksize_bits;
3840 end = byte_end >> sb->s_blocksize_bits;
3841
3842 /* Handle partial zero within the single block */
3843 if (start == end &&
3844 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3845 err = ext4_block_zero_page_range(handle, mapping,
3846 lstart, length);
3847 return err;
3848 }
3849 /* Handle partial zero out on the start of the range */
3850 if (partial_start) {
3851 err = ext4_block_zero_page_range(handle, mapping,
3852 lstart, sb->s_blocksize);
3853 if (err)
3854 return err;
3855 }
3856 /* Handle partial zero out on the end of the range */
3857 if (partial_end != sb->s_blocksize - 1)
3858 err = ext4_block_zero_page_range(handle, mapping,
3859 byte_end - partial_end,
3860 partial_end + 1);
3861 return err;
3862 }
3863
3864 int ext4_can_truncate(struct inode *inode)
3865 {
3866 if (S_ISREG(inode->i_mode))
3867 return 1;
3868 if (S_ISDIR(inode->i_mode))
3869 return 1;
3870 if (S_ISLNK(inode->i_mode))
3871 return !ext4_inode_is_fast_symlink(inode);
3872 return 0;
3873 }
3874
3875 /*
3876 * We have to make sure i_disksize gets properly updated before we truncate
3877 * page cache due to hole punching or zero range. Otherwise i_disksize update
3878 * can get lost as it may have been postponed to submission of writeback but
3879 * that will never happen after we truncate page cache.
3880 */
3881 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3882 loff_t len)
3883 {
3884 handle_t *handle;
3885 int ret;
3886
3887 loff_t size = i_size_read(inode);
3888
3889 WARN_ON(!inode_is_locked(inode));
3890 if (offset > size || offset + len < size)
3891 return 0;
3892
3893 if (EXT4_I(inode)->i_disksize >= size)
3894 return 0;
3895
3896 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3897 if (IS_ERR(handle))
3898 return PTR_ERR(handle);
3899 ext4_update_i_disksize(inode, size);
3900 ret = ext4_mark_inode_dirty(handle, inode);
3901 ext4_journal_stop(handle);
3902
3903 return ret;
3904 }
3905
3906 static void ext4_wait_dax_page(struct inode *inode)
3907 {
3908 filemap_invalidate_unlock(inode->i_mapping);
3909 schedule();
3910 filemap_invalidate_lock(inode->i_mapping);
3911 }
3912
3913 int ext4_break_layouts(struct inode *inode)
3914 {
3915 struct page *page;
3916 int error;
3917
3918 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3919 return -EINVAL;
3920
3921 do {
3922 page = dax_layout_busy_page(inode->i_mapping);
3923 if (!page)
3924 return 0;
3925
3926 error = ___wait_var_event(&page->_refcount,
3927 atomic_read(&page->_refcount) == 1,
3928 TASK_INTERRUPTIBLE, 0, 0,
3929 ext4_wait_dax_page(inode));
3930 } while (error == 0);
3931
3932 return error;
3933 }
3934
3935 /*
3936 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3937 * associated with the given offset and length
3938 *
3939 * @inode: File inode
3940 * @offset: The offset where the hole will begin
3941 * @len: The length of the hole
3942 *
3943 * Returns: 0 on success or negative on failure
3944 */
3945
3946 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3947 {
3948 struct inode *inode = file_inode(file);
3949 struct super_block *sb = inode->i_sb;
3950 ext4_lblk_t first_block, stop_block;
3951 struct address_space *mapping = inode->i_mapping;
3952 loff_t first_block_offset, last_block_offset, max_length;
3953 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3954 handle_t *handle;
3955 unsigned int credits;
3956 int ret = 0, ret2 = 0;
3957
3958 trace_ext4_punch_hole(inode, offset, length, 0);
3959
3960 /*
3961 * Write out all dirty pages to avoid race conditions
3962 * Then release them.
3963 */
3964 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3965 ret = filemap_write_and_wait_range(mapping, offset,
3966 offset + length - 1);
3967 if (ret)
3968 return ret;
3969 }
3970
3971 inode_lock(inode);
3972
3973 /* No need to punch hole beyond i_size */
3974 if (offset >= inode->i_size)
3975 goto out_mutex;
3976
3977 /*
3978 * If the hole extends beyond i_size, set the hole
3979 * to end after the page that contains i_size
3980 */
3981 if (offset + length > inode->i_size) {
3982 length = inode->i_size +
3983 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3984 offset;
3985 }
3986
3987 /*
3988 * For punch hole the length + offset needs to be within one block
3989 * before last range. Adjust the length if it goes beyond that limit.
3990 */
3991 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3992 if (offset + length > max_length)
3993 length = max_length - offset;
3994
3995 if (offset & (sb->s_blocksize - 1) ||
3996 (offset + length) & (sb->s_blocksize - 1)) {
3997 /*
3998 * Attach jinode to inode for jbd2 if we do any zeroing of
3999 * partial block
4000 */
4001 ret = ext4_inode_attach_jinode(inode);
4002 if (ret < 0)
4003 goto out_mutex;
4004
4005 }
4006
4007 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4008 inode_dio_wait(inode);
4009
4010 ret = file_modified(file);
4011 if (ret)
4012 goto out_mutex;
4013
4014 /*
4015 * Prevent page faults from reinstantiating pages we have released from
4016 * page cache.
4017 */
4018 filemap_invalidate_lock(mapping);
4019
4020 ret = ext4_break_layouts(inode);
4021 if (ret)
4022 goto out_dio;
4023
4024 first_block_offset = round_up(offset, sb->s_blocksize);
4025 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4026
4027 /* Now release the pages and zero block aligned part of pages*/
4028 if (last_block_offset > first_block_offset) {
4029 ret = ext4_update_disksize_before_punch(inode, offset, length);
4030 if (ret)
4031 goto out_dio;
4032 truncate_pagecache_range(inode, first_block_offset,
4033 last_block_offset);
4034 }
4035
4036 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4037 credits = ext4_writepage_trans_blocks(inode);
4038 else
4039 credits = ext4_blocks_for_truncate(inode);
4040 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4041 if (IS_ERR(handle)) {
4042 ret = PTR_ERR(handle);
4043 ext4_std_error(sb, ret);
4044 goto out_dio;
4045 }
4046
4047 ret = ext4_zero_partial_blocks(handle, inode, offset,
4048 length);
4049 if (ret)
4050 goto out_stop;
4051
4052 first_block = (offset + sb->s_blocksize - 1) >>
4053 EXT4_BLOCK_SIZE_BITS(sb);
4054 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4055
4056 /* If there are blocks to remove, do it */
4057 if (stop_block > first_block) {
4058 ext4_lblk_t hole_len = stop_block - first_block;
4059
4060 down_write(&EXT4_I(inode)->i_data_sem);
4061 ext4_discard_preallocations(inode);
4062
4063 ext4_es_remove_extent(inode, first_block, hole_len);
4064
4065 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4066 ret = ext4_ext_remove_space(inode, first_block,
4067 stop_block - 1);
4068 else
4069 ret = ext4_ind_remove_space(handle, inode, first_block,
4070 stop_block);
4071
4072 ext4_es_insert_extent(inode, first_block, hole_len, ~0,
4073 EXTENT_STATUS_HOLE);
4074 up_write(&EXT4_I(inode)->i_data_sem);
4075 }
4076 ext4_fc_track_range(handle, inode, first_block, stop_block);
4077 if (IS_SYNC(inode))
4078 ext4_handle_sync(handle);
4079
4080 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4081 ret2 = ext4_mark_inode_dirty(handle, inode);
4082 if (unlikely(ret2))
4083 ret = ret2;
4084 if (ret >= 0)
4085 ext4_update_inode_fsync_trans(handle, inode, 1);
4086 out_stop:
4087 ext4_journal_stop(handle);
4088 out_dio:
4089 filemap_invalidate_unlock(mapping);
4090 out_mutex:
4091 inode_unlock(inode);
4092 return ret;
4093 }
4094
4095 int ext4_inode_attach_jinode(struct inode *inode)
4096 {
4097 struct ext4_inode_info *ei = EXT4_I(inode);
4098 struct jbd2_inode *jinode;
4099
4100 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4101 return 0;
4102
4103 jinode = jbd2_alloc_inode(GFP_KERNEL);
4104 spin_lock(&inode->i_lock);
4105 if (!ei->jinode) {
4106 if (!jinode) {
4107 spin_unlock(&inode->i_lock);
4108 return -ENOMEM;
4109 }
4110 ei->jinode = jinode;
4111 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4112 jinode = NULL;
4113 }
4114 spin_unlock(&inode->i_lock);
4115 if (unlikely(jinode != NULL))
4116 jbd2_free_inode(jinode);
4117 return 0;
4118 }
4119
4120 /*
4121 * ext4_truncate()
4122 *
4123 * We block out ext4_get_block() block instantiations across the entire
4124 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4125 * simultaneously on behalf of the same inode.
4126 *
4127 * As we work through the truncate and commit bits of it to the journal there
4128 * is one core, guiding principle: the file's tree must always be consistent on
4129 * disk. We must be able to restart the truncate after a crash.
4130 *
4131 * The file's tree may be transiently inconsistent in memory (although it
4132 * probably isn't), but whenever we close off and commit a journal transaction,
4133 * the contents of (the filesystem + the journal) must be consistent and
4134 * restartable. It's pretty simple, really: bottom up, right to left (although
4135 * left-to-right works OK too).
4136 *
4137 * Note that at recovery time, journal replay occurs *before* the restart of
4138 * truncate against the orphan inode list.
4139 *
4140 * The committed inode has the new, desired i_size (which is the same as
4141 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4142 * that this inode's truncate did not complete and it will again call
4143 * ext4_truncate() to have another go. So there will be instantiated blocks
4144 * to the right of the truncation point in a crashed ext4 filesystem. But
4145 * that's fine - as long as they are linked from the inode, the post-crash
4146 * ext4_truncate() run will find them and release them.
4147 */
4148 int ext4_truncate(struct inode *inode)
4149 {
4150 struct ext4_inode_info *ei = EXT4_I(inode);
4151 unsigned int credits;
4152 int err = 0, err2;
4153 handle_t *handle;
4154 struct address_space *mapping = inode->i_mapping;
4155
4156 /*
4157 * There is a possibility that we're either freeing the inode
4158 * or it's a completely new inode. In those cases we might not
4159 * have i_rwsem locked because it's not necessary.
4160 */
4161 if (!(inode->i_state & (I_NEW|I_FREEING)))
4162 WARN_ON(!inode_is_locked(inode));
4163 trace_ext4_truncate_enter(inode);
4164
4165 if (!ext4_can_truncate(inode))
4166 goto out_trace;
4167
4168 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4169 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4170
4171 if (ext4_has_inline_data(inode)) {
4172 int has_inline = 1;
4173
4174 err = ext4_inline_data_truncate(inode, &has_inline);
4175 if (err || has_inline)
4176 goto out_trace;
4177 }
4178
4179 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4180 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4181 err = ext4_inode_attach_jinode(inode);
4182 if (err)
4183 goto out_trace;
4184 }
4185
4186 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4187 credits = ext4_writepage_trans_blocks(inode);
4188 else
4189 credits = ext4_blocks_for_truncate(inode);
4190
4191 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4192 if (IS_ERR(handle)) {
4193 err = PTR_ERR(handle);
4194 goto out_trace;
4195 }
4196
4197 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4198 ext4_block_truncate_page(handle, mapping, inode->i_size);
4199
4200 /*
4201 * We add the inode to the orphan list, so that if this
4202 * truncate spans multiple transactions, and we crash, we will
4203 * resume the truncate when the filesystem recovers. It also
4204 * marks the inode dirty, to catch the new size.
4205 *
4206 * Implication: the file must always be in a sane, consistent
4207 * truncatable state while each transaction commits.
4208 */
4209 err = ext4_orphan_add(handle, inode);
4210 if (err)
4211 goto out_stop;
4212
4213 down_write(&EXT4_I(inode)->i_data_sem);
4214
4215 ext4_discard_preallocations(inode);
4216
4217 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4218 err = ext4_ext_truncate(handle, inode);
4219 else
4220 ext4_ind_truncate(handle, inode);
4221
4222 up_write(&ei->i_data_sem);
4223 if (err)
4224 goto out_stop;
4225
4226 if (IS_SYNC(inode))
4227 ext4_handle_sync(handle);
4228
4229 out_stop:
4230 /*
4231 * If this was a simple ftruncate() and the file will remain alive,
4232 * then we need to clear up the orphan record which we created above.
4233 * However, if this was a real unlink then we were called by
4234 * ext4_evict_inode(), and we allow that function to clean up the
4235 * orphan info for us.
4236 */
4237 if (inode->i_nlink)
4238 ext4_orphan_del(handle, inode);
4239
4240 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4241 err2 = ext4_mark_inode_dirty(handle, inode);
4242 if (unlikely(err2 && !err))
4243 err = err2;
4244 ext4_journal_stop(handle);
4245
4246 out_trace:
4247 trace_ext4_truncate_exit(inode);
4248 return err;
4249 }
4250
4251 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4252 {
4253 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4254 return inode_peek_iversion_raw(inode);
4255 else
4256 return inode_peek_iversion(inode);
4257 }
4258
4259 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4260 struct ext4_inode_info *ei)
4261 {
4262 struct inode *inode = &(ei->vfs_inode);
4263 u64 i_blocks = READ_ONCE(inode->i_blocks);
4264 struct super_block *sb = inode->i_sb;
4265
4266 if (i_blocks <= ~0U) {
4267 /*
4268 * i_blocks can be represented in a 32 bit variable
4269 * as multiple of 512 bytes
4270 */
4271 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4272 raw_inode->i_blocks_high = 0;
4273 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4274 return 0;
4275 }
4276
4277 /*
4278 * This should never happen since sb->s_maxbytes should not have
4279 * allowed this, sb->s_maxbytes was set according to the huge_file
4280 * feature in ext4_fill_super().
4281 */
4282 if (!ext4_has_feature_huge_file(sb))
4283 return -EFSCORRUPTED;
4284
4285 if (i_blocks <= 0xffffffffffffULL) {
4286 /*
4287 * i_blocks can be represented in a 48 bit variable
4288 * as multiple of 512 bytes
4289 */
4290 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4291 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4292 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4293 } else {
4294 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4295 /* i_block is stored in file system block size */
4296 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4297 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4298 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4299 }
4300 return 0;
4301 }
4302
4303 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4304 {
4305 struct ext4_inode_info *ei = EXT4_I(inode);
4306 uid_t i_uid;
4307 gid_t i_gid;
4308 projid_t i_projid;
4309 int block;
4310 int err;
4311
4312 err = ext4_inode_blocks_set(raw_inode, ei);
4313
4314 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4315 i_uid = i_uid_read(inode);
4316 i_gid = i_gid_read(inode);
4317 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4318 if (!(test_opt(inode->i_sb, NO_UID32))) {
4319 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4320 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4321 /*
4322 * Fix up interoperability with old kernels. Otherwise,
4323 * old inodes get re-used with the upper 16 bits of the
4324 * uid/gid intact.
4325 */
4326 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4327 raw_inode->i_uid_high = 0;
4328 raw_inode->i_gid_high = 0;
4329 } else {
4330 raw_inode->i_uid_high =
4331 cpu_to_le16(high_16_bits(i_uid));
4332 raw_inode->i_gid_high =
4333 cpu_to_le16(high_16_bits(i_gid));
4334 }
4335 } else {
4336 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4337 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4338 raw_inode->i_uid_high = 0;
4339 raw_inode->i_gid_high = 0;
4340 }
4341 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4342
4343 EXT4_INODE_SET_CTIME(inode, raw_inode);
4344 EXT4_INODE_SET_MTIME(inode, raw_inode);
4345 EXT4_INODE_SET_ATIME(inode, raw_inode);
4346 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4347
4348 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4349 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4350 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4351 raw_inode->i_file_acl_high =
4352 cpu_to_le16(ei->i_file_acl >> 32);
4353 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4354 ext4_isize_set(raw_inode, ei->i_disksize);
4355
4356 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4357 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4358 if (old_valid_dev(inode->i_rdev)) {
4359 raw_inode->i_block[0] =
4360 cpu_to_le32(old_encode_dev(inode->i_rdev));
4361 raw_inode->i_block[1] = 0;
4362 } else {
4363 raw_inode->i_block[0] = 0;
4364 raw_inode->i_block[1] =
4365 cpu_to_le32(new_encode_dev(inode->i_rdev));
4366 raw_inode->i_block[2] = 0;
4367 }
4368 } else if (!ext4_has_inline_data(inode)) {
4369 for (block = 0; block < EXT4_N_BLOCKS; block++)
4370 raw_inode->i_block[block] = ei->i_data[block];
4371 }
4372
4373 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4374 u64 ivers = ext4_inode_peek_iversion(inode);
4375
4376 raw_inode->i_disk_version = cpu_to_le32(ivers);
4377 if (ei->i_extra_isize) {
4378 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4379 raw_inode->i_version_hi =
4380 cpu_to_le32(ivers >> 32);
4381 raw_inode->i_extra_isize =
4382 cpu_to_le16(ei->i_extra_isize);
4383 }
4384 }
4385
4386 if (i_projid != EXT4_DEF_PROJID &&
4387 !ext4_has_feature_project(inode->i_sb))
4388 err = err ?: -EFSCORRUPTED;
4389
4390 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4391 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4392 raw_inode->i_projid = cpu_to_le32(i_projid);
4393
4394 ext4_inode_csum_set(inode, raw_inode, ei);
4395 return err;
4396 }
4397
4398 /*
4399 * ext4_get_inode_loc returns with an extra refcount against the inode's
4400 * underlying buffer_head on success. If we pass 'inode' and it does not
4401 * have in-inode xattr, we have all inode data in memory that is needed
4402 * to recreate the on-disk version of this inode.
4403 */
4404 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4405 struct inode *inode, struct ext4_iloc *iloc,
4406 ext4_fsblk_t *ret_block)
4407 {
4408 struct ext4_group_desc *gdp;
4409 struct buffer_head *bh;
4410 ext4_fsblk_t block;
4411 struct blk_plug plug;
4412 int inodes_per_block, inode_offset;
4413
4414 iloc->bh = NULL;
4415 if (ino < EXT4_ROOT_INO ||
4416 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4417 return -EFSCORRUPTED;
4418
4419 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4420 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4421 if (!gdp)
4422 return -EIO;
4423
4424 /*
4425 * Figure out the offset within the block group inode table
4426 */
4427 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4428 inode_offset = ((ino - 1) %
4429 EXT4_INODES_PER_GROUP(sb));
4430 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4431
4432 block = ext4_inode_table(sb, gdp);
4433 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4434 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4435 ext4_error(sb, "Invalid inode table block %llu in "
4436 "block_group %u", block, iloc->block_group);
4437 return -EFSCORRUPTED;
4438 }
4439 block += (inode_offset / inodes_per_block);
4440
4441 bh = sb_getblk(sb, block);
4442 if (unlikely(!bh))
4443 return -ENOMEM;
4444 if (ext4_buffer_uptodate(bh))
4445 goto has_buffer;
4446
4447 lock_buffer(bh);
4448 if (ext4_buffer_uptodate(bh)) {
4449 /* Someone brought it uptodate while we waited */
4450 unlock_buffer(bh);
4451 goto has_buffer;
4452 }
4453
4454 /*
4455 * If we have all information of the inode in memory and this
4456 * is the only valid inode in the block, we need not read the
4457 * block.
4458 */
4459 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4460 struct buffer_head *bitmap_bh;
4461 int i, start;
4462
4463 start = inode_offset & ~(inodes_per_block - 1);
4464
4465 /* Is the inode bitmap in cache? */
4466 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4467 if (unlikely(!bitmap_bh))
4468 goto make_io;
4469
4470 /*
4471 * If the inode bitmap isn't in cache then the
4472 * optimisation may end up performing two reads instead
4473 * of one, so skip it.
4474 */
4475 if (!buffer_uptodate(bitmap_bh)) {
4476 brelse(bitmap_bh);
4477 goto make_io;
4478 }
4479 for (i = start; i < start + inodes_per_block; i++) {
4480 if (i == inode_offset)
4481 continue;
4482 if (ext4_test_bit(i, bitmap_bh->b_data))
4483 break;
4484 }
4485 brelse(bitmap_bh);
4486 if (i == start + inodes_per_block) {
4487 struct ext4_inode *raw_inode =
4488 (struct ext4_inode *) (bh->b_data + iloc->offset);
4489
4490 /* all other inodes are free, so skip I/O */
4491 memset(bh->b_data, 0, bh->b_size);
4492 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4493 ext4_fill_raw_inode(inode, raw_inode);
4494 set_buffer_uptodate(bh);
4495 unlock_buffer(bh);
4496 goto has_buffer;
4497 }
4498 }
4499
4500 make_io:
4501 /*
4502 * If we need to do any I/O, try to pre-readahead extra
4503 * blocks from the inode table.
4504 */
4505 blk_start_plug(&plug);
4506 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4507 ext4_fsblk_t b, end, table;
4508 unsigned num;
4509 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4510
4511 table = ext4_inode_table(sb, gdp);
4512 /* s_inode_readahead_blks is always a power of 2 */
4513 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4514 if (table > b)
4515 b = table;
4516 end = b + ra_blks;
4517 num = EXT4_INODES_PER_GROUP(sb);
4518 if (ext4_has_group_desc_csum(sb))
4519 num -= ext4_itable_unused_count(sb, gdp);
4520 table += num / inodes_per_block;
4521 if (end > table)
4522 end = table;
4523 while (b <= end)
4524 ext4_sb_breadahead_unmovable(sb, b++);
4525 }
4526
4527 /*
4528 * There are other valid inodes in the buffer, this inode
4529 * has in-inode xattrs, or we don't have this inode in memory.
4530 * Read the block from disk.
4531 */
4532 trace_ext4_load_inode(sb, ino);
4533 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4534 blk_finish_plug(&plug);
4535 wait_on_buffer(bh);
4536 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4537 if (!buffer_uptodate(bh)) {
4538 if (ret_block)
4539 *ret_block = block;
4540 brelse(bh);
4541 return -EIO;
4542 }
4543 has_buffer:
4544 iloc->bh = bh;
4545 return 0;
4546 }
4547
4548 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4549 struct ext4_iloc *iloc)
4550 {
4551 ext4_fsblk_t err_blk = 0;
4552 int ret;
4553
4554 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4555 &err_blk);
4556
4557 if (ret == -EIO)
4558 ext4_error_inode_block(inode, err_blk, EIO,
4559 "unable to read itable block");
4560
4561 return ret;
4562 }
4563
4564 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4565 {
4566 ext4_fsblk_t err_blk = 0;
4567 int ret;
4568
4569 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4570 &err_blk);
4571
4572 if (ret == -EIO)
4573 ext4_error_inode_block(inode, err_blk, EIO,
4574 "unable to read itable block");
4575
4576 return ret;
4577 }
4578
4579
4580 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4581 struct ext4_iloc *iloc)
4582 {
4583 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4584 }
4585
4586 static bool ext4_should_enable_dax(struct inode *inode)
4587 {
4588 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4589
4590 if (test_opt2(inode->i_sb, DAX_NEVER))
4591 return false;
4592 if (!S_ISREG(inode->i_mode))
4593 return false;
4594 if (ext4_should_journal_data(inode))
4595 return false;
4596 if (ext4_has_inline_data(inode))
4597 return false;
4598 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4599 return false;
4600 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4601 return false;
4602 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4603 return false;
4604 if (test_opt(inode->i_sb, DAX_ALWAYS))
4605 return true;
4606
4607 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4608 }
4609
4610 void ext4_set_inode_flags(struct inode *inode, bool init)
4611 {
4612 unsigned int flags = EXT4_I(inode)->i_flags;
4613 unsigned int new_fl = 0;
4614
4615 WARN_ON_ONCE(IS_DAX(inode) && init);
4616
4617 if (flags & EXT4_SYNC_FL)
4618 new_fl |= S_SYNC;
4619 if (flags & EXT4_APPEND_FL)
4620 new_fl |= S_APPEND;
4621 if (flags & EXT4_IMMUTABLE_FL)
4622 new_fl |= S_IMMUTABLE;
4623 if (flags & EXT4_NOATIME_FL)
4624 new_fl |= S_NOATIME;
4625 if (flags & EXT4_DIRSYNC_FL)
4626 new_fl |= S_DIRSYNC;
4627
4628 /* Because of the way inode_set_flags() works we must preserve S_DAX
4629 * here if already set. */
4630 new_fl |= (inode->i_flags & S_DAX);
4631 if (init && ext4_should_enable_dax(inode))
4632 new_fl |= S_DAX;
4633
4634 if (flags & EXT4_ENCRYPT_FL)
4635 new_fl |= S_ENCRYPTED;
4636 if (flags & EXT4_CASEFOLD_FL)
4637 new_fl |= S_CASEFOLD;
4638 if (flags & EXT4_VERITY_FL)
4639 new_fl |= S_VERITY;
4640 inode_set_flags(inode, new_fl,
4641 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4642 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4643 }
4644
4645 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4646 struct ext4_inode_info *ei)
4647 {
4648 blkcnt_t i_blocks ;
4649 struct inode *inode = &(ei->vfs_inode);
4650 struct super_block *sb = inode->i_sb;
4651
4652 if (ext4_has_feature_huge_file(sb)) {
4653 /* we are using combined 48 bit field */
4654 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4655 le32_to_cpu(raw_inode->i_blocks_lo);
4656 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4657 /* i_blocks represent file system block size */
4658 return i_blocks << (inode->i_blkbits - 9);
4659 } else {
4660 return i_blocks;
4661 }
4662 } else {
4663 return le32_to_cpu(raw_inode->i_blocks_lo);
4664 }
4665 }
4666
4667 static inline int ext4_iget_extra_inode(struct inode *inode,
4668 struct ext4_inode *raw_inode,
4669 struct ext4_inode_info *ei)
4670 {
4671 __le32 *magic = (void *)raw_inode +
4672 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4673
4674 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4675 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4676 int err;
4677
4678 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4679 err = ext4_find_inline_data_nolock(inode);
4680 if (!err && ext4_has_inline_data(inode))
4681 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4682 return err;
4683 } else
4684 EXT4_I(inode)->i_inline_off = 0;
4685 return 0;
4686 }
4687
4688 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4689 {
4690 if (!ext4_has_feature_project(inode->i_sb))
4691 return -EOPNOTSUPP;
4692 *projid = EXT4_I(inode)->i_projid;
4693 return 0;
4694 }
4695
4696 /*
4697 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4698 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4699 * set.
4700 */
4701 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4702 {
4703 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4704 inode_set_iversion_raw(inode, val);
4705 else
4706 inode_set_iversion_queried(inode, val);
4707 }
4708
4709 static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4710
4711 {
4712 if (flags & EXT4_IGET_EA_INODE) {
4713 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4714 return "missing EA_INODE flag";
4715 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4716 EXT4_I(inode)->i_file_acl)
4717 return "ea_inode with extended attributes";
4718 } else {
4719 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4720 return "unexpected EA_INODE flag";
4721 }
4722 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4723 return "unexpected bad inode w/o EXT4_IGET_BAD";
4724 return NULL;
4725 }
4726
4727 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4728 ext4_iget_flags flags, const char *function,
4729 unsigned int line)
4730 {
4731 struct ext4_iloc iloc;
4732 struct ext4_inode *raw_inode;
4733 struct ext4_inode_info *ei;
4734 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4735 struct inode *inode;
4736 const char *err_str;
4737 journal_t *journal = EXT4_SB(sb)->s_journal;
4738 long ret;
4739 loff_t size;
4740 int block;
4741 uid_t i_uid;
4742 gid_t i_gid;
4743 projid_t i_projid;
4744
4745 if ((!(flags & EXT4_IGET_SPECIAL) &&
4746 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4747 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4748 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4749 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4750 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4751 (ino < EXT4_ROOT_INO) ||
4752 (ino > le32_to_cpu(es->s_inodes_count))) {
4753 if (flags & EXT4_IGET_HANDLE)
4754 return ERR_PTR(-ESTALE);
4755 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4756 "inode #%lu: comm %s: iget: illegal inode #",
4757 ino, current->comm);
4758 return ERR_PTR(-EFSCORRUPTED);
4759 }
4760
4761 inode = iget_locked(sb, ino);
4762 if (!inode)
4763 return ERR_PTR(-ENOMEM);
4764 if (!(inode->i_state & I_NEW)) {
4765 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4766 ext4_error_inode(inode, function, line, 0, err_str);
4767 iput(inode);
4768 return ERR_PTR(-EFSCORRUPTED);
4769 }
4770 return inode;
4771 }
4772
4773 ei = EXT4_I(inode);
4774 iloc.bh = NULL;
4775
4776 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4777 if (ret < 0)
4778 goto bad_inode;
4779 raw_inode = ext4_raw_inode(&iloc);
4780
4781 if ((flags & EXT4_IGET_HANDLE) &&
4782 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4783 ret = -ESTALE;
4784 goto bad_inode;
4785 }
4786
4787 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4788 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4789 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4790 EXT4_INODE_SIZE(inode->i_sb) ||
4791 (ei->i_extra_isize & 3)) {
4792 ext4_error_inode(inode, function, line, 0,
4793 "iget: bad extra_isize %u "
4794 "(inode size %u)",
4795 ei->i_extra_isize,
4796 EXT4_INODE_SIZE(inode->i_sb));
4797 ret = -EFSCORRUPTED;
4798 goto bad_inode;
4799 }
4800 } else
4801 ei->i_extra_isize = 0;
4802
4803 /* Precompute checksum seed for inode metadata */
4804 if (ext4_has_metadata_csum(sb)) {
4805 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4806 __u32 csum;
4807 __le32 inum = cpu_to_le32(inode->i_ino);
4808 __le32 gen = raw_inode->i_generation;
4809 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4810 sizeof(inum));
4811 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4812 sizeof(gen));
4813 }
4814
4815 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4816 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4817 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4818 ext4_error_inode_err(inode, function, line, 0,
4819 EFSBADCRC, "iget: checksum invalid");
4820 ret = -EFSBADCRC;
4821 goto bad_inode;
4822 }
4823
4824 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4825 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4826 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4827 if (ext4_has_feature_project(sb) &&
4828 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4829 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4830 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4831 else
4832 i_projid = EXT4_DEF_PROJID;
4833
4834 if (!(test_opt(inode->i_sb, NO_UID32))) {
4835 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4836 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4837 }
4838 i_uid_write(inode, i_uid);
4839 i_gid_write(inode, i_gid);
4840 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4841 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4842
4843 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4844 ei->i_inline_off = 0;
4845 ei->i_dir_start_lookup = 0;
4846 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4847 /* We now have enough fields to check if the inode was active or not.
4848 * This is needed because nfsd might try to access dead inodes
4849 * the test is that same one that e2fsck uses
4850 * NeilBrown 1999oct15
4851 */
4852 if (inode->i_nlink == 0) {
4853 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4854 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4855 ino != EXT4_BOOT_LOADER_INO) {
4856 /* this inode is deleted or unallocated */
4857 if (flags & EXT4_IGET_SPECIAL) {
4858 ext4_error_inode(inode, function, line, 0,
4859 "iget: special inode unallocated");
4860 ret = -EFSCORRUPTED;
4861 } else
4862 ret = -ESTALE;
4863 goto bad_inode;
4864 }
4865 /* The only unlinked inodes we let through here have
4866 * valid i_mode and are being read by the orphan
4867 * recovery code: that's fine, we're about to complete
4868 * the process of deleting those.
4869 * OR it is the EXT4_BOOT_LOADER_INO which is
4870 * not initialized on a new filesystem. */
4871 }
4872 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4873 ext4_set_inode_flags(inode, true);
4874 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4875 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4876 if (ext4_has_feature_64bit(sb))
4877 ei->i_file_acl |=
4878 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4879 inode->i_size = ext4_isize(sb, raw_inode);
4880 if ((size = i_size_read(inode)) < 0) {
4881 ext4_error_inode(inode, function, line, 0,
4882 "iget: bad i_size value: %lld", size);
4883 ret = -EFSCORRUPTED;
4884 goto bad_inode;
4885 }
4886 /*
4887 * If dir_index is not enabled but there's dir with INDEX flag set,
4888 * we'd normally treat htree data as empty space. But with metadata
4889 * checksumming that corrupts checksums so forbid that.
4890 */
4891 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4892 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4893 ext4_error_inode(inode, function, line, 0,
4894 "iget: Dir with htree data on filesystem without dir_index feature.");
4895 ret = -EFSCORRUPTED;
4896 goto bad_inode;
4897 }
4898 ei->i_disksize = inode->i_size;
4899 #ifdef CONFIG_QUOTA
4900 ei->i_reserved_quota = 0;
4901 #endif
4902 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4903 ei->i_block_group = iloc.block_group;
4904 ei->i_last_alloc_group = ~0;
4905 /*
4906 * NOTE! The in-memory inode i_data array is in little-endian order
4907 * even on big-endian machines: we do NOT byteswap the block numbers!
4908 */
4909 for (block = 0; block < EXT4_N_BLOCKS; block++)
4910 ei->i_data[block] = raw_inode->i_block[block];
4911 INIT_LIST_HEAD(&ei->i_orphan);
4912 ext4_fc_init_inode(&ei->vfs_inode);
4913
4914 /*
4915 * Set transaction id's of transactions that have to be committed
4916 * to finish f[data]sync. We set them to currently running transaction
4917 * as we cannot be sure that the inode or some of its metadata isn't
4918 * part of the transaction - the inode could have been reclaimed and
4919 * now it is reread from disk.
4920 */
4921 if (journal) {
4922 transaction_t *transaction;
4923 tid_t tid;
4924
4925 read_lock(&journal->j_state_lock);
4926 if (journal->j_running_transaction)
4927 transaction = journal->j_running_transaction;
4928 else
4929 transaction = journal->j_committing_transaction;
4930 if (transaction)
4931 tid = transaction->t_tid;
4932 else
4933 tid = journal->j_commit_sequence;
4934 read_unlock(&journal->j_state_lock);
4935 ei->i_sync_tid = tid;
4936 ei->i_datasync_tid = tid;
4937 }
4938
4939 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4940 if (ei->i_extra_isize == 0) {
4941 /* The extra space is currently unused. Use it. */
4942 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4943 ei->i_extra_isize = sizeof(struct ext4_inode) -
4944 EXT4_GOOD_OLD_INODE_SIZE;
4945 } else {
4946 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4947 if (ret)
4948 goto bad_inode;
4949 }
4950 }
4951
4952 EXT4_INODE_GET_CTIME(inode, raw_inode);
4953 EXT4_INODE_GET_ATIME(inode, raw_inode);
4954 EXT4_INODE_GET_MTIME(inode, raw_inode);
4955 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4956
4957 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4958 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4959
4960 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4961 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4962 ivers |=
4963 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4964 }
4965 ext4_inode_set_iversion_queried(inode, ivers);
4966 }
4967
4968 ret = 0;
4969 if (ei->i_file_acl &&
4970 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4971 ext4_error_inode(inode, function, line, 0,
4972 "iget: bad extended attribute block %llu",
4973 ei->i_file_acl);
4974 ret = -EFSCORRUPTED;
4975 goto bad_inode;
4976 } else if (!ext4_has_inline_data(inode)) {
4977 /* validate the block references in the inode */
4978 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4979 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4980 (S_ISLNK(inode->i_mode) &&
4981 !ext4_inode_is_fast_symlink(inode)))) {
4982 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4983 ret = ext4_ext_check_inode(inode);
4984 else
4985 ret = ext4_ind_check_inode(inode);
4986 }
4987 }
4988 if (ret)
4989 goto bad_inode;
4990
4991 if (S_ISREG(inode->i_mode)) {
4992 inode->i_op = &ext4_file_inode_operations;
4993 inode->i_fop = &ext4_file_operations;
4994 ext4_set_aops(inode);
4995 } else if (S_ISDIR(inode->i_mode)) {
4996 inode->i_op = &ext4_dir_inode_operations;
4997 inode->i_fop = &ext4_dir_operations;
4998 } else if (S_ISLNK(inode->i_mode)) {
4999 /* VFS does not allow setting these so must be corruption */
5000 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5001 ext4_error_inode(inode, function, line, 0,
5002 "iget: immutable or append flags "
5003 "not allowed on symlinks");
5004 ret = -EFSCORRUPTED;
5005 goto bad_inode;
5006 }
5007 if (IS_ENCRYPTED(inode)) {
5008 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5009 } else if (ext4_inode_is_fast_symlink(inode)) {
5010 inode->i_link = (char *)ei->i_data;
5011 inode->i_op = &ext4_fast_symlink_inode_operations;
5012 nd_terminate_link(ei->i_data, inode->i_size,
5013 sizeof(ei->i_data) - 1);
5014 } else {
5015 inode->i_op = &ext4_symlink_inode_operations;
5016 }
5017 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5018 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5019 inode->i_op = &ext4_special_inode_operations;
5020 if (raw_inode->i_block[0])
5021 init_special_inode(inode, inode->i_mode,
5022 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5023 else
5024 init_special_inode(inode, inode->i_mode,
5025 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5026 } else if (ino == EXT4_BOOT_LOADER_INO) {
5027 make_bad_inode(inode);
5028 } else {
5029 ret = -EFSCORRUPTED;
5030 ext4_error_inode(inode, function, line, 0,
5031 "iget: bogus i_mode (%o)", inode->i_mode);
5032 goto bad_inode;
5033 }
5034 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) {
5035 ext4_error_inode(inode, function, line, 0,
5036 "casefold flag without casefold feature");
5037 ret = -EFSCORRUPTED;
5038 goto bad_inode;
5039 }
5040 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
5041 ext4_error_inode(inode, function, line, 0, err_str);
5042 ret = -EFSCORRUPTED;
5043 goto bad_inode;
5044 }
5045
5046 brelse(iloc.bh);
5047 unlock_new_inode(inode);
5048 return inode;
5049
5050 bad_inode:
5051 brelse(iloc.bh);
5052 iget_failed(inode);
5053 return ERR_PTR(ret);
5054 }
5055
5056 static void __ext4_update_other_inode_time(struct super_block *sb,
5057 unsigned long orig_ino,
5058 unsigned long ino,
5059 struct ext4_inode *raw_inode)
5060 {
5061 struct inode *inode;
5062
5063 inode = find_inode_by_ino_rcu(sb, ino);
5064 if (!inode)
5065 return;
5066
5067 if (!inode_is_dirtytime_only(inode))
5068 return;
5069
5070 spin_lock(&inode->i_lock);
5071 if (inode_is_dirtytime_only(inode)) {
5072 struct ext4_inode_info *ei = EXT4_I(inode);
5073
5074 inode->i_state &= ~I_DIRTY_TIME;
5075 spin_unlock(&inode->i_lock);
5076
5077 spin_lock(&ei->i_raw_lock);
5078 EXT4_INODE_SET_CTIME(inode, raw_inode);
5079 EXT4_INODE_SET_MTIME(inode, raw_inode);
5080 EXT4_INODE_SET_ATIME(inode, raw_inode);
5081 ext4_inode_csum_set(inode, raw_inode, ei);
5082 spin_unlock(&ei->i_raw_lock);
5083 trace_ext4_other_inode_update_time(inode, orig_ino);
5084 return;
5085 }
5086 spin_unlock(&inode->i_lock);
5087 }
5088
5089 /*
5090 * Opportunistically update the other time fields for other inodes in
5091 * the same inode table block.
5092 */
5093 static void ext4_update_other_inodes_time(struct super_block *sb,
5094 unsigned long orig_ino, char *buf)
5095 {
5096 unsigned long ino;
5097 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5098 int inode_size = EXT4_INODE_SIZE(sb);
5099
5100 /*
5101 * Calculate the first inode in the inode table block. Inode
5102 * numbers are one-based. That is, the first inode in a block
5103 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5104 */
5105 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5106 rcu_read_lock();
5107 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5108 if (ino == orig_ino)
5109 continue;
5110 __ext4_update_other_inode_time(sb, orig_ino, ino,
5111 (struct ext4_inode *)buf);
5112 }
5113 rcu_read_unlock();
5114 }
5115
5116 /*
5117 * Post the struct inode info into an on-disk inode location in the
5118 * buffer-cache. This gobbles the caller's reference to the
5119 * buffer_head in the inode location struct.
5120 *
5121 * The caller must have write access to iloc->bh.
5122 */
5123 static int ext4_do_update_inode(handle_t *handle,
5124 struct inode *inode,
5125 struct ext4_iloc *iloc)
5126 {
5127 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5128 struct ext4_inode_info *ei = EXT4_I(inode);
5129 struct buffer_head *bh = iloc->bh;
5130 struct super_block *sb = inode->i_sb;
5131 int err;
5132 int need_datasync = 0, set_large_file = 0;
5133
5134 spin_lock(&ei->i_raw_lock);
5135
5136 /*
5137 * For fields not tracked in the in-memory inode, initialise them
5138 * to zero for new inodes.
5139 */
5140 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5141 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5142
5143 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5144 need_datasync = 1;
5145 if (ei->i_disksize > 0x7fffffffULL) {
5146 if (!ext4_has_feature_large_file(sb) ||
5147 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5148 set_large_file = 1;
5149 }
5150
5151 err = ext4_fill_raw_inode(inode, raw_inode);
5152 spin_unlock(&ei->i_raw_lock);
5153 if (err) {
5154 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5155 goto out_brelse;
5156 }
5157
5158 if (inode->i_sb->s_flags & SB_LAZYTIME)
5159 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5160 bh->b_data);
5161
5162 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5163 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5164 if (err)
5165 goto out_error;
5166 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5167 if (set_large_file) {
5168 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5169 err = ext4_journal_get_write_access(handle, sb,
5170 EXT4_SB(sb)->s_sbh,
5171 EXT4_JTR_NONE);
5172 if (err)
5173 goto out_error;
5174 lock_buffer(EXT4_SB(sb)->s_sbh);
5175 ext4_set_feature_large_file(sb);
5176 ext4_superblock_csum_set(sb);
5177 unlock_buffer(EXT4_SB(sb)->s_sbh);
5178 ext4_handle_sync(handle);
5179 err = ext4_handle_dirty_metadata(handle, NULL,
5180 EXT4_SB(sb)->s_sbh);
5181 }
5182 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5183 out_error:
5184 ext4_std_error(inode->i_sb, err);
5185 out_brelse:
5186 brelse(bh);
5187 return err;
5188 }
5189
5190 /*
5191 * ext4_write_inode()
5192 *
5193 * We are called from a few places:
5194 *
5195 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5196 * Here, there will be no transaction running. We wait for any running
5197 * transaction to commit.
5198 *
5199 * - Within flush work (sys_sync(), kupdate and such).
5200 * We wait on commit, if told to.
5201 *
5202 * - Within iput_final() -> write_inode_now()
5203 * We wait on commit, if told to.
5204 *
5205 * In all cases it is actually safe for us to return without doing anything,
5206 * because the inode has been copied into a raw inode buffer in
5207 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5208 * writeback.
5209 *
5210 * Note that we are absolutely dependent upon all inode dirtiers doing the
5211 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5212 * which we are interested.
5213 *
5214 * It would be a bug for them to not do this. The code:
5215 *
5216 * mark_inode_dirty(inode)
5217 * stuff();
5218 * inode->i_size = expr;
5219 *
5220 * is in error because write_inode() could occur while `stuff()' is running,
5221 * and the new i_size will be lost. Plus the inode will no longer be on the
5222 * superblock's dirty inode list.
5223 */
5224 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5225 {
5226 int err;
5227
5228 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5229 return 0;
5230
5231 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5232 return -EIO;
5233
5234 if (EXT4_SB(inode->i_sb)->s_journal) {
5235 if (ext4_journal_current_handle()) {
5236 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5237 dump_stack();
5238 return -EIO;
5239 }
5240
5241 /*
5242 * No need to force transaction in WB_SYNC_NONE mode. Also
5243 * ext4_sync_fs() will force the commit after everything is
5244 * written.
5245 */
5246 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5247 return 0;
5248
5249 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5250 EXT4_I(inode)->i_sync_tid);
5251 } else {
5252 struct ext4_iloc iloc;
5253
5254 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5255 if (err)
5256 return err;
5257 /*
5258 * sync(2) will flush the whole buffer cache. No need to do
5259 * it here separately for each inode.
5260 */
5261 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5262 sync_dirty_buffer(iloc.bh);
5263 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5264 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5265 "IO error syncing inode");
5266 err = -EIO;
5267 }
5268 brelse(iloc.bh);
5269 }
5270 return err;
5271 }
5272
5273 /*
5274 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5275 * buffers that are attached to a folio straddling i_size and are undergoing
5276 * commit. In that case we have to wait for commit to finish and try again.
5277 */
5278 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5279 {
5280 unsigned offset;
5281 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5282 tid_t commit_tid;
5283 int ret;
5284 bool has_transaction;
5285
5286 offset = inode->i_size & (PAGE_SIZE - 1);
5287 /*
5288 * If the folio is fully truncated, we don't need to wait for any commit
5289 * (and we even should not as __ext4_journalled_invalidate_folio() may
5290 * strip all buffers from the folio but keep the folio dirty which can then
5291 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5292 * buffers). Also we don't need to wait for any commit if all buffers in
5293 * the folio remain valid. This is most beneficial for the common case of
5294 * blocksize == PAGESIZE.
5295 */
5296 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5297 return;
5298 while (1) {
5299 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5300 inode->i_size >> PAGE_SHIFT);
5301 if (IS_ERR(folio))
5302 return;
5303 ret = __ext4_journalled_invalidate_folio(folio, offset,
5304 folio_size(folio) - offset);
5305 folio_unlock(folio);
5306 folio_put(folio);
5307 if (ret != -EBUSY)
5308 return;
5309 has_transaction = false;
5310 read_lock(&journal->j_state_lock);
5311 if (journal->j_committing_transaction) {
5312 commit_tid = journal->j_committing_transaction->t_tid;
5313 has_transaction = true;
5314 }
5315 read_unlock(&journal->j_state_lock);
5316 if (has_transaction)
5317 jbd2_log_wait_commit(journal, commit_tid);
5318 }
5319 }
5320
5321 /*
5322 * ext4_setattr()
5323 *
5324 * Called from notify_change.
5325 *
5326 * We want to trap VFS attempts to truncate the file as soon as
5327 * possible. In particular, we want to make sure that when the VFS
5328 * shrinks i_size, we put the inode on the orphan list and modify
5329 * i_disksize immediately, so that during the subsequent flushing of
5330 * dirty pages and freeing of disk blocks, we can guarantee that any
5331 * commit will leave the blocks being flushed in an unused state on
5332 * disk. (On recovery, the inode will get truncated and the blocks will
5333 * be freed, so we have a strong guarantee that no future commit will
5334 * leave these blocks visible to the user.)
5335 *
5336 * Another thing we have to assure is that if we are in ordered mode
5337 * and inode is still attached to the committing transaction, we must
5338 * we start writeout of all the dirty pages which are being truncated.
5339 * This way we are sure that all the data written in the previous
5340 * transaction are already on disk (truncate waits for pages under
5341 * writeback).
5342 *
5343 * Called with inode->i_rwsem down.
5344 */
5345 int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5346 struct iattr *attr)
5347 {
5348 struct inode *inode = d_inode(dentry);
5349 int error, rc = 0;
5350 int orphan = 0;
5351 const unsigned int ia_valid = attr->ia_valid;
5352 bool inc_ivers = true;
5353
5354 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5355 return -EIO;
5356
5357 if (unlikely(IS_IMMUTABLE(inode)))
5358 return -EPERM;
5359
5360 if (unlikely(IS_APPEND(inode) &&
5361 (ia_valid & (ATTR_MODE | ATTR_UID |
5362 ATTR_GID | ATTR_TIMES_SET))))
5363 return -EPERM;
5364
5365 error = setattr_prepare(idmap, dentry, attr);
5366 if (error)
5367 return error;
5368
5369 error = fscrypt_prepare_setattr(dentry, attr);
5370 if (error)
5371 return error;
5372
5373 error = fsverity_prepare_setattr(dentry, attr);
5374 if (error)
5375 return error;
5376
5377 if (is_quota_modification(idmap, inode, attr)) {
5378 error = dquot_initialize(inode);
5379 if (error)
5380 return error;
5381 }
5382
5383 if (i_uid_needs_update(idmap, attr, inode) ||
5384 i_gid_needs_update(idmap, attr, inode)) {
5385 handle_t *handle;
5386
5387 /* (user+group)*(old+new) structure, inode write (sb,
5388 * inode block, ? - but truncate inode update has it) */
5389 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5390 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5391 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5392 if (IS_ERR(handle)) {
5393 error = PTR_ERR(handle);
5394 goto err_out;
5395 }
5396
5397 /* dquot_transfer() calls back ext4_get_inode_usage() which
5398 * counts xattr inode references.
5399 */
5400 down_read(&EXT4_I(inode)->xattr_sem);
5401 error = dquot_transfer(idmap, inode, attr);
5402 up_read(&EXT4_I(inode)->xattr_sem);
5403
5404 if (error) {
5405 ext4_journal_stop(handle);
5406 return error;
5407 }
5408 /* Update corresponding info in inode so that everything is in
5409 * one transaction */
5410 i_uid_update(idmap, attr, inode);
5411 i_gid_update(idmap, attr, inode);
5412 error = ext4_mark_inode_dirty(handle, inode);
5413 ext4_journal_stop(handle);
5414 if (unlikely(error)) {
5415 return error;
5416 }
5417 }
5418
5419 if (attr->ia_valid & ATTR_SIZE) {
5420 handle_t *handle;
5421 loff_t oldsize = inode->i_size;
5422 loff_t old_disksize;
5423 int shrink = (attr->ia_size < inode->i_size);
5424
5425 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5426 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5427
5428 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5429 return -EFBIG;
5430 }
5431 }
5432 if (!S_ISREG(inode->i_mode)) {
5433 return -EINVAL;
5434 }
5435
5436 if (attr->ia_size == inode->i_size)
5437 inc_ivers = false;
5438
5439 if (shrink) {
5440 if (ext4_should_order_data(inode)) {
5441 error = ext4_begin_ordered_truncate(inode,
5442 attr->ia_size);
5443 if (error)
5444 goto err_out;
5445 }
5446 /*
5447 * Blocks are going to be removed from the inode. Wait
5448 * for dio in flight.
5449 */
5450 inode_dio_wait(inode);
5451 }
5452
5453 filemap_invalidate_lock(inode->i_mapping);
5454
5455 rc = ext4_break_layouts(inode);
5456 if (rc) {
5457 filemap_invalidate_unlock(inode->i_mapping);
5458 goto err_out;
5459 }
5460
5461 if (attr->ia_size != inode->i_size) {
5462 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5463 if (IS_ERR(handle)) {
5464 error = PTR_ERR(handle);
5465 goto out_mmap_sem;
5466 }
5467 if (ext4_handle_valid(handle) && shrink) {
5468 error = ext4_orphan_add(handle, inode);
5469 orphan = 1;
5470 }
5471 /*
5472 * Update c/mtime on truncate up, ext4_truncate() will
5473 * update c/mtime in shrink case below
5474 */
5475 if (!shrink)
5476 inode_set_mtime_to_ts(inode,
5477 inode_set_ctime_current(inode));
5478
5479 if (shrink)
5480 ext4_fc_track_range(handle, inode,
5481 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5482 inode->i_sb->s_blocksize_bits,
5483 EXT_MAX_BLOCKS - 1);
5484 else
5485 ext4_fc_track_range(
5486 handle, inode,
5487 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5488 inode->i_sb->s_blocksize_bits,
5489 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5490 inode->i_sb->s_blocksize_bits);
5491
5492 down_write(&EXT4_I(inode)->i_data_sem);
5493 old_disksize = EXT4_I(inode)->i_disksize;
5494 EXT4_I(inode)->i_disksize = attr->ia_size;
5495 rc = ext4_mark_inode_dirty(handle, inode);
5496 if (!error)
5497 error = rc;
5498 /*
5499 * We have to update i_size under i_data_sem together
5500 * with i_disksize to avoid races with writeback code
5501 * running ext4_wb_update_i_disksize().
5502 */
5503 if (!error)
5504 i_size_write(inode, attr->ia_size);
5505 else
5506 EXT4_I(inode)->i_disksize = old_disksize;
5507 up_write(&EXT4_I(inode)->i_data_sem);
5508 ext4_journal_stop(handle);
5509 if (error)
5510 goto out_mmap_sem;
5511 if (!shrink) {
5512 pagecache_isize_extended(inode, oldsize,
5513 inode->i_size);
5514 } else if (ext4_should_journal_data(inode)) {
5515 ext4_wait_for_tail_page_commit(inode);
5516 }
5517 }
5518
5519 /*
5520 * Truncate pagecache after we've waited for commit
5521 * in data=journal mode to make pages freeable.
5522 */
5523 truncate_pagecache(inode, inode->i_size);
5524 /*
5525 * Call ext4_truncate() even if i_size didn't change to
5526 * truncate possible preallocated blocks.
5527 */
5528 if (attr->ia_size <= oldsize) {
5529 rc = ext4_truncate(inode);
5530 if (rc)
5531 error = rc;
5532 }
5533 out_mmap_sem:
5534 filemap_invalidate_unlock(inode->i_mapping);
5535 }
5536
5537 if (!error) {
5538 if (inc_ivers)
5539 inode_inc_iversion(inode);
5540 setattr_copy(idmap, inode, attr);
5541 mark_inode_dirty(inode);
5542 }
5543
5544 /*
5545 * If the call to ext4_truncate failed to get a transaction handle at
5546 * all, we need to clean up the in-core orphan list manually.
5547 */
5548 if (orphan && inode->i_nlink)
5549 ext4_orphan_del(NULL, inode);
5550
5551 if (!error && (ia_valid & ATTR_MODE))
5552 rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5553
5554 err_out:
5555 if (error)
5556 ext4_std_error(inode->i_sb, error);
5557 if (!error)
5558 error = rc;
5559 return error;
5560 }
5561
5562 u32 ext4_dio_alignment(struct inode *inode)
5563 {
5564 if (fsverity_active(inode))
5565 return 0;
5566 if (ext4_should_journal_data(inode))
5567 return 0;
5568 if (ext4_has_inline_data(inode))
5569 return 0;
5570 if (IS_ENCRYPTED(inode)) {
5571 if (!fscrypt_dio_supported(inode))
5572 return 0;
5573 return i_blocksize(inode);
5574 }
5575 return 1; /* use the iomap defaults */
5576 }
5577
5578 int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5579 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5580 {
5581 struct inode *inode = d_inode(path->dentry);
5582 struct ext4_inode *raw_inode;
5583 struct ext4_inode_info *ei = EXT4_I(inode);
5584 unsigned int flags;
5585
5586 if ((request_mask & STATX_BTIME) &&
5587 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5588 stat->result_mask |= STATX_BTIME;
5589 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5590 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5591 }
5592
5593 /*
5594 * Return the DIO alignment restrictions if requested. We only return
5595 * this information when requested, since on encrypted files it might
5596 * take a fair bit of work to get if the file wasn't opened recently.
5597 */
5598 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5599 u32 dio_align = ext4_dio_alignment(inode);
5600
5601 stat->result_mask |= STATX_DIOALIGN;
5602 if (dio_align == 1) {
5603 struct block_device *bdev = inode->i_sb->s_bdev;
5604
5605 /* iomap defaults */
5606 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5607 stat->dio_offset_align = bdev_logical_block_size(bdev);
5608 } else {
5609 stat->dio_mem_align = dio_align;
5610 stat->dio_offset_align = dio_align;
5611 }
5612 }
5613
5614 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5615 if (flags & EXT4_APPEND_FL)
5616 stat->attributes |= STATX_ATTR_APPEND;
5617 if (flags & EXT4_COMPR_FL)
5618 stat->attributes |= STATX_ATTR_COMPRESSED;
5619 if (flags & EXT4_ENCRYPT_FL)
5620 stat->attributes |= STATX_ATTR_ENCRYPTED;
5621 if (flags & EXT4_IMMUTABLE_FL)
5622 stat->attributes |= STATX_ATTR_IMMUTABLE;
5623 if (flags & EXT4_NODUMP_FL)
5624 stat->attributes |= STATX_ATTR_NODUMP;
5625 if (flags & EXT4_VERITY_FL)
5626 stat->attributes |= STATX_ATTR_VERITY;
5627
5628 stat->attributes_mask |= (STATX_ATTR_APPEND |
5629 STATX_ATTR_COMPRESSED |
5630 STATX_ATTR_ENCRYPTED |
5631 STATX_ATTR_IMMUTABLE |
5632 STATX_ATTR_NODUMP |
5633 STATX_ATTR_VERITY);
5634
5635 generic_fillattr(idmap, request_mask, inode, stat);
5636 return 0;
5637 }
5638
5639 int ext4_file_getattr(struct mnt_idmap *idmap,
5640 const struct path *path, struct kstat *stat,
5641 u32 request_mask, unsigned int query_flags)
5642 {
5643 struct inode *inode = d_inode(path->dentry);
5644 u64 delalloc_blocks;
5645
5646 ext4_getattr(idmap, path, stat, request_mask, query_flags);
5647
5648 /*
5649 * If there is inline data in the inode, the inode will normally not
5650 * have data blocks allocated (it may have an external xattr block).
5651 * Report at least one sector for such files, so tools like tar, rsync,
5652 * others don't incorrectly think the file is completely sparse.
5653 */
5654 if (unlikely(ext4_has_inline_data(inode)))
5655 stat->blocks += (stat->size + 511) >> 9;
5656
5657 /*
5658 * We can't update i_blocks if the block allocation is delayed
5659 * otherwise in the case of system crash before the real block
5660 * allocation is done, we will have i_blocks inconsistent with
5661 * on-disk file blocks.
5662 * We always keep i_blocks updated together with real
5663 * allocation. But to not confuse with user, stat
5664 * will return the blocks that include the delayed allocation
5665 * blocks for this file.
5666 */
5667 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5668 EXT4_I(inode)->i_reserved_data_blocks);
5669 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5670 return 0;
5671 }
5672
5673 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5674 int pextents)
5675 {
5676 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5677 return ext4_ind_trans_blocks(inode, lblocks);
5678 return ext4_ext_index_trans_blocks(inode, pextents);
5679 }
5680
5681 /*
5682 * Account for index blocks, block groups bitmaps and block group
5683 * descriptor blocks if modify datablocks and index blocks
5684 * worse case, the indexs blocks spread over different block groups
5685 *
5686 * If datablocks are discontiguous, they are possible to spread over
5687 * different block groups too. If they are contiguous, with flexbg,
5688 * they could still across block group boundary.
5689 *
5690 * Also account for superblock, inode, quota and xattr blocks
5691 */
5692 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5693 int pextents)
5694 {
5695 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5696 int gdpblocks;
5697 int idxblocks;
5698 int ret;
5699
5700 /*
5701 * How many index blocks need to touch to map @lblocks logical blocks
5702 * to @pextents physical extents?
5703 */
5704 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5705
5706 ret = idxblocks;
5707
5708 /*
5709 * Now let's see how many group bitmaps and group descriptors need
5710 * to account
5711 */
5712 groups = idxblocks + pextents;
5713 gdpblocks = groups;
5714 if (groups > ngroups)
5715 groups = ngroups;
5716 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5717 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5718
5719 /* bitmaps and block group descriptor blocks */
5720 ret += groups + gdpblocks;
5721
5722 /* Blocks for super block, inode, quota and xattr blocks */
5723 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5724
5725 return ret;
5726 }
5727
5728 /*
5729 * Calculate the total number of credits to reserve to fit
5730 * the modification of a single pages into a single transaction,
5731 * which may include multiple chunks of block allocations.
5732 *
5733 * This could be called via ext4_write_begin()
5734 *
5735 * We need to consider the worse case, when
5736 * one new block per extent.
5737 */
5738 int ext4_writepage_trans_blocks(struct inode *inode)
5739 {
5740 int bpp = ext4_journal_blocks_per_page(inode);
5741 int ret;
5742
5743 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5744
5745 /* Account for data blocks for journalled mode */
5746 if (ext4_should_journal_data(inode))
5747 ret += bpp;
5748 return ret;
5749 }
5750
5751 /*
5752 * Calculate the journal credits for a chunk of data modification.
5753 *
5754 * This is called from DIO, fallocate or whoever calling
5755 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5756 *
5757 * journal buffers for data blocks are not included here, as DIO
5758 * and fallocate do no need to journal data buffers.
5759 */
5760 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5761 {
5762 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5763 }
5764
5765 /*
5766 * The caller must have previously called ext4_reserve_inode_write().
5767 * Give this, we know that the caller already has write access to iloc->bh.
5768 */
5769 int ext4_mark_iloc_dirty(handle_t *handle,
5770 struct inode *inode, struct ext4_iloc *iloc)
5771 {
5772 int err = 0;
5773
5774 if (unlikely(ext4_forced_shutdown(inode->i_sb))) {
5775 put_bh(iloc->bh);
5776 return -EIO;
5777 }
5778 ext4_fc_track_inode(handle, inode);
5779
5780 /* the do_update_inode consumes one bh->b_count */
5781 get_bh(iloc->bh);
5782
5783 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5784 err = ext4_do_update_inode(handle, inode, iloc);
5785 put_bh(iloc->bh);
5786 return err;
5787 }
5788
5789 /*
5790 * On success, We end up with an outstanding reference count against
5791 * iloc->bh. This _must_ be cleaned up later.
5792 */
5793
5794 int
5795 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5796 struct ext4_iloc *iloc)
5797 {
5798 int err;
5799
5800 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5801 return -EIO;
5802
5803 err = ext4_get_inode_loc(inode, iloc);
5804 if (!err) {
5805 BUFFER_TRACE(iloc->bh, "get_write_access");
5806 err = ext4_journal_get_write_access(handle, inode->i_sb,
5807 iloc->bh, EXT4_JTR_NONE);
5808 if (err) {
5809 brelse(iloc->bh);
5810 iloc->bh = NULL;
5811 }
5812 }
5813 ext4_std_error(inode->i_sb, err);
5814 return err;
5815 }
5816
5817 static int __ext4_expand_extra_isize(struct inode *inode,
5818 unsigned int new_extra_isize,
5819 struct ext4_iloc *iloc,
5820 handle_t *handle, int *no_expand)
5821 {
5822 struct ext4_inode *raw_inode;
5823 struct ext4_xattr_ibody_header *header;
5824 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5825 struct ext4_inode_info *ei = EXT4_I(inode);
5826 int error;
5827
5828 /* this was checked at iget time, but double check for good measure */
5829 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5830 (ei->i_extra_isize & 3)) {
5831 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5832 ei->i_extra_isize,
5833 EXT4_INODE_SIZE(inode->i_sb));
5834 return -EFSCORRUPTED;
5835 }
5836 if ((new_extra_isize < ei->i_extra_isize) ||
5837 (new_extra_isize < 4) ||
5838 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5839 return -EINVAL; /* Should never happen */
5840
5841 raw_inode = ext4_raw_inode(iloc);
5842
5843 header = IHDR(inode, raw_inode);
5844
5845 /* No extended attributes present */
5846 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5847 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5848 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5849 EXT4_I(inode)->i_extra_isize, 0,
5850 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5851 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5852 return 0;
5853 }
5854
5855 /*
5856 * We may need to allocate external xattr block so we need quotas
5857 * initialized. Here we can be called with various locks held so we
5858 * cannot affort to initialize quotas ourselves. So just bail.
5859 */
5860 if (dquot_initialize_needed(inode))
5861 return -EAGAIN;
5862
5863 /* try to expand with EAs present */
5864 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5865 raw_inode, handle);
5866 if (error) {
5867 /*
5868 * Inode size expansion failed; don't try again
5869 */
5870 *no_expand = 1;
5871 }
5872
5873 return error;
5874 }
5875
5876 /*
5877 * Expand an inode by new_extra_isize bytes.
5878 * Returns 0 on success or negative error number on failure.
5879 */
5880 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5881 unsigned int new_extra_isize,
5882 struct ext4_iloc iloc,
5883 handle_t *handle)
5884 {
5885 int no_expand;
5886 int error;
5887
5888 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5889 return -EOVERFLOW;
5890
5891 /*
5892 * In nojournal mode, we can immediately attempt to expand
5893 * the inode. When journaled, we first need to obtain extra
5894 * buffer credits since we may write into the EA block
5895 * with this same handle. If journal_extend fails, then it will
5896 * only result in a minor loss of functionality for that inode.
5897 * If this is felt to be critical, then e2fsck should be run to
5898 * force a large enough s_min_extra_isize.
5899 */
5900 if (ext4_journal_extend(handle,
5901 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5902 return -ENOSPC;
5903
5904 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5905 return -EBUSY;
5906
5907 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5908 handle, &no_expand);
5909 ext4_write_unlock_xattr(inode, &no_expand);
5910
5911 return error;
5912 }
5913
5914 int ext4_expand_extra_isize(struct inode *inode,
5915 unsigned int new_extra_isize,
5916 struct ext4_iloc *iloc)
5917 {
5918 handle_t *handle;
5919 int no_expand;
5920 int error, rc;
5921
5922 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5923 brelse(iloc->bh);
5924 return -EOVERFLOW;
5925 }
5926
5927 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5928 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5929 if (IS_ERR(handle)) {
5930 error = PTR_ERR(handle);
5931 brelse(iloc->bh);
5932 return error;
5933 }
5934
5935 ext4_write_lock_xattr(inode, &no_expand);
5936
5937 BUFFER_TRACE(iloc->bh, "get_write_access");
5938 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5939 EXT4_JTR_NONE);
5940 if (error) {
5941 brelse(iloc->bh);
5942 goto out_unlock;
5943 }
5944
5945 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5946 handle, &no_expand);
5947
5948 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5949 if (!error)
5950 error = rc;
5951
5952 out_unlock:
5953 ext4_write_unlock_xattr(inode, &no_expand);
5954 ext4_journal_stop(handle);
5955 return error;
5956 }
5957
5958 /*
5959 * What we do here is to mark the in-core inode as clean with respect to inode
5960 * dirtiness (it may still be data-dirty).
5961 * This means that the in-core inode may be reaped by prune_icache
5962 * without having to perform any I/O. This is a very good thing,
5963 * because *any* task may call prune_icache - even ones which
5964 * have a transaction open against a different journal.
5965 *
5966 * Is this cheating? Not really. Sure, we haven't written the
5967 * inode out, but prune_icache isn't a user-visible syncing function.
5968 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5969 * we start and wait on commits.
5970 */
5971 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5972 const char *func, unsigned int line)
5973 {
5974 struct ext4_iloc iloc;
5975 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5976 int err;
5977
5978 might_sleep();
5979 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5980 err = ext4_reserve_inode_write(handle, inode, &iloc);
5981 if (err)
5982 goto out;
5983
5984 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5985 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5986 iloc, handle);
5987
5988 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5989 out:
5990 if (unlikely(err))
5991 ext4_error_inode_err(inode, func, line, 0, err,
5992 "mark_inode_dirty error");
5993 return err;
5994 }
5995
5996 /*
5997 * ext4_dirty_inode() is called from __mark_inode_dirty()
5998 *
5999 * We're really interested in the case where a file is being extended.
6000 * i_size has been changed by generic_commit_write() and we thus need
6001 * to include the updated inode in the current transaction.
6002 *
6003 * Also, dquot_alloc_block() will always dirty the inode when blocks
6004 * are allocated to the file.
6005 *
6006 * If the inode is marked synchronous, we don't honour that here - doing
6007 * so would cause a commit on atime updates, which we don't bother doing.
6008 * We handle synchronous inodes at the highest possible level.
6009 */
6010 void ext4_dirty_inode(struct inode *inode, int flags)
6011 {
6012 handle_t *handle;
6013
6014 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6015 if (IS_ERR(handle))
6016 return;
6017 ext4_mark_inode_dirty(handle, inode);
6018 ext4_journal_stop(handle);
6019 }
6020
6021 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6022 {
6023 journal_t *journal;
6024 handle_t *handle;
6025 int err;
6026 int alloc_ctx;
6027
6028 /*
6029 * We have to be very careful here: changing a data block's
6030 * journaling status dynamically is dangerous. If we write a
6031 * data block to the journal, change the status and then delete
6032 * that block, we risk forgetting to revoke the old log record
6033 * from the journal and so a subsequent replay can corrupt data.
6034 * So, first we make sure that the journal is empty and that
6035 * nobody is changing anything.
6036 */
6037
6038 journal = EXT4_JOURNAL(inode);
6039 if (!journal)
6040 return 0;
6041 if (is_journal_aborted(journal))
6042 return -EROFS;
6043
6044 /* Wait for all existing dio workers */
6045 inode_dio_wait(inode);
6046
6047 /*
6048 * Before flushing the journal and switching inode's aops, we have
6049 * to flush all dirty data the inode has. There can be outstanding
6050 * delayed allocations, there can be unwritten extents created by
6051 * fallocate or buffered writes in dioread_nolock mode covered by
6052 * dirty data which can be converted only after flushing the dirty
6053 * data (and journalled aops don't know how to handle these cases).
6054 */
6055 if (val) {
6056 filemap_invalidate_lock(inode->i_mapping);
6057 err = filemap_write_and_wait(inode->i_mapping);
6058 if (err < 0) {
6059 filemap_invalidate_unlock(inode->i_mapping);
6060 return err;
6061 }
6062 }
6063
6064 alloc_ctx = ext4_writepages_down_write(inode->i_sb);
6065 jbd2_journal_lock_updates(journal);
6066
6067 /*
6068 * OK, there are no updates running now, and all cached data is
6069 * synced to disk. We are now in a completely consistent state
6070 * which doesn't have anything in the journal, and we know that
6071 * no filesystem updates are running, so it is safe to modify
6072 * the inode's in-core data-journaling state flag now.
6073 */
6074
6075 if (val)
6076 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6077 else {
6078 err = jbd2_journal_flush(journal, 0);
6079 if (err < 0) {
6080 jbd2_journal_unlock_updates(journal);
6081 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6082 return err;
6083 }
6084 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6085 }
6086 ext4_set_aops(inode);
6087
6088 jbd2_journal_unlock_updates(journal);
6089 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6090
6091 if (val)
6092 filemap_invalidate_unlock(inode->i_mapping);
6093
6094 /* Finally we can mark the inode as dirty. */
6095
6096 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6097 if (IS_ERR(handle))
6098 return PTR_ERR(handle);
6099
6100 ext4_fc_mark_ineligible(inode->i_sb,
6101 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6102 err = ext4_mark_inode_dirty(handle, inode);
6103 ext4_handle_sync(handle);
6104 ext4_journal_stop(handle);
6105 ext4_std_error(inode->i_sb, err);
6106
6107 return err;
6108 }
6109
6110 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6111 struct buffer_head *bh)
6112 {
6113 return !buffer_mapped(bh);
6114 }
6115
6116 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6117 {
6118 struct vm_area_struct *vma = vmf->vma;
6119 struct folio *folio = page_folio(vmf->page);
6120 loff_t size;
6121 unsigned long len;
6122 int err;
6123 vm_fault_t ret;
6124 struct file *file = vma->vm_file;
6125 struct inode *inode = file_inode(file);
6126 struct address_space *mapping = inode->i_mapping;
6127 handle_t *handle;
6128 get_block_t *get_block;
6129 int retries = 0;
6130
6131 if (unlikely(IS_IMMUTABLE(inode)))
6132 return VM_FAULT_SIGBUS;
6133
6134 sb_start_pagefault(inode->i_sb);
6135 file_update_time(vma->vm_file);
6136
6137 filemap_invalidate_lock_shared(mapping);
6138
6139 err = ext4_convert_inline_data(inode);
6140 if (err)
6141 goto out_ret;
6142
6143 /*
6144 * On data journalling we skip straight to the transaction handle:
6145 * there's no delalloc; page truncated will be checked later; the
6146 * early return w/ all buffers mapped (calculates size/len) can't
6147 * be used; and there's no dioread_nolock, so only ext4_get_block.
6148 */
6149 if (ext4_should_journal_data(inode))
6150 goto retry_alloc;
6151
6152 /* Delalloc case is easy... */
6153 if (test_opt(inode->i_sb, DELALLOC) &&
6154 !ext4_nonda_switch(inode->i_sb)) {
6155 do {
6156 err = block_page_mkwrite(vma, vmf,
6157 ext4_da_get_block_prep);
6158 } while (err == -ENOSPC &&
6159 ext4_should_retry_alloc(inode->i_sb, &retries));
6160 goto out_ret;
6161 }
6162
6163 folio_lock(folio);
6164 size = i_size_read(inode);
6165 /* Page got truncated from under us? */
6166 if (folio->mapping != mapping || folio_pos(folio) > size) {
6167 folio_unlock(folio);
6168 ret = VM_FAULT_NOPAGE;
6169 goto out;
6170 }
6171
6172 len = folio_size(folio);
6173 if (folio_pos(folio) + len > size)
6174 len = size - folio_pos(folio);
6175 /*
6176 * Return if we have all the buffers mapped. This avoids the need to do
6177 * journal_start/journal_stop which can block and take a long time
6178 *
6179 * This cannot be done for data journalling, as we have to add the
6180 * inode to the transaction's list to writeprotect pages on commit.
6181 */
6182 if (folio_buffers(folio)) {
6183 if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6184 0, len, NULL,
6185 ext4_bh_unmapped)) {
6186 /* Wait so that we don't change page under IO */
6187 folio_wait_stable(folio);
6188 ret = VM_FAULT_LOCKED;
6189 goto out;
6190 }
6191 }
6192 folio_unlock(folio);
6193 /* OK, we need to fill the hole... */
6194 if (ext4_should_dioread_nolock(inode))
6195 get_block = ext4_get_block_unwritten;
6196 else
6197 get_block = ext4_get_block;
6198 retry_alloc:
6199 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6200 ext4_writepage_trans_blocks(inode));
6201 if (IS_ERR(handle)) {
6202 ret = VM_FAULT_SIGBUS;
6203 goto out;
6204 }
6205 /*
6206 * Data journalling can't use block_page_mkwrite() because it
6207 * will set_buffer_dirty() before do_journal_get_write_access()
6208 * thus might hit warning messages for dirty metadata buffers.
6209 */
6210 if (!ext4_should_journal_data(inode)) {
6211 err = block_page_mkwrite(vma, vmf, get_block);
6212 } else {
6213 folio_lock(folio);
6214 size = i_size_read(inode);
6215 /* Page got truncated from under us? */
6216 if (folio->mapping != mapping || folio_pos(folio) > size) {
6217 ret = VM_FAULT_NOPAGE;
6218 goto out_error;
6219 }
6220
6221 len = folio_size(folio);
6222 if (folio_pos(folio) + len > size)
6223 len = size - folio_pos(folio);
6224
6225 err = __block_write_begin(&folio->page, 0, len, ext4_get_block);
6226 if (!err) {
6227 ret = VM_FAULT_SIGBUS;
6228 if (ext4_journal_folio_buffers(handle, folio, len))
6229 goto out_error;
6230 } else {
6231 folio_unlock(folio);
6232 }
6233 }
6234 ext4_journal_stop(handle);
6235 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6236 goto retry_alloc;
6237 out_ret:
6238 ret = vmf_fs_error(err);
6239 out:
6240 filemap_invalidate_unlock_shared(mapping);
6241 sb_end_pagefault(inode->i_sb);
6242 return ret;
6243 out_error:
6244 folio_unlock(folio);
6245 ext4_journal_stop(handle);
6246 goto out;
6247 }
6248
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.