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 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
571 map->m_pblk, status);
573 up_read((&EXT4_I(inode)->i_data_sem));
576 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
577 ret = check_block_validity(inode, map);
582 /* If it is only a block(s) look up */
583 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
587 * Returns if the blocks have already allocated
589 * Note that if blocks have been preallocated
590 * ext4_ext_get_block() returns the create = 0
591 * with buffer head unmapped.
593 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
595 * If we need to convert extent to unwritten
596 * we continue and do the actual work in
597 * ext4_ext_map_blocks()
599 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
603 * Here we clear m_flags because after allocating an new extent,
604 * it will be set again.
606 map->m_flags &= ~EXT4_MAP_FLAGS;
609 * New blocks allocate and/or writing to unwritten extent
610 * will possibly result in updating i_data, so we take
611 * the write lock of i_data_sem, and call get_block()
612 * with create == 1 flag.
614 down_write(&EXT4_I(inode)->i_data_sem);
617 * We need to check for EXT4 here because migrate
618 * could have changed the inode type in between
620 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
621 retval = ext4_ext_map_blocks(handle, inode, map, flags);
623 retval = ext4_ind_map_blocks(handle, inode, map, flags);
625 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
627 * We allocated new blocks which will result in
628 * i_data's format changing. Force the migrate
629 * to fail by clearing migrate flags
631 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
638 if (unlikely(retval != map->m_len)) {
639 ext4_warning(inode->i_sb,
640 "ES len assertion failed for inode "
641 "%lu: retval %d != map->m_len %d",
642 inode->i_ino, retval, map->m_len);
647 * We have to zeroout blocks before inserting them into extent
648 * status tree. Otherwise someone could look them up there and
649 * use them before they are really zeroed. We also have to
650 * unmap metadata before zeroing as otherwise writeback can
651 * overwrite zeros with stale data from block device.
653 if (flags & EXT4_GET_BLOCKS_ZERO &&
654 map->m_flags & EXT4_MAP_MAPPED &&
655 map->m_flags & EXT4_MAP_NEW) {
656 ret = ext4_issue_zeroout(inode, map->m_lblk,
657 map->m_pblk, map->m_len);
665 * If the extent has been zeroed out, we don't need to update
666 * extent status tree.
668 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
669 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
670 if (ext4_es_is_written(&es))
673 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
674 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
675 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
676 !(status & EXTENT_STATUS_WRITTEN) &&
677 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
678 map->m_lblk + map->m_len - 1))
679 status |= EXTENT_STATUS_DELAYED;
680 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
681 map->m_pblk, status);
685 up_write((&EXT4_I(inode)->i_data_sem));
686 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
687 ret = check_block_validity(inode, map);
692 * Inodes with freshly allocated blocks where contents will be
693 * visible after transaction commit must be on transaction's
696 if (map->m_flags & EXT4_MAP_NEW &&
697 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
698 !(flags & EXT4_GET_BLOCKS_ZERO) &&
699 !ext4_is_quota_file(inode) &&
700 ext4_should_order_data(inode)) {
702 (loff_t)map->m_lblk << inode->i_blkbits;
703 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
705 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
706 ret = ext4_jbd2_inode_add_wait(handle, inode,
709 ret = ext4_jbd2_inode_add_write(handle, inode,
715 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
716 map->m_flags & EXT4_MAP_MAPPED))
717 ext4_fc_track_range(handle, inode, map->m_lblk,
718 map->m_lblk + map->m_len - 1);
720 ext_debug(inode, "failed with err %d\n", retval);
725 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
726 * we have to be careful as someone else may be manipulating b_state as well.
728 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
730 unsigned long old_state;
731 unsigned long new_state;
733 flags &= EXT4_MAP_FLAGS;
735 /* Dummy buffer_head? Set non-atomically. */
737 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
741 * Someone else may be modifying b_state. Be careful! This is ugly but
742 * once we get rid of using bh as a container for mapping information
743 * to pass to / from get_block functions, this can go away.
745 old_state = READ_ONCE(bh->b_state);
747 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
748 } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
751 static int _ext4_get_block(struct inode *inode, sector_t iblock,
752 struct buffer_head *bh, int flags)
754 struct ext4_map_blocks map;
757 if (ext4_has_inline_data(inode))
761 map.m_len = bh->b_size >> inode->i_blkbits;
763 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
766 map_bh(bh, inode->i_sb, map.m_pblk);
767 ext4_update_bh_state(bh, map.m_flags);
768 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
770 } else if (ret == 0) {
771 /* hole case, need to fill in bh->b_size */
772 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
777 int ext4_get_block(struct inode *inode, sector_t iblock,
778 struct buffer_head *bh, int create)
780 return _ext4_get_block(inode, iblock, bh,
781 create ? EXT4_GET_BLOCKS_CREATE : 0);
785 * Get block function used when preparing for buffered write if we require
786 * creating an unwritten extent if blocks haven't been allocated. The extent
787 * will be converted to written after the IO is complete.
789 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
790 struct buffer_head *bh_result, int create)
794 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
795 inode->i_ino, create);
796 ret = _ext4_get_block(inode, iblock, bh_result,
797 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
800 * If the buffer is marked unwritten, mark it as new to make sure it is
801 * zeroed out correctly in case of partial writes. Otherwise, there is
802 * a chance of stale data getting exposed.
804 if (ret == 0 && buffer_unwritten(bh_result))
805 set_buffer_new(bh_result);
810 /* Maximum number of blocks we map for direct IO at once. */
811 #define DIO_MAX_BLOCKS 4096
814 * `handle' can be NULL if create is zero
816 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
817 ext4_lblk_t block, int map_flags)
819 struct ext4_map_blocks map;
820 struct buffer_head *bh;
821 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
822 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
825 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
826 || handle != NULL || create == 0);
827 ASSERT(create == 0 || !nowait);
831 err = ext4_map_blocks(handle, inode, &map, map_flags);
834 return create ? ERR_PTR(-ENOSPC) : NULL;
839 return sb_find_get_block(inode->i_sb, map.m_pblk);
841 bh = sb_getblk(inode->i_sb, map.m_pblk);
843 return ERR_PTR(-ENOMEM);
844 if (map.m_flags & EXT4_MAP_NEW) {
846 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
847 || (handle != NULL));
850 * Now that we do not always journal data, we should
851 * keep in mind whether this should always journal the
852 * new buffer as metadata. For now, regular file
853 * writes use ext4_get_block instead, so it's not a
857 BUFFER_TRACE(bh, "call get_create_access");
858 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
864 if (!buffer_uptodate(bh)) {
865 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
866 set_buffer_uptodate(bh);
869 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
870 err = ext4_handle_dirty_metadata(handle, inode, bh);
874 BUFFER_TRACE(bh, "not a new buffer");
881 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
882 ext4_lblk_t block, int map_flags)
884 struct buffer_head *bh;
887 bh = ext4_getblk(handle, inode, block, map_flags);
890 if (!bh || ext4_buffer_uptodate(bh))
893 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
901 /* Read a contiguous batch of blocks. */
902 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
903 bool wait, struct buffer_head **bhs)
907 for (i = 0; i < bh_count; i++) {
908 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
909 if (IS_ERR(bhs[i])) {
910 err = PTR_ERR(bhs[i]);
916 for (i = 0; i < bh_count; i++)
917 /* Note that NULL bhs[i] is valid because of holes. */
918 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
919 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
924 for (i = 0; i < bh_count; i++)
926 wait_on_buffer(bhs[i]);
928 for (i = 0; i < bh_count; i++) {
929 if (bhs[i] && !buffer_uptodate(bhs[i])) {
937 for (i = 0; i < bh_count; i++) {
944 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
945 struct buffer_head *head,
949 int (*fn)(handle_t *handle, struct inode *inode,
950 struct buffer_head *bh))
952 struct buffer_head *bh;
953 unsigned block_start, block_end;
954 unsigned blocksize = head->b_size;
956 struct buffer_head *next;
958 for (bh = head, block_start = 0;
959 ret == 0 && (bh != head || !block_start);
960 block_start = block_end, bh = next) {
961 next = bh->b_this_page;
962 block_end = block_start + blocksize;
963 if (block_end <= from || block_start >= to) {
964 if (partial && !buffer_uptodate(bh))
968 err = (*fn)(handle, inode, bh);
976 * Helper for handling dirtying of journalled data. We also mark the folio as
977 * dirty so that writeback code knows about this page (and inode) contains
978 * dirty data. ext4_writepages() then commits appropriate transaction to
981 static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
983 folio_mark_dirty(bh->b_folio);
984 return ext4_handle_dirty_metadata(handle, NULL, bh);
987 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
988 struct buffer_head *bh)
990 int dirty = buffer_dirty(bh);
993 if (!buffer_mapped(bh) || buffer_freed(bh))
996 * __block_write_begin() could have dirtied some buffers. Clean
997 * the dirty bit as jbd2_journal_get_write_access() could complain
998 * otherwise about fs integrity issues. Setting of the dirty bit
999 * by __block_write_begin() isn't a real problem here as we clear
1000 * the bit before releasing a page lock and thus writeback cannot
1001 * ever write the buffer.
1004 clear_buffer_dirty(bh);
1005 BUFFER_TRACE(bh, "get write access");
1006 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1009 ret = ext4_dirty_journalled_data(handle, bh);
1013 #ifdef CONFIG_FS_ENCRYPTION
1014 static int ext4_block_write_begin(struct folio *folio, loff_t pos, unsigned len,
1015 get_block_t *get_block)
1017 unsigned from = pos & (PAGE_SIZE - 1);
1018 unsigned to = from + len;
1019 struct inode *inode = folio->mapping->host;
1020 unsigned block_start, block_end;
1023 unsigned blocksize = inode->i_sb->s_blocksize;
1025 struct buffer_head *bh, *head, *wait[2];
1029 BUG_ON(!folio_test_locked(folio));
1030 BUG_ON(from > PAGE_SIZE);
1031 BUG_ON(to > PAGE_SIZE);
1034 head = folio_buffers(folio);
1036 create_empty_buffers(&folio->page, blocksize, 0);
1037 head = folio_buffers(folio);
1039 bbits = ilog2(blocksize);
1040 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1042 for (bh = head, block_start = 0; bh != head || !block_start;
1043 block++, block_start = block_end, bh = bh->b_this_page) {
1044 block_end = block_start + blocksize;
1045 if (block_end <= from || block_start >= to) {
1046 if (folio_test_uptodate(folio)) {
1047 set_buffer_uptodate(bh);
1052 clear_buffer_new(bh);
1053 if (!buffer_mapped(bh)) {
1054 WARN_ON(bh->b_size != blocksize);
1055 err = get_block(inode, block, bh, 1);
1058 if (buffer_new(bh)) {
1059 if (folio_test_uptodate(folio)) {
1060 clear_buffer_new(bh);
1061 set_buffer_uptodate(bh);
1062 mark_buffer_dirty(bh);
1065 if (block_end > to || block_start < from)
1066 folio_zero_segments(folio, to,
1072 if (folio_test_uptodate(folio)) {
1073 set_buffer_uptodate(bh);
1076 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1077 !buffer_unwritten(bh) &&
1078 (block_start < from || block_end > to)) {
1079 ext4_read_bh_lock(bh, 0, false);
1080 wait[nr_wait++] = bh;
1084 * If we issued read requests, let them complete.
1086 for (i = 0; i < nr_wait; i++) {
1087 wait_on_buffer(wait[i]);
1088 if (!buffer_uptodate(wait[i]))
1091 if (unlikely(err)) {
1092 folio_zero_new_buffers(folio, from, to);
1093 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1094 for (i = 0; i < nr_wait; i++) {
1097 err2 = fscrypt_decrypt_pagecache_blocks(folio,
1098 blocksize, bh_offset(wait[i]));
1100 clear_buffer_uptodate(wait[i]);
1111 * To preserve ordering, it is essential that the hole instantiation and
1112 * the data write be encapsulated in a single transaction. We cannot
1113 * close off a transaction and start a new one between the ext4_get_block()
1114 * and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1115 * ext4_write_begin() is the right place.
1117 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1118 loff_t pos, unsigned len,
1119 struct page **pagep, void **fsdata)
1121 struct inode *inode = mapping->host;
1122 int ret, needed_blocks;
1125 struct folio *folio;
1129 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
1132 trace_ext4_write_begin(inode, pos, len);
1134 * Reserve one block more for addition to orphan list in case
1135 * we allocate blocks but write fails for some reason
1137 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1138 index = pos >> PAGE_SHIFT;
1139 from = pos & (PAGE_SIZE - 1);
1142 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1143 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1152 * __filemap_get_folio() can take a long time if the
1153 * system is thrashing due to memory pressure, or if the folio
1154 * is being written back. So grab it first before we start
1155 * the transaction handle. This also allows us to allocate
1156 * the folio (if needed) without using GFP_NOFS.
1159 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1160 mapping_gfp_mask(mapping));
1162 return PTR_ERR(folio);
1164 * The same as page allocation, we prealloc buffer heads before
1165 * starting the handle.
1167 if (!folio_buffers(folio))
1168 create_empty_buffers(&folio->page, inode->i_sb->s_blocksize, 0);
1170 folio_unlock(folio);
1173 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1174 if (IS_ERR(handle)) {
1176 return PTR_ERR(handle);
1180 if (folio->mapping != mapping) {
1181 /* The folio got truncated from under us */
1182 folio_unlock(folio);
1184 ext4_journal_stop(handle);
1187 /* In case writeback began while the folio was unlocked */
1188 folio_wait_stable(folio);
1190 #ifdef CONFIG_FS_ENCRYPTION
1191 if (ext4_should_dioread_nolock(inode))
1192 ret = ext4_block_write_begin(folio, pos, len,
1193 ext4_get_block_unwritten);
1195 ret = ext4_block_write_begin(folio, pos, len, ext4_get_block);
1197 if (ext4_should_dioread_nolock(inode))
1198 ret = __block_write_begin(&folio->page, pos, len,
1199 ext4_get_block_unwritten);
1201 ret = __block_write_begin(&folio->page, pos, len, ext4_get_block);
1203 if (!ret && ext4_should_journal_data(inode)) {
1204 ret = ext4_walk_page_buffers(handle, inode,
1205 folio_buffers(folio), from, to,
1206 NULL, do_journal_get_write_access);
1210 bool extended = (pos + len > inode->i_size) &&
1211 !ext4_verity_in_progress(inode);
1213 folio_unlock(folio);
1215 * __block_write_begin may have instantiated a few blocks
1216 * outside i_size. Trim these off again. Don't need
1217 * i_size_read because we hold i_rwsem.
1219 * Add inode to orphan list in case we crash before
1222 if (extended && ext4_can_truncate(inode))
1223 ext4_orphan_add(handle, inode);
1225 ext4_journal_stop(handle);
1227 ext4_truncate_failed_write(inode);
1229 * If truncate failed early the inode might
1230 * still be on the orphan list; we need to
1231 * make sure the inode is removed from the
1232 * orphan list in that case.
1235 ext4_orphan_del(NULL, inode);
1238 if (ret == -ENOSPC &&
1239 ext4_should_retry_alloc(inode->i_sb, &retries))
1244 *pagep = &folio->page;
1248 /* For write_end() in data=journal mode */
1249 static int write_end_fn(handle_t *handle, struct inode *inode,
1250 struct buffer_head *bh)
1253 if (!buffer_mapped(bh) || buffer_freed(bh))
1255 set_buffer_uptodate(bh);
1256 ret = ext4_dirty_journalled_data(handle, bh);
1257 clear_buffer_meta(bh);
1258 clear_buffer_prio(bh);
1263 * We need to pick up the new inode size which generic_commit_write gave us
1264 * `file' can be NULL - eg, when called from page_symlink().
1266 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1267 * buffers are managed internally.
1269 static int ext4_write_end(struct file *file,
1270 struct address_space *mapping,
1271 loff_t pos, unsigned len, unsigned copied,
1272 struct page *page, void *fsdata)
1274 struct folio *folio = page_folio(page);
1275 handle_t *handle = ext4_journal_current_handle();
1276 struct inode *inode = mapping->host;
1277 loff_t old_size = inode->i_size;
1279 int i_size_changed = 0;
1280 bool verity = ext4_verity_in_progress(inode);
1282 trace_ext4_write_end(inode, pos, len, copied);
1284 if (ext4_has_inline_data(inode) &&
1285 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1286 return ext4_write_inline_data_end(inode, pos, len, copied,
1289 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1291 * it's important to update i_size while still holding folio lock:
1292 * page writeout could otherwise come in and zero beyond i_size.
1294 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1295 * blocks are being written past EOF, so skip the i_size update.
1298 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1299 folio_unlock(folio);
1302 if (old_size < pos && !verity)
1303 pagecache_isize_extended(inode, old_size, pos);
1305 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1306 * makes the holding time of folio lock longer. Second, it forces lock
1307 * ordering of folio lock and transaction start for journaling
1311 ret = ext4_mark_inode_dirty(handle, inode);
1313 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1314 /* if we have allocated more blocks and copied
1315 * less. We will have blocks allocated outside
1316 * inode->i_size. So truncate them
1318 ext4_orphan_add(handle, inode);
1320 ret2 = ext4_journal_stop(handle);
1324 if (pos + len > inode->i_size && !verity) {
1325 ext4_truncate_failed_write(inode);
1327 * If truncate failed early the inode might still be
1328 * on the orphan list; we need to make sure the inode
1329 * is removed from the orphan list in that case.
1332 ext4_orphan_del(NULL, inode);
1335 return ret ? ret : copied;
1339 * This is a private version of folio_zero_new_buffers() which doesn't
1340 * set the buffer to be dirty, since in data=journalled mode we need
1341 * to call ext4_dirty_journalled_data() instead.
1343 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1344 struct inode *inode,
1345 struct folio *folio,
1346 unsigned from, unsigned to)
1348 unsigned int block_start = 0, block_end;
1349 struct buffer_head *head, *bh;
1351 bh = head = folio_buffers(folio);
1353 block_end = block_start + bh->b_size;
1354 if (buffer_new(bh)) {
1355 if (block_end > from && block_start < to) {
1356 if (!folio_test_uptodate(folio)) {
1357 unsigned start, size;
1359 start = max(from, block_start);
1360 size = min(to, block_end) - start;
1362 folio_zero_range(folio, start, size);
1363 write_end_fn(handle, inode, bh);
1365 clear_buffer_new(bh);
1368 block_start = block_end;
1369 bh = bh->b_this_page;
1370 } while (bh != head);
1373 static int ext4_journalled_write_end(struct file *file,
1374 struct address_space *mapping,
1375 loff_t pos, unsigned len, unsigned copied,
1376 struct page *page, void *fsdata)
1378 struct folio *folio = page_folio(page);
1379 handle_t *handle = ext4_journal_current_handle();
1380 struct inode *inode = mapping->host;
1381 loff_t old_size = inode->i_size;
1385 int size_changed = 0;
1386 bool verity = ext4_verity_in_progress(inode);
1388 trace_ext4_journalled_write_end(inode, pos, len, copied);
1389 from = pos & (PAGE_SIZE - 1);
1392 BUG_ON(!ext4_handle_valid(handle));
1394 if (ext4_has_inline_data(inode))
1395 return ext4_write_inline_data_end(inode, pos, len, copied,
1398 if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1400 ext4_journalled_zero_new_buffers(handle, inode, folio,
1403 if (unlikely(copied < len))
1404 ext4_journalled_zero_new_buffers(handle, inode, folio,
1406 ret = ext4_walk_page_buffers(handle, inode,
1407 folio_buffers(folio),
1408 from, from + copied, &partial,
1411 folio_mark_uptodate(folio);
1414 size_changed = ext4_update_inode_size(inode, pos + copied);
1415 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1416 folio_unlock(folio);
1419 if (old_size < pos && !verity)
1420 pagecache_isize_extended(inode, old_size, pos);
1423 ret2 = ext4_mark_inode_dirty(handle, inode);
1428 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1429 /* if we have allocated more blocks and copied
1430 * less. We will have blocks allocated outside
1431 * inode->i_size. So truncate them
1433 ext4_orphan_add(handle, inode);
1435 ret2 = ext4_journal_stop(handle);
1438 if (pos + len > inode->i_size && !verity) {
1439 ext4_truncate_failed_write(inode);
1441 * If truncate failed early the inode might still be
1442 * on the orphan list; we need to make sure the inode
1443 * is removed from the orphan list in that case.
1446 ext4_orphan_del(NULL, inode);
1449 return ret ? ret : copied;
1453 * Reserve space for a single cluster
1455 static int ext4_da_reserve_space(struct inode *inode)
1457 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1458 struct ext4_inode_info *ei = EXT4_I(inode);
1462 * We will charge metadata quota at writeout time; this saves
1463 * us from metadata over-estimation, though we may go over by
1464 * a small amount in the end. Here we just reserve for data.
1466 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1470 spin_lock(&ei->i_block_reservation_lock);
1471 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1472 spin_unlock(&ei->i_block_reservation_lock);
1473 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1476 ei->i_reserved_data_blocks++;
1477 trace_ext4_da_reserve_space(inode);
1478 spin_unlock(&ei->i_block_reservation_lock);
1480 return 0; /* success */
1483 void ext4_da_release_space(struct inode *inode, int to_free)
1485 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1486 struct ext4_inode_info *ei = EXT4_I(inode);
1489 return; /* Nothing to release, exit */
1491 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1493 trace_ext4_da_release_space(inode, to_free);
1494 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1496 * if there aren't enough reserved blocks, then the
1497 * counter is messed up somewhere. Since this
1498 * function is called from invalidate page, it's
1499 * harmless to return without any action.
1501 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1502 "ino %lu, to_free %d with only %d reserved "
1503 "data blocks", inode->i_ino, to_free,
1504 ei->i_reserved_data_blocks);
1506 to_free = ei->i_reserved_data_blocks;
1508 ei->i_reserved_data_blocks -= to_free;
1510 /* update fs dirty data blocks counter */
1511 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1513 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1515 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1519 * Delayed allocation stuff
1522 struct mpage_da_data {
1523 /* These are input fields for ext4_do_writepages() */
1524 struct inode *inode;
1525 struct writeback_control *wbc;
1526 unsigned int can_map:1; /* Can writepages call map blocks? */
1528 /* These are internal state of ext4_do_writepages() */
1529 pgoff_t first_page; /* The first page to write */
1530 pgoff_t next_page; /* Current page to examine */
1531 pgoff_t last_page; /* Last page to examine */
1533 * Extent to map - this can be after first_page because that can be
1534 * fully mapped. We somewhat abuse m_flags to store whether the extent
1535 * is delalloc or unwritten.
1537 struct ext4_map_blocks map;
1538 struct ext4_io_submit io_submit; /* IO submission data */
1539 unsigned int do_map:1;
1540 unsigned int scanned_until_end:1;
1541 unsigned int journalled_more_data:1;
1544 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1549 struct folio_batch fbatch;
1550 struct inode *inode = mpd->inode;
1551 struct address_space *mapping = inode->i_mapping;
1553 /* This is necessary when next_page == 0. */
1554 if (mpd->first_page >= mpd->next_page)
1557 mpd->scanned_until_end = 0;
1558 index = mpd->first_page;
1559 end = mpd->next_page - 1;
1561 ext4_lblk_t start, last;
1562 start = index << (PAGE_SHIFT - inode->i_blkbits);
1563 last = end << (PAGE_SHIFT - inode->i_blkbits);
1566 * avoid racing with extent status tree scans made by
1567 * ext4_insert_delayed_block()
1569 down_write(&EXT4_I(inode)->i_data_sem);
1570 ext4_es_remove_extent(inode, start, last - start + 1);
1571 up_write(&EXT4_I(inode)->i_data_sem);
1574 folio_batch_init(&fbatch);
1575 while (index <= end) {
1576 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1579 for (i = 0; i < nr; i++) {
1580 struct folio *folio = fbatch.folios[i];
1582 if (folio->index < mpd->first_page)
1584 if (folio_next_index(folio) - 1 > end)
1586 BUG_ON(!folio_test_locked(folio));
1587 BUG_ON(folio_test_writeback(folio));
1589 if (folio_mapped(folio))
1590 folio_clear_dirty_for_io(folio);
1591 block_invalidate_folio(folio, 0,
1593 folio_clear_uptodate(folio);
1595 folio_unlock(folio);
1597 folio_batch_release(&fbatch);
1601 static void ext4_print_free_blocks(struct inode *inode)
1603 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1604 struct super_block *sb = inode->i_sb;
1605 struct ext4_inode_info *ei = EXT4_I(inode);
1607 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1608 EXT4_C2B(EXT4_SB(inode->i_sb),
1609 ext4_count_free_clusters(sb)));
1610 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1611 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1612 (long long) EXT4_C2B(EXT4_SB(sb),
1613 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1614 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1615 (long long) EXT4_C2B(EXT4_SB(sb),
1616 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1617 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1618 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1619 ei->i_reserved_data_blocks);
1624 * ext4_insert_delayed_block - adds a delayed block to the extents status
1625 * tree, incrementing the reserved cluster/block
1626 * count or making a pending reservation
1629 * @inode - file containing the newly added block
1630 * @lblk - logical block to be added
1632 * Returns 0 on success, negative error code on failure.
1634 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1636 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1638 bool allocated = false;
1641 * If the cluster containing lblk is shared with a delayed,
1642 * written, or unwritten extent in a bigalloc file system, it's
1643 * already been accounted for and does not need to be reserved.
1644 * A pending reservation must be made for the cluster if it's
1645 * shared with a written or unwritten extent and doesn't already
1646 * have one. Written and unwritten extents can be purged from the
1647 * extents status tree if the system is under memory pressure, so
1648 * it's necessary to examine the extent tree if a search of the
1649 * extents status tree doesn't get a match.
1651 if (sbi->s_cluster_ratio == 1) {
1652 ret = ext4_da_reserve_space(inode);
1653 if (ret != 0) /* ENOSPC */
1655 } else { /* bigalloc */
1656 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1657 if (!ext4_es_scan_clu(inode,
1658 &ext4_es_is_mapped, lblk)) {
1659 ret = ext4_clu_mapped(inode,
1660 EXT4_B2C(sbi, lblk));
1664 ret = ext4_da_reserve_space(inode);
1665 if (ret != 0) /* ENOSPC */
1676 ext4_es_insert_delayed_block(inode, lblk, allocated);
1681 * This function is grabs code from the very beginning of
1682 * ext4_map_blocks, but assumes that the caller is from delayed write
1683 * time. This function looks up the requested blocks and sets the
1684 * buffer delay bit under the protection of i_data_sem.
1686 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1687 struct ext4_map_blocks *map,
1688 struct buffer_head *bh)
1690 struct extent_status es;
1692 sector_t invalid_block = ~((sector_t) 0xffff);
1693 #ifdef ES_AGGRESSIVE_TEST
1694 struct ext4_map_blocks orig_map;
1696 memcpy(&orig_map, map, sizeof(*map));
1699 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1703 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1704 (unsigned long) map->m_lblk);
1706 /* Lookup extent status tree firstly */
1707 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1708 if (ext4_es_is_hole(&es)) {
1710 down_read(&EXT4_I(inode)->i_data_sem);
1715 * Delayed extent could be allocated by fallocate.
1716 * So we need to check it.
1718 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1719 map_bh(bh, inode->i_sb, invalid_block);
1721 set_buffer_delay(bh);
1725 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1726 retval = es.es_len - (iblock - es.es_lblk);
1727 if (retval > map->m_len)
1728 retval = map->m_len;
1729 map->m_len = retval;
1730 if (ext4_es_is_written(&es))
1731 map->m_flags |= EXT4_MAP_MAPPED;
1732 else if (ext4_es_is_unwritten(&es))
1733 map->m_flags |= EXT4_MAP_UNWRITTEN;
1737 #ifdef ES_AGGRESSIVE_TEST
1738 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1744 * Try to see if we can get the block without requesting a new
1745 * file system block.
1747 down_read(&EXT4_I(inode)->i_data_sem);
1748 if (ext4_has_inline_data(inode))
1750 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1751 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1753 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1760 * XXX: __block_prepare_write() unmaps passed block,
1764 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1770 map_bh(bh, inode->i_sb, invalid_block);
1772 set_buffer_delay(bh);
1773 } else if (retval > 0) {
1774 unsigned int status;
1776 if (unlikely(retval != map->m_len)) {
1777 ext4_warning(inode->i_sb,
1778 "ES len assertion failed for inode "
1779 "%lu: retval %d != map->m_len %d",
1780 inode->i_ino, retval, map->m_len);
1784 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1785 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1786 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1787 map->m_pblk, status);
1791 up_read((&EXT4_I(inode)->i_data_sem));
1797 * This is a special get_block_t callback which is used by
1798 * ext4_da_write_begin(). It will either return mapped block or
1799 * reserve space for a single block.
1801 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1802 * We also have b_blocknr = -1 and b_bdev initialized properly
1804 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1805 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1806 * initialized properly.
1808 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1809 struct buffer_head *bh, int create)
1811 struct ext4_map_blocks map;
1814 BUG_ON(create == 0);
1815 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1817 map.m_lblk = iblock;
1821 * first, we need to know whether the block is allocated already
1822 * preallocated blocks are unmapped but should treated
1823 * the same as allocated blocks.
1825 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1829 map_bh(bh, inode->i_sb, map.m_pblk);
1830 ext4_update_bh_state(bh, map.m_flags);
1832 if (buffer_unwritten(bh)) {
1833 /* A delayed write to unwritten bh should be marked
1834 * new and mapped. Mapped ensures that we don't do
1835 * get_block multiple times when we write to the same
1836 * offset and new ensures that we do proper zero out
1837 * for partial write.
1840 set_buffer_mapped(bh);
1845 static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
1847 mpd->first_page += folio_nr_pages(folio);
1848 folio_unlock(folio);
1851 static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1857 BUG_ON(folio->index != mpd->first_page);
1858 folio_clear_dirty_for_io(folio);
1860 * We have to be very careful here! Nothing protects writeback path
1861 * against i_size changes and the page can be writeably mapped into
1862 * page tables. So an application can be growing i_size and writing
1863 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1864 * write-protects our page in page tables and the page cannot get
1865 * written to again until we release folio lock. So only after
1866 * folio_clear_dirty_for_io() we are safe to sample i_size for
1867 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1868 * on the barrier provided by folio_test_clear_dirty() in
1869 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1870 * after page tables are updated.
1872 size = i_size_read(mpd->inode);
1873 len = folio_size(folio);
1874 if (folio_pos(folio) + len > size &&
1875 !ext4_verity_in_progress(mpd->inode))
1876 len = size & ~PAGE_MASK;
1877 err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
1879 mpd->wbc->nr_to_write--;
1884 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1887 * mballoc gives us at most this number of blocks...
1888 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1889 * The rest of mballoc seems to handle chunks up to full group size.
1891 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1894 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1896 * @mpd - extent of blocks
1897 * @lblk - logical number of the block in the file
1898 * @bh - buffer head we want to add to the extent
1900 * The function is used to collect contig. blocks in the same state. If the
1901 * buffer doesn't require mapping for writeback and we haven't started the
1902 * extent of buffers to map yet, the function returns 'true' immediately - the
1903 * caller can write the buffer right away. Otherwise the function returns true
1904 * if the block has been added to the extent, false if the block couldn't be
1907 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1908 struct buffer_head *bh)
1910 struct ext4_map_blocks *map = &mpd->map;
1912 /* Buffer that doesn't need mapping for writeback? */
1913 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1914 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1915 /* So far no extent to map => we write the buffer right away */
1916 if (map->m_len == 0)
1921 /* First block in the extent? */
1922 if (map->m_len == 0) {
1923 /* We cannot map unless handle is started... */
1928 map->m_flags = bh->b_state & BH_FLAGS;
1932 /* Don't go larger than mballoc is willing to allocate */
1933 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1936 /* Can we merge the block to our big extent? */
1937 if (lblk == map->m_lblk + map->m_len &&
1938 (bh->b_state & BH_FLAGS) == map->m_flags) {
1946 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1948 * @mpd - extent of blocks for mapping
1949 * @head - the first buffer in the page
1950 * @bh - buffer we should start processing from
1951 * @lblk - logical number of the block in the file corresponding to @bh
1953 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1954 * the page for IO if all buffers in this page were mapped and there's no
1955 * accumulated extent of buffers to map or add buffers in the page to the
1956 * extent of buffers to map. The function returns 1 if the caller can continue
1957 * by processing the next page, 0 if it should stop adding buffers to the
1958 * extent to map because we cannot extend it anymore. It can also return value
1959 * < 0 in case of error during IO submission.
1961 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1962 struct buffer_head *head,
1963 struct buffer_head *bh,
1966 struct inode *inode = mpd->inode;
1968 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
1969 >> inode->i_blkbits;
1971 if (ext4_verity_in_progress(inode))
1972 blocks = EXT_MAX_BLOCKS;
1975 BUG_ON(buffer_locked(bh));
1977 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1978 /* Found extent to map? */
1981 /* Buffer needs mapping and handle is not started? */
1984 /* Everything mapped so far and we hit EOF */
1987 } while (lblk++, (bh = bh->b_this_page) != head);
1988 /* So far everything mapped? Submit the page for IO. */
1989 if (mpd->map.m_len == 0) {
1990 err = mpage_submit_folio(mpd, head->b_folio);
1993 mpage_folio_done(mpd, head->b_folio);
1995 if (lblk >= blocks) {
1996 mpd->scanned_until_end = 1;
2003 * mpage_process_folio - update folio buffers corresponding to changed extent
2004 * and may submit fully mapped page for IO
2005 * @mpd: description of extent to map, on return next extent to map
2006 * @folio: Contains these buffers.
2007 * @m_lblk: logical block mapping.
2008 * @m_pblk: corresponding physical mapping.
2009 * @map_bh: determines on return whether this page requires any further
2012 * Scan given folio buffers corresponding to changed extent and update buffer
2013 * state according to new extent state.
2014 * We map delalloc buffers to their physical location, clear unwritten bits.
2015 * If the given folio is not fully mapped, we update @mpd to the next extent in
2016 * the given folio that needs mapping & return @map_bh as true.
2018 static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2019 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2022 struct buffer_head *head, *bh;
2023 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2024 ext4_lblk_t lblk = *m_lblk;
2025 ext4_fsblk_t pblock = *m_pblk;
2027 int blkbits = mpd->inode->i_blkbits;
2028 ssize_t io_end_size = 0;
2029 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2031 bh = head = folio_buffers(folio);
2033 if (lblk < mpd->map.m_lblk)
2035 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2037 * Buffer after end of mapped extent.
2038 * Find next buffer in the folio to map.
2041 mpd->map.m_flags = 0;
2042 io_end_vec->size += io_end_size;
2044 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2047 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2048 io_end_vec = ext4_alloc_io_end_vec(io_end);
2049 if (IS_ERR(io_end_vec)) {
2050 err = PTR_ERR(io_end_vec);
2053 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2058 if (buffer_delay(bh)) {
2059 clear_buffer_delay(bh);
2060 bh->b_blocknr = pblock++;
2062 clear_buffer_unwritten(bh);
2063 io_end_size += (1 << blkbits);
2064 } while (lblk++, (bh = bh->b_this_page) != head);
2066 io_end_vec->size += io_end_size;
2075 * mpage_map_buffers - update buffers corresponding to changed extent and
2076 * submit fully mapped pages for IO
2078 * @mpd - description of extent to map, on return next extent to map
2080 * Scan buffers corresponding to changed extent (we expect corresponding pages
2081 * to be already locked) and update buffer state according to new extent state.
2082 * We map delalloc buffers to their physical location, clear unwritten bits,
2083 * and mark buffers as uninit when we perform writes to unwritten extents
2084 * and do extent conversion after IO is finished. If the last page is not fully
2085 * mapped, we update @map to the next extent in the last page that needs
2086 * mapping. Otherwise we submit the page for IO.
2088 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2090 struct folio_batch fbatch;
2092 struct inode *inode = mpd->inode;
2093 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2096 ext4_fsblk_t pblock;
2098 bool map_bh = false;
2100 start = mpd->map.m_lblk >> bpp_bits;
2101 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2102 lblk = start << bpp_bits;
2103 pblock = mpd->map.m_pblk;
2105 folio_batch_init(&fbatch);
2106 while (start <= end) {
2107 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2110 for (i = 0; i < nr; i++) {
2111 struct folio *folio = fbatch.folios[i];
2113 err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2116 * If map_bh is true, means page may require further bh
2117 * mapping, or maybe the page was submitted for IO.
2118 * So we return to call further extent mapping.
2120 if (err < 0 || map_bh)
2122 /* Page fully mapped - let IO run! */
2123 err = mpage_submit_folio(mpd, folio);
2126 mpage_folio_done(mpd, folio);
2128 folio_batch_release(&fbatch);
2130 /* Extent fully mapped and matches with page boundary. We are done. */
2132 mpd->map.m_flags = 0;
2135 folio_batch_release(&fbatch);
2139 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2141 struct inode *inode = mpd->inode;
2142 struct ext4_map_blocks *map = &mpd->map;
2143 int get_blocks_flags;
2144 int err, dioread_nolock;
2146 trace_ext4_da_write_pages_extent(inode, map);
2148 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2149 * to convert an unwritten extent to be initialized (in the case
2150 * where we have written into one or more preallocated blocks). It is
2151 * possible that we're going to need more metadata blocks than
2152 * previously reserved. However we must not fail because we're in
2153 * writeback and there is nothing we can do about it so it might result
2154 * in data loss. So use reserved blocks to allocate metadata if
2157 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2158 * the blocks in question are delalloc blocks. This indicates
2159 * that the blocks and quotas has already been checked when
2160 * the data was copied into the page cache.
2162 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2163 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2164 EXT4_GET_BLOCKS_IO_SUBMIT;
2165 dioread_nolock = ext4_should_dioread_nolock(inode);
2167 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2168 if (map->m_flags & BIT(BH_Delay))
2169 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2171 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2174 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2175 if (!mpd->io_submit.io_end->handle &&
2176 ext4_handle_valid(handle)) {
2177 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2178 handle->h_rsv_handle = NULL;
2180 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2183 BUG_ON(map->m_len == 0);
2188 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2189 * mpd->len and submit pages underlying it for IO
2191 * @handle - handle for journal operations
2192 * @mpd - extent to map
2193 * @give_up_on_write - we set this to true iff there is a fatal error and there
2194 * is no hope of writing the data. The caller should discard
2195 * dirty pages to avoid infinite loops.
2197 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2198 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2199 * them to initialized or split the described range from larger unwritten
2200 * extent. Note that we need not map all the described range since allocation
2201 * can return less blocks or the range is covered by more unwritten extents. We
2202 * cannot map more because we are limited by reserved transaction credits. On
2203 * the other hand we always make sure that the last touched page is fully
2204 * mapped so that it can be written out (and thus forward progress is
2205 * guaranteed). After mapping we submit all mapped pages for IO.
2207 static int mpage_map_and_submit_extent(handle_t *handle,
2208 struct mpage_da_data *mpd,
2209 bool *give_up_on_write)
2211 struct inode *inode = mpd->inode;
2212 struct ext4_map_blocks *map = &mpd->map;
2216 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2217 struct ext4_io_end_vec *io_end_vec;
2219 io_end_vec = ext4_alloc_io_end_vec(io_end);
2220 if (IS_ERR(io_end_vec))
2221 return PTR_ERR(io_end_vec);
2222 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2224 err = mpage_map_one_extent(handle, mpd);
2226 struct super_block *sb = inode->i_sb;
2228 if (ext4_forced_shutdown(sb))
2229 goto invalidate_dirty_pages;
2231 * Let the uper layers retry transient errors.
2232 * In the case of ENOSPC, if ext4_count_free_blocks()
2233 * is non-zero, a commit should free up blocks.
2235 if ((err == -ENOMEM) ||
2236 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2238 goto update_disksize;
2241 ext4_msg(sb, KERN_CRIT,
2242 "Delayed block allocation failed for "
2243 "inode %lu at logical offset %llu with"
2244 " max blocks %u with error %d",
2246 (unsigned long long)map->m_lblk,
2247 (unsigned)map->m_len, -err);
2248 ext4_msg(sb, KERN_CRIT,
2249 "This should not happen!! Data will "
2252 ext4_print_free_blocks(inode);
2253 invalidate_dirty_pages:
2254 *give_up_on_write = true;
2259 * Update buffer state, submit mapped pages, and get us new
2262 err = mpage_map_and_submit_buffers(mpd);
2264 goto update_disksize;
2265 } while (map->m_len);
2269 * Update on-disk size after IO is submitted. Races with
2270 * truncate are avoided by checking i_size under i_data_sem.
2272 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2273 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2277 down_write(&EXT4_I(inode)->i_data_sem);
2278 i_size = i_size_read(inode);
2279 if (disksize > i_size)
2281 if (disksize > EXT4_I(inode)->i_disksize)
2282 EXT4_I(inode)->i_disksize = disksize;
2283 up_write(&EXT4_I(inode)->i_data_sem);
2284 err2 = ext4_mark_inode_dirty(handle, inode);
2286 ext4_error_err(inode->i_sb, -err2,
2287 "Failed to mark inode %lu dirty",
2297 * Calculate the total number of credits to reserve for one writepages
2298 * iteration. This is called from ext4_writepages(). We map an extent of
2299 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2300 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2301 * bpp - 1 blocks in bpp different extents.
2303 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2305 int bpp = ext4_journal_blocks_per_page(inode);
2307 return ext4_meta_trans_blocks(inode,
2308 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2311 static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2314 struct buffer_head *page_bufs = folio_buffers(folio);
2315 struct inode *inode = folio->mapping->host;
2318 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2319 NULL, do_journal_get_write_access);
2320 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2321 NULL, write_end_fn);
2324 err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len);
2327 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2332 static int mpage_journal_page_buffers(handle_t *handle,
2333 struct mpage_da_data *mpd,
2334 struct folio *folio)
2336 struct inode *inode = mpd->inode;
2337 loff_t size = i_size_read(inode);
2338 size_t len = folio_size(folio);
2340 folio_clear_checked(folio);
2341 mpd->wbc->nr_to_write--;
2343 if (folio_pos(folio) + len > size &&
2344 !ext4_verity_in_progress(inode))
2345 len = size - folio_pos(folio);
2347 return ext4_journal_folio_buffers(handle, folio, len);
2351 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2352 * needing mapping, submit mapped pages
2354 * @mpd - where to look for pages
2356 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2357 * IO immediately. If we cannot map blocks, we submit just already mapped
2358 * buffers in the page for IO and keep page dirty. When we can map blocks and
2359 * we find a page which isn't mapped we start accumulating extent of buffers
2360 * underlying these pages that needs mapping (formed by either delayed or
2361 * unwritten buffers). We also lock the pages containing these buffers. The
2362 * extent found is returned in @mpd structure (starting at mpd->lblk with
2363 * length mpd->len blocks).
2365 * Note that this function can attach bios to one io_end structure which are
2366 * neither logically nor physically contiguous. Although it may seem as an
2367 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2368 * case as we need to track IO to all buffers underlying a page in one io_end.
2370 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2372 struct address_space *mapping = mpd->inode->i_mapping;
2373 struct folio_batch fbatch;
2374 unsigned int nr_folios;
2375 pgoff_t index = mpd->first_page;
2376 pgoff_t end = mpd->last_page;
2379 int blkbits = mpd->inode->i_blkbits;
2381 struct buffer_head *head;
2382 handle_t *handle = NULL;
2383 int bpp = ext4_journal_blocks_per_page(mpd->inode);
2385 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2386 tag = PAGECACHE_TAG_TOWRITE;
2388 tag = PAGECACHE_TAG_DIRTY;
2391 mpd->next_page = index;
2392 if (ext4_should_journal_data(mpd->inode)) {
2393 handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2396 return PTR_ERR(handle);
2398 folio_batch_init(&fbatch);
2399 while (index <= end) {
2400 nr_folios = filemap_get_folios_tag(mapping, &index, end,
2405 for (i = 0; i < nr_folios; i++) {
2406 struct folio *folio = fbatch.folios[i];
2409 * Accumulated enough dirty pages? This doesn't apply
2410 * to WB_SYNC_ALL mode. For integrity sync we have to
2411 * keep going because someone may be concurrently
2412 * dirtying pages, and we might have synced a lot of
2413 * newly appeared dirty pages, but have not synced all
2414 * of the old dirty pages.
2416 if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2417 mpd->wbc->nr_to_write <=
2418 mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2421 /* If we can't merge this page, we are done. */
2422 if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2426 err = ext4_journal_ensure_credits(handle, bpp,
2434 * If the page is no longer dirty, or its mapping no
2435 * longer corresponds to inode we are writing (which
2436 * means it has been truncated or invalidated), or the
2437 * page is already under writeback and we are not doing
2438 * a data integrity writeback, skip the page
2440 if (!folio_test_dirty(folio) ||
2441 (folio_test_writeback(folio) &&
2442 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2443 unlikely(folio->mapping != mapping)) {
2444 folio_unlock(folio);
2448 folio_wait_writeback(folio);
2449 BUG_ON(folio_test_writeback(folio));
2452 * Should never happen but for buggy code in
2453 * other subsystems that call
2454 * set_page_dirty() without properly warning
2455 * the file system first. See [1] for more
2458 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2460 if (!folio_buffers(folio)) {
2461 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2462 folio_clear_dirty(folio);
2463 folio_unlock(folio);
2467 if (mpd->map.m_len == 0)
2468 mpd->first_page = folio->index;
2469 mpd->next_page = folio_next_index(folio);
2471 * Writeout when we cannot modify metadata is simple.
2472 * Just submit the page. For data=journal mode we
2473 * first handle writeout of the page for checkpoint and
2474 * only after that handle delayed page dirtying. This
2475 * makes sure current data is checkpointed to the final
2476 * location before possibly journalling it again which
2477 * is desirable when the page is frequently dirtied
2480 if (!mpd->can_map) {
2481 err = mpage_submit_folio(mpd, folio);
2484 /* Pending dirtying of journalled data? */
2485 if (folio_test_checked(folio)) {
2486 err = mpage_journal_page_buffers(handle,
2490 mpd->journalled_more_data = 1;
2492 mpage_folio_done(mpd, folio);
2494 /* Add all dirty buffers to mpd */
2495 lblk = ((ext4_lblk_t)folio->index) <<
2496 (PAGE_SHIFT - blkbits);
2497 head = folio_buffers(folio);
2498 err = mpage_process_page_bufs(mpd, head, head,
2505 folio_batch_release(&fbatch);
2508 mpd->scanned_until_end = 1;
2510 ext4_journal_stop(handle);
2513 folio_batch_release(&fbatch);
2515 ext4_journal_stop(handle);
2519 static int ext4_do_writepages(struct mpage_da_data *mpd)
2521 struct writeback_control *wbc = mpd->wbc;
2522 pgoff_t writeback_index = 0;
2523 long nr_to_write = wbc->nr_to_write;
2524 int range_whole = 0;
2526 handle_t *handle = NULL;
2527 struct inode *inode = mpd->inode;
2528 struct address_space *mapping = inode->i_mapping;
2529 int needed_blocks, rsv_blocks = 0, ret = 0;
2530 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2531 struct blk_plug plug;
2532 bool give_up_on_write = false;
2534 trace_ext4_writepages(inode, wbc);
2537 * No pages to write? This is mainly a kludge to avoid starting
2538 * a transaction for special inodes like journal inode on last iput()
2539 * because that could violate lock ordering on umount
2541 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2542 goto out_writepages;
2545 * If the filesystem has aborted, it is read-only, so return
2546 * right away instead of dumping stack traces later on that
2547 * will obscure the real source of the problem. We test
2548 * fs shutdown state instead of sb->s_flag's SB_RDONLY because
2549 * the latter could be true if the filesystem is mounted
2550 * read-only, and in that case, ext4_writepages should
2551 * *never* be called, so if that ever happens, we would want
2554 if (unlikely(ext4_forced_shutdown(mapping->host->i_sb))) {
2556 goto out_writepages;
2560 * If we have inline data and arrive here, it means that
2561 * we will soon create the block for the 1st page, so
2562 * we'd better clear the inline data here.
2564 if (ext4_has_inline_data(inode)) {
2565 /* Just inode will be modified... */
2566 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2567 if (IS_ERR(handle)) {
2568 ret = PTR_ERR(handle);
2569 goto out_writepages;
2571 BUG_ON(ext4_test_inode_state(inode,
2572 EXT4_STATE_MAY_INLINE_DATA));
2573 ext4_destroy_inline_data(handle, inode);
2574 ext4_journal_stop(handle);
2578 * data=journal mode does not do delalloc so we just need to writeout /
2579 * journal already mapped buffers. On the other hand we need to commit
2580 * transaction to make data stable. We expect all the data to be
2581 * already in the journal (the only exception are DMA pinned pages
2582 * dirtied behind our back) so we commit transaction here and run the
2583 * writeback loop to checkpoint them. The checkpointing is not actually
2584 * necessary to make data persistent *but* quite a few places (extent
2585 * shifting operations, fsverity, ...) depend on being able to drop
2586 * pagecache pages after calling filemap_write_and_wait() and for that
2587 * checkpointing needs to happen.
2589 if (ext4_should_journal_data(inode)) {
2591 if (wbc->sync_mode == WB_SYNC_ALL)
2592 ext4_fc_commit(sbi->s_journal,
2593 EXT4_I(inode)->i_datasync_tid);
2595 mpd->journalled_more_data = 0;
2597 if (ext4_should_dioread_nolock(inode)) {
2599 * We may need to convert up to one extent per block in
2600 * the page and we may dirty the inode.
2602 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2603 PAGE_SIZE >> inode->i_blkbits);
2606 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2609 if (wbc->range_cyclic) {
2610 writeback_index = mapping->writeback_index;
2611 if (writeback_index)
2613 mpd->first_page = writeback_index;
2614 mpd->last_page = -1;
2616 mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2617 mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2620 ext4_io_submit_init(&mpd->io_submit, wbc);
2622 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2623 tag_pages_for_writeback(mapping, mpd->first_page,
2625 blk_start_plug(&plug);
2628 * First writeback pages that don't need mapping - we can avoid
2629 * starting a transaction unnecessarily and also avoid being blocked
2630 * in the block layer on device congestion while having transaction
2634 mpd->scanned_until_end = 0;
2635 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2636 if (!mpd->io_submit.io_end) {
2640 ret = mpage_prepare_extent_to_map(mpd);
2641 /* Unlock pages we didn't use */
2642 mpage_release_unused_pages(mpd, false);
2643 /* Submit prepared bio */
2644 ext4_io_submit(&mpd->io_submit);
2645 ext4_put_io_end_defer(mpd->io_submit.io_end);
2646 mpd->io_submit.io_end = NULL;
2650 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2651 /* For each extent of pages we use new io_end */
2652 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2653 if (!mpd->io_submit.io_end) {
2658 WARN_ON_ONCE(!mpd->can_map);
2660 * We have two constraints: We find one extent to map and we
2661 * must always write out whole page (makes a difference when
2662 * blocksize < pagesize) so that we don't block on IO when we
2663 * try to write out the rest of the page. Journalled mode is
2664 * not supported by delalloc.
2666 BUG_ON(ext4_should_journal_data(inode));
2667 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2669 /* start a new transaction */
2670 handle = ext4_journal_start_with_reserve(inode,
2671 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2672 if (IS_ERR(handle)) {
2673 ret = PTR_ERR(handle);
2674 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2675 "%ld pages, ino %lu; err %d", __func__,
2676 wbc->nr_to_write, inode->i_ino, ret);
2677 /* Release allocated io_end */
2678 ext4_put_io_end(mpd->io_submit.io_end);
2679 mpd->io_submit.io_end = NULL;
2684 trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2685 ret = mpage_prepare_extent_to_map(mpd);
2686 if (!ret && mpd->map.m_len)
2687 ret = mpage_map_and_submit_extent(handle, mpd,
2690 * Caution: If the handle is synchronous,
2691 * ext4_journal_stop() can wait for transaction commit
2692 * to finish which may depend on writeback of pages to
2693 * complete or on page lock to be released. In that
2694 * case, we have to wait until after we have
2695 * submitted all the IO, released page locks we hold,
2696 * and dropped io_end reference (for extent conversion
2697 * to be able to complete) before stopping the handle.
2699 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2700 ext4_journal_stop(handle);
2704 /* Unlock pages we didn't use */
2705 mpage_release_unused_pages(mpd, give_up_on_write);
2706 /* Submit prepared bio */
2707 ext4_io_submit(&mpd->io_submit);
2710 * Drop our io_end reference we got from init. We have
2711 * to be careful and use deferred io_end finishing if
2712 * we are still holding the transaction as we can
2713 * release the last reference to io_end which may end
2714 * up doing unwritten extent conversion.
2717 ext4_put_io_end_defer(mpd->io_submit.io_end);
2718 ext4_journal_stop(handle);
2720 ext4_put_io_end(mpd->io_submit.io_end);
2721 mpd->io_submit.io_end = NULL;
2723 if (ret == -ENOSPC && sbi->s_journal) {
2725 * Commit the transaction which would
2726 * free blocks released in the transaction
2729 jbd2_journal_force_commit_nested(sbi->s_journal);
2733 /* Fatal error - ENOMEM, EIO... */
2738 blk_finish_plug(&plug);
2739 if (!ret && !cycled && wbc->nr_to_write > 0) {
2741 mpd->last_page = writeback_index - 1;
2742 mpd->first_page = 0;
2747 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2749 * Set the writeback_index so that range_cyclic
2750 * mode will write it back later
2752 mapping->writeback_index = mpd->first_page;
2755 trace_ext4_writepages_result(inode, wbc, ret,
2756 nr_to_write - wbc->nr_to_write);
2760 static int ext4_writepages(struct address_space *mapping,
2761 struct writeback_control *wbc)
2763 struct super_block *sb = mapping->host->i_sb;
2764 struct mpage_da_data mpd = {
2765 .inode = mapping->host,
2772 if (unlikely(ext4_forced_shutdown(sb)))
2775 alloc_ctx = ext4_writepages_down_read(sb);
2776 ret = ext4_do_writepages(&mpd);
2778 * For data=journal writeback we could have come across pages marked
2779 * for delayed dirtying (PageChecked) which were just added to the
2780 * running transaction. Try once more to get them to stable storage.
2782 if (!ret && mpd.journalled_more_data)
2783 ret = ext4_do_writepages(&mpd);
2784 ext4_writepages_up_read(sb, alloc_ctx);
2789 int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2791 struct writeback_control wbc = {
2792 .sync_mode = WB_SYNC_ALL,
2793 .nr_to_write = LONG_MAX,
2794 .range_start = jinode->i_dirty_start,
2795 .range_end = jinode->i_dirty_end,
2797 struct mpage_da_data mpd = {
2798 .inode = jinode->i_vfs_inode,
2802 return ext4_do_writepages(&mpd);
2805 static int ext4_dax_writepages(struct address_space *mapping,
2806 struct writeback_control *wbc)
2809 long nr_to_write = wbc->nr_to_write;
2810 struct inode *inode = mapping->host;
2813 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2816 alloc_ctx = ext4_writepages_down_read(inode->i_sb);
2817 trace_ext4_writepages(inode, wbc);
2819 ret = dax_writeback_mapping_range(mapping,
2820 EXT4_SB(inode->i_sb)->s_daxdev, wbc);
2821 trace_ext4_writepages_result(inode, wbc, ret,
2822 nr_to_write - wbc->nr_to_write);
2823 ext4_writepages_up_read(inode->i_sb, alloc_ctx);
2827 static int ext4_nonda_switch(struct super_block *sb)
2829 s64 free_clusters, dirty_clusters;
2830 struct ext4_sb_info *sbi = EXT4_SB(sb);
2833 * switch to non delalloc mode if we are running low
2834 * on free block. The free block accounting via percpu
2835 * counters can get slightly wrong with percpu_counter_batch getting
2836 * accumulated on each CPU without updating global counters
2837 * Delalloc need an accurate free block accounting. So switch
2838 * to non delalloc when we are near to error range.
2841 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2843 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2845 * Start pushing delalloc when 1/2 of free blocks are dirty.
2847 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2848 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2850 if (2 * free_clusters < 3 * dirty_clusters ||
2851 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2853 * free block count is less than 150% of dirty blocks
2854 * or free blocks is less than watermark
2861 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2862 loff_t pos, unsigned len,
2863 struct page **pagep, void **fsdata)
2865 int ret, retries = 0;
2866 struct folio *folio;
2868 struct inode *inode = mapping->host;
2870 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2873 index = pos >> PAGE_SHIFT;
2875 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2876 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2877 return ext4_write_begin(file, mapping, pos,
2878 len, pagep, fsdata);
2880 *fsdata = (void *)0;
2881 trace_ext4_da_write_begin(inode, pos, len);
2883 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2884 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2893 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2894 mapping_gfp_mask(mapping));
2896 return PTR_ERR(folio);
2898 /* In case writeback began while the folio was unlocked */
2899 folio_wait_stable(folio);
2901 #ifdef CONFIG_FS_ENCRYPTION
2902 ret = ext4_block_write_begin(folio, pos, len, ext4_da_get_block_prep);
2904 ret = __block_write_begin(&folio->page, pos, len, ext4_da_get_block_prep);
2907 folio_unlock(folio);
2910 * block_write_begin may have instantiated a few blocks
2911 * outside i_size. Trim these off again. Don't need
2912 * i_size_read because we hold inode lock.
2914 if (pos + len > inode->i_size)
2915 ext4_truncate_failed_write(inode);
2917 if (ret == -ENOSPC &&
2918 ext4_should_retry_alloc(inode->i_sb, &retries))
2923 *pagep = &folio->page;
2928 * Check if we should update i_disksize
2929 * when write to the end of file but not require block allocation
2931 static int ext4_da_should_update_i_disksize(struct folio *folio,
2932 unsigned long offset)
2934 struct buffer_head *bh;
2935 struct inode *inode = folio->mapping->host;
2939 bh = folio_buffers(folio);
2940 idx = offset >> inode->i_blkbits;
2942 for (i = 0; i < idx; i++)
2943 bh = bh->b_this_page;
2945 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2950 static int ext4_da_do_write_end(struct address_space *mapping,
2951 loff_t pos, unsigned len, unsigned copied,
2954 struct inode *inode = mapping->host;
2955 loff_t old_size = inode->i_size;
2956 bool disksize_changed = false;
2960 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
2961 * flag, which all that's needed to trigger page writeback.
2963 copied = block_write_end(NULL, mapping, pos, len, copied, page, NULL);
2964 new_i_size = pos + copied;
2967 * It's important to update i_size while still holding page lock,
2968 * because page writeout could otherwise come in and zero beyond
2971 * Since we are holding inode lock, we are sure i_disksize <=
2972 * i_size. We also know that if i_disksize < i_size, there are
2973 * delalloc writes pending in the range up to i_size. If the end of
2974 * the current write is <= i_size, there's no need to touch
2975 * i_disksize since writeback will push i_disksize up to i_size
2976 * eventually. If the end of the current write is > i_size and
2977 * inside an allocated block which ext4_da_should_update_i_disksize()
2978 * checked, we need to update i_disksize here as certain
2979 * ext4_writepages() paths not allocating blocks and update i_disksize.
2981 if (new_i_size > inode->i_size) {
2984 i_size_write(inode, new_i_size);
2985 end = (new_i_size - 1) & (PAGE_SIZE - 1);
2986 if (copied && ext4_da_should_update_i_disksize(page_folio(page), end)) {
2987 ext4_update_i_disksize(inode, new_i_size);
2988 disksize_changed = true;
2996 pagecache_isize_extended(inode, old_size, pos);
2998 if (disksize_changed) {
3001 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3003 return PTR_ERR(handle);
3004 ext4_mark_inode_dirty(handle, inode);
3005 ext4_journal_stop(handle);
3011 static int ext4_da_write_end(struct file *file,
3012 struct address_space *mapping,
3013 loff_t pos, unsigned len, unsigned copied,
3014 struct page *page, void *fsdata)
3016 struct inode *inode = mapping->host;
3017 int write_mode = (int)(unsigned long)fsdata;
3018 struct folio *folio = page_folio(page);
3020 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3021 return ext4_write_end(file, mapping, pos,
3022 len, copied, &folio->page, fsdata);
3024 trace_ext4_da_write_end(inode, pos, len, copied);
3026 if (write_mode != CONVERT_INLINE_DATA &&
3027 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3028 ext4_has_inline_data(inode))
3029 return ext4_write_inline_data_end(inode, pos, len, copied,
3032 if (unlikely(copied < len) && !PageUptodate(page))
3035 return ext4_da_do_write_end(mapping, pos, len, copied, &folio->page);
3039 * Force all delayed allocation blocks to be allocated for a given inode.
3041 int ext4_alloc_da_blocks(struct inode *inode)
3043 trace_ext4_alloc_da_blocks(inode);
3045 if (!EXT4_I(inode)->i_reserved_data_blocks)
3049 * We do something simple for now. The filemap_flush() will
3050 * also start triggering a write of the data blocks, which is
3051 * not strictly speaking necessary (and for users of
3052 * laptop_mode, not even desirable). However, to do otherwise
3053 * would require replicating code paths in:
3055 * ext4_writepages() ->
3056 * write_cache_pages() ---> (via passed in callback function)
3057 * __mpage_da_writepage() -->
3058 * mpage_add_bh_to_extent()
3059 * mpage_da_map_blocks()
3061 * The problem is that write_cache_pages(), located in
3062 * mm/page-writeback.c, marks pages clean in preparation for
3063 * doing I/O, which is not desirable if we're not planning on
3066 * We could call write_cache_pages(), and then redirty all of
3067 * the pages by calling redirty_page_for_writepage() but that
3068 * would be ugly in the extreme. So instead we would need to
3069 * replicate parts of the code in the above functions,
3070 * simplifying them because we wouldn't actually intend to
3071 * write out the pages, but rather only collect contiguous
3072 * logical block extents, call the multi-block allocator, and
3073 * then update the buffer heads with the block allocations.
3075 * For now, though, we'll cheat by calling filemap_flush(),
3076 * which will map the blocks, and start the I/O, but not
3077 * actually wait for the I/O to complete.
3079 return filemap_flush(inode->i_mapping);
3083 * bmap() is special. It gets used by applications such as lilo and by
3084 * the swapper to find the on-disk block of a specific piece of data.
3086 * Naturally, this is dangerous if the block concerned is still in the
3087 * journal. If somebody makes a swapfile on an ext4 data-journaling
3088 * filesystem and enables swap, then they may get a nasty shock when the
3089 * data getting swapped to that swapfile suddenly gets overwritten by
3090 * the original zero's written out previously to the journal and
3091 * awaiting writeback in the kernel's buffer cache.
3093 * So, if we see any bmap calls here on a modified, data-journaled file,
3094 * take extra steps to flush any blocks which might be in the cache.
3096 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3098 struct inode *inode = mapping->host;
3101 inode_lock_shared(inode);
3103 * We can get here for an inline file via the FIBMAP ioctl
3105 if (ext4_has_inline_data(inode))
3108 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3109 (test_opt(inode->i_sb, DELALLOC) ||
3110 ext4_should_journal_data(inode))) {
3112 * With delalloc or journalled data we want to sync the file so
3113 * that we can make sure we allocate blocks for file and data
3114 * is in place for the user to see it
3116 filemap_write_and_wait(mapping);
3119 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3122 inode_unlock_shared(inode);
3126 static int ext4_read_folio(struct file *file, struct folio *folio)
3129 struct inode *inode = folio->mapping->host;
3131 trace_ext4_read_folio(inode, folio);
3133 if (ext4_has_inline_data(inode))
3134 ret = ext4_readpage_inline(inode, folio);
3137 return ext4_mpage_readpages(inode, NULL, folio);
3142 static void ext4_readahead(struct readahead_control *rac)
3144 struct inode *inode = rac->mapping->host;
3146 /* If the file has inline data, no need to do readahead. */
3147 if (ext4_has_inline_data(inode))
3150 ext4_mpage_readpages(inode, rac, NULL);
3153 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3156 trace_ext4_invalidate_folio(folio, offset, length);
3158 /* No journalling happens on data buffers when this function is used */
3159 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3161 block_invalidate_folio(folio, offset, length);
3164 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3165 size_t offset, size_t length)
3167 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3169 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3172 * If it's a full truncate we just forget about the pending dirtying
3174 if (offset == 0 && length == folio_size(folio))
3175 folio_clear_checked(folio);
3177 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3180 /* Wrapper for aops... */
3181 static void ext4_journalled_invalidate_folio(struct folio *folio,
3185 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3188 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3190 struct inode *inode = folio->mapping->host;
3191 journal_t *journal = EXT4_JOURNAL(inode);
3193 trace_ext4_release_folio(inode, folio);
3195 /* Page has dirty journalled data -> cannot release */
3196 if (folio_test_checked(folio))
3199 return jbd2_journal_try_to_free_buffers(journal, folio);
3201 return try_to_free_buffers(folio);
3204 static bool ext4_inode_datasync_dirty(struct inode *inode)
3206 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3209 if (jbd2_transaction_committed(journal,
3210 EXT4_I(inode)->i_datasync_tid))
3212 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3213 return !list_empty(&EXT4_I(inode)->i_fc_list);
3217 /* Any metadata buffers to write? */
3218 if (!list_empty(&inode->i_mapping->private_list))
3220 return inode->i_state & I_DIRTY_DATASYNC;
3223 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3224 struct ext4_map_blocks *map, loff_t offset,
3225 loff_t length, unsigned int flags)
3227 u8 blkbits = inode->i_blkbits;
3230 * Writes that span EOF might trigger an I/O size update on completion,
3231 * so consider them to be dirty for the purpose of O_DSYNC, even if
3232 * there is no other metadata changes being made or are pending.
3235 if (ext4_inode_datasync_dirty(inode) ||
3236 offset + length > i_size_read(inode))
3237 iomap->flags |= IOMAP_F_DIRTY;
3239 if (map->m_flags & EXT4_MAP_NEW)
3240 iomap->flags |= IOMAP_F_NEW;
3242 if (flags & IOMAP_DAX)
3243 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3245 iomap->bdev = inode->i_sb->s_bdev;
3246 iomap->offset = (u64) map->m_lblk << blkbits;
3247 iomap->length = (u64) map->m_len << blkbits;
3249 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3250 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3251 iomap->flags |= IOMAP_F_MERGED;
3254 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3255 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3256 * set. In order for any allocated unwritten extents to be converted
3257 * into written extents correctly within the ->end_io() handler, we
3258 * need to ensure that the iomap->type is set appropriately. Hence, the
3259 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3262 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3263 iomap->type = IOMAP_UNWRITTEN;
3264 iomap->addr = (u64) map->m_pblk << blkbits;
3265 if (flags & IOMAP_DAX)
3266 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3267 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3268 iomap->type = IOMAP_MAPPED;
3269 iomap->addr = (u64) map->m_pblk << blkbits;
3270 if (flags & IOMAP_DAX)
3271 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3273 iomap->type = IOMAP_HOLE;
3274 iomap->addr = IOMAP_NULL_ADDR;
3278 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3282 u8 blkbits = inode->i_blkbits;
3283 int ret, dio_credits, m_flags = 0, retries = 0;
3286 * Trim the mapping request to the maximum value that we can map at
3287 * once for direct I/O.
3289 if (map->m_len > DIO_MAX_BLOCKS)
3290 map->m_len = DIO_MAX_BLOCKS;
3291 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3295 * Either we allocate blocks and then don't get an unwritten extent, so
3296 * in that case we have reserved enough credits. Or, the blocks are
3297 * already allocated and unwritten. In that case, the extent conversion
3298 * fits into the credits as well.
3300 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3302 return PTR_ERR(handle);
3305 * DAX and direct I/O are the only two operations that are currently
3306 * supported with IOMAP_WRITE.
3308 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3309 if (flags & IOMAP_DAX)
3310 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3312 * We use i_size instead of i_disksize here because delalloc writeback
3313 * can complete at any point during the I/O and subsequently push the
3314 * i_disksize out to i_size. This could be beyond where direct I/O is
3315 * happening and thus expose allocated blocks to direct I/O reads.
3317 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3318 m_flags = EXT4_GET_BLOCKS_CREATE;
3319 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3320 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3322 ret = ext4_map_blocks(handle, inode, map, m_flags);
3325 * We cannot fill holes in indirect tree based inodes as that could
3326 * expose stale data in the case of a crash. Use the magic error code
3327 * to fallback to buffered I/O.
3329 if (!m_flags && !ret)
3332 ext4_journal_stop(handle);
3333 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3340 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3341 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3344 struct ext4_map_blocks map;
3345 u8 blkbits = inode->i_blkbits;
3347 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3350 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3354 * Calculate the first and last logical blocks respectively.
3356 map.m_lblk = offset >> blkbits;
3357 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3358 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3360 if (flags & IOMAP_WRITE) {
3362 * We check here if the blocks are already allocated, then we
3363 * don't need to start a journal txn and we can directly return
3364 * the mapping information. This could boost performance
3365 * especially in multi-threaded overwrite requests.
3367 if (offset + length <= i_size_read(inode)) {
3368 ret = ext4_map_blocks(NULL, inode, &map, 0);
3369 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3372 ret = ext4_iomap_alloc(inode, &map, flags);
3374 ret = ext4_map_blocks(NULL, inode, &map, 0);
3381 * When inline encryption is enabled, sometimes I/O to an encrypted file
3382 * has to be broken up to guarantee DUN contiguity. Handle this by
3383 * limiting the length of the mapping returned.
3385 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3387 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3392 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3393 loff_t length, unsigned flags, struct iomap *iomap,
3394 struct iomap *srcmap)
3399 * Even for writes we don't need to allocate blocks, so just pretend
3400 * we are reading to save overhead of starting a transaction.
3402 flags &= ~IOMAP_WRITE;
3403 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3404 WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3408 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3409 ssize_t written, unsigned flags, struct iomap *iomap)
3412 * Check to see whether an error occurred while writing out the data to
3413 * the allocated blocks. If so, return the magic error code so that we
3414 * fallback to buffered I/O and attempt to complete the remainder of
3415 * the I/O. Any blocks that may have been allocated in preparation for
3416 * the direct I/O will be reused during buffered I/O.
3418 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3424 const struct iomap_ops ext4_iomap_ops = {
3425 .iomap_begin = ext4_iomap_begin,
3426 .iomap_end = ext4_iomap_end,
3429 const struct iomap_ops ext4_iomap_overwrite_ops = {
3430 .iomap_begin = ext4_iomap_overwrite_begin,
3431 .iomap_end = ext4_iomap_end,
3434 static bool ext4_iomap_is_delalloc(struct inode *inode,
3435 struct ext4_map_blocks *map)
3437 struct extent_status es;
3438 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3440 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3441 map->m_lblk, end, &es);
3443 if (!es.es_len || es.es_lblk > end)
3446 if (es.es_lblk > map->m_lblk) {
3447 map->m_len = es.es_lblk - map->m_lblk;
3451 offset = map->m_lblk - es.es_lblk;
3452 map->m_len = es.es_len - offset;
3457 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3458 loff_t length, unsigned int flags,
3459 struct iomap *iomap, struct iomap *srcmap)
3462 bool delalloc = false;
3463 struct ext4_map_blocks map;
3464 u8 blkbits = inode->i_blkbits;
3466 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3469 if (ext4_has_inline_data(inode)) {
3470 ret = ext4_inline_data_iomap(inode, iomap);
3471 if (ret != -EAGAIN) {
3472 if (ret == 0 && offset >= iomap->length)
3479 * Calculate the first and last logical block respectively.
3481 map.m_lblk = offset >> blkbits;
3482 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3483 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3486 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3487 * So handle it here itself instead of querying ext4_map_blocks().
3488 * Since ext4_map_blocks() will warn about it and will return
3491 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3492 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3494 if (offset >= sbi->s_bitmap_maxbytes) {
3500 ret = ext4_map_blocks(NULL, inode, &map, 0);
3504 delalloc = ext4_iomap_is_delalloc(inode, &map);
3507 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3508 if (delalloc && iomap->type == IOMAP_HOLE)
3509 iomap->type = IOMAP_DELALLOC;
3514 const struct iomap_ops ext4_iomap_report_ops = {
3515 .iomap_begin = ext4_iomap_begin_report,
3519 * For data=journal mode, folio should be marked dirty only when it was
3520 * writeably mapped. When that happens, it was already attached to the
3521 * transaction and marked as jbddirty (we take care of this in
3522 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3523 * so we should have nothing to do here, except for the case when someone
3524 * had the page pinned and dirtied the page through this pin (e.g. by doing
3525 * direct IO to it). In that case we'd need to attach buffers here to the
3526 * transaction but we cannot due to lock ordering. We cannot just dirty the
3527 * folio and leave attached buffers clean, because the buffers' dirty state is
3528 * "definitive". We cannot just set the buffers dirty or jbddirty because all
3529 * the journalling code will explode. So what we do is to mark the folio
3530 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3531 * to the transaction appropriately.
3533 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3534 struct folio *folio)
3536 WARN_ON_ONCE(!folio_buffers(folio));
3537 if (folio_maybe_dma_pinned(folio))
3538 folio_set_checked(folio);
3539 return filemap_dirty_folio(mapping, folio);
3542 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3544 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3545 WARN_ON_ONCE(!folio_buffers(folio));
3546 return block_dirty_folio(mapping, folio);
3549 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3550 struct file *file, sector_t *span)
3552 return iomap_swapfile_activate(sis, file, span,
3553 &ext4_iomap_report_ops);
3556 static const struct address_space_operations ext4_aops = {
3557 .read_folio = ext4_read_folio,
3558 .readahead = ext4_readahead,
3559 .writepages = ext4_writepages,
3560 .write_begin = ext4_write_begin,
3561 .write_end = ext4_write_end,
3562 .dirty_folio = ext4_dirty_folio,
3564 .invalidate_folio = ext4_invalidate_folio,
3565 .release_folio = ext4_release_folio,
3566 .direct_IO = noop_direct_IO,
3567 .migrate_folio = buffer_migrate_folio,
3568 .is_partially_uptodate = block_is_partially_uptodate,
3569 .error_remove_page = generic_error_remove_page,
3570 .swap_activate = ext4_iomap_swap_activate,
3573 static const struct address_space_operations ext4_journalled_aops = {
3574 .read_folio = ext4_read_folio,
3575 .readahead = ext4_readahead,
3576 .writepages = ext4_writepages,
3577 .write_begin = ext4_write_begin,
3578 .write_end = ext4_journalled_write_end,
3579 .dirty_folio = ext4_journalled_dirty_folio,
3581 .invalidate_folio = ext4_journalled_invalidate_folio,
3582 .release_folio = ext4_release_folio,
3583 .direct_IO = noop_direct_IO,
3584 .migrate_folio = buffer_migrate_folio_norefs,
3585 .is_partially_uptodate = block_is_partially_uptodate,
3586 .error_remove_page = generic_error_remove_page,
3587 .swap_activate = ext4_iomap_swap_activate,
3590 static const struct address_space_operations ext4_da_aops = {
3591 .read_folio = ext4_read_folio,
3592 .readahead = ext4_readahead,
3593 .writepages = ext4_writepages,
3594 .write_begin = ext4_da_write_begin,
3595 .write_end = ext4_da_write_end,
3596 .dirty_folio = ext4_dirty_folio,
3598 .invalidate_folio = ext4_invalidate_folio,
3599 .release_folio = ext4_release_folio,
3600 .direct_IO = noop_direct_IO,
3601 .migrate_folio = buffer_migrate_folio,
3602 .is_partially_uptodate = block_is_partially_uptodate,
3603 .error_remove_page = generic_error_remove_page,
3604 .swap_activate = ext4_iomap_swap_activate,
3607 static const struct address_space_operations ext4_dax_aops = {
3608 .writepages = ext4_dax_writepages,
3609 .direct_IO = noop_direct_IO,
3610 .dirty_folio = noop_dirty_folio,
3612 .swap_activate = ext4_iomap_swap_activate,
3615 void ext4_set_aops(struct inode *inode)
3617 switch (ext4_inode_journal_mode(inode)) {
3618 case EXT4_INODE_ORDERED_DATA_MODE:
3619 case EXT4_INODE_WRITEBACK_DATA_MODE:
3621 case EXT4_INODE_JOURNAL_DATA_MODE:
3622 inode->i_mapping->a_ops = &ext4_journalled_aops;
3628 inode->i_mapping->a_ops = &ext4_dax_aops;
3629 else if (test_opt(inode->i_sb, DELALLOC))
3630 inode->i_mapping->a_ops = &ext4_da_aops;
3632 inode->i_mapping->a_ops = &ext4_aops;
3635 static int __ext4_block_zero_page_range(handle_t *handle,
3636 struct address_space *mapping, loff_t from, loff_t length)
3638 ext4_fsblk_t index = from >> PAGE_SHIFT;
3639 unsigned offset = from & (PAGE_SIZE-1);
3640 unsigned blocksize, pos;
3642 struct inode *inode = mapping->host;
3643 struct buffer_head *bh;
3644 struct folio *folio;
3647 folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3648 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3649 mapping_gfp_constraint(mapping, ~__GFP_FS));
3651 return PTR_ERR(folio);
3653 blocksize = inode->i_sb->s_blocksize;
3655 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3657 bh = folio_buffers(folio);
3659 create_empty_buffers(&folio->page, blocksize, 0);
3660 bh = folio_buffers(folio);
3663 /* Find the buffer that contains "offset" */
3665 while (offset >= pos) {
3666 bh = bh->b_this_page;
3670 if (buffer_freed(bh)) {
3671 BUFFER_TRACE(bh, "freed: skip");
3674 if (!buffer_mapped(bh)) {
3675 BUFFER_TRACE(bh, "unmapped");
3676 ext4_get_block(inode, iblock, bh, 0);
3677 /* unmapped? It's a hole - nothing to do */
3678 if (!buffer_mapped(bh)) {
3679 BUFFER_TRACE(bh, "still unmapped");
3684 /* Ok, it's mapped. Make sure it's up-to-date */
3685 if (folio_test_uptodate(folio))
3686 set_buffer_uptodate(bh);
3688 if (!buffer_uptodate(bh)) {
3689 err = ext4_read_bh_lock(bh, 0, true);
3692 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3693 /* We expect the key to be set. */
3694 BUG_ON(!fscrypt_has_encryption_key(inode));
3695 err = fscrypt_decrypt_pagecache_blocks(folio,
3699 clear_buffer_uptodate(bh);
3704 if (ext4_should_journal_data(inode)) {
3705 BUFFER_TRACE(bh, "get write access");
3706 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3711 folio_zero_range(folio, offset, length);
3712 BUFFER_TRACE(bh, "zeroed end of block");
3714 if (ext4_should_journal_data(inode)) {
3715 err = ext4_dirty_journalled_data(handle, bh);
3718 mark_buffer_dirty(bh);
3719 if (ext4_should_order_data(inode))
3720 err = ext4_jbd2_inode_add_write(handle, inode, from,
3725 folio_unlock(folio);
3731 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3732 * starting from file offset 'from'. The range to be zero'd must
3733 * be contained with in one block. If the specified range exceeds
3734 * the end of the block it will be shortened to end of the block
3735 * that corresponds to 'from'
3737 static int ext4_block_zero_page_range(handle_t *handle,
3738 struct address_space *mapping, loff_t from, loff_t length)
3740 struct inode *inode = mapping->host;
3741 unsigned offset = from & (PAGE_SIZE-1);
3742 unsigned blocksize = inode->i_sb->s_blocksize;
3743 unsigned max = blocksize - (offset & (blocksize - 1));
3746 * correct length if it does not fall between
3747 * 'from' and the end of the block
3749 if (length > max || length < 0)
3752 if (IS_DAX(inode)) {
3753 return dax_zero_range(inode, from, length, NULL,
3756 return __ext4_block_zero_page_range(handle, mapping, from, length);
3760 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3761 * up to the end of the block which corresponds to `from'.
3762 * This required during truncate. We need to physically zero the tail end
3763 * of that block so it doesn't yield old data if the file is later grown.
3765 static int ext4_block_truncate_page(handle_t *handle,
3766 struct address_space *mapping, loff_t from)
3768 unsigned offset = from & (PAGE_SIZE-1);
3771 struct inode *inode = mapping->host;
3773 /* If we are processing an encrypted inode during orphan list handling */
3774 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3777 blocksize = inode->i_sb->s_blocksize;
3778 length = blocksize - (offset & (blocksize - 1));
3780 return ext4_block_zero_page_range(handle, mapping, from, length);
3783 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3784 loff_t lstart, loff_t length)
3786 struct super_block *sb = inode->i_sb;
3787 struct address_space *mapping = inode->i_mapping;
3788 unsigned partial_start, partial_end;
3789 ext4_fsblk_t start, end;
3790 loff_t byte_end = (lstart + length - 1);
3793 partial_start = lstart & (sb->s_blocksize - 1);
3794 partial_end = byte_end & (sb->s_blocksize - 1);
3796 start = lstart >> sb->s_blocksize_bits;
3797 end = byte_end >> sb->s_blocksize_bits;
3799 /* Handle partial zero within the single block */
3801 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3802 err = ext4_block_zero_page_range(handle, mapping,
3806 /* Handle partial zero out on the start of the range */
3807 if (partial_start) {
3808 err = ext4_block_zero_page_range(handle, mapping,
3809 lstart, sb->s_blocksize);
3813 /* Handle partial zero out on the end of the range */
3814 if (partial_end != sb->s_blocksize - 1)
3815 err = ext4_block_zero_page_range(handle, mapping,
3816 byte_end - partial_end,
3821 int ext4_can_truncate(struct inode *inode)
3823 if (S_ISREG(inode->i_mode))
3825 if (S_ISDIR(inode->i_mode))
3827 if (S_ISLNK(inode->i_mode))
3828 return !ext4_inode_is_fast_symlink(inode);
3833 * We have to make sure i_disksize gets properly updated before we truncate
3834 * page cache due to hole punching or zero range. Otherwise i_disksize update
3835 * can get lost as it may have been postponed to submission of writeback but
3836 * that will never happen after we truncate page cache.
3838 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3844 loff_t size = i_size_read(inode);
3846 WARN_ON(!inode_is_locked(inode));
3847 if (offset > size || offset + len < size)
3850 if (EXT4_I(inode)->i_disksize >= size)
3853 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3855 return PTR_ERR(handle);
3856 ext4_update_i_disksize(inode, size);
3857 ret = ext4_mark_inode_dirty(handle, inode);
3858 ext4_journal_stop(handle);
3863 static void ext4_wait_dax_page(struct inode *inode)
3865 filemap_invalidate_unlock(inode->i_mapping);
3867 filemap_invalidate_lock(inode->i_mapping);
3870 int ext4_break_layouts(struct inode *inode)
3875 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3879 page = dax_layout_busy_page(inode->i_mapping);
3883 error = ___wait_var_event(&page->_refcount,
3884 atomic_read(&page->_refcount) == 1,
3885 TASK_INTERRUPTIBLE, 0, 0,
3886 ext4_wait_dax_page(inode));
3887 } while (error == 0);
3893 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3894 * associated with the given offset and length
3896 * @inode: File inode
3897 * @offset: The offset where the hole will begin
3898 * @len: The length of the hole
3900 * Returns: 0 on success or negative on failure
3903 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3905 struct inode *inode = file_inode(file);
3906 struct super_block *sb = inode->i_sb;
3907 ext4_lblk_t first_block, stop_block;
3908 struct address_space *mapping = inode->i_mapping;
3909 loff_t first_block_offset, last_block_offset, max_length;
3910 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3912 unsigned int credits;
3913 int ret = 0, ret2 = 0;
3915 trace_ext4_punch_hole(inode, offset, length, 0);
3918 * Write out all dirty pages to avoid race conditions
3919 * Then release them.
3921 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3922 ret = filemap_write_and_wait_range(mapping, offset,
3923 offset + length - 1);
3930 /* No need to punch hole beyond i_size */
3931 if (offset >= inode->i_size)
3935 * If the hole extends beyond i_size, set the hole
3936 * to end after the page that contains i_size
3938 if (offset + length > inode->i_size) {
3939 length = inode->i_size +
3940 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3945 * For punch hole the length + offset needs to be within one block
3946 * before last range. Adjust the length if it goes beyond that limit.
3948 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3949 if (offset + length > max_length)
3950 length = max_length - offset;
3952 if (offset & (sb->s_blocksize - 1) ||
3953 (offset + length) & (sb->s_blocksize - 1)) {
3955 * Attach jinode to inode for jbd2 if we do any zeroing of
3958 ret = ext4_inode_attach_jinode(inode);
3964 /* Wait all existing dio workers, newcomers will block on i_rwsem */
3965 inode_dio_wait(inode);
3967 ret = file_modified(file);
3972 * Prevent page faults from reinstantiating pages we have released from
3975 filemap_invalidate_lock(mapping);
3977 ret = ext4_break_layouts(inode);
3981 first_block_offset = round_up(offset, sb->s_blocksize);
3982 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3984 /* Now release the pages and zero block aligned part of pages*/
3985 if (last_block_offset > first_block_offset) {
3986 ret = ext4_update_disksize_before_punch(inode, offset, length);
3989 truncate_pagecache_range(inode, first_block_offset,
3993 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3994 credits = ext4_writepage_trans_blocks(inode);
3996 credits = ext4_blocks_for_truncate(inode);
3997 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3998 if (IS_ERR(handle)) {
3999 ret = PTR_ERR(handle);
4000 ext4_std_error(sb, ret);
4004 ret = ext4_zero_partial_blocks(handle, inode, offset,
4009 first_block = (offset + sb->s_blocksize - 1) >>
4010 EXT4_BLOCK_SIZE_BITS(sb);
4011 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4013 /* If there are blocks to remove, do it */
4014 if (stop_block > first_block) {
4016 down_write(&EXT4_I(inode)->i_data_sem);
4017 ext4_discard_preallocations(inode, 0);
4019 ext4_es_remove_extent(inode, first_block,
4020 stop_block - first_block);
4022 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4023 ret = ext4_ext_remove_space(inode, first_block,
4026 ret = ext4_ind_remove_space(handle, inode, first_block,
4029 up_write(&EXT4_I(inode)->i_data_sem);
4031 ext4_fc_track_range(handle, inode, first_block, stop_block);
4033 ext4_handle_sync(handle);
4035 inode->i_mtime = inode_set_ctime_current(inode);
4036 ret2 = ext4_mark_inode_dirty(handle, inode);
4040 ext4_update_inode_fsync_trans(handle, inode, 1);
4042 ext4_journal_stop(handle);
4044 filemap_invalidate_unlock(mapping);
4046 inode_unlock(inode);
4050 int ext4_inode_attach_jinode(struct inode *inode)
4052 struct ext4_inode_info *ei = EXT4_I(inode);
4053 struct jbd2_inode *jinode;
4055 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4058 jinode = jbd2_alloc_inode(GFP_KERNEL);
4059 spin_lock(&inode->i_lock);
4062 spin_unlock(&inode->i_lock);
4065 ei->jinode = jinode;
4066 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4069 spin_unlock(&inode->i_lock);
4070 if (unlikely(jinode != NULL))
4071 jbd2_free_inode(jinode);
4078 * We block out ext4_get_block() block instantiations across the entire
4079 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4080 * simultaneously on behalf of the same inode.
4082 * As we work through the truncate and commit bits of it to the journal there
4083 * is one core, guiding principle: the file's tree must always be consistent on
4084 * disk. We must be able to restart the truncate after a crash.
4086 * The file's tree may be transiently inconsistent in memory (although it
4087 * probably isn't), but whenever we close off and commit a journal transaction,
4088 * the contents of (the filesystem + the journal) must be consistent and
4089 * restartable. It's pretty simple, really: bottom up, right to left (although
4090 * left-to-right works OK too).
4092 * Note that at recovery time, journal replay occurs *before* the restart of
4093 * truncate against the orphan inode list.
4095 * The committed inode has the new, desired i_size (which is the same as
4096 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4097 * that this inode's truncate did not complete and it will again call
4098 * ext4_truncate() to have another go. So there will be instantiated blocks
4099 * to the right of the truncation point in a crashed ext4 filesystem. But
4100 * that's fine - as long as they are linked from the inode, the post-crash
4101 * ext4_truncate() run will find them and release them.
4103 int ext4_truncate(struct inode *inode)
4105 struct ext4_inode_info *ei = EXT4_I(inode);
4106 unsigned int credits;
4109 struct address_space *mapping = inode->i_mapping;
4112 * There is a possibility that we're either freeing the inode
4113 * or it's a completely new inode. In those cases we might not
4114 * have i_rwsem locked because it's not necessary.
4116 if (!(inode->i_state & (I_NEW|I_FREEING)))
4117 WARN_ON(!inode_is_locked(inode));
4118 trace_ext4_truncate_enter(inode);
4120 if (!ext4_can_truncate(inode))
4123 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4124 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4126 if (ext4_has_inline_data(inode)) {
4129 err = ext4_inline_data_truncate(inode, &has_inline);
4130 if (err || has_inline)
4134 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4135 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4136 err = ext4_inode_attach_jinode(inode);
4141 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4142 credits = ext4_writepage_trans_blocks(inode);
4144 credits = ext4_blocks_for_truncate(inode);
4146 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4147 if (IS_ERR(handle)) {
4148 err = PTR_ERR(handle);
4152 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4153 ext4_block_truncate_page(handle, mapping, inode->i_size);
4156 * We add the inode to the orphan list, so that if this
4157 * truncate spans multiple transactions, and we crash, we will
4158 * resume the truncate when the filesystem recovers. It also
4159 * marks the inode dirty, to catch the new size.
4161 * Implication: the file must always be in a sane, consistent
4162 * truncatable state while each transaction commits.
4164 err = ext4_orphan_add(handle, inode);
4168 down_write(&EXT4_I(inode)->i_data_sem);
4170 ext4_discard_preallocations(inode, 0);
4172 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4173 err = ext4_ext_truncate(handle, inode);
4175 ext4_ind_truncate(handle, inode);
4177 up_write(&ei->i_data_sem);
4182 ext4_handle_sync(handle);
4186 * If this was a simple ftruncate() and the file will remain alive,
4187 * then we need to clear up the orphan record which we created above.
4188 * However, if this was a real unlink then we were called by
4189 * ext4_evict_inode(), and we allow that function to clean up the
4190 * orphan info for us.
4193 ext4_orphan_del(handle, inode);
4195 inode->i_mtime = inode_set_ctime_current(inode);
4196 err2 = ext4_mark_inode_dirty(handle, inode);
4197 if (unlikely(err2 && !err))
4199 ext4_journal_stop(handle);
4202 trace_ext4_truncate_exit(inode);
4206 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4208 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4209 return inode_peek_iversion_raw(inode);
4211 return inode_peek_iversion(inode);
4214 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4215 struct ext4_inode_info *ei)
4217 struct inode *inode = &(ei->vfs_inode);
4218 u64 i_blocks = READ_ONCE(inode->i_blocks);
4219 struct super_block *sb = inode->i_sb;
4221 if (i_blocks <= ~0U) {
4223 * i_blocks can be represented in a 32 bit variable
4224 * as multiple of 512 bytes
4226 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4227 raw_inode->i_blocks_high = 0;
4228 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4233 * This should never happen since sb->s_maxbytes should not have
4234 * allowed this, sb->s_maxbytes was set according to the huge_file
4235 * feature in ext4_fill_super().
4237 if (!ext4_has_feature_huge_file(sb))
4238 return -EFSCORRUPTED;
4240 if (i_blocks <= 0xffffffffffffULL) {
4242 * i_blocks can be represented in a 48 bit variable
4243 * as multiple of 512 bytes
4245 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4246 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4247 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4249 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4250 /* i_block is stored in file system block size */
4251 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4252 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4253 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4258 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4260 struct ext4_inode_info *ei = EXT4_I(inode);
4267 err = ext4_inode_blocks_set(raw_inode, ei);
4269 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4270 i_uid = i_uid_read(inode);
4271 i_gid = i_gid_read(inode);
4272 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4273 if (!(test_opt(inode->i_sb, NO_UID32))) {
4274 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4275 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4277 * Fix up interoperability with old kernels. Otherwise,
4278 * old inodes get re-used with the upper 16 bits of the
4281 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4282 raw_inode->i_uid_high = 0;
4283 raw_inode->i_gid_high = 0;
4285 raw_inode->i_uid_high =
4286 cpu_to_le16(high_16_bits(i_uid));
4287 raw_inode->i_gid_high =
4288 cpu_to_le16(high_16_bits(i_gid));
4291 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4292 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4293 raw_inode->i_uid_high = 0;
4294 raw_inode->i_gid_high = 0;
4296 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4298 EXT4_INODE_SET_CTIME(inode, raw_inode);
4299 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4300 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4301 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4303 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4304 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4305 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4306 raw_inode->i_file_acl_high =
4307 cpu_to_le16(ei->i_file_acl >> 32);
4308 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4309 ext4_isize_set(raw_inode, ei->i_disksize);
4311 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4312 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4313 if (old_valid_dev(inode->i_rdev)) {
4314 raw_inode->i_block[0] =
4315 cpu_to_le32(old_encode_dev(inode->i_rdev));
4316 raw_inode->i_block[1] = 0;
4318 raw_inode->i_block[0] = 0;
4319 raw_inode->i_block[1] =
4320 cpu_to_le32(new_encode_dev(inode->i_rdev));
4321 raw_inode->i_block[2] = 0;
4323 } else if (!ext4_has_inline_data(inode)) {
4324 for (block = 0; block < EXT4_N_BLOCKS; block++)
4325 raw_inode->i_block[block] = ei->i_data[block];
4328 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4329 u64 ivers = ext4_inode_peek_iversion(inode);
4331 raw_inode->i_disk_version = cpu_to_le32(ivers);
4332 if (ei->i_extra_isize) {
4333 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4334 raw_inode->i_version_hi =
4335 cpu_to_le32(ivers >> 32);
4336 raw_inode->i_extra_isize =
4337 cpu_to_le16(ei->i_extra_isize);
4341 if (i_projid != EXT4_DEF_PROJID &&
4342 !ext4_has_feature_project(inode->i_sb))
4343 err = err ?: -EFSCORRUPTED;
4345 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4346 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4347 raw_inode->i_projid = cpu_to_le32(i_projid);
4349 ext4_inode_csum_set(inode, raw_inode, ei);
4354 * ext4_get_inode_loc returns with an extra refcount against the inode's
4355 * underlying buffer_head on success. If we pass 'inode' and it does not
4356 * have in-inode xattr, we have all inode data in memory that is needed
4357 * to recreate the on-disk version of this inode.
4359 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4360 struct inode *inode, struct ext4_iloc *iloc,
4361 ext4_fsblk_t *ret_block)
4363 struct ext4_group_desc *gdp;
4364 struct buffer_head *bh;
4366 struct blk_plug plug;
4367 int inodes_per_block, inode_offset;
4370 if (ino < EXT4_ROOT_INO ||
4371 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4372 return -EFSCORRUPTED;
4374 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4375 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4380 * Figure out the offset within the block group inode table
4382 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4383 inode_offset = ((ino - 1) %
4384 EXT4_INODES_PER_GROUP(sb));
4385 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4387 block = ext4_inode_table(sb, gdp);
4388 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4389 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4390 ext4_error(sb, "Invalid inode table block %llu in "
4391 "block_group %u", block, iloc->block_group);
4392 return -EFSCORRUPTED;
4394 block += (inode_offset / inodes_per_block);
4396 bh = sb_getblk(sb, block);
4399 if (ext4_buffer_uptodate(bh))
4403 if (ext4_buffer_uptodate(bh)) {
4404 /* Someone brought it uptodate while we waited */
4410 * If we have all information of the inode in memory and this
4411 * is the only valid inode in the block, we need not read the
4414 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4415 struct buffer_head *bitmap_bh;
4418 start = inode_offset & ~(inodes_per_block - 1);
4420 /* Is the inode bitmap in cache? */
4421 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4422 if (unlikely(!bitmap_bh))
4426 * If the inode bitmap isn't in cache then the
4427 * optimisation may end up performing two reads instead
4428 * of one, so skip it.
4430 if (!buffer_uptodate(bitmap_bh)) {
4434 for (i = start; i < start + inodes_per_block; i++) {
4435 if (i == inode_offset)
4437 if (ext4_test_bit(i, bitmap_bh->b_data))
4441 if (i == start + inodes_per_block) {
4442 struct ext4_inode *raw_inode =
4443 (struct ext4_inode *) (bh->b_data + iloc->offset);
4445 /* all other inodes are free, so skip I/O */
4446 memset(bh->b_data, 0, bh->b_size);
4447 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4448 ext4_fill_raw_inode(inode, raw_inode);
4449 set_buffer_uptodate(bh);
4457 * If we need to do any I/O, try to pre-readahead extra
4458 * blocks from the inode table.
4460 blk_start_plug(&plug);
4461 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4462 ext4_fsblk_t b, end, table;
4464 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4466 table = ext4_inode_table(sb, gdp);
4467 /* s_inode_readahead_blks is always a power of 2 */
4468 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4472 num = EXT4_INODES_PER_GROUP(sb);
4473 if (ext4_has_group_desc_csum(sb))
4474 num -= ext4_itable_unused_count(sb, gdp);
4475 table += num / inodes_per_block;
4479 ext4_sb_breadahead_unmovable(sb, b++);
4483 * There are other valid inodes in the buffer, this inode
4484 * has in-inode xattrs, or we don't have this inode in memory.
4485 * Read the block from disk.
4487 trace_ext4_load_inode(sb, ino);
4488 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4489 blk_finish_plug(&plug);
4491 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4492 if (!buffer_uptodate(bh)) {
4503 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4504 struct ext4_iloc *iloc)
4506 ext4_fsblk_t err_blk = 0;
4509 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4513 ext4_error_inode_block(inode, err_blk, EIO,
4514 "unable to read itable block");
4519 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4521 ext4_fsblk_t err_blk = 0;
4524 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4528 ext4_error_inode_block(inode, err_blk, EIO,
4529 "unable to read itable block");
4535 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4536 struct ext4_iloc *iloc)
4538 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4541 static bool ext4_should_enable_dax(struct inode *inode)
4543 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4545 if (test_opt2(inode->i_sb, DAX_NEVER))
4547 if (!S_ISREG(inode->i_mode))
4549 if (ext4_should_journal_data(inode))
4551 if (ext4_has_inline_data(inode))
4553 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4555 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4557 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4559 if (test_opt(inode->i_sb, DAX_ALWAYS))
4562 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4565 void ext4_set_inode_flags(struct inode *inode, bool init)
4567 unsigned int flags = EXT4_I(inode)->i_flags;
4568 unsigned int new_fl = 0;
4570 WARN_ON_ONCE(IS_DAX(inode) && init);
4572 if (flags & EXT4_SYNC_FL)
4574 if (flags & EXT4_APPEND_FL)
4576 if (flags & EXT4_IMMUTABLE_FL)
4577 new_fl |= S_IMMUTABLE;
4578 if (flags & EXT4_NOATIME_FL)
4579 new_fl |= S_NOATIME;
4580 if (flags & EXT4_DIRSYNC_FL)
4581 new_fl |= S_DIRSYNC;
4583 /* Because of the way inode_set_flags() works we must preserve S_DAX
4584 * here if already set. */
4585 new_fl |= (inode->i_flags & S_DAX);
4586 if (init && ext4_should_enable_dax(inode))
4589 if (flags & EXT4_ENCRYPT_FL)
4590 new_fl |= S_ENCRYPTED;
4591 if (flags & EXT4_CASEFOLD_FL)
4592 new_fl |= S_CASEFOLD;
4593 if (flags & EXT4_VERITY_FL)
4595 inode_set_flags(inode, new_fl,
4596 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4597 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4600 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4601 struct ext4_inode_info *ei)
4604 struct inode *inode = &(ei->vfs_inode);
4605 struct super_block *sb = inode->i_sb;
4607 if (ext4_has_feature_huge_file(sb)) {
4608 /* we are using combined 48 bit field */
4609 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4610 le32_to_cpu(raw_inode->i_blocks_lo);
4611 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4612 /* i_blocks represent file system block size */
4613 return i_blocks << (inode->i_blkbits - 9);
4618 return le32_to_cpu(raw_inode->i_blocks_lo);
4622 static inline int ext4_iget_extra_inode(struct inode *inode,
4623 struct ext4_inode *raw_inode,
4624 struct ext4_inode_info *ei)
4626 __le32 *magic = (void *)raw_inode +
4627 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4629 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4630 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4633 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4634 err = ext4_find_inline_data_nolock(inode);
4635 if (!err && ext4_has_inline_data(inode))
4636 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4639 EXT4_I(inode)->i_inline_off = 0;
4643 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4645 if (!ext4_has_feature_project(inode->i_sb))
4647 *projid = EXT4_I(inode)->i_projid;
4652 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4653 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4656 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4658 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4659 inode_set_iversion_raw(inode, val);
4661 inode_set_iversion_queried(inode, val);
4664 static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4667 if (flags & EXT4_IGET_EA_INODE) {
4668 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4669 return "missing EA_INODE flag";
4670 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4671 EXT4_I(inode)->i_file_acl)
4672 return "ea_inode with extended attributes";
4674 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4675 return "unexpected EA_INODE flag";
4677 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4678 return "unexpected bad inode w/o EXT4_IGET_BAD";
4682 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4683 ext4_iget_flags flags, const char *function,
4686 struct ext4_iloc iloc;
4687 struct ext4_inode *raw_inode;
4688 struct ext4_inode_info *ei;
4689 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4690 struct inode *inode;
4691 const char *err_str;
4692 journal_t *journal = EXT4_SB(sb)->s_journal;
4700 if ((!(flags & EXT4_IGET_SPECIAL) &&
4701 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4702 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4703 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4704 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4705 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4706 (ino < EXT4_ROOT_INO) ||
4707 (ino > le32_to_cpu(es->s_inodes_count))) {
4708 if (flags & EXT4_IGET_HANDLE)
4709 return ERR_PTR(-ESTALE);
4710 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4711 "inode #%lu: comm %s: iget: illegal inode #",
4712 ino, current->comm);
4713 return ERR_PTR(-EFSCORRUPTED);
4716 inode = iget_locked(sb, ino);
4718 return ERR_PTR(-ENOMEM);
4719 if (!(inode->i_state & I_NEW)) {
4720 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4721 ext4_error_inode(inode, function, line, 0, err_str);
4723 return ERR_PTR(-EFSCORRUPTED);
4731 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4734 raw_inode = ext4_raw_inode(&iloc);
4736 if ((flags & EXT4_IGET_HANDLE) &&
4737 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4742 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4743 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4744 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4745 EXT4_INODE_SIZE(inode->i_sb) ||
4746 (ei->i_extra_isize & 3)) {
4747 ext4_error_inode(inode, function, line, 0,
4748 "iget: bad extra_isize %u "
4751 EXT4_INODE_SIZE(inode->i_sb));
4752 ret = -EFSCORRUPTED;
4756 ei->i_extra_isize = 0;
4758 /* Precompute checksum seed for inode metadata */
4759 if (ext4_has_metadata_csum(sb)) {
4760 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4762 __le32 inum = cpu_to_le32(inode->i_ino);
4763 __le32 gen = raw_inode->i_generation;
4764 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4766 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4770 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4771 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4772 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4773 ext4_error_inode_err(inode, function, line, 0,
4774 EFSBADCRC, "iget: checksum invalid");
4779 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4780 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4781 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4782 if (ext4_has_feature_project(sb) &&
4783 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4784 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4785 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4787 i_projid = EXT4_DEF_PROJID;
4789 if (!(test_opt(inode->i_sb, NO_UID32))) {
4790 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4791 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4793 i_uid_write(inode, i_uid);
4794 i_gid_write(inode, i_gid);
4795 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4796 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4798 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4799 ei->i_inline_off = 0;
4800 ei->i_dir_start_lookup = 0;
4801 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4802 /* We now have enough fields to check if the inode was active or not.
4803 * This is needed because nfsd might try to access dead inodes
4804 * the test is that same one that e2fsck uses
4805 * NeilBrown 1999oct15
4807 if (inode->i_nlink == 0) {
4808 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4809 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4810 ino != EXT4_BOOT_LOADER_INO) {
4811 /* this inode is deleted or unallocated */
4812 if (flags & EXT4_IGET_SPECIAL) {
4813 ext4_error_inode(inode, function, line, 0,
4814 "iget: special inode unallocated");
4815 ret = -EFSCORRUPTED;
4820 /* The only unlinked inodes we let through here have
4821 * valid i_mode and are being read by the orphan
4822 * recovery code: that's fine, we're about to complete
4823 * the process of deleting those.
4824 * OR it is the EXT4_BOOT_LOADER_INO which is
4825 * not initialized on a new filesystem. */
4827 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4828 ext4_set_inode_flags(inode, true);
4829 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4830 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4831 if (ext4_has_feature_64bit(sb))
4833 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4834 inode->i_size = ext4_isize(sb, raw_inode);
4835 if ((size = i_size_read(inode)) < 0) {
4836 ext4_error_inode(inode, function, line, 0,
4837 "iget: bad i_size value: %lld", size);
4838 ret = -EFSCORRUPTED;
4842 * If dir_index is not enabled but there's dir with INDEX flag set,
4843 * we'd normally treat htree data as empty space. But with metadata
4844 * checksumming that corrupts checksums so forbid that.
4846 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4847 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4848 ext4_error_inode(inode, function, line, 0,
4849 "iget: Dir with htree data on filesystem without dir_index feature.");
4850 ret = -EFSCORRUPTED;
4853 ei->i_disksize = inode->i_size;
4855 ei->i_reserved_quota = 0;
4857 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4858 ei->i_block_group = iloc.block_group;
4859 ei->i_last_alloc_group = ~0;
4861 * NOTE! The in-memory inode i_data array is in little-endian order
4862 * even on big-endian machines: we do NOT byteswap the block numbers!
4864 for (block = 0; block < EXT4_N_BLOCKS; block++)
4865 ei->i_data[block] = raw_inode->i_block[block];
4866 INIT_LIST_HEAD(&ei->i_orphan);
4867 ext4_fc_init_inode(&ei->vfs_inode);
4870 * Set transaction id's of transactions that have to be committed
4871 * to finish f[data]sync. We set them to currently running transaction
4872 * as we cannot be sure that the inode or some of its metadata isn't
4873 * part of the transaction - the inode could have been reclaimed and
4874 * now it is reread from disk.
4877 transaction_t *transaction;
4880 read_lock(&journal->j_state_lock);
4881 if (journal->j_running_transaction)
4882 transaction = journal->j_running_transaction;
4884 transaction = journal->j_committing_transaction;
4886 tid = transaction->t_tid;
4888 tid = journal->j_commit_sequence;
4889 read_unlock(&journal->j_state_lock);
4890 ei->i_sync_tid = tid;
4891 ei->i_datasync_tid = tid;
4894 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4895 if (ei->i_extra_isize == 0) {
4896 /* The extra space is currently unused. Use it. */
4897 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4898 ei->i_extra_isize = sizeof(struct ext4_inode) -
4899 EXT4_GOOD_OLD_INODE_SIZE;
4901 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4907 EXT4_INODE_GET_CTIME(inode, raw_inode);
4908 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4909 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4910 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4912 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4913 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4915 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4916 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4918 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4920 ext4_inode_set_iversion_queried(inode, ivers);
4924 if (ei->i_file_acl &&
4925 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4926 ext4_error_inode(inode, function, line, 0,
4927 "iget: bad extended attribute block %llu",
4929 ret = -EFSCORRUPTED;
4931 } else if (!ext4_has_inline_data(inode)) {
4932 /* validate the block references in the inode */
4933 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4934 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4935 (S_ISLNK(inode->i_mode) &&
4936 !ext4_inode_is_fast_symlink(inode)))) {
4937 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4938 ret = ext4_ext_check_inode(inode);
4940 ret = ext4_ind_check_inode(inode);
4946 if (S_ISREG(inode->i_mode)) {
4947 inode->i_op = &ext4_file_inode_operations;
4948 inode->i_fop = &ext4_file_operations;
4949 ext4_set_aops(inode);
4950 } else if (S_ISDIR(inode->i_mode)) {
4951 inode->i_op = &ext4_dir_inode_operations;
4952 inode->i_fop = &ext4_dir_operations;
4953 } else if (S_ISLNK(inode->i_mode)) {
4954 /* VFS does not allow setting these so must be corruption */
4955 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4956 ext4_error_inode(inode, function, line, 0,
4957 "iget: immutable or append flags "
4958 "not allowed on symlinks");
4959 ret = -EFSCORRUPTED;
4962 if (IS_ENCRYPTED(inode)) {
4963 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4964 } else if (ext4_inode_is_fast_symlink(inode)) {
4965 inode->i_link = (char *)ei->i_data;
4966 inode->i_op = &ext4_fast_symlink_inode_operations;
4967 nd_terminate_link(ei->i_data, inode->i_size,
4968 sizeof(ei->i_data) - 1);
4970 inode->i_op = &ext4_symlink_inode_operations;
4972 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4973 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4974 inode->i_op = &ext4_special_inode_operations;
4975 if (raw_inode->i_block[0])
4976 init_special_inode(inode, inode->i_mode,
4977 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4979 init_special_inode(inode, inode->i_mode,
4980 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4981 } else if (ino == EXT4_BOOT_LOADER_INO) {
4982 make_bad_inode(inode);
4984 ret = -EFSCORRUPTED;
4985 ext4_error_inode(inode, function, line, 0,
4986 "iget: bogus i_mode (%o)", inode->i_mode);
4989 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) {
4990 ext4_error_inode(inode, function, line, 0,
4991 "casefold flag without casefold feature");
4992 ret = -EFSCORRUPTED;
4995 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4996 ext4_error_inode(inode, function, line, 0, err_str);
4997 ret = -EFSCORRUPTED;
5002 unlock_new_inode(inode);
5008 return ERR_PTR(ret);
5011 static void __ext4_update_other_inode_time(struct super_block *sb,
5012 unsigned long orig_ino,
5014 struct ext4_inode *raw_inode)
5016 struct inode *inode;
5018 inode = find_inode_by_ino_rcu(sb, ino);
5022 if (!inode_is_dirtytime_only(inode))
5025 spin_lock(&inode->i_lock);
5026 if (inode_is_dirtytime_only(inode)) {
5027 struct ext4_inode_info *ei = EXT4_I(inode);
5029 inode->i_state &= ~I_DIRTY_TIME;
5030 spin_unlock(&inode->i_lock);
5032 spin_lock(&ei->i_raw_lock);
5033 EXT4_INODE_SET_CTIME(inode, raw_inode);
5034 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5035 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5036 ext4_inode_csum_set(inode, raw_inode, ei);
5037 spin_unlock(&ei->i_raw_lock);
5038 trace_ext4_other_inode_update_time(inode, orig_ino);
5041 spin_unlock(&inode->i_lock);
5045 * Opportunistically update the other time fields for other inodes in
5046 * the same inode table block.
5048 static void ext4_update_other_inodes_time(struct super_block *sb,
5049 unsigned long orig_ino, char *buf)
5052 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5053 int inode_size = EXT4_INODE_SIZE(sb);
5056 * Calculate the first inode in the inode table block. Inode
5057 * numbers are one-based. That is, the first inode in a block
5058 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5060 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5062 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5063 if (ino == orig_ino)
5065 __ext4_update_other_inode_time(sb, orig_ino, ino,
5066 (struct ext4_inode *)buf);
5072 * Post the struct inode info into an on-disk inode location in the
5073 * buffer-cache. This gobbles the caller's reference to the
5074 * buffer_head in the inode location struct.
5076 * The caller must have write access to iloc->bh.
5078 static int ext4_do_update_inode(handle_t *handle,
5079 struct inode *inode,
5080 struct ext4_iloc *iloc)
5082 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5083 struct ext4_inode_info *ei = EXT4_I(inode);
5084 struct buffer_head *bh = iloc->bh;
5085 struct super_block *sb = inode->i_sb;
5087 int need_datasync = 0, set_large_file = 0;
5089 spin_lock(&ei->i_raw_lock);
5092 * For fields not tracked in the in-memory inode, initialise them
5093 * to zero for new inodes.
5095 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5096 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5098 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5100 if (ei->i_disksize > 0x7fffffffULL) {
5101 if (!ext4_has_feature_large_file(sb) ||
5102 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5106 err = ext4_fill_raw_inode(inode, raw_inode);
5107 spin_unlock(&ei->i_raw_lock);
5109 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5113 if (inode->i_sb->s_flags & SB_LAZYTIME)
5114 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5117 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5118 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5121 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5122 if (set_large_file) {
5123 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5124 err = ext4_journal_get_write_access(handle, sb,
5129 lock_buffer(EXT4_SB(sb)->s_sbh);
5130 ext4_set_feature_large_file(sb);
5131 ext4_superblock_csum_set(sb);
5132 unlock_buffer(EXT4_SB(sb)->s_sbh);
5133 ext4_handle_sync(handle);
5134 err = ext4_handle_dirty_metadata(handle, NULL,
5135 EXT4_SB(sb)->s_sbh);
5137 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5139 ext4_std_error(inode->i_sb, err);
5146 * ext4_write_inode()
5148 * We are called from a few places:
5150 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5151 * Here, there will be no transaction running. We wait for any running
5152 * transaction to commit.
5154 * - Within flush work (sys_sync(), kupdate and such).
5155 * We wait on commit, if told to.
5157 * - Within iput_final() -> write_inode_now()
5158 * We wait on commit, if told to.
5160 * In all cases it is actually safe for us to return without doing anything,
5161 * because the inode has been copied into a raw inode buffer in
5162 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5165 * Note that we are absolutely dependent upon all inode dirtiers doing the
5166 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5167 * which we are interested.
5169 * It would be a bug for them to not do this. The code:
5171 * mark_inode_dirty(inode)
5173 * inode->i_size = expr;
5175 * is in error because write_inode() could occur while `stuff()' is running,
5176 * and the new i_size will be lost. Plus the inode will no longer be on the
5177 * superblock's dirty inode list.
5179 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5183 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5186 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5189 if (EXT4_SB(inode->i_sb)->s_journal) {
5190 if (ext4_journal_current_handle()) {
5191 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5197 * No need to force transaction in WB_SYNC_NONE mode. Also
5198 * ext4_sync_fs() will force the commit after everything is
5201 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5204 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5205 EXT4_I(inode)->i_sync_tid);
5207 struct ext4_iloc iloc;
5209 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5213 * sync(2) will flush the whole buffer cache. No need to do
5214 * it here separately for each inode.
5216 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5217 sync_dirty_buffer(iloc.bh);
5218 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5219 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5220 "IO error syncing inode");
5229 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5230 * buffers that are attached to a folio straddling i_size and are undergoing
5231 * commit. In that case we have to wait for commit to finish and try again.
5233 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5236 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5237 tid_t commit_tid = 0;
5240 offset = inode->i_size & (PAGE_SIZE - 1);
5242 * If the folio is fully truncated, we don't need to wait for any commit
5243 * (and we even should not as __ext4_journalled_invalidate_folio() may
5244 * strip all buffers from the folio but keep the folio dirty which can then
5245 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5246 * buffers). Also we don't need to wait for any commit if all buffers in
5247 * the folio remain valid. This is most beneficial for the common case of
5248 * blocksize == PAGESIZE.
5250 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5253 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5254 inode->i_size >> PAGE_SHIFT);
5257 ret = __ext4_journalled_invalidate_folio(folio, offset,
5258 folio_size(folio) - offset);
5259 folio_unlock(folio);
5264 read_lock(&journal->j_state_lock);
5265 if (journal->j_committing_transaction)
5266 commit_tid = journal->j_committing_transaction->t_tid;
5267 read_unlock(&journal->j_state_lock);
5269 jbd2_log_wait_commit(journal, commit_tid);
5276 * Called from notify_change.
5278 * We want to trap VFS attempts to truncate the file as soon as
5279 * possible. In particular, we want to make sure that when the VFS
5280 * shrinks i_size, we put the inode on the orphan list and modify
5281 * i_disksize immediately, so that during the subsequent flushing of
5282 * dirty pages and freeing of disk blocks, we can guarantee that any
5283 * commit will leave the blocks being flushed in an unused state on
5284 * disk. (On recovery, the inode will get truncated and the blocks will
5285 * be freed, so we have a strong guarantee that no future commit will
5286 * leave these blocks visible to the user.)
5288 * Another thing we have to assure is that if we are in ordered mode
5289 * and inode is still attached to the committing transaction, we must
5290 * we start writeout of all the dirty pages which are being truncated.
5291 * This way we are sure that all the data written in the previous
5292 * transaction are already on disk (truncate waits for pages under
5295 * Called with inode->i_rwsem down.
5297 int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5300 struct inode *inode = d_inode(dentry);
5303 const unsigned int ia_valid = attr->ia_valid;
5304 bool inc_ivers = true;
5306 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5309 if (unlikely(IS_IMMUTABLE(inode)))
5312 if (unlikely(IS_APPEND(inode) &&
5313 (ia_valid & (ATTR_MODE | ATTR_UID |
5314 ATTR_GID | ATTR_TIMES_SET))))
5317 error = setattr_prepare(idmap, dentry, attr);
5321 error = fscrypt_prepare_setattr(dentry, attr);
5325 error = fsverity_prepare_setattr(dentry, attr);
5329 if (is_quota_modification(idmap, inode, attr)) {
5330 error = dquot_initialize(inode);
5335 if (i_uid_needs_update(idmap, attr, inode) ||
5336 i_gid_needs_update(idmap, attr, inode)) {
5339 /* (user+group)*(old+new) structure, inode write (sb,
5340 * inode block, ? - but truncate inode update has it) */
5341 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5342 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5343 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5344 if (IS_ERR(handle)) {
5345 error = PTR_ERR(handle);
5349 /* dquot_transfer() calls back ext4_get_inode_usage() which
5350 * counts xattr inode references.
5352 down_read(&EXT4_I(inode)->xattr_sem);
5353 error = dquot_transfer(idmap, inode, attr);
5354 up_read(&EXT4_I(inode)->xattr_sem);
5357 ext4_journal_stop(handle);
5360 /* Update corresponding info in inode so that everything is in
5361 * one transaction */
5362 i_uid_update(idmap, attr, inode);
5363 i_gid_update(idmap, attr, inode);
5364 error = ext4_mark_inode_dirty(handle, inode);
5365 ext4_journal_stop(handle);
5366 if (unlikely(error)) {
5371 if (attr->ia_valid & ATTR_SIZE) {
5373 loff_t oldsize = inode->i_size;
5374 loff_t old_disksize;
5375 int shrink = (attr->ia_size < inode->i_size);
5377 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5378 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5380 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5384 if (!S_ISREG(inode->i_mode)) {
5388 if (attr->ia_size == inode->i_size)
5392 if (ext4_should_order_data(inode)) {
5393 error = ext4_begin_ordered_truncate(inode,
5399 * Blocks are going to be removed from the inode. Wait
5400 * for dio in flight.
5402 inode_dio_wait(inode);
5405 filemap_invalidate_lock(inode->i_mapping);
5407 rc = ext4_break_layouts(inode);
5409 filemap_invalidate_unlock(inode->i_mapping);
5413 if (attr->ia_size != inode->i_size) {
5414 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5415 if (IS_ERR(handle)) {
5416 error = PTR_ERR(handle);
5419 if (ext4_handle_valid(handle) && shrink) {
5420 error = ext4_orphan_add(handle, inode);
5424 * Update c/mtime on truncate up, ext4_truncate() will
5425 * update c/mtime in shrink case below
5428 inode->i_mtime = inode_set_ctime_current(inode);
5431 ext4_fc_track_range(handle, inode,
5432 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5433 inode->i_sb->s_blocksize_bits,
5434 EXT_MAX_BLOCKS - 1);
5436 ext4_fc_track_range(
5438 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5439 inode->i_sb->s_blocksize_bits,
5440 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5441 inode->i_sb->s_blocksize_bits);
5443 down_write(&EXT4_I(inode)->i_data_sem);
5444 old_disksize = EXT4_I(inode)->i_disksize;
5445 EXT4_I(inode)->i_disksize = attr->ia_size;
5446 rc = ext4_mark_inode_dirty(handle, inode);
5450 * We have to update i_size under i_data_sem together
5451 * with i_disksize to avoid races with writeback code
5452 * running ext4_wb_update_i_disksize().
5455 i_size_write(inode, attr->ia_size);
5457 EXT4_I(inode)->i_disksize = old_disksize;
5458 up_write(&EXT4_I(inode)->i_data_sem);
5459 ext4_journal_stop(handle);
5463 pagecache_isize_extended(inode, oldsize,
5465 } else if (ext4_should_journal_data(inode)) {
5466 ext4_wait_for_tail_page_commit(inode);
5471 * Truncate pagecache after we've waited for commit
5472 * in data=journal mode to make pages freeable.
5474 truncate_pagecache(inode, inode->i_size);
5476 * Call ext4_truncate() even if i_size didn't change to
5477 * truncate possible preallocated blocks.
5479 if (attr->ia_size <= oldsize) {
5480 rc = ext4_truncate(inode);
5485 filemap_invalidate_unlock(inode->i_mapping);
5490 inode_inc_iversion(inode);
5491 setattr_copy(idmap, inode, attr);
5492 mark_inode_dirty(inode);
5496 * If the call to ext4_truncate failed to get a transaction handle at
5497 * all, we need to clean up the in-core orphan list manually.
5499 if (orphan && inode->i_nlink)
5500 ext4_orphan_del(NULL, inode);
5502 if (!error && (ia_valid & ATTR_MODE))
5503 rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5507 ext4_std_error(inode->i_sb, error);
5513 u32 ext4_dio_alignment(struct inode *inode)
5515 if (fsverity_active(inode))
5517 if (ext4_should_journal_data(inode))
5519 if (ext4_has_inline_data(inode))
5521 if (IS_ENCRYPTED(inode)) {
5522 if (!fscrypt_dio_supported(inode))
5524 return i_blocksize(inode);
5526 return 1; /* use the iomap defaults */
5529 int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5530 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5532 struct inode *inode = d_inode(path->dentry);
5533 struct ext4_inode *raw_inode;
5534 struct ext4_inode_info *ei = EXT4_I(inode);
5537 if ((request_mask & STATX_BTIME) &&
5538 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5539 stat->result_mask |= STATX_BTIME;
5540 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5541 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5545 * Return the DIO alignment restrictions if requested. We only return
5546 * this information when requested, since on encrypted files it might
5547 * take a fair bit of work to get if the file wasn't opened recently.
5549 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5550 u32 dio_align = ext4_dio_alignment(inode);
5552 stat->result_mask |= STATX_DIOALIGN;
5553 if (dio_align == 1) {
5554 struct block_device *bdev = inode->i_sb->s_bdev;
5556 /* iomap defaults */
5557 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5558 stat->dio_offset_align = bdev_logical_block_size(bdev);
5560 stat->dio_mem_align = dio_align;
5561 stat->dio_offset_align = dio_align;
5565 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5566 if (flags & EXT4_APPEND_FL)
5567 stat->attributes |= STATX_ATTR_APPEND;
5568 if (flags & EXT4_COMPR_FL)
5569 stat->attributes |= STATX_ATTR_COMPRESSED;
5570 if (flags & EXT4_ENCRYPT_FL)
5571 stat->attributes |= STATX_ATTR_ENCRYPTED;
5572 if (flags & EXT4_IMMUTABLE_FL)
5573 stat->attributes |= STATX_ATTR_IMMUTABLE;
5574 if (flags & EXT4_NODUMP_FL)
5575 stat->attributes |= STATX_ATTR_NODUMP;
5576 if (flags & EXT4_VERITY_FL)
5577 stat->attributes |= STATX_ATTR_VERITY;
5579 stat->attributes_mask |= (STATX_ATTR_APPEND |
5580 STATX_ATTR_COMPRESSED |
5581 STATX_ATTR_ENCRYPTED |
5582 STATX_ATTR_IMMUTABLE |
5586 generic_fillattr(idmap, request_mask, inode, stat);
5590 int ext4_file_getattr(struct mnt_idmap *idmap,
5591 const struct path *path, struct kstat *stat,
5592 u32 request_mask, unsigned int query_flags)
5594 struct inode *inode = d_inode(path->dentry);
5595 u64 delalloc_blocks;
5597 ext4_getattr(idmap, path, stat, request_mask, query_flags);
5600 * If there is inline data in the inode, the inode will normally not
5601 * have data blocks allocated (it may have an external xattr block).
5602 * Report at least one sector for such files, so tools like tar, rsync,
5603 * others don't incorrectly think the file is completely sparse.
5605 if (unlikely(ext4_has_inline_data(inode)))
5606 stat->blocks += (stat->size + 511) >> 9;
5609 * We can't update i_blocks if the block allocation is delayed
5610 * otherwise in the case of system crash before the real block
5611 * allocation is done, we will have i_blocks inconsistent with
5612 * on-disk file blocks.
5613 * We always keep i_blocks updated together with real
5614 * allocation. But to not confuse with user, stat
5615 * will return the blocks that include the delayed allocation
5616 * blocks for this file.
5618 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5619 EXT4_I(inode)->i_reserved_data_blocks);
5620 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5624 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5627 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5628 return ext4_ind_trans_blocks(inode, lblocks);
5629 return ext4_ext_index_trans_blocks(inode, pextents);
5633 * Account for index blocks, block groups bitmaps and block group
5634 * descriptor blocks if modify datablocks and index blocks
5635 * worse case, the indexs blocks spread over different block groups
5637 * If datablocks are discontiguous, they are possible to spread over
5638 * different block groups too. If they are contiguous, with flexbg,
5639 * they could still across block group boundary.
5641 * Also account for superblock, inode, quota and xattr blocks
5643 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5646 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5652 * How many index blocks need to touch to map @lblocks logical blocks
5653 * to @pextents physical extents?
5655 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5660 * Now let's see how many group bitmaps and group descriptors need
5663 groups = idxblocks + pextents;
5665 if (groups > ngroups)
5667 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5668 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5670 /* bitmaps and block group descriptor blocks */
5671 ret += groups + gdpblocks;
5673 /* Blocks for super block, inode, quota and xattr blocks */
5674 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5680 * Calculate the total number of credits to reserve to fit
5681 * the modification of a single pages into a single transaction,
5682 * which may include multiple chunks of block allocations.
5684 * This could be called via ext4_write_begin()
5686 * We need to consider the worse case, when
5687 * one new block per extent.
5689 int ext4_writepage_trans_blocks(struct inode *inode)
5691 int bpp = ext4_journal_blocks_per_page(inode);
5694 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5696 /* Account for data blocks for journalled mode */
5697 if (ext4_should_journal_data(inode))
5703 * Calculate the journal credits for a chunk of data modification.
5705 * This is called from DIO, fallocate or whoever calling
5706 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5708 * journal buffers for data blocks are not included here, as DIO
5709 * and fallocate do no need to journal data buffers.
5711 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5713 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5717 * The caller must have previously called ext4_reserve_inode_write().
5718 * Give this, we know that the caller already has write access to iloc->bh.
5720 int ext4_mark_iloc_dirty(handle_t *handle,
5721 struct inode *inode, struct ext4_iloc *iloc)
5725 if (unlikely(ext4_forced_shutdown(inode->i_sb))) {
5729 ext4_fc_track_inode(handle, inode);
5731 /* the do_update_inode consumes one bh->b_count */
5734 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5735 err = ext4_do_update_inode(handle, inode, iloc);
5741 * On success, We end up with an outstanding reference count against
5742 * iloc->bh. This _must_ be cleaned up later.
5746 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5747 struct ext4_iloc *iloc)
5751 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5754 err = ext4_get_inode_loc(inode, iloc);
5756 BUFFER_TRACE(iloc->bh, "get_write_access");
5757 err = ext4_journal_get_write_access(handle, inode->i_sb,
5758 iloc->bh, EXT4_JTR_NONE);
5764 ext4_std_error(inode->i_sb, err);
5768 static int __ext4_expand_extra_isize(struct inode *inode,
5769 unsigned int new_extra_isize,
5770 struct ext4_iloc *iloc,
5771 handle_t *handle, int *no_expand)
5773 struct ext4_inode *raw_inode;
5774 struct ext4_xattr_ibody_header *header;
5775 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5776 struct ext4_inode_info *ei = EXT4_I(inode);
5779 /* this was checked at iget time, but double check for good measure */
5780 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5781 (ei->i_extra_isize & 3)) {
5782 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5784 EXT4_INODE_SIZE(inode->i_sb));
5785 return -EFSCORRUPTED;
5787 if ((new_extra_isize < ei->i_extra_isize) ||
5788 (new_extra_isize < 4) ||
5789 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5790 return -EINVAL; /* Should never happen */
5792 raw_inode = ext4_raw_inode(iloc);
5794 header = IHDR(inode, raw_inode);
5796 /* No extended attributes present */
5797 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5798 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5799 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5800 EXT4_I(inode)->i_extra_isize, 0,
5801 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5802 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5807 * We may need to allocate external xattr block so we need quotas
5808 * initialized. Here we can be called with various locks held so we
5809 * cannot affort to initialize quotas ourselves. So just bail.
5811 if (dquot_initialize_needed(inode))
5814 /* try to expand with EAs present */
5815 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5819 * Inode size expansion failed; don't try again
5828 * Expand an inode by new_extra_isize bytes.
5829 * Returns 0 on success or negative error number on failure.
5831 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5832 unsigned int new_extra_isize,
5833 struct ext4_iloc iloc,
5839 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5843 * In nojournal mode, we can immediately attempt to expand
5844 * the inode. When journaled, we first need to obtain extra
5845 * buffer credits since we may write into the EA block
5846 * with this same handle. If journal_extend fails, then it will
5847 * only result in a minor loss of functionality for that inode.
5848 * If this is felt to be critical, then e2fsck should be run to
5849 * force a large enough s_min_extra_isize.
5851 if (ext4_journal_extend(handle,
5852 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5855 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5858 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5859 handle, &no_expand);
5860 ext4_write_unlock_xattr(inode, &no_expand);
5865 int ext4_expand_extra_isize(struct inode *inode,
5866 unsigned int new_extra_isize,
5867 struct ext4_iloc *iloc)
5873 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5878 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5879 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5880 if (IS_ERR(handle)) {
5881 error = PTR_ERR(handle);
5886 ext4_write_lock_xattr(inode, &no_expand);
5888 BUFFER_TRACE(iloc->bh, "get_write_access");
5889 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5896 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5897 handle, &no_expand);
5899 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5904 ext4_write_unlock_xattr(inode, &no_expand);
5905 ext4_journal_stop(handle);
5910 * What we do here is to mark the in-core inode as clean with respect to inode
5911 * dirtiness (it may still be data-dirty).
5912 * This means that the in-core inode may be reaped by prune_icache
5913 * without having to perform any I/O. This is a very good thing,
5914 * because *any* task may call prune_icache - even ones which
5915 * have a transaction open against a different journal.
5917 * Is this cheating? Not really. Sure, we haven't written the
5918 * inode out, but prune_icache isn't a user-visible syncing function.
5919 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5920 * we start and wait on commits.
5922 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5923 const char *func, unsigned int line)
5925 struct ext4_iloc iloc;
5926 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5930 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5931 err = ext4_reserve_inode_write(handle, inode, &iloc);
5935 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5936 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5939 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5942 ext4_error_inode_err(inode, func, line, 0, err,
5943 "mark_inode_dirty error");
5948 * ext4_dirty_inode() is called from __mark_inode_dirty()
5950 * We're really interested in the case where a file is being extended.
5951 * i_size has been changed by generic_commit_write() and we thus need
5952 * to include the updated inode in the current transaction.
5954 * Also, dquot_alloc_block() will always dirty the inode when blocks
5955 * are allocated to the file.
5957 * If the inode is marked synchronous, we don't honour that here - doing
5958 * so would cause a commit on atime updates, which we don't bother doing.
5959 * We handle synchronous inodes at the highest possible level.
5961 void ext4_dirty_inode(struct inode *inode, int flags)
5965 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5968 ext4_mark_inode_dirty(handle, inode);
5969 ext4_journal_stop(handle);
5972 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5980 * We have to be very careful here: changing a data block's
5981 * journaling status dynamically is dangerous. If we write a
5982 * data block to the journal, change the status and then delete
5983 * that block, we risk forgetting to revoke the old log record
5984 * from the journal and so a subsequent replay can corrupt data.
5985 * So, first we make sure that the journal is empty and that
5986 * nobody is changing anything.
5989 journal = EXT4_JOURNAL(inode);
5992 if (is_journal_aborted(journal))
5995 /* Wait for all existing dio workers */
5996 inode_dio_wait(inode);
5999 * Before flushing the journal and switching inode's aops, we have
6000 * to flush all dirty data the inode has. There can be outstanding
6001 * delayed allocations, there can be unwritten extents created by
6002 * fallocate or buffered writes in dioread_nolock mode covered by
6003 * dirty data which can be converted only after flushing the dirty
6004 * data (and journalled aops don't know how to handle these cases).
6007 filemap_invalidate_lock(inode->i_mapping);
6008 err = filemap_write_and_wait(inode->i_mapping);
6010 filemap_invalidate_unlock(inode->i_mapping);
6015 alloc_ctx = ext4_writepages_down_write(inode->i_sb);
6016 jbd2_journal_lock_updates(journal);
6019 * OK, there are no updates running now, and all cached data is
6020 * synced to disk. We are now in a completely consistent state
6021 * which doesn't have anything in the journal, and we know that
6022 * no filesystem updates are running, so it is safe to modify
6023 * the inode's in-core data-journaling state flag now.
6027 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6029 err = jbd2_journal_flush(journal, 0);
6031 jbd2_journal_unlock_updates(journal);
6032 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6035 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6037 ext4_set_aops(inode);
6039 jbd2_journal_unlock_updates(journal);
6040 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6043 filemap_invalidate_unlock(inode->i_mapping);
6045 /* Finally we can mark the inode as dirty. */
6047 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6049 return PTR_ERR(handle);
6051 ext4_fc_mark_ineligible(inode->i_sb,
6052 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6053 err = ext4_mark_inode_dirty(handle, inode);
6054 ext4_handle_sync(handle);
6055 ext4_journal_stop(handle);
6056 ext4_std_error(inode->i_sb, err);
6061 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6062 struct buffer_head *bh)
6064 return !buffer_mapped(bh);
6067 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6069 struct vm_area_struct *vma = vmf->vma;
6070 struct folio *folio = page_folio(vmf->page);
6075 struct file *file = vma->vm_file;
6076 struct inode *inode = file_inode(file);
6077 struct address_space *mapping = inode->i_mapping;
6079 get_block_t *get_block;
6082 if (unlikely(IS_IMMUTABLE(inode)))
6083 return VM_FAULT_SIGBUS;
6085 sb_start_pagefault(inode->i_sb);
6086 file_update_time(vma->vm_file);
6088 filemap_invalidate_lock_shared(mapping);
6090 err = ext4_convert_inline_data(inode);
6095 * On data journalling we skip straight to the transaction handle:
6096 * there's no delalloc; page truncated will be checked later; the
6097 * early return w/ all buffers mapped (calculates size/len) can't
6098 * be used; and there's no dioread_nolock, so only ext4_get_block.
6100 if (ext4_should_journal_data(inode))
6103 /* Delalloc case is easy... */
6104 if (test_opt(inode->i_sb, DELALLOC) &&
6105 !ext4_nonda_switch(inode->i_sb)) {
6107 err = block_page_mkwrite(vma, vmf,
6108 ext4_da_get_block_prep);
6109 } while (err == -ENOSPC &&
6110 ext4_should_retry_alloc(inode->i_sb, &retries));
6115 size = i_size_read(inode);
6116 /* Page got truncated from under us? */
6117 if (folio->mapping != mapping || folio_pos(folio) > size) {
6118 folio_unlock(folio);
6119 ret = VM_FAULT_NOPAGE;
6123 len = folio_size(folio);
6124 if (folio_pos(folio) + len > size)
6125 len = size - folio_pos(folio);
6127 * Return if we have all the buffers mapped. This avoids the need to do
6128 * journal_start/journal_stop which can block and take a long time
6130 * This cannot be done for data journalling, as we have to add the
6131 * inode to the transaction's list to writeprotect pages on commit.
6133 if (folio_buffers(folio)) {
6134 if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6136 ext4_bh_unmapped)) {
6137 /* Wait so that we don't change page under IO */
6138 folio_wait_stable(folio);
6139 ret = VM_FAULT_LOCKED;
6143 folio_unlock(folio);
6144 /* OK, we need to fill the hole... */
6145 if (ext4_should_dioread_nolock(inode))
6146 get_block = ext4_get_block_unwritten;
6148 get_block = ext4_get_block;
6150 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6151 ext4_writepage_trans_blocks(inode));
6152 if (IS_ERR(handle)) {
6153 ret = VM_FAULT_SIGBUS;
6157 * Data journalling can't use block_page_mkwrite() because it
6158 * will set_buffer_dirty() before do_journal_get_write_access()
6159 * thus might hit warning messages for dirty metadata buffers.
6161 if (!ext4_should_journal_data(inode)) {
6162 err = block_page_mkwrite(vma, vmf, get_block);
6165 size = i_size_read(inode);
6166 /* Page got truncated from under us? */
6167 if (folio->mapping != mapping || folio_pos(folio) > size) {
6168 ret = VM_FAULT_NOPAGE;
6172 len = folio_size(folio);
6173 if (folio_pos(folio) + len > size)
6174 len = size - folio_pos(folio);
6176 err = __block_write_begin(&folio->page, 0, len, ext4_get_block);
6178 ret = VM_FAULT_SIGBUS;
6179 if (ext4_journal_folio_buffers(handle, folio, len))
6182 folio_unlock(folio);
6185 ext4_journal_stop(handle);
6186 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6189 ret = vmf_fs_error(err);
6191 filemap_invalidate_unlock_shared(mapping);
6192 sb_end_pagefault(inode->i_sb);
6195 folio_unlock(folio);
6196 ext4_journal_stop(handle);