1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
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)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
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>
45 #include "ext4_jbd2.h"
50 #include <trace/events/ext4.h>
52 static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
58 int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 unsigned int csum_size = sizeof(dummy_csum);
61 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
64 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 EXT4_GOOD_OLD_INODE_SIZE - offset);
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,
77 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 EXT4_INODE_SIZE(inode->i_sb) - offset);
84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
87 __u32 provided, calculated;
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))
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;
100 calculated &= 0xFFFF;
102 return provided == calculated;
105 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
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))
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);
122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
125 trace_ext4_begin_ordered_truncate(inode, new_size);
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.
132 if (!EXT4_I(inode)->jinode)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
139 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
143 * Test whether an inode is a fast symlink.
144 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
146 int ext4_inode_is_fast_symlink(struct inode *inode)
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;
152 if (ext4_has_inline_data(inode))
155 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
157 return S_ISLNK(inode->i_mode) && inode->i_size &&
158 (inode->i_size < EXT4_N_BLOCKS * 4);
162 * Called at the last iput() if i_nlink is zero.
164 void ext4_evict_inode(struct inode *inode)
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)
173 int extra_credits = 6;
174 struct ext4_xattr_inode_array *ea_inode_array = NULL;
175 bool freeze_protected = false;
177 trace_ext4_evict_inode(inode);
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);
187 if (is_bad_inode(inode))
189 dquot_initialize(inode);
191 if (ext4_should_order_data(inode))
192 ext4_begin_ordered_truncate(inode, 0);
193 truncate_inode_pages_final(&inode->i_data);
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.
202 if (!list_empty_careful(&inode->i_io_list))
203 inode_io_list_del(inode);
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.
211 if (!ext4_journal_current_handle()) {
212 sb_start_intwrite(inode->i_sb);
213 freeze_protected = true;
216 if (!IS_NOQUOTA(inode))
217 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
220 * Block bitmap, group descriptor, and inode are accounted in both
221 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
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));
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
232 ext4_orphan_del(NULL, inode);
233 if (freeze_protected)
234 sb_end_intwrite(inode->i_sb);
239 ext4_handle_sync(handle);
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.
248 if (ext4_inode_is_fast_symlink(inode))
249 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
251 err = ext4_mark_inode_dirty(handle, inode);
253 ext4_warning(inode->i_sb,
254 "couldn't mark inode dirty (err %d)", err);
257 if (inode->i_blocks) {
258 err = ext4_truncate(inode);
260 ext4_error_err(inode->i_sb, -err,
261 "couldn't truncate inode %lu (err %d)",
267 /* Remove xattr references. */
268 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
271 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
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);
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)
289 ext4_orphan_del(handle, inode);
290 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
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
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);
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);
311 * Check out some where else accidentally dirty the evicting inode,
312 * which may probably cause inode use-after-free issues later.
314 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
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... */
322 qsize_t *ext4_get_reserved_space(struct inode *inode)
324 return &EXT4_I(inode)->i_reserved_quota;
329 * Called with i_data_sem down, which is important since we can call
330 * ext4_discard_preallocations() from here.
332 void ext4_da_update_reserve_space(struct inode *inode,
333 int used, int quota_claim)
335 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
336 struct ext4_inode_info *ei = EXT4_I(inode);
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);
346 used = ei->i_reserved_data_blocks;
349 /* Update per-inode reservations */
350 ei->i_reserved_data_blocks -= used;
351 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
353 spin_unlock(&ei->i_block_reservation_lock);
355 /* Update quota subsystem for data blocks */
357 dquot_claim_block(inode, EXT4_C2B(sbi, used));
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.
364 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
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.
372 if ((ei->i_reserved_data_blocks == 0) &&
373 !inode_is_open_for_write(inode))
374 ext4_discard_preallocations(inode, 0);
377 static int __check_block_validity(struct inode *inode, const char *func,
379 struct ext4_map_blocks *map)
381 if (ext4_has_feature_journal(inode->i_sb) &&
383 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
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;
395 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
400 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
401 return fscrypt_zeroout_range(inode, lblk, pblk, len);
403 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
410 #define check_block_validity(inode, map) \
411 __check_block_validity((inode), __func__, __LINE__, (map))
413 #ifdef ES_AGGRESSIVE_TEST
414 static void ext4_map_blocks_es_recheck(handle_t *handle,
416 struct ext4_map_blocks *es_map,
417 struct ext4_map_blocks *map,
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.
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);
434 retval = ext4_ind_map_blocks(handle, inode, map, 0);
436 up_read((&EXT4_I(inode)->i_data_sem));
439 * We don't check m_len because extent will be collpased in status
440 * tree. So the m_len might not equal.
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,
454 #endif /* ES_AGGRESSIVE_TEST */
457 * The ext4_map_blocks() function tries to look up the requested blocks,
458 * and returns if the blocks are already mapped.
460 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
461 * and store the allocated blocks in the result buffer head and mark it
464 * If file type is extents based, it will call ext4_ext_map_blocks(),
465 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
468 * On success, it returns the number of blocks being mapped or allocated. if
469 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
470 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
472 * It returns 0 if plain look up failed (blocks have not been allocated), in
473 * that case, @map is returned as unmapped but we still do fill map->m_len to
474 * indicate the length of a hole starting at map->m_lblk.
476 * It returns the error in case of allocation failure.
478 int ext4_map_blocks(handle_t *handle, struct inode *inode,
479 struct ext4_map_blocks *map, int flags)
481 struct extent_status es;
484 #ifdef ES_AGGRESSIVE_TEST
485 struct ext4_map_blocks orig_map;
487 memcpy(&orig_map, map, sizeof(*map));
491 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
492 flags, map->m_len, (unsigned long) map->m_lblk);
495 * ext4_map_blocks returns an int, and m_len is an unsigned int
497 if (unlikely(map->m_len > INT_MAX))
498 map->m_len = INT_MAX;
500 /* We can handle the block number less than EXT_MAX_BLOCKS */
501 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
502 return -EFSCORRUPTED;
504 /* Lookup extent status tree firstly */
505 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
506 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
507 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
508 map->m_pblk = ext4_es_pblock(&es) +
509 map->m_lblk - es.es_lblk;
510 map->m_flags |= ext4_es_is_written(&es) ?
511 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
512 retval = es.es_len - (map->m_lblk - es.es_lblk);
513 if (retval > map->m_len)
516 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
518 retval = es.es_len - (map->m_lblk - es.es_lblk);
519 if (retval > map->m_len)
527 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
529 #ifdef ES_AGGRESSIVE_TEST
530 ext4_map_blocks_es_recheck(handle, inode, map,
536 * In the query cache no-wait mode, nothing we can do more if we
537 * cannot find extent in the cache.
539 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
543 * Try to see if we can get the block without requesting a new
546 down_read(&EXT4_I(inode)->i_data_sem);
547 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
548 retval = ext4_ext_map_blocks(handle, inode, map, 0);
550 retval = ext4_ind_map_blocks(handle, inode, map, 0);
555 if (unlikely(retval != map->m_len)) {
556 ext4_warning(inode->i_sb,
557 "ES len assertion failed for inode "
558 "%lu: retval %d != map->m_len %d",
559 inode->i_ino, retval, map->m_len);
563 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
564 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
565 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
566 !(status & EXTENT_STATUS_WRITTEN) &&
567 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
568 map->m_lblk + map->m_len - 1))
569 status |= EXTENT_STATUS_DELAYED;
570 ret = ext4_es_insert_extent(inode, map->m_lblk,
571 map->m_len, map->m_pblk, status);
575 up_read((&EXT4_I(inode)->i_data_sem));
578 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
579 ret = check_block_validity(inode, map);
584 /* If it is only a block(s) look up */
585 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
589 * Returns if the blocks have already allocated
591 * Note that if blocks have been preallocated
592 * ext4_ext_get_block() returns the create = 0
593 * with buffer head unmapped.
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
597 * If we need to convert extent to unwritten
598 * we continue and do the actual work in
599 * ext4_ext_map_blocks()
601 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
605 * Here we clear m_flags because after allocating an new extent,
606 * it will be set again.
608 map->m_flags &= ~EXT4_MAP_FLAGS;
611 * New blocks allocate and/or writing to unwritten extent
612 * will possibly result in updating i_data, so we take
613 * the write lock of i_data_sem, and call get_block()
614 * with create == 1 flag.
616 down_write(&EXT4_I(inode)->i_data_sem);
619 * We need to check for EXT4 here because migrate
620 * could have changed the inode type in between
622 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
623 retval = ext4_ext_map_blocks(handle, inode, map, flags);
625 retval = ext4_ind_map_blocks(handle, inode, map, flags);
627 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
629 * We allocated new blocks which will result in
630 * i_data's format changing. Force the migrate
631 * to fail by clearing migrate flags
633 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
637 * Update reserved blocks/metadata blocks after successful
638 * block allocation which had been deferred till now. We don't
639 * support fallocate for non extent files. So we can update
640 * reserve space here.
643 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
644 ext4_da_update_reserve_space(inode, retval, 1);
650 if (unlikely(retval != map->m_len)) {
651 ext4_warning(inode->i_sb,
652 "ES len assertion failed for inode "
653 "%lu: retval %d != map->m_len %d",
654 inode->i_ino, retval, map->m_len);
659 * We have to zeroout blocks before inserting them into extent
660 * status tree. Otherwise someone could look them up there and
661 * use them before they are really zeroed. We also have to
662 * unmap metadata before zeroing as otherwise writeback can
663 * overwrite zeros with stale data from block device.
665 if (flags & EXT4_GET_BLOCKS_ZERO &&
666 map->m_flags & EXT4_MAP_MAPPED &&
667 map->m_flags & EXT4_MAP_NEW) {
668 ret = ext4_issue_zeroout(inode, map->m_lblk,
669 map->m_pblk, map->m_len);
677 * If the extent has been zeroed out, we don't need to update
678 * extent status tree.
680 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
681 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
682 if (ext4_es_is_written(&es))
685 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
686 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
687 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
688 !(status & EXTENT_STATUS_WRITTEN) &&
689 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
690 map->m_lblk + map->m_len - 1))
691 status |= EXTENT_STATUS_DELAYED;
692 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
693 map->m_pblk, status);
701 up_write((&EXT4_I(inode)->i_data_sem));
702 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
703 ret = check_block_validity(inode, map);
708 * Inodes with freshly allocated blocks where contents will be
709 * visible after transaction commit must be on transaction's
712 if (map->m_flags & EXT4_MAP_NEW &&
713 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
714 !(flags & EXT4_GET_BLOCKS_ZERO) &&
715 !ext4_is_quota_file(inode) &&
716 ext4_should_order_data(inode)) {
718 (loff_t)map->m_lblk << inode->i_blkbits;
719 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
721 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
722 ret = ext4_jbd2_inode_add_wait(handle, inode,
725 ret = ext4_jbd2_inode_add_write(handle, inode,
731 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
732 map->m_flags & EXT4_MAP_MAPPED))
733 ext4_fc_track_range(handle, inode, map->m_lblk,
734 map->m_lblk + map->m_len - 1);
736 ext_debug(inode, "failed with err %d\n", retval);
741 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
742 * we have to be careful as someone else may be manipulating b_state as well.
744 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
746 unsigned long old_state;
747 unsigned long new_state;
749 flags &= EXT4_MAP_FLAGS;
751 /* Dummy buffer_head? Set non-atomically. */
753 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
757 * Someone else may be modifying b_state. Be careful! This is ugly but
758 * once we get rid of using bh as a container for mapping information
759 * to pass to / from get_block functions, this can go away.
761 old_state = READ_ONCE(bh->b_state);
763 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
764 } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
767 static int _ext4_get_block(struct inode *inode, sector_t iblock,
768 struct buffer_head *bh, int flags)
770 struct ext4_map_blocks map;
773 if (ext4_has_inline_data(inode))
777 map.m_len = bh->b_size >> inode->i_blkbits;
779 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
782 map_bh(bh, inode->i_sb, map.m_pblk);
783 ext4_update_bh_state(bh, map.m_flags);
784 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
786 } else if (ret == 0) {
787 /* hole case, need to fill in bh->b_size */
788 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
793 int ext4_get_block(struct inode *inode, sector_t iblock,
794 struct buffer_head *bh, int create)
796 return _ext4_get_block(inode, iblock, bh,
797 create ? EXT4_GET_BLOCKS_CREATE : 0);
801 * Get block function used when preparing for buffered write if we require
802 * creating an unwritten extent if blocks haven't been allocated. The extent
803 * will be converted to written after the IO is complete.
805 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
806 struct buffer_head *bh_result, int create)
808 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
809 inode->i_ino, create);
810 return _ext4_get_block(inode, iblock, bh_result,
811 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
814 /* Maximum number of blocks we map for direct IO at once. */
815 #define DIO_MAX_BLOCKS 4096
818 * `handle' can be NULL if create is zero
820 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
821 ext4_lblk_t block, int map_flags)
823 struct ext4_map_blocks map;
824 struct buffer_head *bh;
825 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
826 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
829 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
830 || handle != NULL || create == 0);
831 ASSERT(create == 0 || !nowait);
835 err = ext4_map_blocks(handle, inode, &map, map_flags);
838 return create ? ERR_PTR(-ENOSPC) : NULL;
843 return sb_find_get_block(inode->i_sb, map.m_pblk);
845 bh = sb_getblk(inode->i_sb, map.m_pblk);
847 return ERR_PTR(-ENOMEM);
848 if (map.m_flags & EXT4_MAP_NEW) {
850 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
851 || (handle != NULL));
854 * Now that we do not always journal data, we should
855 * keep in mind whether this should always journal the
856 * new buffer as metadata. For now, regular file
857 * writes use ext4_get_block instead, so it's not a
861 BUFFER_TRACE(bh, "call get_create_access");
862 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
868 if (!buffer_uptodate(bh)) {
869 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
870 set_buffer_uptodate(bh);
873 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
874 err = ext4_handle_dirty_metadata(handle, inode, bh);
878 BUFFER_TRACE(bh, "not a new buffer");
885 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
886 ext4_lblk_t block, int map_flags)
888 struct buffer_head *bh;
891 bh = ext4_getblk(handle, inode, block, map_flags);
894 if (!bh || ext4_buffer_uptodate(bh))
897 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
905 /* Read a contiguous batch of blocks. */
906 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
907 bool wait, struct buffer_head **bhs)
911 for (i = 0; i < bh_count; i++) {
912 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
913 if (IS_ERR(bhs[i])) {
914 err = PTR_ERR(bhs[i]);
920 for (i = 0; i < bh_count; i++)
921 /* Note that NULL bhs[i] is valid because of holes. */
922 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
923 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
928 for (i = 0; i < bh_count; i++)
930 wait_on_buffer(bhs[i]);
932 for (i = 0; i < bh_count; i++) {
933 if (bhs[i] && !buffer_uptodate(bhs[i])) {
941 for (i = 0; i < bh_count; i++) {
948 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
949 struct buffer_head *head,
953 int (*fn)(handle_t *handle, struct inode *inode,
954 struct buffer_head *bh))
956 struct buffer_head *bh;
957 unsigned block_start, block_end;
958 unsigned blocksize = head->b_size;
960 struct buffer_head *next;
962 for (bh = head, block_start = 0;
963 ret == 0 && (bh != head || !block_start);
964 block_start = block_end, bh = next) {
965 next = bh->b_this_page;
966 block_end = block_start + blocksize;
967 if (block_end <= from || block_start >= to) {
968 if (partial && !buffer_uptodate(bh))
972 err = (*fn)(handle, inode, bh);
980 * Helper for handling dirtying of journalled data. We also mark the folio as
981 * dirty so that writeback code knows about this page (and inode) contains
982 * dirty data. ext4_writepages() then commits appropriate transaction to
985 static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
987 folio_mark_dirty(bh->b_folio);
988 return ext4_handle_dirty_metadata(handle, NULL, bh);
991 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
992 struct buffer_head *bh)
994 int dirty = buffer_dirty(bh);
997 if (!buffer_mapped(bh) || buffer_freed(bh))
1000 * __block_write_begin() could have dirtied some buffers. Clean
1001 * the dirty bit as jbd2_journal_get_write_access() could complain
1002 * otherwise about fs integrity issues. Setting of the dirty bit
1003 * by __block_write_begin() isn't a real problem here as we clear
1004 * the bit before releasing a page lock and thus writeback cannot
1005 * ever write the buffer.
1008 clear_buffer_dirty(bh);
1009 BUFFER_TRACE(bh, "get write access");
1010 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1013 ret = ext4_dirty_journalled_data(handle, bh);
1017 #ifdef CONFIG_FS_ENCRYPTION
1018 static int ext4_block_write_begin(struct folio *folio, loff_t pos, unsigned len,
1019 get_block_t *get_block)
1021 unsigned from = pos & (PAGE_SIZE - 1);
1022 unsigned to = from + len;
1023 struct inode *inode = folio->mapping->host;
1024 unsigned block_start, block_end;
1027 unsigned blocksize = inode->i_sb->s_blocksize;
1029 struct buffer_head *bh, *head, *wait[2];
1033 BUG_ON(!folio_test_locked(folio));
1034 BUG_ON(from > PAGE_SIZE);
1035 BUG_ON(to > PAGE_SIZE);
1038 head = folio_buffers(folio);
1040 create_empty_buffers(&folio->page, blocksize, 0);
1041 head = folio_buffers(folio);
1043 bbits = ilog2(blocksize);
1044 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1046 for (bh = head, block_start = 0; bh != head || !block_start;
1047 block++, block_start = block_end, bh = bh->b_this_page) {
1048 block_end = block_start + blocksize;
1049 if (block_end <= from || block_start >= to) {
1050 if (folio_test_uptodate(folio)) {
1051 set_buffer_uptodate(bh);
1056 clear_buffer_new(bh);
1057 if (!buffer_mapped(bh)) {
1058 WARN_ON(bh->b_size != blocksize);
1059 err = get_block(inode, block, bh, 1);
1062 if (buffer_new(bh)) {
1063 if (folio_test_uptodate(folio)) {
1064 clear_buffer_new(bh);
1065 set_buffer_uptodate(bh);
1066 mark_buffer_dirty(bh);
1069 if (block_end > to || block_start < from)
1070 folio_zero_segments(folio, to,
1076 if (folio_test_uptodate(folio)) {
1077 set_buffer_uptodate(bh);
1080 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1081 !buffer_unwritten(bh) &&
1082 (block_start < from || block_end > to)) {
1083 ext4_read_bh_lock(bh, 0, false);
1084 wait[nr_wait++] = bh;
1088 * If we issued read requests, let them complete.
1090 for (i = 0; i < nr_wait; i++) {
1091 wait_on_buffer(wait[i]);
1092 if (!buffer_uptodate(wait[i]))
1095 if (unlikely(err)) {
1096 page_zero_new_buffers(&folio->page, from, to);
1097 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1098 for (i = 0; i < nr_wait; i++) {
1101 err2 = fscrypt_decrypt_pagecache_blocks(folio,
1102 blocksize, bh_offset(wait[i]));
1104 clear_buffer_uptodate(wait[i]);
1115 * To preserve ordering, it is essential that the hole instantiation and
1116 * the data write be encapsulated in a single transaction. We cannot
1117 * close off a transaction and start a new one between the ext4_get_block()
1118 * and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1119 * ext4_write_begin() is the right place.
1121 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1122 loff_t pos, unsigned len,
1123 struct page **pagep, void **fsdata)
1125 struct inode *inode = mapping->host;
1126 int ret, needed_blocks;
1129 struct folio *folio;
1133 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1136 trace_ext4_write_begin(inode, pos, len);
1138 * Reserve one block more for addition to orphan list in case
1139 * we allocate blocks but write fails for some reason
1141 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1142 index = pos >> PAGE_SHIFT;
1143 from = pos & (PAGE_SIZE - 1);
1146 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1147 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1156 * __filemap_get_folio() can take a long time if the
1157 * system is thrashing due to memory pressure, or if the folio
1158 * is being written back. So grab it first before we start
1159 * the transaction handle. This also allows us to allocate
1160 * the folio (if needed) without using GFP_NOFS.
1163 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1164 mapping_gfp_mask(mapping));
1166 return PTR_ERR(folio);
1168 * The same as page allocation, we prealloc buffer heads before
1169 * starting the handle.
1171 if (!folio_buffers(folio))
1172 create_empty_buffers(&folio->page, inode->i_sb->s_blocksize, 0);
1174 folio_unlock(folio);
1177 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1178 if (IS_ERR(handle)) {
1180 return PTR_ERR(handle);
1184 if (folio->mapping != mapping) {
1185 /* The folio got truncated from under us */
1186 folio_unlock(folio);
1188 ext4_journal_stop(handle);
1191 /* In case writeback began while the folio was unlocked */
1192 folio_wait_stable(folio);
1194 #ifdef CONFIG_FS_ENCRYPTION
1195 if (ext4_should_dioread_nolock(inode))
1196 ret = ext4_block_write_begin(folio, pos, len,
1197 ext4_get_block_unwritten);
1199 ret = ext4_block_write_begin(folio, pos, len, ext4_get_block);
1201 if (ext4_should_dioread_nolock(inode))
1202 ret = __block_write_begin(&folio->page, pos, len,
1203 ext4_get_block_unwritten);
1205 ret = __block_write_begin(&folio->page, pos, len, ext4_get_block);
1207 if (!ret && ext4_should_journal_data(inode)) {
1208 ret = ext4_walk_page_buffers(handle, inode,
1209 folio_buffers(folio), from, to,
1210 NULL, do_journal_get_write_access);
1214 bool extended = (pos + len > inode->i_size) &&
1215 !ext4_verity_in_progress(inode);
1217 folio_unlock(folio);
1219 * __block_write_begin may have instantiated a few blocks
1220 * outside i_size. Trim these off again. Don't need
1221 * i_size_read because we hold i_rwsem.
1223 * Add inode to orphan list in case we crash before
1226 if (extended && ext4_can_truncate(inode))
1227 ext4_orphan_add(handle, inode);
1229 ext4_journal_stop(handle);
1231 ext4_truncate_failed_write(inode);
1233 * If truncate failed early the inode might
1234 * still be on the orphan list; we need to
1235 * make sure the inode is removed from the
1236 * orphan list in that case.
1239 ext4_orphan_del(NULL, inode);
1242 if (ret == -ENOSPC &&
1243 ext4_should_retry_alloc(inode->i_sb, &retries))
1248 *pagep = &folio->page;
1252 /* For write_end() in data=journal mode */
1253 static int write_end_fn(handle_t *handle, struct inode *inode,
1254 struct buffer_head *bh)
1257 if (!buffer_mapped(bh) || buffer_freed(bh))
1259 set_buffer_uptodate(bh);
1260 ret = ext4_dirty_journalled_data(handle, bh);
1261 clear_buffer_meta(bh);
1262 clear_buffer_prio(bh);
1267 * We need to pick up the new inode size which generic_commit_write gave us
1268 * `file' can be NULL - eg, when called from page_symlink().
1270 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1271 * buffers are managed internally.
1273 static int ext4_write_end(struct file *file,
1274 struct address_space *mapping,
1275 loff_t pos, unsigned len, unsigned copied,
1276 struct page *page, void *fsdata)
1278 struct folio *folio = page_folio(page);
1279 handle_t *handle = ext4_journal_current_handle();
1280 struct inode *inode = mapping->host;
1281 loff_t old_size = inode->i_size;
1283 int i_size_changed = 0;
1284 bool verity = ext4_verity_in_progress(inode);
1286 trace_ext4_write_end(inode, pos, len, copied);
1288 if (ext4_has_inline_data(inode) &&
1289 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1290 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1292 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1294 * it's important to update i_size while still holding folio lock:
1295 * page writeout could otherwise come in and zero beyond i_size.
1297 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1298 * blocks are being written past EOF, so skip the i_size update.
1301 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1302 folio_unlock(folio);
1305 if (old_size < pos && !verity)
1306 pagecache_isize_extended(inode, old_size, pos);
1308 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1309 * makes the holding time of folio lock longer. Second, it forces lock
1310 * ordering of folio lock and transaction start for journaling
1314 ret = ext4_mark_inode_dirty(handle, inode);
1316 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1317 /* if we have allocated more blocks and copied
1318 * less. We will have blocks allocated outside
1319 * inode->i_size. So truncate them
1321 ext4_orphan_add(handle, inode);
1323 ret2 = ext4_journal_stop(handle);
1327 if (pos + len > inode->i_size && !verity) {
1328 ext4_truncate_failed_write(inode);
1330 * If truncate failed early the inode might still be
1331 * on the orphan list; we need to make sure the inode
1332 * is removed from the orphan list in that case.
1335 ext4_orphan_del(NULL, inode);
1338 return ret ? ret : copied;
1342 * This is a private version of page_zero_new_buffers() which doesn't
1343 * set the buffer to be dirty, since in data=journalled mode we need
1344 * to call ext4_dirty_journalled_data() instead.
1346 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1347 struct inode *inode,
1348 struct folio *folio,
1349 unsigned from, unsigned to)
1351 unsigned int block_start = 0, block_end;
1352 struct buffer_head *head, *bh;
1354 bh = head = folio_buffers(folio);
1356 block_end = block_start + bh->b_size;
1357 if (buffer_new(bh)) {
1358 if (block_end > from && block_start < to) {
1359 if (!folio_test_uptodate(folio)) {
1360 unsigned start, size;
1362 start = max(from, block_start);
1363 size = min(to, block_end) - start;
1365 folio_zero_range(folio, start, size);
1366 write_end_fn(handle, inode, bh);
1368 clear_buffer_new(bh);
1371 block_start = block_end;
1372 bh = bh->b_this_page;
1373 } while (bh != head);
1376 static int ext4_journalled_write_end(struct file *file,
1377 struct address_space *mapping,
1378 loff_t pos, unsigned len, unsigned copied,
1379 struct page *page, void *fsdata)
1381 struct folio *folio = page_folio(page);
1382 handle_t *handle = ext4_journal_current_handle();
1383 struct inode *inode = mapping->host;
1384 loff_t old_size = inode->i_size;
1388 int size_changed = 0;
1389 bool verity = ext4_verity_in_progress(inode);
1391 trace_ext4_journalled_write_end(inode, pos, len, copied);
1392 from = pos & (PAGE_SIZE - 1);
1395 BUG_ON(!ext4_handle_valid(handle));
1397 if (ext4_has_inline_data(inode))
1398 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1400 if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1402 ext4_journalled_zero_new_buffers(handle, inode, folio,
1405 if (unlikely(copied < len))
1406 ext4_journalled_zero_new_buffers(handle, inode, folio,
1408 ret = ext4_walk_page_buffers(handle, inode,
1409 folio_buffers(folio),
1410 from, from + copied, &partial,
1413 folio_mark_uptodate(folio);
1416 size_changed = ext4_update_inode_size(inode, pos + copied);
1417 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1418 folio_unlock(folio);
1421 if (old_size < pos && !verity)
1422 pagecache_isize_extended(inode, old_size, pos);
1425 ret2 = ext4_mark_inode_dirty(handle, inode);
1430 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1431 /* if we have allocated more blocks and copied
1432 * less. We will have blocks allocated outside
1433 * inode->i_size. So truncate them
1435 ext4_orphan_add(handle, inode);
1437 ret2 = ext4_journal_stop(handle);
1440 if (pos + len > inode->i_size && !verity) {
1441 ext4_truncate_failed_write(inode);
1443 * If truncate failed early the inode might still be
1444 * on the orphan list; we need to make sure the inode
1445 * is removed from the orphan list in that case.
1448 ext4_orphan_del(NULL, inode);
1451 return ret ? ret : copied;
1455 * Reserve space for a single cluster
1457 static int ext4_da_reserve_space(struct inode *inode)
1459 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1460 struct ext4_inode_info *ei = EXT4_I(inode);
1464 * We will charge metadata quota at writeout time; this saves
1465 * us from metadata over-estimation, though we may go over by
1466 * a small amount in the end. Here we just reserve for data.
1468 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1472 spin_lock(&ei->i_block_reservation_lock);
1473 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1474 spin_unlock(&ei->i_block_reservation_lock);
1475 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1478 ei->i_reserved_data_blocks++;
1479 trace_ext4_da_reserve_space(inode);
1480 spin_unlock(&ei->i_block_reservation_lock);
1482 return 0; /* success */
1485 void ext4_da_release_space(struct inode *inode, int to_free)
1487 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1488 struct ext4_inode_info *ei = EXT4_I(inode);
1491 return; /* Nothing to release, exit */
1493 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1495 trace_ext4_da_release_space(inode, to_free);
1496 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1498 * if there aren't enough reserved blocks, then the
1499 * counter is messed up somewhere. Since this
1500 * function is called from invalidate page, it's
1501 * harmless to return without any action.
1503 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1504 "ino %lu, to_free %d with only %d reserved "
1505 "data blocks", inode->i_ino, to_free,
1506 ei->i_reserved_data_blocks);
1508 to_free = ei->i_reserved_data_blocks;
1510 ei->i_reserved_data_blocks -= to_free;
1512 /* update fs dirty data blocks counter */
1513 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1515 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1517 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1521 * Delayed allocation stuff
1524 struct mpage_da_data {
1525 /* These are input fields for ext4_do_writepages() */
1526 struct inode *inode;
1527 struct writeback_control *wbc;
1528 unsigned int can_map:1; /* Can writepages call map blocks? */
1530 /* These are internal state of ext4_do_writepages() */
1531 pgoff_t first_page; /* The first page to write */
1532 pgoff_t next_page; /* Current page to examine */
1533 pgoff_t last_page; /* Last page to examine */
1535 * Extent to map - this can be after first_page because that can be
1536 * fully mapped. We somewhat abuse m_flags to store whether the extent
1537 * is delalloc or unwritten.
1539 struct ext4_map_blocks map;
1540 struct ext4_io_submit io_submit; /* IO submission data */
1541 unsigned int do_map:1;
1542 unsigned int scanned_until_end:1;
1543 unsigned int journalled_more_data:1;
1546 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1551 struct folio_batch fbatch;
1552 struct inode *inode = mpd->inode;
1553 struct address_space *mapping = inode->i_mapping;
1555 /* This is necessary when next_page == 0. */
1556 if (mpd->first_page >= mpd->next_page)
1559 mpd->scanned_until_end = 0;
1560 index = mpd->first_page;
1561 end = mpd->next_page - 1;
1563 ext4_lblk_t start, last;
1564 start = index << (PAGE_SHIFT - inode->i_blkbits);
1565 last = end << (PAGE_SHIFT - inode->i_blkbits);
1568 * avoid racing with extent status tree scans made by
1569 * ext4_insert_delayed_block()
1571 down_write(&EXT4_I(inode)->i_data_sem);
1572 ext4_es_remove_extent(inode, start, last - start + 1);
1573 up_write(&EXT4_I(inode)->i_data_sem);
1576 folio_batch_init(&fbatch);
1577 while (index <= end) {
1578 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1581 for (i = 0; i < nr; i++) {
1582 struct folio *folio = fbatch.folios[i];
1584 if (folio->index < mpd->first_page)
1586 if (folio->index + folio_nr_pages(folio) - 1 > end)
1588 BUG_ON(!folio_test_locked(folio));
1589 BUG_ON(folio_test_writeback(folio));
1591 if (folio_mapped(folio))
1592 folio_clear_dirty_for_io(folio);
1593 block_invalidate_folio(folio, 0,
1595 folio_clear_uptodate(folio);
1597 folio_unlock(folio);
1599 folio_batch_release(&fbatch);
1603 static void ext4_print_free_blocks(struct inode *inode)
1605 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1606 struct super_block *sb = inode->i_sb;
1607 struct ext4_inode_info *ei = EXT4_I(inode);
1609 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1610 EXT4_C2B(EXT4_SB(inode->i_sb),
1611 ext4_count_free_clusters(sb)));
1612 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1613 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1614 (long long) EXT4_C2B(EXT4_SB(sb),
1615 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1616 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1617 (long long) EXT4_C2B(EXT4_SB(sb),
1618 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1619 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1620 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1621 ei->i_reserved_data_blocks);
1626 * ext4_insert_delayed_block - adds a delayed block to the extents status
1627 * tree, incrementing the reserved cluster/block
1628 * count or making a pending reservation
1631 * @inode - file containing the newly added block
1632 * @lblk - logical block to be added
1634 * Returns 0 on success, negative error code on failure.
1636 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1638 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1640 bool allocated = false;
1641 bool reserved = false;
1644 * If the cluster containing lblk is shared with a delayed,
1645 * written, or unwritten extent in a bigalloc file system, it's
1646 * already been accounted for and does not need to be reserved.
1647 * A pending reservation must be made for the cluster if it's
1648 * shared with a written or unwritten extent and doesn't already
1649 * have one. Written and unwritten extents can be purged from the
1650 * extents status tree if the system is under memory pressure, so
1651 * it's necessary to examine the extent tree if a search of the
1652 * extents status tree doesn't get a match.
1654 if (sbi->s_cluster_ratio == 1) {
1655 ret = ext4_da_reserve_space(inode);
1656 if (ret != 0) /* ENOSPC */
1659 } else { /* bigalloc */
1660 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1661 if (!ext4_es_scan_clu(inode,
1662 &ext4_es_is_mapped, lblk)) {
1663 ret = ext4_clu_mapped(inode,
1664 EXT4_B2C(sbi, lblk));
1668 ret = ext4_da_reserve_space(inode);
1669 if (ret != 0) /* ENOSPC */
1681 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1682 if (ret && reserved)
1683 ext4_da_release_space(inode, 1);
1690 * This function is grabs code from the very beginning of
1691 * ext4_map_blocks, but assumes that the caller is from delayed write
1692 * time. This function looks up the requested blocks and sets the
1693 * buffer delay bit under the protection of i_data_sem.
1695 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1696 struct ext4_map_blocks *map,
1697 struct buffer_head *bh)
1699 struct extent_status es;
1701 sector_t invalid_block = ~((sector_t) 0xffff);
1702 #ifdef ES_AGGRESSIVE_TEST
1703 struct ext4_map_blocks orig_map;
1705 memcpy(&orig_map, map, sizeof(*map));
1708 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1712 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1713 (unsigned long) map->m_lblk);
1715 /* Lookup extent status tree firstly */
1716 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1717 if (ext4_es_is_hole(&es)) {
1719 down_read(&EXT4_I(inode)->i_data_sem);
1724 * Delayed extent could be allocated by fallocate.
1725 * So we need to check it.
1727 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1728 map_bh(bh, inode->i_sb, invalid_block);
1730 set_buffer_delay(bh);
1734 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1735 retval = es.es_len - (iblock - es.es_lblk);
1736 if (retval > map->m_len)
1737 retval = map->m_len;
1738 map->m_len = retval;
1739 if (ext4_es_is_written(&es))
1740 map->m_flags |= EXT4_MAP_MAPPED;
1741 else if (ext4_es_is_unwritten(&es))
1742 map->m_flags |= EXT4_MAP_UNWRITTEN;
1746 #ifdef ES_AGGRESSIVE_TEST
1747 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1753 * Try to see if we can get the block without requesting a new
1754 * file system block.
1756 down_read(&EXT4_I(inode)->i_data_sem);
1757 if (ext4_has_inline_data(inode))
1759 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1760 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1762 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1769 * XXX: __block_prepare_write() unmaps passed block,
1773 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1779 map_bh(bh, inode->i_sb, invalid_block);
1781 set_buffer_delay(bh);
1782 } else if (retval > 0) {
1784 unsigned int status;
1786 if (unlikely(retval != map->m_len)) {
1787 ext4_warning(inode->i_sb,
1788 "ES len assertion failed for inode "
1789 "%lu: retval %d != map->m_len %d",
1790 inode->i_ino, retval, map->m_len);
1794 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1795 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1796 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1797 map->m_pblk, status);
1803 up_read((&EXT4_I(inode)->i_data_sem));
1809 * This is a special get_block_t callback which is used by
1810 * ext4_da_write_begin(). It will either return mapped block or
1811 * reserve space for a single block.
1813 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1814 * We also have b_blocknr = -1 and b_bdev initialized properly
1816 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1817 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1818 * initialized properly.
1820 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1821 struct buffer_head *bh, int create)
1823 struct ext4_map_blocks map;
1826 BUG_ON(create == 0);
1827 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1829 map.m_lblk = iblock;
1833 * first, we need to know whether the block is allocated already
1834 * preallocated blocks are unmapped but should treated
1835 * the same as allocated blocks.
1837 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1841 map_bh(bh, inode->i_sb, map.m_pblk);
1842 ext4_update_bh_state(bh, map.m_flags);
1844 if (buffer_unwritten(bh)) {
1845 /* A delayed write to unwritten bh should be marked
1846 * new and mapped. Mapped ensures that we don't do
1847 * get_block multiple times when we write to the same
1848 * offset and new ensures that we do proper zero out
1849 * for partial write.
1852 set_buffer_mapped(bh);
1857 static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
1859 mpd->first_page += folio_nr_pages(folio);
1860 folio_unlock(folio);
1863 static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1869 BUG_ON(folio->index != mpd->first_page);
1870 folio_clear_dirty_for_io(folio);
1872 * We have to be very careful here! Nothing protects writeback path
1873 * against i_size changes and the page can be writeably mapped into
1874 * page tables. So an application can be growing i_size and writing
1875 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1876 * write-protects our page in page tables and the page cannot get
1877 * written to again until we release folio lock. So only after
1878 * folio_clear_dirty_for_io() we are safe to sample i_size for
1879 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1880 * on the barrier provided by folio_test_clear_dirty() in
1881 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1882 * after page tables are updated.
1884 size = i_size_read(mpd->inode);
1885 len = folio_size(folio);
1886 if (folio_pos(folio) + len > size &&
1887 !ext4_verity_in_progress(mpd->inode))
1888 len = size & ~PAGE_MASK;
1889 err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
1891 mpd->wbc->nr_to_write--;
1896 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1899 * mballoc gives us at most this number of blocks...
1900 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1901 * The rest of mballoc seems to handle chunks up to full group size.
1903 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1906 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1908 * @mpd - extent of blocks
1909 * @lblk - logical number of the block in the file
1910 * @bh - buffer head we want to add to the extent
1912 * The function is used to collect contig. blocks in the same state. If the
1913 * buffer doesn't require mapping for writeback and we haven't started the
1914 * extent of buffers to map yet, the function returns 'true' immediately - the
1915 * caller can write the buffer right away. Otherwise the function returns true
1916 * if the block has been added to the extent, false if the block couldn't be
1919 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1920 struct buffer_head *bh)
1922 struct ext4_map_blocks *map = &mpd->map;
1924 /* Buffer that doesn't need mapping for writeback? */
1925 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1926 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1927 /* So far no extent to map => we write the buffer right away */
1928 if (map->m_len == 0)
1933 /* First block in the extent? */
1934 if (map->m_len == 0) {
1935 /* We cannot map unless handle is started... */
1940 map->m_flags = bh->b_state & BH_FLAGS;
1944 /* Don't go larger than mballoc is willing to allocate */
1945 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1948 /* Can we merge the block to our big extent? */
1949 if (lblk == map->m_lblk + map->m_len &&
1950 (bh->b_state & BH_FLAGS) == map->m_flags) {
1958 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1960 * @mpd - extent of blocks for mapping
1961 * @head - the first buffer in the page
1962 * @bh - buffer we should start processing from
1963 * @lblk - logical number of the block in the file corresponding to @bh
1965 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1966 * the page for IO if all buffers in this page were mapped and there's no
1967 * accumulated extent of buffers to map or add buffers in the page to the
1968 * extent of buffers to map. The function returns 1 if the caller can continue
1969 * by processing the next page, 0 if it should stop adding buffers to the
1970 * extent to map because we cannot extend it anymore. It can also return value
1971 * < 0 in case of error during IO submission.
1973 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1974 struct buffer_head *head,
1975 struct buffer_head *bh,
1978 struct inode *inode = mpd->inode;
1980 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
1981 >> inode->i_blkbits;
1983 if (ext4_verity_in_progress(inode))
1984 blocks = EXT_MAX_BLOCKS;
1987 BUG_ON(buffer_locked(bh));
1989 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1990 /* Found extent to map? */
1993 /* Buffer needs mapping and handle is not started? */
1996 /* Everything mapped so far and we hit EOF */
1999 } while (lblk++, (bh = bh->b_this_page) != head);
2000 /* So far everything mapped? Submit the page for IO. */
2001 if (mpd->map.m_len == 0) {
2002 err = mpage_submit_folio(mpd, head->b_folio);
2005 mpage_folio_done(mpd, head->b_folio);
2007 if (lblk >= blocks) {
2008 mpd->scanned_until_end = 1;
2015 * mpage_process_folio - update folio buffers corresponding to changed extent
2016 * and may submit fully mapped page for IO
2017 * @mpd: description of extent to map, on return next extent to map
2018 * @folio: Contains these buffers.
2019 * @m_lblk: logical block mapping.
2020 * @m_pblk: corresponding physical mapping.
2021 * @map_bh: determines on return whether this page requires any further
2024 * Scan given folio buffers corresponding to changed extent and update buffer
2025 * state according to new extent state.
2026 * We map delalloc buffers to their physical location, clear unwritten bits.
2027 * If the given folio is not fully mapped, we update @mpd to the next extent in
2028 * the given folio that needs mapping & return @map_bh as true.
2030 static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2031 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2034 struct buffer_head *head, *bh;
2035 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2036 ext4_lblk_t lblk = *m_lblk;
2037 ext4_fsblk_t pblock = *m_pblk;
2039 int blkbits = mpd->inode->i_blkbits;
2040 ssize_t io_end_size = 0;
2041 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2043 bh = head = folio_buffers(folio);
2045 if (lblk < mpd->map.m_lblk)
2047 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2049 * Buffer after end of mapped extent.
2050 * Find next buffer in the folio to map.
2053 mpd->map.m_flags = 0;
2054 io_end_vec->size += io_end_size;
2056 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2059 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2060 io_end_vec = ext4_alloc_io_end_vec(io_end);
2061 if (IS_ERR(io_end_vec)) {
2062 err = PTR_ERR(io_end_vec);
2065 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2070 if (buffer_delay(bh)) {
2071 clear_buffer_delay(bh);
2072 bh->b_blocknr = pblock++;
2074 clear_buffer_unwritten(bh);
2075 io_end_size += (1 << blkbits);
2076 } while (lblk++, (bh = bh->b_this_page) != head);
2078 io_end_vec->size += io_end_size;
2087 * mpage_map_buffers - update buffers corresponding to changed extent and
2088 * submit fully mapped pages for IO
2090 * @mpd - description of extent to map, on return next extent to map
2092 * Scan buffers corresponding to changed extent (we expect corresponding pages
2093 * to be already locked) and update buffer state according to new extent state.
2094 * We map delalloc buffers to their physical location, clear unwritten bits,
2095 * and mark buffers as uninit when we perform writes to unwritten extents
2096 * and do extent conversion after IO is finished. If the last page is not fully
2097 * mapped, we update @map to the next extent in the last page that needs
2098 * mapping. Otherwise we submit the page for IO.
2100 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2102 struct folio_batch fbatch;
2104 struct inode *inode = mpd->inode;
2105 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2108 ext4_fsblk_t pblock;
2110 bool map_bh = false;
2112 start = mpd->map.m_lblk >> bpp_bits;
2113 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2114 lblk = start << bpp_bits;
2115 pblock = mpd->map.m_pblk;
2117 folio_batch_init(&fbatch);
2118 while (start <= end) {
2119 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2122 for (i = 0; i < nr; i++) {
2123 struct folio *folio = fbatch.folios[i];
2125 err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2128 * If map_bh is true, means page may require further bh
2129 * mapping, or maybe the page was submitted for IO.
2130 * So we return to call further extent mapping.
2132 if (err < 0 || map_bh)
2134 /* Page fully mapped - let IO run! */
2135 err = mpage_submit_folio(mpd, folio);
2138 mpage_folio_done(mpd, folio);
2140 folio_batch_release(&fbatch);
2142 /* Extent fully mapped and matches with page boundary. We are done. */
2144 mpd->map.m_flags = 0;
2147 folio_batch_release(&fbatch);
2151 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2153 struct inode *inode = mpd->inode;
2154 struct ext4_map_blocks *map = &mpd->map;
2155 int get_blocks_flags;
2156 int err, dioread_nolock;
2158 trace_ext4_da_write_pages_extent(inode, map);
2160 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2161 * to convert an unwritten extent to be initialized (in the case
2162 * where we have written into one or more preallocated blocks). It is
2163 * possible that we're going to need more metadata blocks than
2164 * previously reserved. However we must not fail because we're in
2165 * writeback and there is nothing we can do about it so it might result
2166 * in data loss. So use reserved blocks to allocate metadata if
2169 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2170 * the blocks in question are delalloc blocks. This indicates
2171 * that the blocks and quotas has already been checked when
2172 * the data was copied into the page cache.
2174 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2175 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2176 EXT4_GET_BLOCKS_IO_SUBMIT;
2177 dioread_nolock = ext4_should_dioread_nolock(inode);
2179 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2180 if (map->m_flags & BIT(BH_Delay))
2181 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2183 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2186 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2187 if (!mpd->io_submit.io_end->handle &&
2188 ext4_handle_valid(handle)) {
2189 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2190 handle->h_rsv_handle = NULL;
2192 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2195 BUG_ON(map->m_len == 0);
2200 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2201 * mpd->len and submit pages underlying it for IO
2203 * @handle - handle for journal operations
2204 * @mpd - extent to map
2205 * @give_up_on_write - we set this to true iff there is a fatal error and there
2206 * is no hope of writing the data. The caller should discard
2207 * dirty pages to avoid infinite loops.
2209 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2210 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2211 * them to initialized or split the described range from larger unwritten
2212 * extent. Note that we need not map all the described range since allocation
2213 * can return less blocks or the range is covered by more unwritten extents. We
2214 * cannot map more because we are limited by reserved transaction credits. On
2215 * the other hand we always make sure that the last touched page is fully
2216 * mapped so that it can be written out (and thus forward progress is
2217 * guaranteed). After mapping we submit all mapped pages for IO.
2219 static int mpage_map_and_submit_extent(handle_t *handle,
2220 struct mpage_da_data *mpd,
2221 bool *give_up_on_write)
2223 struct inode *inode = mpd->inode;
2224 struct ext4_map_blocks *map = &mpd->map;
2228 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2229 struct ext4_io_end_vec *io_end_vec;
2231 io_end_vec = ext4_alloc_io_end_vec(io_end);
2232 if (IS_ERR(io_end_vec))
2233 return PTR_ERR(io_end_vec);
2234 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2236 err = mpage_map_one_extent(handle, mpd);
2238 struct super_block *sb = inode->i_sb;
2240 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2241 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2242 goto invalidate_dirty_pages;
2244 * Let the uper layers retry transient errors.
2245 * In the case of ENOSPC, if ext4_count_free_blocks()
2246 * is non-zero, a commit should free up blocks.
2248 if ((err == -ENOMEM) ||
2249 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2251 goto update_disksize;
2254 ext4_msg(sb, KERN_CRIT,
2255 "Delayed block allocation failed for "
2256 "inode %lu at logical offset %llu with"
2257 " max blocks %u with error %d",
2259 (unsigned long long)map->m_lblk,
2260 (unsigned)map->m_len, -err);
2261 ext4_msg(sb, KERN_CRIT,
2262 "This should not happen!! Data will "
2265 ext4_print_free_blocks(inode);
2266 invalidate_dirty_pages:
2267 *give_up_on_write = true;
2272 * Update buffer state, submit mapped pages, and get us new
2275 err = mpage_map_and_submit_buffers(mpd);
2277 goto update_disksize;
2278 } while (map->m_len);
2282 * Update on-disk size after IO is submitted. Races with
2283 * truncate are avoided by checking i_size under i_data_sem.
2285 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2286 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2290 down_write(&EXT4_I(inode)->i_data_sem);
2291 i_size = i_size_read(inode);
2292 if (disksize > i_size)
2294 if (disksize > EXT4_I(inode)->i_disksize)
2295 EXT4_I(inode)->i_disksize = disksize;
2296 up_write(&EXT4_I(inode)->i_data_sem);
2297 err2 = ext4_mark_inode_dirty(handle, inode);
2299 ext4_error_err(inode->i_sb, -err2,
2300 "Failed to mark inode %lu dirty",
2310 * Calculate the total number of credits to reserve for one writepages
2311 * iteration. This is called from ext4_writepages(). We map an extent of
2312 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2313 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2314 * bpp - 1 blocks in bpp different extents.
2316 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2318 int bpp = ext4_journal_blocks_per_page(inode);
2320 return ext4_meta_trans_blocks(inode,
2321 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2324 static int ext4_journal_page_buffers(handle_t *handle, struct page *page,
2327 struct buffer_head *page_bufs = page_buffers(page);
2328 struct inode *inode = page->mapping->host;
2331 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2332 NULL, do_journal_get_write_access);
2333 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2334 NULL, write_end_fn);
2337 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
2340 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2345 static int mpage_journal_page_buffers(handle_t *handle,
2346 struct mpage_da_data *mpd,
2349 struct inode *inode = mpd->inode;
2350 loff_t size = i_size_read(inode);
2353 ClearPageChecked(page);
2354 mpd->wbc->nr_to_write--;
2356 if (page->index == size >> PAGE_SHIFT &&
2357 !ext4_verity_in_progress(inode))
2358 len = size & ~PAGE_MASK;
2362 return ext4_journal_page_buffers(handle, page, len);
2366 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2367 * needing mapping, submit mapped pages
2369 * @mpd - where to look for pages
2371 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2372 * IO immediately. If we cannot map blocks, we submit just already mapped
2373 * buffers in the page for IO and keep page dirty. When we can map blocks and
2374 * we find a page which isn't mapped we start accumulating extent of buffers
2375 * underlying these pages that needs mapping (formed by either delayed or
2376 * unwritten buffers). We also lock the pages containing these buffers. The
2377 * extent found is returned in @mpd structure (starting at mpd->lblk with
2378 * length mpd->len blocks).
2380 * Note that this function can attach bios to one io_end structure which are
2381 * neither logically nor physically contiguous. Although it may seem as an
2382 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2383 * case as we need to track IO to all buffers underlying a page in one io_end.
2385 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2387 struct address_space *mapping = mpd->inode->i_mapping;
2388 struct folio_batch fbatch;
2389 unsigned int nr_folios;
2390 pgoff_t index = mpd->first_page;
2391 pgoff_t end = mpd->last_page;
2394 int blkbits = mpd->inode->i_blkbits;
2396 struct buffer_head *head;
2397 handle_t *handle = NULL;
2398 int bpp = ext4_journal_blocks_per_page(mpd->inode);
2400 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2401 tag = PAGECACHE_TAG_TOWRITE;
2403 tag = PAGECACHE_TAG_DIRTY;
2406 mpd->next_page = index;
2407 if (ext4_should_journal_data(mpd->inode)) {
2408 handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2411 return PTR_ERR(handle);
2413 folio_batch_init(&fbatch);
2414 while (index <= end) {
2415 nr_folios = filemap_get_folios_tag(mapping, &index, end,
2420 for (i = 0; i < nr_folios; i++) {
2421 struct folio *folio = fbatch.folios[i];
2424 * Accumulated enough dirty pages? This doesn't apply
2425 * to WB_SYNC_ALL mode. For integrity sync we have to
2426 * keep going because someone may be concurrently
2427 * dirtying pages, and we might have synced a lot of
2428 * newly appeared dirty pages, but have not synced all
2429 * of the old dirty pages.
2431 if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2432 mpd->wbc->nr_to_write <=
2433 mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2436 /* If we can't merge this page, we are done. */
2437 if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2441 err = ext4_journal_ensure_credits(handle, bpp,
2449 * If the page is no longer dirty, or its mapping no
2450 * longer corresponds to inode we are writing (which
2451 * means it has been truncated or invalidated), or the
2452 * page is already under writeback and we are not doing
2453 * a data integrity writeback, skip the page
2455 if (!folio_test_dirty(folio) ||
2456 (folio_test_writeback(folio) &&
2457 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2458 unlikely(folio->mapping != mapping)) {
2459 folio_unlock(folio);
2463 folio_wait_writeback(folio);
2464 BUG_ON(folio_test_writeback(folio));
2467 * Should never happen but for buggy code in
2468 * other subsystems that call
2469 * set_page_dirty() without properly warning
2470 * the file system first. See [1] for more
2473 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2475 if (!folio_buffers(folio)) {
2476 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2477 folio_clear_dirty(folio);
2478 folio_unlock(folio);
2482 if (mpd->map.m_len == 0)
2483 mpd->first_page = folio->index;
2484 mpd->next_page = folio->index + folio_nr_pages(folio);
2486 * Writeout when we cannot modify metadata is simple.
2487 * Just submit the page. For data=journal mode we
2488 * first handle writeout of the page for checkpoint and
2489 * only after that handle delayed page dirtying. This
2490 * makes sure current data is checkpointed to the final
2491 * location before possibly journalling it again which
2492 * is desirable when the page is frequently dirtied
2495 if (!mpd->can_map) {
2496 err = mpage_submit_folio(mpd, folio);
2499 /* Pending dirtying of journalled data? */
2500 if (folio_test_checked(folio)) {
2501 err = mpage_journal_page_buffers(handle,
2505 mpd->journalled_more_data = 1;
2507 mpage_folio_done(mpd, folio);
2509 /* Add all dirty buffers to mpd */
2510 lblk = ((ext4_lblk_t)folio->index) <<
2511 (PAGE_SHIFT - blkbits);
2512 head = folio_buffers(folio);
2513 err = mpage_process_page_bufs(mpd, head, head,
2520 folio_batch_release(&fbatch);
2523 mpd->scanned_until_end = 1;
2525 ext4_journal_stop(handle);
2528 folio_batch_release(&fbatch);
2530 ext4_journal_stop(handle);
2534 static int ext4_do_writepages(struct mpage_da_data *mpd)
2536 struct writeback_control *wbc = mpd->wbc;
2537 pgoff_t writeback_index = 0;
2538 long nr_to_write = wbc->nr_to_write;
2539 int range_whole = 0;
2541 handle_t *handle = NULL;
2542 struct inode *inode = mpd->inode;
2543 struct address_space *mapping = inode->i_mapping;
2544 int needed_blocks, rsv_blocks = 0, ret = 0;
2545 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2546 struct blk_plug plug;
2547 bool give_up_on_write = false;
2549 trace_ext4_writepages(inode, wbc);
2552 * No pages to write? This is mainly a kludge to avoid starting
2553 * a transaction for special inodes like journal inode on last iput()
2554 * because that could violate lock ordering on umount
2556 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2557 goto out_writepages;
2560 * If the filesystem has aborted, it is read-only, so return
2561 * right away instead of dumping stack traces later on that
2562 * will obscure the real source of the problem. We test
2563 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2564 * the latter could be true if the filesystem is mounted
2565 * read-only, and in that case, ext4_writepages should
2566 * *never* be called, so if that ever happens, we would want
2569 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2570 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2572 goto out_writepages;
2576 * If we have inline data and arrive here, it means that
2577 * we will soon create the block for the 1st page, so
2578 * we'd better clear the inline data here.
2580 if (ext4_has_inline_data(inode)) {
2581 /* Just inode will be modified... */
2582 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2583 if (IS_ERR(handle)) {
2584 ret = PTR_ERR(handle);
2585 goto out_writepages;
2587 BUG_ON(ext4_test_inode_state(inode,
2588 EXT4_STATE_MAY_INLINE_DATA));
2589 ext4_destroy_inline_data(handle, inode);
2590 ext4_journal_stop(handle);
2594 * data=journal mode does not do delalloc so we just need to writeout /
2595 * journal already mapped buffers. On the other hand we need to commit
2596 * transaction to make data stable. We expect all the data to be
2597 * already in the journal (the only exception are DMA pinned pages
2598 * dirtied behind our back) so we commit transaction here and run the
2599 * writeback loop to checkpoint them. The checkpointing is not actually
2600 * necessary to make data persistent *but* quite a few places (extent
2601 * shifting operations, fsverity, ...) depend on being able to drop
2602 * pagecache pages after calling filemap_write_and_wait() and for that
2603 * checkpointing needs to happen.
2605 if (ext4_should_journal_data(inode)) {
2607 if (wbc->sync_mode == WB_SYNC_ALL)
2608 ext4_fc_commit(sbi->s_journal,
2609 EXT4_I(inode)->i_datasync_tid);
2611 mpd->journalled_more_data = 0;
2613 if (ext4_should_dioread_nolock(inode)) {
2615 * We may need to convert up to one extent per block in
2616 * the page and we may dirty the inode.
2618 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2619 PAGE_SIZE >> inode->i_blkbits);
2622 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2625 if (wbc->range_cyclic) {
2626 writeback_index = mapping->writeback_index;
2627 if (writeback_index)
2629 mpd->first_page = writeback_index;
2630 mpd->last_page = -1;
2632 mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2633 mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2636 ext4_io_submit_init(&mpd->io_submit, wbc);
2638 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2639 tag_pages_for_writeback(mapping, mpd->first_page,
2641 blk_start_plug(&plug);
2644 * First writeback pages that don't need mapping - we can avoid
2645 * starting a transaction unnecessarily and also avoid being blocked
2646 * in the block layer on device congestion while having transaction
2650 mpd->scanned_until_end = 0;
2651 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2652 if (!mpd->io_submit.io_end) {
2656 ret = mpage_prepare_extent_to_map(mpd);
2657 /* Unlock pages we didn't use */
2658 mpage_release_unused_pages(mpd, false);
2659 /* Submit prepared bio */
2660 ext4_io_submit(&mpd->io_submit);
2661 ext4_put_io_end_defer(mpd->io_submit.io_end);
2662 mpd->io_submit.io_end = NULL;
2666 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2667 /* For each extent of pages we use new io_end */
2668 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2669 if (!mpd->io_submit.io_end) {
2674 WARN_ON_ONCE(!mpd->can_map);
2676 * We have two constraints: We find one extent to map and we
2677 * must always write out whole page (makes a difference when
2678 * blocksize < pagesize) so that we don't block on IO when we
2679 * try to write out the rest of the page. Journalled mode is
2680 * not supported by delalloc.
2682 BUG_ON(ext4_should_journal_data(inode));
2683 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2685 /* start a new transaction */
2686 handle = ext4_journal_start_with_reserve(inode,
2687 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2688 if (IS_ERR(handle)) {
2689 ret = PTR_ERR(handle);
2690 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2691 "%ld pages, ino %lu; err %d", __func__,
2692 wbc->nr_to_write, inode->i_ino, ret);
2693 /* Release allocated io_end */
2694 ext4_put_io_end(mpd->io_submit.io_end);
2695 mpd->io_submit.io_end = NULL;
2700 trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2701 ret = mpage_prepare_extent_to_map(mpd);
2702 if (!ret && mpd->map.m_len)
2703 ret = mpage_map_and_submit_extent(handle, mpd,
2706 * Caution: If the handle is synchronous,
2707 * ext4_journal_stop() can wait for transaction commit
2708 * to finish which may depend on writeback of pages to
2709 * complete or on page lock to be released. In that
2710 * case, we have to wait until after we have
2711 * submitted all the IO, released page locks we hold,
2712 * and dropped io_end reference (for extent conversion
2713 * to be able to complete) before stopping the handle.
2715 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2716 ext4_journal_stop(handle);
2720 /* Unlock pages we didn't use */
2721 mpage_release_unused_pages(mpd, give_up_on_write);
2722 /* Submit prepared bio */
2723 ext4_io_submit(&mpd->io_submit);
2726 * Drop our io_end reference we got from init. We have
2727 * to be careful and use deferred io_end finishing if
2728 * we are still holding the transaction as we can
2729 * release the last reference to io_end which may end
2730 * up doing unwritten extent conversion.
2733 ext4_put_io_end_defer(mpd->io_submit.io_end);
2734 ext4_journal_stop(handle);
2736 ext4_put_io_end(mpd->io_submit.io_end);
2737 mpd->io_submit.io_end = NULL;
2739 if (ret == -ENOSPC && sbi->s_journal) {
2741 * Commit the transaction which would
2742 * free blocks released in the transaction
2745 jbd2_journal_force_commit_nested(sbi->s_journal);
2749 /* Fatal error - ENOMEM, EIO... */
2754 blk_finish_plug(&plug);
2755 if (!ret && !cycled && wbc->nr_to_write > 0) {
2757 mpd->last_page = writeback_index - 1;
2758 mpd->first_page = 0;
2763 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2765 * Set the writeback_index so that range_cyclic
2766 * mode will write it back later
2768 mapping->writeback_index = mpd->first_page;
2771 trace_ext4_writepages_result(inode, wbc, ret,
2772 nr_to_write - wbc->nr_to_write);
2776 static int ext4_writepages(struct address_space *mapping,
2777 struct writeback_control *wbc)
2779 struct super_block *sb = mapping->host->i_sb;
2780 struct mpage_da_data mpd = {
2781 .inode = mapping->host,
2787 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
2790 percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
2791 ret = ext4_do_writepages(&mpd);
2793 * For data=journal writeback we could have come across pages marked
2794 * for delayed dirtying (PageChecked) which were just added to the
2795 * running transaction. Try once more to get them to stable storage.
2797 if (!ret && mpd.journalled_more_data)
2798 ret = ext4_do_writepages(&mpd);
2799 percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
2804 int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2806 struct writeback_control wbc = {
2807 .sync_mode = WB_SYNC_ALL,
2808 .nr_to_write = LONG_MAX,
2809 .range_start = jinode->i_dirty_start,
2810 .range_end = jinode->i_dirty_end,
2812 struct mpage_da_data mpd = {
2813 .inode = jinode->i_vfs_inode,
2817 return ext4_do_writepages(&mpd);
2820 static int ext4_dax_writepages(struct address_space *mapping,
2821 struct writeback_control *wbc)
2824 long nr_to_write = wbc->nr_to_write;
2825 struct inode *inode = mapping->host;
2826 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2828 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2831 percpu_down_read(&sbi->s_writepages_rwsem);
2832 trace_ext4_writepages(inode, wbc);
2834 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2835 trace_ext4_writepages_result(inode, wbc, ret,
2836 nr_to_write - wbc->nr_to_write);
2837 percpu_up_read(&sbi->s_writepages_rwsem);
2841 static int ext4_nonda_switch(struct super_block *sb)
2843 s64 free_clusters, dirty_clusters;
2844 struct ext4_sb_info *sbi = EXT4_SB(sb);
2847 * switch to non delalloc mode if we are running low
2848 * on free block. The free block accounting via percpu
2849 * counters can get slightly wrong with percpu_counter_batch getting
2850 * accumulated on each CPU without updating global counters
2851 * Delalloc need an accurate free block accounting. So switch
2852 * to non delalloc when we are near to error range.
2855 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2857 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2859 * Start pushing delalloc when 1/2 of free blocks are dirty.
2861 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2862 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2864 if (2 * free_clusters < 3 * dirty_clusters ||
2865 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2867 * free block count is less than 150% of dirty blocks
2868 * or free blocks is less than watermark
2875 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2876 loff_t pos, unsigned len,
2877 struct page **pagep, void **fsdata)
2879 int ret, retries = 0;
2880 struct folio *folio;
2882 struct inode *inode = mapping->host;
2884 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2887 index = pos >> PAGE_SHIFT;
2889 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2890 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2891 return ext4_write_begin(file, mapping, pos,
2892 len, pagep, fsdata);
2894 *fsdata = (void *)0;
2895 trace_ext4_da_write_begin(inode, pos, len);
2897 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2898 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2907 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2908 mapping_gfp_mask(mapping));
2910 return PTR_ERR(folio);
2912 /* In case writeback began while the folio was unlocked */
2913 folio_wait_stable(folio);
2915 #ifdef CONFIG_FS_ENCRYPTION
2916 ret = ext4_block_write_begin(folio, pos, len, ext4_da_get_block_prep);
2918 ret = __block_write_begin(&folio->page, pos, len, ext4_da_get_block_prep);
2921 folio_unlock(folio);
2924 * block_write_begin may have instantiated a few blocks
2925 * outside i_size. Trim these off again. Don't need
2926 * i_size_read because we hold inode lock.
2928 if (pos + len > inode->i_size)
2929 ext4_truncate_failed_write(inode);
2931 if (ret == -ENOSPC &&
2932 ext4_should_retry_alloc(inode->i_sb, &retries))
2937 *pagep = &folio->page;
2942 * Check if we should update i_disksize
2943 * when write to the end of file but not require block allocation
2945 static int ext4_da_should_update_i_disksize(struct page *page,
2946 unsigned long offset)
2948 struct buffer_head *bh;
2949 struct inode *inode = page->mapping->host;
2953 bh = page_buffers(page);
2954 idx = offset >> inode->i_blkbits;
2956 for (i = 0; i < idx; i++)
2957 bh = bh->b_this_page;
2959 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2964 static int ext4_da_write_end(struct file *file,
2965 struct address_space *mapping,
2966 loff_t pos, unsigned len, unsigned copied,
2967 struct page *page, void *fsdata)
2969 struct inode *inode = mapping->host;
2971 unsigned long start, end;
2972 int write_mode = (int)(unsigned long)fsdata;
2974 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2975 return ext4_write_end(file, mapping, pos,
2976 len, copied, page, fsdata);
2978 trace_ext4_da_write_end(inode, pos, len, copied);
2980 if (write_mode != CONVERT_INLINE_DATA &&
2981 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2982 ext4_has_inline_data(inode))
2983 return ext4_write_inline_data_end(inode, pos, len, copied, page);
2985 if (unlikely(copied < len) && !PageUptodate(page))
2988 start = pos & (PAGE_SIZE - 1);
2989 end = start + copied - 1;
2992 * Since we are holding inode lock, we are sure i_disksize <=
2993 * i_size. We also know that if i_disksize < i_size, there are
2994 * delalloc writes pending in the range upto i_size. If the end of
2995 * the current write is <= i_size, there's no need to touch
2996 * i_disksize since writeback will push i_disksize upto i_size
2997 * eventually. If the end of the current write is > i_size and
2998 * inside an allocated block (ext4_da_should_update_i_disksize()
2999 * check), we need to update i_disksize here as certain
3000 * ext4_writepages() paths not allocating blocks update i_disksize.
3002 * Note that we defer inode dirtying to generic_write_end() /
3003 * ext4_da_write_inline_data_end().
3005 new_i_size = pos + copied;
3006 if (copied && new_i_size > inode->i_size &&
3007 ext4_da_should_update_i_disksize(page, end))
3008 ext4_update_i_disksize(inode, new_i_size);
3010 return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3014 * Force all delayed allocation blocks to be allocated for a given inode.
3016 int ext4_alloc_da_blocks(struct inode *inode)
3018 trace_ext4_alloc_da_blocks(inode);
3020 if (!EXT4_I(inode)->i_reserved_data_blocks)
3024 * We do something simple for now. The filemap_flush() will
3025 * also start triggering a write of the data blocks, which is
3026 * not strictly speaking necessary (and for users of
3027 * laptop_mode, not even desirable). However, to do otherwise
3028 * would require replicating code paths in:
3030 * ext4_writepages() ->
3031 * write_cache_pages() ---> (via passed in callback function)
3032 * __mpage_da_writepage() -->
3033 * mpage_add_bh_to_extent()
3034 * mpage_da_map_blocks()
3036 * The problem is that write_cache_pages(), located in
3037 * mm/page-writeback.c, marks pages clean in preparation for
3038 * doing I/O, which is not desirable if we're not planning on
3041 * We could call write_cache_pages(), and then redirty all of
3042 * the pages by calling redirty_page_for_writepage() but that
3043 * would be ugly in the extreme. So instead we would need to
3044 * replicate parts of the code in the above functions,
3045 * simplifying them because we wouldn't actually intend to
3046 * write out the pages, but rather only collect contiguous
3047 * logical block extents, call the multi-block allocator, and
3048 * then update the buffer heads with the block allocations.
3050 * For now, though, we'll cheat by calling filemap_flush(),
3051 * which will map the blocks, and start the I/O, but not
3052 * actually wait for the I/O to complete.
3054 return filemap_flush(inode->i_mapping);
3058 * bmap() is special. It gets used by applications such as lilo and by
3059 * the swapper to find the on-disk block of a specific piece of data.
3061 * Naturally, this is dangerous if the block concerned is still in the
3062 * journal. If somebody makes a swapfile on an ext4 data-journaling
3063 * filesystem and enables swap, then they may get a nasty shock when the
3064 * data getting swapped to that swapfile suddenly gets overwritten by
3065 * the original zero's written out previously to the journal and
3066 * awaiting writeback in the kernel's buffer cache.
3068 * So, if we see any bmap calls here on a modified, data-journaled file,
3069 * take extra steps to flush any blocks which might be in the cache.
3071 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3073 struct inode *inode = mapping->host;
3076 inode_lock_shared(inode);
3078 * We can get here for an inline file via the FIBMAP ioctl
3080 if (ext4_has_inline_data(inode))
3083 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3084 (test_opt(inode->i_sb, DELALLOC) ||
3085 ext4_should_journal_data(inode))) {
3087 * With delalloc or journalled data we want to sync the file so
3088 * that we can make sure we allocate blocks for file and data
3089 * is in place for the user to see it
3091 filemap_write_and_wait(mapping);
3094 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3097 inode_unlock_shared(inode);
3101 static int ext4_read_folio(struct file *file, struct folio *folio)
3104 struct inode *inode = folio->mapping->host;
3106 trace_ext4_readpage(&folio->page);
3108 if (ext4_has_inline_data(inode))
3109 ret = ext4_readpage_inline(inode, folio);
3112 return ext4_mpage_readpages(inode, NULL, folio);
3117 static void ext4_readahead(struct readahead_control *rac)
3119 struct inode *inode = rac->mapping->host;
3121 /* If the file has inline data, no need to do readahead. */
3122 if (ext4_has_inline_data(inode))
3125 ext4_mpage_readpages(inode, rac, NULL);
3128 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3131 trace_ext4_invalidate_folio(folio, offset, length);
3133 /* No journalling happens on data buffers when this function is used */
3134 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3136 block_invalidate_folio(folio, offset, length);
3139 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3140 size_t offset, size_t length)
3142 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3144 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3147 * If it's a full truncate we just forget about the pending dirtying
3149 if (offset == 0 && length == folio_size(folio))
3150 folio_clear_checked(folio);
3152 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3155 /* Wrapper for aops... */
3156 static void ext4_journalled_invalidate_folio(struct folio *folio,
3160 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3163 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3165 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3167 trace_ext4_releasepage(&folio->page);
3169 /* Page has dirty journalled data -> cannot release */
3170 if (folio_test_checked(folio))
3173 return jbd2_journal_try_to_free_buffers(journal, folio);
3175 return try_to_free_buffers(folio);
3178 static bool ext4_inode_datasync_dirty(struct inode *inode)
3180 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3183 if (jbd2_transaction_committed(journal,
3184 EXT4_I(inode)->i_datasync_tid))
3186 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3187 return !list_empty(&EXT4_I(inode)->i_fc_list);
3191 /* Any metadata buffers to write? */
3192 if (!list_empty(&inode->i_mapping->private_list))
3194 return inode->i_state & I_DIRTY_DATASYNC;
3197 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3198 struct ext4_map_blocks *map, loff_t offset,
3199 loff_t length, unsigned int flags)
3201 u8 blkbits = inode->i_blkbits;
3204 * Writes that span EOF might trigger an I/O size update on completion,
3205 * so consider them to be dirty for the purpose of O_DSYNC, even if
3206 * there is no other metadata changes being made or are pending.
3209 if (ext4_inode_datasync_dirty(inode) ||
3210 offset + length > i_size_read(inode))
3211 iomap->flags |= IOMAP_F_DIRTY;
3213 if (map->m_flags & EXT4_MAP_NEW)
3214 iomap->flags |= IOMAP_F_NEW;
3216 if (flags & IOMAP_DAX)
3217 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3219 iomap->bdev = inode->i_sb->s_bdev;
3220 iomap->offset = (u64) map->m_lblk << blkbits;
3221 iomap->length = (u64) map->m_len << blkbits;
3223 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3224 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3225 iomap->flags |= IOMAP_F_MERGED;
3228 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3229 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3230 * set. In order for any allocated unwritten extents to be converted
3231 * into written extents correctly within the ->end_io() handler, we
3232 * need to ensure that the iomap->type is set appropriately. Hence, the
3233 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3236 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3237 iomap->type = IOMAP_UNWRITTEN;
3238 iomap->addr = (u64) map->m_pblk << blkbits;
3239 if (flags & IOMAP_DAX)
3240 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3241 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3242 iomap->type = IOMAP_MAPPED;
3243 iomap->addr = (u64) map->m_pblk << blkbits;
3244 if (flags & IOMAP_DAX)
3245 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3247 iomap->type = IOMAP_HOLE;
3248 iomap->addr = IOMAP_NULL_ADDR;
3252 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3256 u8 blkbits = inode->i_blkbits;
3257 int ret, dio_credits, m_flags = 0, retries = 0;
3260 * Trim the mapping request to the maximum value that we can map at
3261 * once for direct I/O.
3263 if (map->m_len > DIO_MAX_BLOCKS)
3264 map->m_len = DIO_MAX_BLOCKS;
3265 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3269 * Either we allocate blocks and then don't get an unwritten extent, so
3270 * in that case we have reserved enough credits. Or, the blocks are
3271 * already allocated and unwritten. In that case, the extent conversion
3272 * fits into the credits as well.
3274 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3276 return PTR_ERR(handle);
3279 * DAX and direct I/O are the only two operations that are currently
3280 * supported with IOMAP_WRITE.
3282 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3283 if (flags & IOMAP_DAX)
3284 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3286 * We use i_size instead of i_disksize here because delalloc writeback
3287 * can complete at any point during the I/O and subsequently push the
3288 * i_disksize out to i_size. This could be beyond where direct I/O is
3289 * happening and thus expose allocated blocks to direct I/O reads.
3291 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3292 m_flags = EXT4_GET_BLOCKS_CREATE;
3293 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3294 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3296 ret = ext4_map_blocks(handle, inode, map, m_flags);
3299 * We cannot fill holes in indirect tree based inodes as that could
3300 * expose stale data in the case of a crash. Use the magic error code
3301 * to fallback to buffered I/O.
3303 if (!m_flags && !ret)
3306 ext4_journal_stop(handle);
3307 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3314 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3315 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3318 struct ext4_map_blocks map;
3319 u8 blkbits = inode->i_blkbits;
3321 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3324 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3328 * Calculate the first and last logical blocks respectively.
3330 map.m_lblk = offset >> blkbits;
3331 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3332 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3334 if (flags & IOMAP_WRITE) {
3336 * We check here if the blocks are already allocated, then we
3337 * don't need to start a journal txn and we can directly return
3338 * the mapping information. This could boost performance
3339 * especially in multi-threaded overwrite requests.
3341 if (offset + length <= i_size_read(inode)) {
3342 ret = ext4_map_blocks(NULL, inode, &map, 0);
3343 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3346 ret = ext4_iomap_alloc(inode, &map, flags);
3348 ret = ext4_map_blocks(NULL, inode, &map, 0);
3355 * When inline encryption is enabled, sometimes I/O to an encrypted file
3356 * has to be broken up to guarantee DUN contiguity. Handle this by
3357 * limiting the length of the mapping returned.
3359 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3361 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3366 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3367 loff_t length, unsigned flags, struct iomap *iomap,
3368 struct iomap *srcmap)
3373 * Even for writes we don't need to allocate blocks, so just pretend
3374 * we are reading to save overhead of starting a transaction.
3376 flags &= ~IOMAP_WRITE;
3377 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3378 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3382 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3383 ssize_t written, unsigned flags, struct iomap *iomap)
3386 * Check to see whether an error occurred while writing out the data to
3387 * the allocated blocks. If so, return the magic error code so that we
3388 * fallback to buffered I/O and attempt to complete the remainder of
3389 * the I/O. Any blocks that may have been allocated in preparation for
3390 * the direct I/O will be reused during buffered I/O.
3392 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3398 const struct iomap_ops ext4_iomap_ops = {
3399 .iomap_begin = ext4_iomap_begin,
3400 .iomap_end = ext4_iomap_end,
3403 const struct iomap_ops ext4_iomap_overwrite_ops = {
3404 .iomap_begin = ext4_iomap_overwrite_begin,
3405 .iomap_end = ext4_iomap_end,
3408 static bool ext4_iomap_is_delalloc(struct inode *inode,
3409 struct ext4_map_blocks *map)
3411 struct extent_status es;
3412 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3414 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3415 map->m_lblk, end, &es);
3417 if (!es.es_len || es.es_lblk > end)
3420 if (es.es_lblk > map->m_lblk) {
3421 map->m_len = es.es_lblk - map->m_lblk;
3425 offset = map->m_lblk - es.es_lblk;
3426 map->m_len = es.es_len - offset;
3431 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3432 loff_t length, unsigned int flags,
3433 struct iomap *iomap, struct iomap *srcmap)
3436 bool delalloc = false;
3437 struct ext4_map_blocks map;
3438 u8 blkbits = inode->i_blkbits;
3440 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3443 if (ext4_has_inline_data(inode)) {
3444 ret = ext4_inline_data_iomap(inode, iomap);
3445 if (ret != -EAGAIN) {
3446 if (ret == 0 && offset >= iomap->length)
3453 * Calculate the first and last logical block respectively.
3455 map.m_lblk = offset >> blkbits;
3456 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3457 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3460 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3461 * So handle it here itself instead of querying ext4_map_blocks().
3462 * Since ext4_map_blocks() will warn about it and will return
3465 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3466 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3468 if (offset >= sbi->s_bitmap_maxbytes) {
3474 ret = ext4_map_blocks(NULL, inode, &map, 0);
3478 delalloc = ext4_iomap_is_delalloc(inode, &map);
3481 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3482 if (delalloc && iomap->type == IOMAP_HOLE)
3483 iomap->type = IOMAP_DELALLOC;
3488 const struct iomap_ops ext4_iomap_report_ops = {
3489 .iomap_begin = ext4_iomap_begin_report,
3493 * For data=journal mode, folio should be marked dirty only when it was
3494 * writeably mapped. When that happens, it was already attached to the
3495 * transaction and marked as jbddirty (we take care of this in
3496 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3497 * so we should have nothing to do here, except for the case when someone
3498 * had the page pinned and dirtied the page through this pin (e.g. by doing
3499 * direct IO to it). In that case we'd need to attach buffers here to the
3500 * transaction but we cannot due to lock ordering. We cannot just dirty the
3501 * folio and leave attached buffers clean, because the buffers' dirty state is
3502 * "definitive". We cannot just set the buffers dirty or jbddirty because all
3503 * the journalling code will explode. So what we do is to mark the folio
3504 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3505 * to the transaction appropriately.
3507 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3508 struct folio *folio)
3510 WARN_ON_ONCE(!folio_buffers(folio));
3511 if (folio_maybe_dma_pinned(folio))
3512 folio_set_checked(folio);
3513 return filemap_dirty_folio(mapping, folio);
3516 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3518 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3519 WARN_ON_ONCE(!folio_buffers(folio));
3520 return block_dirty_folio(mapping, folio);
3523 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3524 struct file *file, sector_t *span)
3526 return iomap_swapfile_activate(sis, file, span,
3527 &ext4_iomap_report_ops);
3530 static const struct address_space_operations ext4_aops = {
3531 .read_folio = ext4_read_folio,
3532 .readahead = ext4_readahead,
3533 .writepages = ext4_writepages,
3534 .write_begin = ext4_write_begin,
3535 .write_end = ext4_write_end,
3536 .dirty_folio = ext4_dirty_folio,
3538 .invalidate_folio = ext4_invalidate_folio,
3539 .release_folio = ext4_release_folio,
3540 .direct_IO = noop_direct_IO,
3541 .migrate_folio = buffer_migrate_folio,
3542 .is_partially_uptodate = block_is_partially_uptodate,
3543 .error_remove_page = generic_error_remove_page,
3544 .swap_activate = ext4_iomap_swap_activate,
3547 static const struct address_space_operations ext4_journalled_aops = {
3548 .read_folio = ext4_read_folio,
3549 .readahead = ext4_readahead,
3550 .writepages = ext4_writepages,
3551 .write_begin = ext4_write_begin,
3552 .write_end = ext4_journalled_write_end,
3553 .dirty_folio = ext4_journalled_dirty_folio,
3555 .invalidate_folio = ext4_journalled_invalidate_folio,
3556 .release_folio = ext4_release_folio,
3557 .direct_IO = noop_direct_IO,
3558 .migrate_folio = buffer_migrate_folio_norefs,
3559 .is_partially_uptodate = block_is_partially_uptodate,
3560 .error_remove_page = generic_error_remove_page,
3561 .swap_activate = ext4_iomap_swap_activate,
3564 static const struct address_space_operations ext4_da_aops = {
3565 .read_folio = ext4_read_folio,
3566 .readahead = ext4_readahead,
3567 .writepages = ext4_writepages,
3568 .write_begin = ext4_da_write_begin,
3569 .write_end = ext4_da_write_end,
3570 .dirty_folio = ext4_dirty_folio,
3572 .invalidate_folio = ext4_invalidate_folio,
3573 .release_folio = ext4_release_folio,
3574 .direct_IO = noop_direct_IO,
3575 .migrate_folio = buffer_migrate_folio,
3576 .is_partially_uptodate = block_is_partially_uptodate,
3577 .error_remove_page = generic_error_remove_page,
3578 .swap_activate = ext4_iomap_swap_activate,
3581 static const struct address_space_operations ext4_dax_aops = {
3582 .writepages = ext4_dax_writepages,
3583 .direct_IO = noop_direct_IO,
3584 .dirty_folio = noop_dirty_folio,
3586 .swap_activate = ext4_iomap_swap_activate,
3589 void ext4_set_aops(struct inode *inode)
3591 switch (ext4_inode_journal_mode(inode)) {
3592 case EXT4_INODE_ORDERED_DATA_MODE:
3593 case EXT4_INODE_WRITEBACK_DATA_MODE:
3595 case EXT4_INODE_JOURNAL_DATA_MODE:
3596 inode->i_mapping->a_ops = &ext4_journalled_aops;
3602 inode->i_mapping->a_ops = &ext4_dax_aops;
3603 else if (test_opt(inode->i_sb, DELALLOC))
3604 inode->i_mapping->a_ops = &ext4_da_aops;
3606 inode->i_mapping->a_ops = &ext4_aops;
3609 static int __ext4_block_zero_page_range(handle_t *handle,
3610 struct address_space *mapping, loff_t from, loff_t length)
3612 ext4_fsblk_t index = from >> PAGE_SHIFT;
3613 unsigned offset = from & (PAGE_SIZE-1);
3614 unsigned blocksize, pos;
3616 struct inode *inode = mapping->host;
3617 struct buffer_head *bh;
3618 struct folio *folio;
3621 folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3622 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3623 mapping_gfp_constraint(mapping, ~__GFP_FS));
3625 return PTR_ERR(folio);
3627 blocksize = inode->i_sb->s_blocksize;
3629 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3631 bh = folio_buffers(folio);
3633 create_empty_buffers(&folio->page, blocksize, 0);
3634 bh = folio_buffers(folio);
3637 /* Find the buffer that contains "offset" */
3639 while (offset >= pos) {
3640 bh = bh->b_this_page;
3644 if (buffer_freed(bh)) {
3645 BUFFER_TRACE(bh, "freed: skip");
3648 if (!buffer_mapped(bh)) {
3649 BUFFER_TRACE(bh, "unmapped");
3650 ext4_get_block(inode, iblock, bh, 0);
3651 /* unmapped? It's a hole - nothing to do */
3652 if (!buffer_mapped(bh)) {
3653 BUFFER_TRACE(bh, "still unmapped");
3658 /* Ok, it's mapped. Make sure it's up-to-date */
3659 if (folio_test_uptodate(folio))
3660 set_buffer_uptodate(bh);
3662 if (!buffer_uptodate(bh)) {
3663 err = ext4_read_bh_lock(bh, 0, true);
3666 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3667 /* We expect the key to be set. */
3668 BUG_ON(!fscrypt_has_encryption_key(inode));
3669 err = fscrypt_decrypt_pagecache_blocks(folio,
3673 clear_buffer_uptodate(bh);
3678 if (ext4_should_journal_data(inode)) {
3679 BUFFER_TRACE(bh, "get write access");
3680 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3685 folio_zero_range(folio, offset, length);
3686 BUFFER_TRACE(bh, "zeroed end of block");
3688 if (ext4_should_journal_data(inode)) {
3689 err = ext4_dirty_journalled_data(handle, bh);
3692 mark_buffer_dirty(bh);
3693 if (ext4_should_order_data(inode))
3694 err = ext4_jbd2_inode_add_write(handle, inode, from,
3699 folio_unlock(folio);
3705 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3706 * starting from file offset 'from'. The range to be zero'd must
3707 * be contained with in one block. If the specified range exceeds
3708 * the end of the block it will be shortened to end of the block
3709 * that corresponds to 'from'
3711 static int ext4_block_zero_page_range(handle_t *handle,
3712 struct address_space *mapping, loff_t from, loff_t length)
3714 struct inode *inode = mapping->host;
3715 unsigned offset = from & (PAGE_SIZE-1);
3716 unsigned blocksize = inode->i_sb->s_blocksize;
3717 unsigned max = blocksize - (offset & (blocksize - 1));
3720 * correct length if it does not fall between
3721 * 'from' and the end of the block
3723 if (length > max || length < 0)
3726 if (IS_DAX(inode)) {
3727 return dax_zero_range(inode, from, length, NULL,
3730 return __ext4_block_zero_page_range(handle, mapping, from, length);
3734 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3735 * up to the end of the block which corresponds to `from'.
3736 * This required during truncate. We need to physically zero the tail end
3737 * of that block so it doesn't yield old data if the file is later grown.
3739 static int ext4_block_truncate_page(handle_t *handle,
3740 struct address_space *mapping, loff_t from)
3742 unsigned offset = from & (PAGE_SIZE-1);
3745 struct inode *inode = mapping->host;
3747 /* If we are processing an encrypted inode during orphan list handling */
3748 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3751 blocksize = inode->i_sb->s_blocksize;
3752 length = blocksize - (offset & (blocksize - 1));
3754 return ext4_block_zero_page_range(handle, mapping, from, length);
3757 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3758 loff_t lstart, loff_t length)
3760 struct super_block *sb = inode->i_sb;
3761 struct address_space *mapping = inode->i_mapping;
3762 unsigned partial_start, partial_end;
3763 ext4_fsblk_t start, end;
3764 loff_t byte_end = (lstart + length - 1);
3767 partial_start = lstart & (sb->s_blocksize - 1);
3768 partial_end = byte_end & (sb->s_blocksize - 1);
3770 start = lstart >> sb->s_blocksize_bits;
3771 end = byte_end >> sb->s_blocksize_bits;
3773 /* Handle partial zero within the single block */
3775 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3776 err = ext4_block_zero_page_range(handle, mapping,
3780 /* Handle partial zero out on the start of the range */
3781 if (partial_start) {
3782 err = ext4_block_zero_page_range(handle, mapping,
3783 lstart, sb->s_blocksize);
3787 /* Handle partial zero out on the end of the range */
3788 if (partial_end != sb->s_blocksize - 1)
3789 err = ext4_block_zero_page_range(handle, mapping,
3790 byte_end - partial_end,
3795 int ext4_can_truncate(struct inode *inode)
3797 if (S_ISREG(inode->i_mode))
3799 if (S_ISDIR(inode->i_mode))
3801 if (S_ISLNK(inode->i_mode))
3802 return !ext4_inode_is_fast_symlink(inode);
3807 * We have to make sure i_disksize gets properly updated before we truncate
3808 * page cache due to hole punching or zero range. Otherwise i_disksize update
3809 * can get lost as it may have been postponed to submission of writeback but
3810 * that will never happen after we truncate page cache.
3812 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3818 loff_t size = i_size_read(inode);
3820 WARN_ON(!inode_is_locked(inode));
3821 if (offset > size || offset + len < size)
3824 if (EXT4_I(inode)->i_disksize >= size)
3827 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3829 return PTR_ERR(handle);
3830 ext4_update_i_disksize(inode, size);
3831 ret = ext4_mark_inode_dirty(handle, inode);
3832 ext4_journal_stop(handle);
3837 static void ext4_wait_dax_page(struct inode *inode)
3839 filemap_invalidate_unlock(inode->i_mapping);
3841 filemap_invalidate_lock(inode->i_mapping);
3844 int ext4_break_layouts(struct inode *inode)
3849 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3853 page = dax_layout_busy_page(inode->i_mapping);
3857 error = ___wait_var_event(&page->_refcount,
3858 atomic_read(&page->_refcount) == 1,
3859 TASK_INTERRUPTIBLE, 0, 0,
3860 ext4_wait_dax_page(inode));
3861 } while (error == 0);
3867 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3868 * associated with the given offset and length
3870 * @inode: File inode
3871 * @offset: The offset where the hole will begin
3872 * @len: The length of the hole
3874 * Returns: 0 on success or negative on failure
3877 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3879 struct inode *inode = file_inode(file);
3880 struct super_block *sb = inode->i_sb;
3881 ext4_lblk_t first_block, stop_block;
3882 struct address_space *mapping = inode->i_mapping;
3883 loff_t first_block_offset, last_block_offset, max_length;
3884 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3886 unsigned int credits;
3887 int ret = 0, ret2 = 0;
3889 trace_ext4_punch_hole(inode, offset, length, 0);
3892 * Write out all dirty pages to avoid race conditions
3893 * Then release them.
3895 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3896 ret = filemap_write_and_wait_range(mapping, offset,
3897 offset + length - 1);
3904 /* No need to punch hole beyond i_size */
3905 if (offset >= inode->i_size)
3909 * If the hole extends beyond i_size, set the hole
3910 * to end after the page that contains i_size
3912 if (offset + length > inode->i_size) {
3913 length = inode->i_size +
3914 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3919 * For punch hole the length + offset needs to be within one block
3920 * before last range. Adjust the length if it goes beyond that limit.
3922 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3923 if (offset + length > max_length)
3924 length = max_length - offset;
3926 if (offset & (sb->s_blocksize - 1) ||
3927 (offset + length) & (sb->s_blocksize - 1)) {
3929 * Attach jinode to inode for jbd2 if we do any zeroing of
3932 ret = ext4_inode_attach_jinode(inode);
3938 /* Wait all existing dio workers, newcomers will block on i_rwsem */
3939 inode_dio_wait(inode);
3941 ret = file_modified(file);
3946 * Prevent page faults from reinstantiating pages we have released from
3949 filemap_invalidate_lock(mapping);
3951 ret = ext4_break_layouts(inode);
3955 first_block_offset = round_up(offset, sb->s_blocksize);
3956 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3958 /* Now release the pages and zero block aligned part of pages*/
3959 if (last_block_offset > first_block_offset) {
3960 ret = ext4_update_disksize_before_punch(inode, offset, length);
3963 truncate_pagecache_range(inode, first_block_offset,
3967 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3968 credits = ext4_writepage_trans_blocks(inode);
3970 credits = ext4_blocks_for_truncate(inode);
3971 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3972 if (IS_ERR(handle)) {
3973 ret = PTR_ERR(handle);
3974 ext4_std_error(sb, ret);
3978 ret = ext4_zero_partial_blocks(handle, inode, offset,
3983 first_block = (offset + sb->s_blocksize - 1) >>
3984 EXT4_BLOCK_SIZE_BITS(sb);
3985 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3987 /* If there are blocks to remove, do it */
3988 if (stop_block > first_block) {
3990 down_write(&EXT4_I(inode)->i_data_sem);
3991 ext4_discard_preallocations(inode, 0);
3993 ret = ext4_es_remove_extent(inode, first_block,
3994 stop_block - first_block);
3996 up_write(&EXT4_I(inode)->i_data_sem);
4000 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4001 ret = ext4_ext_remove_space(inode, first_block,
4004 ret = ext4_ind_remove_space(handle, inode, first_block,
4007 up_write(&EXT4_I(inode)->i_data_sem);
4009 ext4_fc_track_range(handle, inode, first_block, stop_block);
4011 ext4_handle_sync(handle);
4013 inode->i_mtime = inode->i_ctime = current_time(inode);
4014 ret2 = ext4_mark_inode_dirty(handle, inode);
4018 ext4_update_inode_fsync_trans(handle, inode, 1);
4020 ext4_journal_stop(handle);
4022 filemap_invalidate_unlock(mapping);
4024 inode_unlock(inode);
4028 int ext4_inode_attach_jinode(struct inode *inode)
4030 struct ext4_inode_info *ei = EXT4_I(inode);
4031 struct jbd2_inode *jinode;
4033 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4036 jinode = jbd2_alloc_inode(GFP_KERNEL);
4037 spin_lock(&inode->i_lock);
4040 spin_unlock(&inode->i_lock);
4043 ei->jinode = jinode;
4044 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4047 spin_unlock(&inode->i_lock);
4048 if (unlikely(jinode != NULL))
4049 jbd2_free_inode(jinode);
4056 * We block out ext4_get_block() block instantiations across the entire
4057 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4058 * simultaneously on behalf of the same inode.
4060 * As we work through the truncate and commit bits of it to the journal there
4061 * is one core, guiding principle: the file's tree must always be consistent on
4062 * disk. We must be able to restart the truncate after a crash.
4064 * The file's tree may be transiently inconsistent in memory (although it
4065 * probably isn't), but whenever we close off and commit a journal transaction,
4066 * the contents of (the filesystem + the journal) must be consistent and
4067 * restartable. It's pretty simple, really: bottom up, right to left (although
4068 * left-to-right works OK too).
4070 * Note that at recovery time, journal replay occurs *before* the restart of
4071 * truncate against the orphan inode list.
4073 * The committed inode has the new, desired i_size (which is the same as
4074 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4075 * that this inode's truncate did not complete and it will again call
4076 * ext4_truncate() to have another go. So there will be instantiated blocks
4077 * to the right of the truncation point in a crashed ext4 filesystem. But
4078 * that's fine - as long as they are linked from the inode, the post-crash
4079 * ext4_truncate() run will find them and release them.
4081 int ext4_truncate(struct inode *inode)
4083 struct ext4_inode_info *ei = EXT4_I(inode);
4084 unsigned int credits;
4087 struct address_space *mapping = inode->i_mapping;
4090 * There is a possibility that we're either freeing the inode
4091 * or it's a completely new inode. In those cases we might not
4092 * have i_rwsem locked because it's not necessary.
4094 if (!(inode->i_state & (I_NEW|I_FREEING)))
4095 WARN_ON(!inode_is_locked(inode));
4096 trace_ext4_truncate_enter(inode);
4098 if (!ext4_can_truncate(inode))
4101 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4102 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4104 if (ext4_has_inline_data(inode)) {
4107 err = ext4_inline_data_truncate(inode, &has_inline);
4108 if (err || has_inline)
4112 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4113 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4114 err = ext4_inode_attach_jinode(inode);
4119 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4120 credits = ext4_writepage_trans_blocks(inode);
4122 credits = ext4_blocks_for_truncate(inode);
4124 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4125 if (IS_ERR(handle)) {
4126 err = PTR_ERR(handle);
4130 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4131 ext4_block_truncate_page(handle, mapping, inode->i_size);
4134 * We add the inode to the orphan list, so that if this
4135 * truncate spans multiple transactions, and we crash, we will
4136 * resume the truncate when the filesystem recovers. It also
4137 * marks the inode dirty, to catch the new size.
4139 * Implication: the file must always be in a sane, consistent
4140 * truncatable state while each transaction commits.
4142 err = ext4_orphan_add(handle, inode);
4146 down_write(&EXT4_I(inode)->i_data_sem);
4148 ext4_discard_preallocations(inode, 0);
4150 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4151 err = ext4_ext_truncate(handle, inode);
4153 ext4_ind_truncate(handle, inode);
4155 up_write(&ei->i_data_sem);
4160 ext4_handle_sync(handle);
4164 * If this was a simple ftruncate() and the file will remain alive,
4165 * then we need to clear up the orphan record which we created above.
4166 * However, if this was a real unlink then we were called by
4167 * ext4_evict_inode(), and we allow that function to clean up the
4168 * orphan info for us.
4171 ext4_orphan_del(handle, inode);
4173 inode->i_mtime = inode->i_ctime = current_time(inode);
4174 err2 = ext4_mark_inode_dirty(handle, inode);
4175 if (unlikely(err2 && !err))
4177 ext4_journal_stop(handle);
4180 trace_ext4_truncate_exit(inode);
4184 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4186 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4187 return inode_peek_iversion_raw(inode);
4189 return inode_peek_iversion(inode);
4192 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4193 struct ext4_inode_info *ei)
4195 struct inode *inode = &(ei->vfs_inode);
4196 u64 i_blocks = READ_ONCE(inode->i_blocks);
4197 struct super_block *sb = inode->i_sb;
4199 if (i_blocks <= ~0U) {
4201 * i_blocks can be represented in a 32 bit variable
4202 * as multiple of 512 bytes
4204 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4205 raw_inode->i_blocks_high = 0;
4206 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4211 * This should never happen since sb->s_maxbytes should not have
4212 * allowed this, sb->s_maxbytes was set according to the huge_file
4213 * feature in ext4_fill_super().
4215 if (!ext4_has_feature_huge_file(sb))
4216 return -EFSCORRUPTED;
4218 if (i_blocks <= 0xffffffffffffULL) {
4220 * i_blocks can be represented in a 48 bit variable
4221 * as multiple of 512 bytes
4223 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4224 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4225 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4227 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4228 /* i_block is stored in file system block size */
4229 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4230 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4231 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4236 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4238 struct ext4_inode_info *ei = EXT4_I(inode);
4245 err = ext4_inode_blocks_set(raw_inode, ei);
4247 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4248 i_uid = i_uid_read(inode);
4249 i_gid = i_gid_read(inode);
4250 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4251 if (!(test_opt(inode->i_sb, NO_UID32))) {
4252 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4253 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4255 * Fix up interoperability with old kernels. Otherwise,
4256 * old inodes get re-used with the upper 16 bits of the
4259 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4260 raw_inode->i_uid_high = 0;
4261 raw_inode->i_gid_high = 0;
4263 raw_inode->i_uid_high =
4264 cpu_to_le16(high_16_bits(i_uid));
4265 raw_inode->i_gid_high =
4266 cpu_to_le16(high_16_bits(i_gid));
4269 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4270 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4271 raw_inode->i_uid_high = 0;
4272 raw_inode->i_gid_high = 0;
4274 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4276 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4277 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4278 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4279 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4281 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4282 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4283 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4284 raw_inode->i_file_acl_high =
4285 cpu_to_le16(ei->i_file_acl >> 32);
4286 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4287 ext4_isize_set(raw_inode, ei->i_disksize);
4289 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4290 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4291 if (old_valid_dev(inode->i_rdev)) {
4292 raw_inode->i_block[0] =
4293 cpu_to_le32(old_encode_dev(inode->i_rdev));
4294 raw_inode->i_block[1] = 0;
4296 raw_inode->i_block[0] = 0;
4297 raw_inode->i_block[1] =
4298 cpu_to_le32(new_encode_dev(inode->i_rdev));
4299 raw_inode->i_block[2] = 0;
4301 } else if (!ext4_has_inline_data(inode)) {
4302 for (block = 0; block < EXT4_N_BLOCKS; block++)
4303 raw_inode->i_block[block] = ei->i_data[block];
4306 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4307 u64 ivers = ext4_inode_peek_iversion(inode);
4309 raw_inode->i_disk_version = cpu_to_le32(ivers);
4310 if (ei->i_extra_isize) {
4311 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4312 raw_inode->i_version_hi =
4313 cpu_to_le32(ivers >> 32);
4314 raw_inode->i_extra_isize =
4315 cpu_to_le16(ei->i_extra_isize);
4319 if (i_projid != EXT4_DEF_PROJID &&
4320 !ext4_has_feature_project(inode->i_sb))
4321 err = err ?: -EFSCORRUPTED;
4323 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4324 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4325 raw_inode->i_projid = cpu_to_le32(i_projid);
4327 ext4_inode_csum_set(inode, raw_inode, ei);
4332 * ext4_get_inode_loc returns with an extra refcount against the inode's
4333 * underlying buffer_head on success. If we pass 'inode' and it does not
4334 * have in-inode xattr, we have all inode data in memory that is needed
4335 * to recreate the on-disk version of this inode.
4337 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4338 struct inode *inode, struct ext4_iloc *iloc,
4339 ext4_fsblk_t *ret_block)
4341 struct ext4_group_desc *gdp;
4342 struct buffer_head *bh;
4344 struct blk_plug plug;
4345 int inodes_per_block, inode_offset;
4348 if (ino < EXT4_ROOT_INO ||
4349 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4350 return -EFSCORRUPTED;
4352 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4353 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4358 * Figure out the offset within the block group inode table
4360 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4361 inode_offset = ((ino - 1) %
4362 EXT4_INODES_PER_GROUP(sb));
4363 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4365 block = ext4_inode_table(sb, gdp);
4366 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4367 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4368 ext4_error(sb, "Invalid inode table block %llu in "
4369 "block_group %u", block, iloc->block_group);
4370 return -EFSCORRUPTED;
4372 block += (inode_offset / inodes_per_block);
4374 bh = sb_getblk(sb, block);
4377 if (ext4_buffer_uptodate(bh))
4381 if (ext4_buffer_uptodate(bh)) {
4382 /* Someone brought it uptodate while we waited */
4388 * If we have all information of the inode in memory and this
4389 * is the only valid inode in the block, we need not read the
4392 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4393 struct buffer_head *bitmap_bh;
4396 start = inode_offset & ~(inodes_per_block - 1);
4398 /* Is the inode bitmap in cache? */
4399 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4400 if (unlikely(!bitmap_bh))
4404 * If the inode bitmap isn't in cache then the
4405 * optimisation may end up performing two reads instead
4406 * of one, so skip it.
4408 if (!buffer_uptodate(bitmap_bh)) {
4412 for (i = start; i < start + inodes_per_block; i++) {
4413 if (i == inode_offset)
4415 if (ext4_test_bit(i, bitmap_bh->b_data))
4419 if (i == start + inodes_per_block) {
4420 struct ext4_inode *raw_inode =
4421 (struct ext4_inode *) (bh->b_data + iloc->offset);
4423 /* all other inodes are free, so skip I/O */
4424 memset(bh->b_data, 0, bh->b_size);
4425 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4426 ext4_fill_raw_inode(inode, raw_inode);
4427 set_buffer_uptodate(bh);
4435 * If we need to do any I/O, try to pre-readahead extra
4436 * blocks from the inode table.
4438 blk_start_plug(&plug);
4439 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4440 ext4_fsblk_t b, end, table;
4442 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4444 table = ext4_inode_table(sb, gdp);
4445 /* s_inode_readahead_blks is always a power of 2 */
4446 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4450 num = EXT4_INODES_PER_GROUP(sb);
4451 if (ext4_has_group_desc_csum(sb))
4452 num -= ext4_itable_unused_count(sb, gdp);
4453 table += num / inodes_per_block;
4457 ext4_sb_breadahead_unmovable(sb, b++);
4461 * There are other valid inodes in the buffer, this inode
4462 * has in-inode xattrs, or we don't have this inode in memory.
4463 * Read the block from disk.
4465 trace_ext4_load_inode(sb, ino);
4466 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4467 blk_finish_plug(&plug);
4469 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4470 if (!buffer_uptodate(bh)) {
4481 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4482 struct ext4_iloc *iloc)
4484 ext4_fsblk_t err_blk = 0;
4487 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4491 ext4_error_inode_block(inode, err_blk, EIO,
4492 "unable to read itable block");
4497 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4499 ext4_fsblk_t err_blk = 0;
4502 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4506 ext4_error_inode_block(inode, err_blk, EIO,
4507 "unable to read itable block");
4513 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4514 struct ext4_iloc *iloc)
4516 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4519 static bool ext4_should_enable_dax(struct inode *inode)
4521 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4523 if (test_opt2(inode->i_sb, DAX_NEVER))
4525 if (!S_ISREG(inode->i_mode))
4527 if (ext4_should_journal_data(inode))
4529 if (ext4_has_inline_data(inode))
4531 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4533 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4535 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4537 if (test_opt(inode->i_sb, DAX_ALWAYS))
4540 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4543 void ext4_set_inode_flags(struct inode *inode, bool init)
4545 unsigned int flags = EXT4_I(inode)->i_flags;
4546 unsigned int new_fl = 0;
4548 WARN_ON_ONCE(IS_DAX(inode) && init);
4550 if (flags & EXT4_SYNC_FL)
4552 if (flags & EXT4_APPEND_FL)
4554 if (flags & EXT4_IMMUTABLE_FL)
4555 new_fl |= S_IMMUTABLE;
4556 if (flags & EXT4_NOATIME_FL)
4557 new_fl |= S_NOATIME;
4558 if (flags & EXT4_DIRSYNC_FL)
4559 new_fl |= S_DIRSYNC;
4561 /* Because of the way inode_set_flags() works we must preserve S_DAX
4562 * here if already set. */
4563 new_fl |= (inode->i_flags & S_DAX);
4564 if (init && ext4_should_enable_dax(inode))
4567 if (flags & EXT4_ENCRYPT_FL)
4568 new_fl |= S_ENCRYPTED;
4569 if (flags & EXT4_CASEFOLD_FL)
4570 new_fl |= S_CASEFOLD;
4571 if (flags & EXT4_VERITY_FL)
4573 inode_set_flags(inode, new_fl,
4574 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4575 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4578 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4579 struct ext4_inode_info *ei)
4582 struct inode *inode = &(ei->vfs_inode);
4583 struct super_block *sb = inode->i_sb;
4585 if (ext4_has_feature_huge_file(sb)) {
4586 /* we are using combined 48 bit field */
4587 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4588 le32_to_cpu(raw_inode->i_blocks_lo);
4589 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4590 /* i_blocks represent file system block size */
4591 return i_blocks << (inode->i_blkbits - 9);
4596 return le32_to_cpu(raw_inode->i_blocks_lo);
4600 static inline int ext4_iget_extra_inode(struct inode *inode,
4601 struct ext4_inode *raw_inode,
4602 struct ext4_inode_info *ei)
4604 __le32 *magic = (void *)raw_inode +
4605 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4607 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4608 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4611 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4612 err = ext4_find_inline_data_nolock(inode);
4613 if (!err && ext4_has_inline_data(inode))
4614 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4617 EXT4_I(inode)->i_inline_off = 0;
4621 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4623 if (!ext4_has_feature_project(inode->i_sb))
4625 *projid = EXT4_I(inode)->i_projid;
4630 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4631 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4634 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4636 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4637 inode_set_iversion_raw(inode, val);
4639 inode_set_iversion_queried(inode, val);
4642 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4643 ext4_iget_flags flags, const char *function,
4646 struct ext4_iloc iloc;
4647 struct ext4_inode *raw_inode;
4648 struct ext4_inode_info *ei;
4649 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4650 struct inode *inode;
4651 journal_t *journal = EXT4_SB(sb)->s_journal;
4659 if ((!(flags & EXT4_IGET_SPECIAL) &&
4660 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4661 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4662 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4663 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4664 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4665 (ino < EXT4_ROOT_INO) ||
4666 (ino > le32_to_cpu(es->s_inodes_count))) {
4667 if (flags & EXT4_IGET_HANDLE)
4668 return ERR_PTR(-ESTALE);
4669 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4670 "inode #%lu: comm %s: iget: illegal inode #",
4671 ino, current->comm);
4672 return ERR_PTR(-EFSCORRUPTED);
4675 inode = iget_locked(sb, ino);
4677 return ERR_PTR(-ENOMEM);
4678 if (!(inode->i_state & I_NEW))
4684 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4687 raw_inode = ext4_raw_inode(&iloc);
4689 if ((flags & EXT4_IGET_HANDLE) &&
4690 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4695 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4696 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4697 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4698 EXT4_INODE_SIZE(inode->i_sb) ||
4699 (ei->i_extra_isize & 3)) {
4700 ext4_error_inode(inode, function, line, 0,
4701 "iget: bad extra_isize %u "
4704 EXT4_INODE_SIZE(inode->i_sb));
4705 ret = -EFSCORRUPTED;
4709 ei->i_extra_isize = 0;
4711 /* Precompute checksum seed for inode metadata */
4712 if (ext4_has_metadata_csum(sb)) {
4713 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4715 __le32 inum = cpu_to_le32(inode->i_ino);
4716 __le32 gen = raw_inode->i_generation;
4717 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4719 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4723 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4724 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4725 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4726 ext4_error_inode_err(inode, function, line, 0,
4727 EFSBADCRC, "iget: checksum invalid");
4732 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4733 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4734 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4735 if (ext4_has_feature_project(sb) &&
4736 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4737 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4738 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4740 i_projid = EXT4_DEF_PROJID;
4742 if (!(test_opt(inode->i_sb, NO_UID32))) {
4743 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4744 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4746 i_uid_write(inode, i_uid);
4747 i_gid_write(inode, i_gid);
4748 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4749 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4751 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4752 ei->i_inline_off = 0;
4753 ei->i_dir_start_lookup = 0;
4754 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4755 /* We now have enough fields to check if the inode was active or not.
4756 * This is needed because nfsd might try to access dead inodes
4757 * the test is that same one that e2fsck uses
4758 * NeilBrown 1999oct15
4760 if (inode->i_nlink == 0) {
4761 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4762 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4763 ino != EXT4_BOOT_LOADER_INO) {
4764 /* this inode is deleted or unallocated */
4765 if (flags & EXT4_IGET_SPECIAL) {
4766 ext4_error_inode(inode, function, line, 0,
4767 "iget: special inode unallocated");
4768 ret = -EFSCORRUPTED;
4773 /* The only unlinked inodes we let through here have
4774 * valid i_mode and are being read by the orphan
4775 * recovery code: that's fine, we're about to complete
4776 * the process of deleting those.
4777 * OR it is the EXT4_BOOT_LOADER_INO which is
4778 * not initialized on a new filesystem. */
4780 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4781 ext4_set_inode_flags(inode, true);
4782 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4783 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4784 if (ext4_has_feature_64bit(sb))
4786 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4787 inode->i_size = ext4_isize(sb, raw_inode);
4788 if ((size = i_size_read(inode)) < 0) {
4789 ext4_error_inode(inode, function, line, 0,
4790 "iget: bad i_size value: %lld", size);
4791 ret = -EFSCORRUPTED;
4795 * If dir_index is not enabled but there's dir with INDEX flag set,
4796 * we'd normally treat htree data as empty space. But with metadata
4797 * checksumming that corrupts checksums so forbid that.
4799 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4800 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4801 ext4_error_inode(inode, function, line, 0,
4802 "iget: Dir with htree data on filesystem without dir_index feature.");
4803 ret = -EFSCORRUPTED;
4806 ei->i_disksize = inode->i_size;
4808 ei->i_reserved_quota = 0;
4810 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4811 ei->i_block_group = iloc.block_group;
4812 ei->i_last_alloc_group = ~0;
4814 * NOTE! The in-memory inode i_data array is in little-endian order
4815 * even on big-endian machines: we do NOT byteswap the block numbers!
4817 for (block = 0; block < EXT4_N_BLOCKS; block++)
4818 ei->i_data[block] = raw_inode->i_block[block];
4819 INIT_LIST_HEAD(&ei->i_orphan);
4820 ext4_fc_init_inode(&ei->vfs_inode);
4823 * Set transaction id's of transactions that have to be committed
4824 * to finish f[data]sync. We set them to currently running transaction
4825 * as we cannot be sure that the inode or some of its metadata isn't
4826 * part of the transaction - the inode could have been reclaimed and
4827 * now it is reread from disk.
4830 transaction_t *transaction;
4833 read_lock(&journal->j_state_lock);
4834 if (journal->j_running_transaction)
4835 transaction = journal->j_running_transaction;
4837 transaction = journal->j_committing_transaction;
4839 tid = transaction->t_tid;
4841 tid = journal->j_commit_sequence;
4842 read_unlock(&journal->j_state_lock);
4843 ei->i_sync_tid = tid;
4844 ei->i_datasync_tid = tid;
4847 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4848 if (ei->i_extra_isize == 0) {
4849 /* The extra space is currently unused. Use it. */
4850 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4851 ei->i_extra_isize = sizeof(struct ext4_inode) -
4852 EXT4_GOOD_OLD_INODE_SIZE;
4854 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4860 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4861 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4862 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4863 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4865 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4866 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4868 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4869 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4871 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4873 ext4_inode_set_iversion_queried(inode, ivers);
4877 if (ei->i_file_acl &&
4878 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4879 ext4_error_inode(inode, function, line, 0,
4880 "iget: bad extended attribute block %llu",
4882 ret = -EFSCORRUPTED;
4884 } else if (!ext4_has_inline_data(inode)) {
4885 /* validate the block references in the inode */
4886 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4887 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4888 (S_ISLNK(inode->i_mode) &&
4889 !ext4_inode_is_fast_symlink(inode)))) {
4890 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4891 ret = ext4_ext_check_inode(inode);
4893 ret = ext4_ind_check_inode(inode);
4899 if (S_ISREG(inode->i_mode)) {
4900 inode->i_op = &ext4_file_inode_operations;
4901 inode->i_fop = &ext4_file_operations;
4902 ext4_set_aops(inode);
4903 } else if (S_ISDIR(inode->i_mode)) {
4904 inode->i_op = &ext4_dir_inode_operations;
4905 inode->i_fop = &ext4_dir_operations;
4906 } else if (S_ISLNK(inode->i_mode)) {
4907 /* VFS does not allow setting these so must be corruption */
4908 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4909 ext4_error_inode(inode, function, line, 0,
4910 "iget: immutable or append flags "
4911 "not allowed on symlinks");
4912 ret = -EFSCORRUPTED;
4915 if (IS_ENCRYPTED(inode)) {
4916 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4917 } else if (ext4_inode_is_fast_symlink(inode)) {
4918 inode->i_link = (char *)ei->i_data;
4919 inode->i_op = &ext4_fast_symlink_inode_operations;
4920 nd_terminate_link(ei->i_data, inode->i_size,
4921 sizeof(ei->i_data) - 1);
4923 inode->i_op = &ext4_symlink_inode_operations;
4925 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4926 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4927 inode->i_op = &ext4_special_inode_operations;
4928 if (raw_inode->i_block[0])
4929 init_special_inode(inode, inode->i_mode,
4930 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4932 init_special_inode(inode, inode->i_mode,
4933 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4934 } else if (ino == EXT4_BOOT_LOADER_INO) {
4935 make_bad_inode(inode);
4937 ret = -EFSCORRUPTED;
4938 ext4_error_inode(inode, function, line, 0,
4939 "iget: bogus i_mode (%o)", inode->i_mode);
4942 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4943 ext4_error_inode(inode, function, line, 0,
4944 "casefold flag without casefold feature");
4945 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
4946 ext4_error_inode(inode, function, line, 0,
4947 "bad inode without EXT4_IGET_BAD flag");
4953 unlock_new_inode(inode);
4959 return ERR_PTR(ret);
4962 static void __ext4_update_other_inode_time(struct super_block *sb,
4963 unsigned long orig_ino,
4965 struct ext4_inode *raw_inode)
4967 struct inode *inode;
4969 inode = find_inode_by_ino_rcu(sb, ino);
4973 if (!inode_is_dirtytime_only(inode))
4976 spin_lock(&inode->i_lock);
4977 if (inode_is_dirtytime_only(inode)) {
4978 struct ext4_inode_info *ei = EXT4_I(inode);
4980 inode->i_state &= ~I_DIRTY_TIME;
4981 spin_unlock(&inode->i_lock);
4983 spin_lock(&ei->i_raw_lock);
4984 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4985 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4986 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4987 ext4_inode_csum_set(inode, raw_inode, ei);
4988 spin_unlock(&ei->i_raw_lock);
4989 trace_ext4_other_inode_update_time(inode, orig_ino);
4992 spin_unlock(&inode->i_lock);
4996 * Opportunistically update the other time fields for other inodes in
4997 * the same inode table block.
4999 static void ext4_update_other_inodes_time(struct super_block *sb,
5000 unsigned long orig_ino, char *buf)
5003 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5004 int inode_size = EXT4_INODE_SIZE(sb);
5007 * Calculate the first inode in the inode table block. Inode
5008 * numbers are one-based. That is, the first inode in a block
5009 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5011 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5013 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5014 if (ino == orig_ino)
5016 __ext4_update_other_inode_time(sb, orig_ino, ino,
5017 (struct ext4_inode *)buf);
5023 * Post the struct inode info into an on-disk inode location in the
5024 * buffer-cache. This gobbles the caller's reference to the
5025 * buffer_head in the inode location struct.
5027 * The caller must have write access to iloc->bh.
5029 static int ext4_do_update_inode(handle_t *handle,
5030 struct inode *inode,
5031 struct ext4_iloc *iloc)
5033 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5034 struct ext4_inode_info *ei = EXT4_I(inode);
5035 struct buffer_head *bh = iloc->bh;
5036 struct super_block *sb = inode->i_sb;
5038 int need_datasync = 0, set_large_file = 0;
5040 spin_lock(&ei->i_raw_lock);
5043 * For fields not tracked in the in-memory inode, initialise them
5044 * to zero for new inodes.
5046 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5047 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5049 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5051 if (ei->i_disksize > 0x7fffffffULL) {
5052 if (!ext4_has_feature_large_file(sb) ||
5053 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5057 err = ext4_fill_raw_inode(inode, raw_inode);
5058 spin_unlock(&ei->i_raw_lock);
5060 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5064 if (inode->i_sb->s_flags & SB_LAZYTIME)
5065 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5068 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5069 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5072 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5073 if (set_large_file) {
5074 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5075 err = ext4_journal_get_write_access(handle, sb,
5080 lock_buffer(EXT4_SB(sb)->s_sbh);
5081 ext4_set_feature_large_file(sb);
5082 ext4_superblock_csum_set(sb);
5083 unlock_buffer(EXT4_SB(sb)->s_sbh);
5084 ext4_handle_sync(handle);
5085 err = ext4_handle_dirty_metadata(handle, NULL,
5086 EXT4_SB(sb)->s_sbh);
5088 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5090 ext4_std_error(inode->i_sb, err);
5097 * ext4_write_inode()
5099 * We are called from a few places:
5101 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5102 * Here, there will be no transaction running. We wait for any running
5103 * transaction to commit.
5105 * - Within flush work (sys_sync(), kupdate and such).
5106 * We wait on commit, if told to.
5108 * - Within iput_final() -> write_inode_now()
5109 * We wait on commit, if told to.
5111 * In all cases it is actually safe for us to return without doing anything,
5112 * because the inode has been copied into a raw inode buffer in
5113 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5116 * Note that we are absolutely dependent upon all inode dirtiers doing the
5117 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5118 * which we are interested.
5120 * It would be a bug for them to not do this. The code:
5122 * mark_inode_dirty(inode)
5124 * inode->i_size = expr;
5126 * is in error because write_inode() could occur while `stuff()' is running,
5127 * and the new i_size will be lost. Plus the inode will no longer be on the
5128 * superblock's dirty inode list.
5130 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5134 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5135 sb_rdonly(inode->i_sb))
5138 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5141 if (EXT4_SB(inode->i_sb)->s_journal) {
5142 if (ext4_journal_current_handle()) {
5143 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5149 * No need to force transaction in WB_SYNC_NONE mode. Also
5150 * ext4_sync_fs() will force the commit after everything is
5153 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5156 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5157 EXT4_I(inode)->i_sync_tid);
5159 struct ext4_iloc iloc;
5161 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5165 * sync(2) will flush the whole buffer cache. No need to do
5166 * it here separately for each inode.
5168 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5169 sync_dirty_buffer(iloc.bh);
5170 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5171 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5172 "IO error syncing inode");
5181 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5182 * buffers that are attached to a folio straddling i_size and are undergoing
5183 * commit. In that case we have to wait for commit to finish and try again.
5185 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5188 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5189 tid_t commit_tid = 0;
5192 offset = inode->i_size & (PAGE_SIZE - 1);
5194 * If the folio is fully truncated, we don't need to wait for any commit
5195 * (and we even should not as __ext4_journalled_invalidate_folio() may
5196 * strip all buffers from the folio but keep the folio dirty which can then
5197 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5198 * buffers). Also we don't need to wait for any commit if all buffers in
5199 * the folio remain valid. This is most beneficial for the common case of
5200 * blocksize == PAGESIZE.
5202 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5205 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5206 inode->i_size >> PAGE_SHIFT);
5209 ret = __ext4_journalled_invalidate_folio(folio, offset,
5210 folio_size(folio) - offset);
5211 folio_unlock(folio);
5216 read_lock(&journal->j_state_lock);
5217 if (journal->j_committing_transaction)
5218 commit_tid = journal->j_committing_transaction->t_tid;
5219 read_unlock(&journal->j_state_lock);
5221 jbd2_log_wait_commit(journal, commit_tid);
5228 * Called from notify_change.
5230 * We want to trap VFS attempts to truncate the file as soon as
5231 * possible. In particular, we want to make sure that when the VFS
5232 * shrinks i_size, we put the inode on the orphan list and modify
5233 * i_disksize immediately, so that during the subsequent flushing of
5234 * dirty pages and freeing of disk blocks, we can guarantee that any
5235 * commit will leave the blocks being flushed in an unused state on
5236 * disk. (On recovery, the inode will get truncated and the blocks will
5237 * be freed, so we have a strong guarantee that no future commit will
5238 * leave these blocks visible to the user.)
5240 * Another thing we have to assure is that if we are in ordered mode
5241 * and inode is still attached to the committing transaction, we must
5242 * we start writeout of all the dirty pages which are being truncated.
5243 * This way we are sure that all the data written in the previous
5244 * transaction are already on disk (truncate waits for pages under
5247 * Called with inode->i_rwsem down.
5249 int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5252 struct inode *inode = d_inode(dentry);
5255 const unsigned int ia_valid = attr->ia_valid;
5256 bool inc_ivers = true;
5258 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5261 if (unlikely(IS_IMMUTABLE(inode)))
5264 if (unlikely(IS_APPEND(inode) &&
5265 (ia_valid & (ATTR_MODE | ATTR_UID |
5266 ATTR_GID | ATTR_TIMES_SET))))
5269 error = setattr_prepare(idmap, dentry, attr);
5273 error = fscrypt_prepare_setattr(dentry, attr);
5277 error = fsverity_prepare_setattr(dentry, attr);
5281 if (is_quota_modification(idmap, inode, attr)) {
5282 error = dquot_initialize(inode);
5287 if (i_uid_needs_update(idmap, attr, inode) ||
5288 i_gid_needs_update(idmap, attr, inode)) {
5291 /* (user+group)*(old+new) structure, inode write (sb,
5292 * inode block, ? - but truncate inode update has it) */
5293 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5294 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5295 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5296 if (IS_ERR(handle)) {
5297 error = PTR_ERR(handle);
5301 /* dquot_transfer() calls back ext4_get_inode_usage() which
5302 * counts xattr inode references.
5304 down_read(&EXT4_I(inode)->xattr_sem);
5305 error = dquot_transfer(idmap, inode, attr);
5306 up_read(&EXT4_I(inode)->xattr_sem);
5309 ext4_journal_stop(handle);
5312 /* Update corresponding info in inode so that everything is in
5313 * one transaction */
5314 i_uid_update(idmap, attr, inode);
5315 i_gid_update(idmap, attr, inode);
5316 error = ext4_mark_inode_dirty(handle, inode);
5317 ext4_journal_stop(handle);
5318 if (unlikely(error)) {
5323 if (attr->ia_valid & ATTR_SIZE) {
5325 loff_t oldsize = inode->i_size;
5326 loff_t old_disksize;
5327 int shrink = (attr->ia_size < inode->i_size);
5329 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5330 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5332 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5336 if (!S_ISREG(inode->i_mode)) {
5340 if (attr->ia_size == inode->i_size)
5344 if (ext4_should_order_data(inode)) {
5345 error = ext4_begin_ordered_truncate(inode,
5351 * Blocks are going to be removed from the inode. Wait
5352 * for dio in flight.
5354 inode_dio_wait(inode);
5357 filemap_invalidate_lock(inode->i_mapping);
5359 rc = ext4_break_layouts(inode);
5361 filemap_invalidate_unlock(inode->i_mapping);
5365 if (attr->ia_size != inode->i_size) {
5366 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5367 if (IS_ERR(handle)) {
5368 error = PTR_ERR(handle);
5371 if (ext4_handle_valid(handle) && shrink) {
5372 error = ext4_orphan_add(handle, inode);
5376 * Update c/mtime on truncate up, ext4_truncate() will
5377 * update c/mtime in shrink case below
5380 inode->i_mtime = current_time(inode);
5381 inode->i_ctime = inode->i_mtime;
5385 ext4_fc_track_range(handle, inode,
5386 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5387 inode->i_sb->s_blocksize_bits,
5388 EXT_MAX_BLOCKS - 1);
5390 ext4_fc_track_range(
5392 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5393 inode->i_sb->s_blocksize_bits,
5394 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5395 inode->i_sb->s_blocksize_bits);
5397 down_write(&EXT4_I(inode)->i_data_sem);
5398 old_disksize = EXT4_I(inode)->i_disksize;
5399 EXT4_I(inode)->i_disksize = attr->ia_size;
5400 rc = ext4_mark_inode_dirty(handle, inode);
5404 * We have to update i_size under i_data_sem together
5405 * with i_disksize to avoid races with writeback code
5406 * running ext4_wb_update_i_disksize().
5409 i_size_write(inode, attr->ia_size);
5411 EXT4_I(inode)->i_disksize = old_disksize;
5412 up_write(&EXT4_I(inode)->i_data_sem);
5413 ext4_journal_stop(handle);
5417 pagecache_isize_extended(inode, oldsize,
5419 } else if (ext4_should_journal_data(inode)) {
5420 ext4_wait_for_tail_page_commit(inode);
5425 * Truncate pagecache after we've waited for commit
5426 * in data=journal mode to make pages freeable.
5428 truncate_pagecache(inode, inode->i_size);
5430 * Call ext4_truncate() even if i_size didn't change to
5431 * truncate possible preallocated blocks.
5433 if (attr->ia_size <= oldsize) {
5434 rc = ext4_truncate(inode);
5439 filemap_invalidate_unlock(inode->i_mapping);
5444 inode_inc_iversion(inode);
5445 setattr_copy(idmap, inode, attr);
5446 mark_inode_dirty(inode);
5450 * If the call to ext4_truncate failed to get a transaction handle at
5451 * all, we need to clean up the in-core orphan list manually.
5453 if (orphan && inode->i_nlink)
5454 ext4_orphan_del(NULL, inode);
5456 if (!error && (ia_valid & ATTR_MODE))
5457 rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5461 ext4_std_error(inode->i_sb, error);
5467 u32 ext4_dio_alignment(struct inode *inode)
5469 if (fsverity_active(inode))
5471 if (ext4_should_journal_data(inode))
5473 if (ext4_has_inline_data(inode))
5475 if (IS_ENCRYPTED(inode)) {
5476 if (!fscrypt_dio_supported(inode))
5478 return i_blocksize(inode);
5480 return 1; /* use the iomap defaults */
5483 int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5484 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5486 struct inode *inode = d_inode(path->dentry);
5487 struct ext4_inode *raw_inode;
5488 struct ext4_inode_info *ei = EXT4_I(inode);
5491 if ((request_mask & STATX_BTIME) &&
5492 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5493 stat->result_mask |= STATX_BTIME;
5494 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5495 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5499 * Return the DIO alignment restrictions if requested. We only return
5500 * this information when requested, since on encrypted files it might
5501 * take a fair bit of work to get if the file wasn't opened recently.
5503 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5504 u32 dio_align = ext4_dio_alignment(inode);
5506 stat->result_mask |= STATX_DIOALIGN;
5507 if (dio_align == 1) {
5508 struct block_device *bdev = inode->i_sb->s_bdev;
5510 /* iomap defaults */
5511 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5512 stat->dio_offset_align = bdev_logical_block_size(bdev);
5514 stat->dio_mem_align = dio_align;
5515 stat->dio_offset_align = dio_align;
5519 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5520 if (flags & EXT4_APPEND_FL)
5521 stat->attributes |= STATX_ATTR_APPEND;
5522 if (flags & EXT4_COMPR_FL)
5523 stat->attributes |= STATX_ATTR_COMPRESSED;
5524 if (flags & EXT4_ENCRYPT_FL)
5525 stat->attributes |= STATX_ATTR_ENCRYPTED;
5526 if (flags & EXT4_IMMUTABLE_FL)
5527 stat->attributes |= STATX_ATTR_IMMUTABLE;
5528 if (flags & EXT4_NODUMP_FL)
5529 stat->attributes |= STATX_ATTR_NODUMP;
5530 if (flags & EXT4_VERITY_FL)
5531 stat->attributes |= STATX_ATTR_VERITY;
5533 stat->attributes_mask |= (STATX_ATTR_APPEND |
5534 STATX_ATTR_COMPRESSED |
5535 STATX_ATTR_ENCRYPTED |
5536 STATX_ATTR_IMMUTABLE |
5540 generic_fillattr(idmap, inode, stat);
5544 int ext4_file_getattr(struct mnt_idmap *idmap,
5545 const struct path *path, struct kstat *stat,
5546 u32 request_mask, unsigned int query_flags)
5548 struct inode *inode = d_inode(path->dentry);
5549 u64 delalloc_blocks;
5551 ext4_getattr(idmap, path, stat, request_mask, query_flags);
5554 * If there is inline data in the inode, the inode will normally not
5555 * have data blocks allocated (it may have an external xattr block).
5556 * Report at least one sector for such files, so tools like tar, rsync,
5557 * others don't incorrectly think the file is completely sparse.
5559 if (unlikely(ext4_has_inline_data(inode)))
5560 stat->blocks += (stat->size + 511) >> 9;
5563 * We can't update i_blocks if the block allocation is delayed
5564 * otherwise in the case of system crash before the real block
5565 * allocation is done, we will have i_blocks inconsistent with
5566 * on-disk file blocks.
5567 * We always keep i_blocks updated together with real
5568 * allocation. But to not confuse with user, stat
5569 * will return the blocks that include the delayed allocation
5570 * blocks for this file.
5572 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5573 EXT4_I(inode)->i_reserved_data_blocks);
5574 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5578 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5581 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5582 return ext4_ind_trans_blocks(inode, lblocks);
5583 return ext4_ext_index_trans_blocks(inode, pextents);
5587 * Account for index blocks, block groups bitmaps and block group
5588 * descriptor blocks if modify datablocks and index blocks
5589 * worse case, the indexs blocks spread over different block groups
5591 * If datablocks are discontiguous, they are possible to spread over
5592 * different block groups too. If they are contiguous, with flexbg,
5593 * they could still across block group boundary.
5595 * Also account for superblock, inode, quota and xattr blocks
5597 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5600 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5606 * How many index blocks need to touch to map @lblocks logical blocks
5607 * to @pextents physical extents?
5609 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5614 * Now let's see how many group bitmaps and group descriptors need
5617 groups = idxblocks + pextents;
5619 if (groups > ngroups)
5621 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5622 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5624 /* bitmaps and block group descriptor blocks */
5625 ret += groups + gdpblocks;
5627 /* Blocks for super block, inode, quota and xattr blocks */
5628 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5634 * Calculate the total number of credits to reserve to fit
5635 * the modification of a single pages into a single transaction,
5636 * which may include multiple chunks of block allocations.
5638 * This could be called via ext4_write_begin()
5640 * We need to consider the worse case, when
5641 * one new block per extent.
5643 int ext4_writepage_trans_blocks(struct inode *inode)
5645 int bpp = ext4_journal_blocks_per_page(inode);
5648 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5650 /* Account for data blocks for journalled mode */
5651 if (ext4_should_journal_data(inode))
5657 * Calculate the journal credits for a chunk of data modification.
5659 * This is called from DIO, fallocate or whoever calling
5660 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5662 * journal buffers for data blocks are not included here, as DIO
5663 * and fallocate do no need to journal data buffers.
5665 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5667 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5671 * The caller must have previously called ext4_reserve_inode_write().
5672 * Give this, we know that the caller already has write access to iloc->bh.
5674 int ext4_mark_iloc_dirty(handle_t *handle,
5675 struct inode *inode, struct ext4_iloc *iloc)
5679 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5683 ext4_fc_track_inode(handle, inode);
5685 /* the do_update_inode consumes one bh->b_count */
5688 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5689 err = ext4_do_update_inode(handle, inode, iloc);
5695 * On success, We end up with an outstanding reference count against
5696 * iloc->bh. This _must_ be cleaned up later.
5700 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5701 struct ext4_iloc *iloc)
5705 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5708 err = ext4_get_inode_loc(inode, iloc);
5710 BUFFER_TRACE(iloc->bh, "get_write_access");
5711 err = ext4_journal_get_write_access(handle, inode->i_sb,
5712 iloc->bh, EXT4_JTR_NONE);
5718 ext4_std_error(inode->i_sb, err);
5722 static int __ext4_expand_extra_isize(struct inode *inode,
5723 unsigned int new_extra_isize,
5724 struct ext4_iloc *iloc,
5725 handle_t *handle, int *no_expand)
5727 struct ext4_inode *raw_inode;
5728 struct ext4_xattr_ibody_header *header;
5729 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5730 struct ext4_inode_info *ei = EXT4_I(inode);
5733 /* this was checked at iget time, but double check for good measure */
5734 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5735 (ei->i_extra_isize & 3)) {
5736 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5738 EXT4_INODE_SIZE(inode->i_sb));
5739 return -EFSCORRUPTED;
5741 if ((new_extra_isize < ei->i_extra_isize) ||
5742 (new_extra_isize < 4) ||
5743 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5744 return -EINVAL; /* Should never happen */
5746 raw_inode = ext4_raw_inode(iloc);
5748 header = IHDR(inode, raw_inode);
5750 /* No extended attributes present */
5751 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5752 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5753 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5754 EXT4_I(inode)->i_extra_isize, 0,
5755 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5756 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5761 * We may need to allocate external xattr block so we need quotas
5762 * initialized. Here we can be called with various locks held so we
5763 * cannot affort to initialize quotas ourselves. So just bail.
5765 if (dquot_initialize_needed(inode))
5768 /* try to expand with EAs present */
5769 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5773 * Inode size expansion failed; don't try again
5782 * Expand an inode by new_extra_isize bytes.
5783 * Returns 0 on success or negative error number on failure.
5785 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5786 unsigned int new_extra_isize,
5787 struct ext4_iloc iloc,
5793 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5797 * In nojournal mode, we can immediately attempt to expand
5798 * the inode. When journaled, we first need to obtain extra
5799 * buffer credits since we may write into the EA block
5800 * with this same handle. If journal_extend fails, then it will
5801 * only result in a minor loss of functionality for that inode.
5802 * If this is felt to be critical, then e2fsck should be run to
5803 * force a large enough s_min_extra_isize.
5805 if (ext4_journal_extend(handle,
5806 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5809 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5812 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5813 handle, &no_expand);
5814 ext4_write_unlock_xattr(inode, &no_expand);
5819 int ext4_expand_extra_isize(struct inode *inode,
5820 unsigned int new_extra_isize,
5821 struct ext4_iloc *iloc)
5827 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5832 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5833 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5834 if (IS_ERR(handle)) {
5835 error = PTR_ERR(handle);
5840 ext4_write_lock_xattr(inode, &no_expand);
5842 BUFFER_TRACE(iloc->bh, "get_write_access");
5843 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5850 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5851 handle, &no_expand);
5853 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5858 ext4_write_unlock_xattr(inode, &no_expand);
5859 ext4_journal_stop(handle);
5864 * What we do here is to mark the in-core inode as clean with respect to inode
5865 * dirtiness (it may still be data-dirty).
5866 * This means that the in-core inode may be reaped by prune_icache
5867 * without having to perform any I/O. This is a very good thing,
5868 * because *any* task may call prune_icache - even ones which
5869 * have a transaction open against a different journal.
5871 * Is this cheating? Not really. Sure, we haven't written the
5872 * inode out, but prune_icache isn't a user-visible syncing function.
5873 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5874 * we start and wait on commits.
5876 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5877 const char *func, unsigned int line)
5879 struct ext4_iloc iloc;
5880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5884 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5885 err = ext4_reserve_inode_write(handle, inode, &iloc);
5889 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5890 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5893 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5896 ext4_error_inode_err(inode, func, line, 0, err,
5897 "mark_inode_dirty error");
5902 * ext4_dirty_inode() is called from __mark_inode_dirty()
5904 * We're really interested in the case where a file is being extended.
5905 * i_size has been changed by generic_commit_write() and we thus need
5906 * to include the updated inode in the current transaction.
5908 * Also, dquot_alloc_block() will always dirty the inode when blocks
5909 * are allocated to the file.
5911 * If the inode is marked synchronous, we don't honour that here - doing
5912 * so would cause a commit on atime updates, which we don't bother doing.
5913 * We handle synchronous inodes at the highest possible level.
5915 void ext4_dirty_inode(struct inode *inode, int flags)
5919 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5922 ext4_mark_inode_dirty(handle, inode);
5923 ext4_journal_stop(handle);
5926 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5931 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5934 * We have to be very careful here: changing a data block's
5935 * journaling status dynamically is dangerous. If we write a
5936 * data block to the journal, change the status and then delete
5937 * that block, we risk forgetting to revoke the old log record
5938 * from the journal and so a subsequent replay can corrupt data.
5939 * So, first we make sure that the journal is empty and that
5940 * nobody is changing anything.
5943 journal = EXT4_JOURNAL(inode);
5946 if (is_journal_aborted(journal))
5949 /* Wait for all existing dio workers */
5950 inode_dio_wait(inode);
5953 * Before flushing the journal and switching inode's aops, we have
5954 * to flush all dirty data the inode has. There can be outstanding
5955 * delayed allocations, there can be unwritten extents created by
5956 * fallocate or buffered writes in dioread_nolock mode covered by
5957 * dirty data which can be converted only after flushing the dirty
5958 * data (and journalled aops don't know how to handle these cases).
5961 filemap_invalidate_lock(inode->i_mapping);
5962 err = filemap_write_and_wait(inode->i_mapping);
5964 filemap_invalidate_unlock(inode->i_mapping);
5969 percpu_down_write(&sbi->s_writepages_rwsem);
5970 jbd2_journal_lock_updates(journal);
5973 * OK, there are no updates running now, and all cached data is
5974 * synced to disk. We are now in a completely consistent state
5975 * which doesn't have anything in the journal, and we know that
5976 * no filesystem updates are running, so it is safe to modify
5977 * the inode's in-core data-journaling state flag now.
5981 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5983 err = jbd2_journal_flush(journal, 0);
5985 jbd2_journal_unlock_updates(journal);
5986 percpu_up_write(&sbi->s_writepages_rwsem);
5989 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5991 ext4_set_aops(inode);
5993 jbd2_journal_unlock_updates(journal);
5994 percpu_up_write(&sbi->s_writepages_rwsem);
5997 filemap_invalidate_unlock(inode->i_mapping);
5999 /* Finally we can mark the inode as dirty. */
6001 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6003 return PTR_ERR(handle);
6005 ext4_fc_mark_ineligible(inode->i_sb,
6006 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6007 err = ext4_mark_inode_dirty(handle, inode);
6008 ext4_handle_sync(handle);
6009 ext4_journal_stop(handle);
6010 ext4_std_error(inode->i_sb, err);
6015 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6016 struct buffer_head *bh)
6018 return !buffer_mapped(bh);
6021 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6023 struct vm_area_struct *vma = vmf->vma;
6024 struct folio *folio = page_folio(vmf->page);
6029 struct file *file = vma->vm_file;
6030 struct inode *inode = file_inode(file);
6031 struct address_space *mapping = inode->i_mapping;
6033 get_block_t *get_block;
6036 if (unlikely(IS_IMMUTABLE(inode)))
6037 return VM_FAULT_SIGBUS;
6039 sb_start_pagefault(inode->i_sb);
6040 file_update_time(vma->vm_file);
6042 filemap_invalidate_lock_shared(mapping);
6044 err = ext4_convert_inline_data(inode);
6049 * On data journalling we skip straight to the transaction handle:
6050 * there's no delalloc; page truncated will be checked later; the
6051 * early return w/ all buffers mapped (calculates size/len) can't
6052 * be used; and there's no dioread_nolock, so only ext4_get_block.
6054 if (ext4_should_journal_data(inode))
6057 /* Delalloc case is easy... */
6058 if (test_opt(inode->i_sb, DELALLOC) &&
6059 !ext4_nonda_switch(inode->i_sb)) {
6061 err = block_page_mkwrite(vma, vmf,
6062 ext4_da_get_block_prep);
6063 } while (err == -ENOSPC &&
6064 ext4_should_retry_alloc(inode->i_sb, &retries));
6069 size = i_size_read(inode);
6070 /* Page got truncated from under us? */
6071 if (folio->mapping != mapping || folio_pos(folio) > size) {
6072 folio_unlock(folio);
6073 ret = VM_FAULT_NOPAGE;
6077 len = folio_size(folio);
6078 if (folio_pos(folio) + len > size)
6079 len = size - folio_pos(folio);
6081 * Return if we have all the buffers mapped. This avoids the need to do
6082 * journal_start/journal_stop which can block and take a long time
6084 * This cannot be done for data journalling, as we have to add the
6085 * inode to the transaction's list to writeprotect pages on commit.
6087 if (folio_buffers(folio)) {
6088 if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6090 ext4_bh_unmapped)) {
6091 /* Wait so that we don't change page under IO */
6092 folio_wait_stable(folio);
6093 ret = VM_FAULT_LOCKED;
6097 folio_unlock(folio);
6098 /* OK, we need to fill the hole... */
6099 if (ext4_should_dioread_nolock(inode))
6100 get_block = ext4_get_block_unwritten;
6102 get_block = ext4_get_block;
6104 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6105 ext4_writepage_trans_blocks(inode));
6106 if (IS_ERR(handle)) {
6107 ret = VM_FAULT_SIGBUS;
6111 * Data journalling can't use block_page_mkwrite() because it
6112 * will set_buffer_dirty() before do_journal_get_write_access()
6113 * thus might hit warning messages for dirty metadata buffers.
6115 if (!ext4_should_journal_data(inode)) {
6116 err = block_page_mkwrite(vma, vmf, get_block);
6119 size = i_size_read(inode);
6120 /* Page got truncated from under us? */
6121 if (folio->mapping != mapping || folio_pos(folio) > size) {
6122 ret = VM_FAULT_NOPAGE;
6126 len = folio_size(folio);
6127 if (folio_pos(folio) + len > size)
6128 len = size - folio_pos(folio);
6130 err = __block_write_begin(&folio->page, 0, len, ext4_get_block);
6132 ret = VM_FAULT_SIGBUS;
6133 if (ext4_journal_page_buffers(handle, &folio->page, len))
6136 folio_unlock(folio);
6139 ext4_journal_stop(handle);
6140 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6143 ret = block_page_mkwrite_return(err);
6145 filemap_invalidate_unlock_shared(mapping);
6146 sb_end_pagefault(inode->i_sb);
6149 folio_unlock(folio);
6150 ext4_journal_stop(handle);