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_journalled_writepage(struct page *page, unsigned int len);
140 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
144 * Test whether an inode is a fast symlink.
145 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
147 int ext4_inode_is_fast_symlink(struct inode *inode)
149 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
151 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
153 if (ext4_has_inline_data(inode))
156 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
158 return S_ISLNK(inode->i_mode) && inode->i_size &&
159 (inode->i_size < EXT4_N_BLOCKS * 4);
163 * Called at the last iput() if i_nlink is zero.
165 void ext4_evict_inode(struct inode *inode)
170 * Credits for final inode cleanup and freeing:
171 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
172 * (xattr block freeing), bitmap, group descriptor (inode freeing)
174 int extra_credits = 6;
175 struct ext4_xattr_inode_array *ea_inode_array = NULL;
176 bool freeze_protected = false;
178 trace_ext4_evict_inode(inode);
180 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
181 ext4_evict_ea_inode(inode);
182 if (inode->i_nlink) {
184 * When journalling data dirty buffers are tracked only in the
185 * journal. So although mm thinks everything is clean and
186 * ready for reaping the inode might still have some pages to
187 * write in the running transaction or waiting to be
188 * checkpointed. Thus calling jbd2_journal_invalidate_folio()
189 * (via truncate_inode_pages()) to discard these buffers can
190 * cause data loss. Also even if we did not discard these
191 * buffers, we would have no way to find them after the inode
192 * is reaped and thus user could see stale data if he tries to
193 * read them before the transaction is checkpointed. So be
194 * careful and force everything to disk here... We use
195 * ei->i_datasync_tid to store the newest transaction
196 * containing inode's data.
198 * Note that directories do not have this problem because they
199 * don't use page cache.
201 if (inode->i_ino != EXT4_JOURNAL_INO &&
202 ext4_should_journal_data(inode) &&
203 S_ISREG(inode->i_mode) && inode->i_data.nrpages) {
204 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
205 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
207 jbd2_complete_transaction(journal, commit_tid);
208 filemap_write_and_wait(&inode->i_data);
210 truncate_inode_pages_final(&inode->i_data);
215 if (is_bad_inode(inode))
217 dquot_initialize(inode);
219 if (ext4_should_order_data(inode))
220 ext4_begin_ordered_truncate(inode, 0);
221 truncate_inode_pages_final(&inode->i_data);
224 * For inodes with journalled data, transaction commit could have
225 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
226 * flag but we still need to remove the inode from the writeback lists.
228 if (!list_empty_careful(&inode->i_io_list)) {
229 WARN_ON_ONCE(!ext4_should_journal_data(inode));
230 inode_io_list_del(inode);
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it. When we are in a running transaction though,
236 * we are already protected against freezing and we cannot grab further
237 * protection due to lock ordering constraints.
239 if (!ext4_journal_current_handle()) {
240 sb_start_intwrite(inode->i_sb);
241 freeze_protected = true;
244 if (!IS_NOQUOTA(inode))
245 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
248 * Block bitmap, group descriptor, and inode are accounted in both
249 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
251 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
252 ext4_blocks_for_truncate(inode) + extra_credits - 3);
253 if (IS_ERR(handle)) {
254 ext4_std_error(inode->i_sb, PTR_ERR(handle));
256 * If we're going to skip the normal cleanup, we still need to
257 * make sure that the in-core orphan linked list is properly
260 ext4_orphan_del(NULL, inode);
261 if (freeze_protected)
262 sb_end_intwrite(inode->i_sb);
267 ext4_handle_sync(handle);
270 * Set inode->i_size to 0 before calling ext4_truncate(). We need
271 * special handling of symlinks here because i_size is used to
272 * determine whether ext4_inode_info->i_data contains symlink data or
273 * block mappings. Setting i_size to 0 will remove its fast symlink
274 * status. Erase i_data so that it becomes a valid empty block map.
276 if (ext4_inode_is_fast_symlink(inode))
277 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
279 err = ext4_mark_inode_dirty(handle, inode);
281 ext4_warning(inode->i_sb,
282 "couldn't mark inode dirty (err %d)", err);
285 if (inode->i_blocks) {
286 err = ext4_truncate(inode);
288 ext4_error_err(inode->i_sb, -err,
289 "couldn't truncate inode %lu (err %d)",
295 /* Remove xattr references. */
296 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
299 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
301 ext4_journal_stop(handle);
302 ext4_orphan_del(NULL, inode);
303 if (freeze_protected)
304 sb_end_intwrite(inode->i_sb);
305 ext4_xattr_inode_array_free(ea_inode_array);
310 * Kill off the orphan record which ext4_truncate created.
311 * AKPM: I think this can be inside the above `if'.
312 * Note that ext4_orphan_del() has to be able to cope with the
313 * deletion of a non-existent orphan - this is because we don't
314 * know if ext4_truncate() actually created an orphan record.
315 * (Well, we could do this if we need to, but heck - it works)
317 ext4_orphan_del(handle, inode);
318 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
321 * One subtle ordering requirement: if anything has gone wrong
322 * (transaction abort, IO errors, whatever), then we can still
323 * do these next steps (the fs will already have been marked as
324 * having errors), but we can't free the inode if the mark_dirty
327 if (ext4_mark_inode_dirty(handle, inode))
328 /* If that failed, just do the required in-core inode clear. */
329 ext4_clear_inode(inode);
331 ext4_free_inode(handle, inode);
332 ext4_journal_stop(handle);
333 if (freeze_protected)
334 sb_end_intwrite(inode->i_sb);
335 ext4_xattr_inode_array_free(ea_inode_array);
338 if (!list_empty(&EXT4_I(inode)->i_fc_list))
339 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
340 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
344 qsize_t *ext4_get_reserved_space(struct inode *inode)
346 return &EXT4_I(inode)->i_reserved_quota;
351 * Called with i_data_sem down, which is important since we can call
352 * ext4_discard_preallocations() from here.
354 void ext4_da_update_reserve_space(struct inode *inode,
355 int used, int quota_claim)
357 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
358 struct ext4_inode_info *ei = EXT4_I(inode);
360 spin_lock(&ei->i_block_reservation_lock);
361 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
362 if (unlikely(used > ei->i_reserved_data_blocks)) {
363 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
364 "with only %d reserved data blocks",
365 __func__, inode->i_ino, used,
366 ei->i_reserved_data_blocks);
368 used = ei->i_reserved_data_blocks;
371 /* Update per-inode reservations */
372 ei->i_reserved_data_blocks -= used;
373 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
375 spin_unlock(&ei->i_block_reservation_lock);
377 /* Update quota subsystem for data blocks */
379 dquot_claim_block(inode, EXT4_C2B(sbi, used));
382 * We did fallocate with an offset that is already delayed
383 * allocated. So on delayed allocated writeback we should
384 * not re-claim the quota for fallocated blocks.
386 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
390 * If we have done all the pending block allocations and if
391 * there aren't any writers on the inode, we can discard the
392 * inode's preallocations.
394 if ((ei->i_reserved_data_blocks == 0) &&
395 !inode_is_open_for_write(inode))
396 ext4_discard_preallocations(inode, 0);
399 static int __check_block_validity(struct inode *inode, const char *func,
401 struct ext4_map_blocks *map)
403 if (ext4_has_feature_journal(inode->i_sb) &&
405 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
407 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
408 ext4_error_inode(inode, func, line, map->m_pblk,
409 "lblock %lu mapped to illegal pblock %llu "
410 "(length %d)", (unsigned long) map->m_lblk,
411 map->m_pblk, map->m_len);
412 return -EFSCORRUPTED;
417 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
422 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
423 return fscrypt_zeroout_range(inode, lblk, pblk, len);
425 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
432 #define check_block_validity(inode, map) \
433 __check_block_validity((inode), __func__, __LINE__, (map))
435 #ifdef ES_AGGRESSIVE_TEST
436 static void ext4_map_blocks_es_recheck(handle_t *handle,
438 struct ext4_map_blocks *es_map,
439 struct ext4_map_blocks *map,
446 * There is a race window that the result is not the same.
447 * e.g. xfstests #223 when dioread_nolock enables. The reason
448 * is that we lookup a block mapping in extent status tree with
449 * out taking i_data_sem. So at the time the unwritten extent
450 * could be converted.
452 down_read(&EXT4_I(inode)->i_data_sem);
453 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
454 retval = ext4_ext_map_blocks(handle, inode, map, 0);
456 retval = ext4_ind_map_blocks(handle, inode, map, 0);
458 up_read((&EXT4_I(inode)->i_data_sem));
461 * We don't check m_len because extent will be collpased in status
462 * tree. So the m_len might not equal.
464 if (es_map->m_lblk != map->m_lblk ||
465 es_map->m_flags != map->m_flags ||
466 es_map->m_pblk != map->m_pblk) {
467 printk("ES cache assertion failed for inode: %lu "
468 "es_cached ex [%d/%d/%llu/%x] != "
469 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
470 inode->i_ino, es_map->m_lblk, es_map->m_len,
471 es_map->m_pblk, es_map->m_flags, map->m_lblk,
472 map->m_len, map->m_pblk, map->m_flags,
476 #endif /* ES_AGGRESSIVE_TEST */
479 * The ext4_map_blocks() function tries to look up the requested blocks,
480 * and returns if the blocks are already mapped.
482 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
483 * and store the allocated blocks in the result buffer head and mark it
486 * If file type is extents based, it will call ext4_ext_map_blocks(),
487 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
490 * On success, it returns the number of blocks being mapped or allocated. if
491 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
492 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
494 * It returns 0 if plain look up failed (blocks have not been allocated), in
495 * that case, @map is returned as unmapped but we still do fill map->m_len to
496 * indicate the length of a hole starting at map->m_lblk.
498 * It returns the error in case of allocation failure.
500 int ext4_map_blocks(handle_t *handle, struct inode *inode,
501 struct ext4_map_blocks *map, int flags)
503 struct extent_status es;
506 #ifdef ES_AGGRESSIVE_TEST
507 struct ext4_map_blocks orig_map;
509 memcpy(&orig_map, map, sizeof(*map));
513 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
514 flags, map->m_len, (unsigned long) map->m_lblk);
517 * ext4_map_blocks returns an int, and m_len is an unsigned int
519 if (unlikely(map->m_len > INT_MAX))
520 map->m_len = INT_MAX;
522 /* We can handle the block number less than EXT_MAX_BLOCKS */
523 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
524 return -EFSCORRUPTED;
526 /* Lookup extent status tree firstly */
527 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
528 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
529 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
530 map->m_pblk = ext4_es_pblock(&es) +
531 map->m_lblk - es.es_lblk;
532 map->m_flags |= ext4_es_is_written(&es) ?
533 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
534 retval = es.es_len - (map->m_lblk - es.es_lblk);
535 if (retval > map->m_len)
538 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
540 retval = es.es_len - (map->m_lblk - es.es_lblk);
541 if (retval > map->m_len)
549 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
551 #ifdef ES_AGGRESSIVE_TEST
552 ext4_map_blocks_es_recheck(handle, inode, map,
558 * In the query cache no-wait mode, nothing we can do more if we
559 * cannot find extent in the cache.
561 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
565 * Try to see if we can get the block without requesting a new
568 down_read(&EXT4_I(inode)->i_data_sem);
569 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
570 retval = ext4_ext_map_blocks(handle, inode, map, 0);
572 retval = ext4_ind_map_blocks(handle, inode, map, 0);
577 if (unlikely(retval != map->m_len)) {
578 ext4_warning(inode->i_sb,
579 "ES len assertion failed for inode "
580 "%lu: retval %d != map->m_len %d",
581 inode->i_ino, retval, map->m_len);
585 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
586 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
587 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
588 !(status & EXTENT_STATUS_WRITTEN) &&
589 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
590 map->m_lblk + map->m_len - 1))
591 status |= EXTENT_STATUS_DELAYED;
592 ret = ext4_es_insert_extent(inode, map->m_lblk,
593 map->m_len, map->m_pblk, status);
597 up_read((&EXT4_I(inode)->i_data_sem));
600 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
601 ret = check_block_validity(inode, map);
606 /* If it is only a block(s) look up */
607 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
611 * Returns if the blocks have already allocated
613 * Note that if blocks have been preallocated
614 * ext4_ext_get_block() returns the create = 0
615 * with buffer head unmapped.
617 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
619 * If we need to convert extent to unwritten
620 * we continue and do the actual work in
621 * ext4_ext_map_blocks()
623 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
627 * Here we clear m_flags because after allocating an new extent,
628 * it will be set again.
630 map->m_flags &= ~EXT4_MAP_FLAGS;
633 * New blocks allocate and/or writing to unwritten extent
634 * will possibly result in updating i_data, so we take
635 * the write lock of i_data_sem, and call get_block()
636 * with create == 1 flag.
638 down_write(&EXT4_I(inode)->i_data_sem);
641 * We need to check for EXT4 here because migrate
642 * could have changed the inode type in between
644 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
645 retval = ext4_ext_map_blocks(handle, inode, map, flags);
647 retval = ext4_ind_map_blocks(handle, inode, map, flags);
649 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
651 * We allocated new blocks which will result in
652 * i_data's format changing. Force the migrate
653 * to fail by clearing migrate flags
655 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
659 * Update reserved blocks/metadata blocks after successful
660 * block allocation which had been deferred till now. We don't
661 * support fallocate for non extent files. So we can update
662 * reserve space here.
665 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
666 ext4_da_update_reserve_space(inode, retval, 1);
672 if (unlikely(retval != map->m_len)) {
673 ext4_warning(inode->i_sb,
674 "ES len assertion failed for inode "
675 "%lu: retval %d != map->m_len %d",
676 inode->i_ino, retval, map->m_len);
681 * We have to zeroout blocks before inserting them into extent
682 * status tree. Otherwise someone could look them up there and
683 * use them before they are really zeroed. We also have to
684 * unmap metadata before zeroing as otherwise writeback can
685 * overwrite zeros with stale data from block device.
687 if (flags & EXT4_GET_BLOCKS_ZERO &&
688 map->m_flags & EXT4_MAP_MAPPED &&
689 map->m_flags & EXT4_MAP_NEW) {
690 ret = ext4_issue_zeroout(inode, map->m_lblk,
691 map->m_pblk, map->m_len);
699 * If the extent has been zeroed out, we don't need to update
700 * extent status tree.
702 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
703 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
704 if (ext4_es_is_written(&es))
707 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
708 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
709 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
710 !(status & EXTENT_STATUS_WRITTEN) &&
711 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
712 map->m_lblk + map->m_len - 1))
713 status |= EXTENT_STATUS_DELAYED;
714 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
715 map->m_pblk, status);
723 up_write((&EXT4_I(inode)->i_data_sem));
724 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
725 ret = check_block_validity(inode, map);
730 * Inodes with freshly allocated blocks where contents will be
731 * visible after transaction commit must be on transaction's
734 if (map->m_flags & EXT4_MAP_NEW &&
735 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
736 !(flags & EXT4_GET_BLOCKS_ZERO) &&
737 !ext4_is_quota_file(inode) &&
738 ext4_should_order_data(inode)) {
740 (loff_t)map->m_lblk << inode->i_blkbits;
741 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
743 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
744 ret = ext4_jbd2_inode_add_wait(handle, inode,
747 ret = ext4_jbd2_inode_add_write(handle, inode,
753 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
754 map->m_flags & EXT4_MAP_MAPPED))
755 ext4_fc_track_range(handle, inode, map->m_lblk,
756 map->m_lblk + map->m_len - 1);
758 ext_debug(inode, "failed with err %d\n", retval);
763 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
764 * we have to be careful as someone else may be manipulating b_state as well.
766 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
768 unsigned long old_state;
769 unsigned long new_state;
771 flags &= EXT4_MAP_FLAGS;
773 /* Dummy buffer_head? Set non-atomically. */
775 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
779 * Someone else may be modifying b_state. Be careful! This is ugly but
780 * once we get rid of using bh as a container for mapping information
781 * to pass to / from get_block functions, this can go away.
784 old_state = READ_ONCE(bh->b_state);
785 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
787 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
790 static int _ext4_get_block(struct inode *inode, sector_t iblock,
791 struct buffer_head *bh, int flags)
793 struct ext4_map_blocks map;
796 if (ext4_has_inline_data(inode))
800 map.m_len = bh->b_size >> inode->i_blkbits;
802 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
805 map_bh(bh, inode->i_sb, map.m_pblk);
806 ext4_update_bh_state(bh, map.m_flags);
807 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
809 } else if (ret == 0) {
810 /* hole case, need to fill in bh->b_size */
811 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
816 int ext4_get_block(struct inode *inode, sector_t iblock,
817 struct buffer_head *bh, int create)
819 return _ext4_get_block(inode, iblock, bh,
820 create ? EXT4_GET_BLOCKS_CREATE : 0);
824 * Get block function used when preparing for buffered write if we require
825 * creating an unwritten extent if blocks haven't been allocated. The extent
826 * will be converted to written after the IO is complete.
828 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
829 struct buffer_head *bh_result, int create)
831 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
832 inode->i_ino, create);
833 return _ext4_get_block(inode, iblock, bh_result,
834 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
837 /* Maximum number of blocks we map for direct IO at once. */
838 #define DIO_MAX_BLOCKS 4096
841 * `handle' can be NULL if create is zero
843 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
844 ext4_lblk_t block, int map_flags)
846 struct ext4_map_blocks map;
847 struct buffer_head *bh;
848 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
849 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
852 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
853 || handle != NULL || create == 0);
854 ASSERT(create == 0 || !nowait);
858 err = ext4_map_blocks(handle, inode, &map, map_flags);
861 return create ? ERR_PTR(-ENOSPC) : NULL;
866 return sb_find_get_block(inode->i_sb, map.m_pblk);
868 bh = sb_getblk(inode->i_sb, map.m_pblk);
870 return ERR_PTR(-ENOMEM);
871 if (map.m_flags & EXT4_MAP_NEW) {
873 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
874 || (handle != NULL));
877 * Now that we do not always journal data, we should
878 * keep in mind whether this should always journal the
879 * new buffer as metadata. For now, regular file
880 * writes use ext4_get_block instead, so it's not a
884 BUFFER_TRACE(bh, "call get_create_access");
885 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
891 if (!buffer_uptodate(bh)) {
892 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
893 set_buffer_uptodate(bh);
896 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
897 err = ext4_handle_dirty_metadata(handle, inode, bh);
901 BUFFER_TRACE(bh, "not a new buffer");
908 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
909 ext4_lblk_t block, int map_flags)
911 struct buffer_head *bh;
914 bh = ext4_getblk(handle, inode, block, map_flags);
917 if (!bh || ext4_buffer_uptodate(bh))
920 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
928 /* Read a contiguous batch of blocks. */
929 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
930 bool wait, struct buffer_head **bhs)
934 for (i = 0; i < bh_count; i++) {
935 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
936 if (IS_ERR(bhs[i])) {
937 err = PTR_ERR(bhs[i]);
943 for (i = 0; i < bh_count; i++)
944 /* Note that NULL bhs[i] is valid because of holes. */
945 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
946 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
951 for (i = 0; i < bh_count; i++)
953 wait_on_buffer(bhs[i]);
955 for (i = 0; i < bh_count; i++) {
956 if (bhs[i] && !buffer_uptodate(bhs[i])) {
964 for (i = 0; i < bh_count; i++) {
971 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
972 struct buffer_head *head,
976 int (*fn)(handle_t *handle, struct inode *inode,
977 struct buffer_head *bh))
979 struct buffer_head *bh;
980 unsigned block_start, block_end;
981 unsigned blocksize = head->b_size;
983 struct buffer_head *next;
985 for (bh = head, block_start = 0;
986 ret == 0 && (bh != head || !block_start);
987 block_start = block_end, bh = next) {
988 next = bh->b_this_page;
989 block_end = block_start + blocksize;
990 if (block_end <= from || block_start >= to) {
991 if (partial && !buffer_uptodate(bh))
995 err = (*fn)(handle, inode, bh);
1003 * To preserve ordering, it is essential that the hole instantiation and
1004 * the data write be encapsulated in a single transaction. We cannot
1005 * close off a transaction and start a new one between the ext4_get_block()
1006 * and the commit_write(). So doing the jbd2_journal_start at the start of
1007 * prepare_write() is the right place.
1009 * Also, this function can nest inside ext4_writepage(). In that case, we
1010 * *know* that ext4_writepage() has generated enough buffer credits to do the
1011 * whole page. So we won't block on the journal in that case, which is good,
1012 * because the caller may be PF_MEMALLOC.
1014 * By accident, ext4 can be reentered when a transaction is open via
1015 * quota file writes. If we were to commit the transaction while thus
1016 * reentered, there can be a deadlock - we would be holding a quota
1017 * lock, and the commit would never complete if another thread had a
1018 * transaction open and was blocking on the quota lock - a ranking
1021 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1022 * will _not_ run commit under these circumstances because handle->h_ref
1023 * is elevated. We'll still have enough credits for the tiny quotafile
1026 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1027 struct buffer_head *bh)
1029 int dirty = buffer_dirty(bh);
1032 if (!buffer_mapped(bh) || buffer_freed(bh))
1035 * __block_write_begin() could have dirtied some buffers. Clean
1036 * the dirty bit as jbd2_journal_get_write_access() could complain
1037 * otherwise about fs integrity issues. Setting of the dirty bit
1038 * by __block_write_begin() isn't a real problem here as we clear
1039 * the bit before releasing a page lock and thus writeback cannot
1040 * ever write the buffer.
1043 clear_buffer_dirty(bh);
1044 BUFFER_TRACE(bh, "get write access");
1045 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1048 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1052 #ifdef CONFIG_FS_ENCRYPTION
1053 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1054 get_block_t *get_block)
1056 unsigned from = pos & (PAGE_SIZE - 1);
1057 unsigned to = from + len;
1058 struct inode *inode = page->mapping->host;
1059 unsigned block_start, block_end;
1062 unsigned blocksize = inode->i_sb->s_blocksize;
1064 struct buffer_head *bh, *head, *wait[2];
1068 BUG_ON(!PageLocked(page));
1069 BUG_ON(from > PAGE_SIZE);
1070 BUG_ON(to > PAGE_SIZE);
1073 if (!page_has_buffers(page))
1074 create_empty_buffers(page, blocksize, 0);
1075 head = page_buffers(page);
1076 bbits = ilog2(blocksize);
1077 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1079 for (bh = head, block_start = 0; bh != head || !block_start;
1080 block++, block_start = block_end, bh = bh->b_this_page) {
1081 block_end = block_start + blocksize;
1082 if (block_end <= from || block_start >= to) {
1083 if (PageUptodate(page)) {
1084 set_buffer_uptodate(bh);
1089 clear_buffer_new(bh);
1090 if (!buffer_mapped(bh)) {
1091 WARN_ON(bh->b_size != blocksize);
1092 err = get_block(inode, block, bh, 1);
1095 if (buffer_new(bh)) {
1096 if (PageUptodate(page)) {
1097 clear_buffer_new(bh);
1098 set_buffer_uptodate(bh);
1099 mark_buffer_dirty(bh);
1102 if (block_end > to || block_start < from)
1103 zero_user_segments(page, to, block_end,
1108 if (PageUptodate(page)) {
1109 set_buffer_uptodate(bh);
1112 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1113 !buffer_unwritten(bh) &&
1114 (block_start < from || block_end > to)) {
1115 ext4_read_bh_lock(bh, 0, false);
1116 wait[nr_wait++] = bh;
1120 * If we issued read requests, let them complete.
1122 for (i = 0; i < nr_wait; i++) {
1123 wait_on_buffer(wait[i]);
1124 if (!buffer_uptodate(wait[i]))
1127 if (unlikely(err)) {
1128 page_zero_new_buffers(page, from, to);
1129 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1130 for (i = 0; i < nr_wait; i++) {
1133 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1134 bh_offset(wait[i]));
1136 clear_buffer_uptodate(wait[i]);
1146 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1147 loff_t pos, unsigned len,
1148 struct page **pagep, void **fsdata)
1150 struct inode *inode = mapping->host;
1151 int ret, needed_blocks;
1158 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1161 trace_ext4_write_begin(inode, pos, len);
1163 * Reserve one block more for addition to orphan list in case
1164 * we allocate blocks but write fails for some reason
1166 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1167 index = pos >> PAGE_SHIFT;
1168 from = pos & (PAGE_SIZE - 1);
1171 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1172 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1181 * grab_cache_page_write_begin() can take a long time if the
1182 * system is thrashing due to memory pressure, or if the page
1183 * is being written back. So grab it first before we start
1184 * the transaction handle. This also allows us to allocate
1185 * the page (if needed) without using GFP_NOFS.
1188 page = grab_cache_page_write_begin(mapping, index);
1194 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1195 if (IS_ERR(handle)) {
1197 return PTR_ERR(handle);
1201 if (page->mapping != mapping) {
1202 /* The page got truncated from under us */
1205 ext4_journal_stop(handle);
1208 /* In case writeback began while the page was unlocked */
1209 wait_for_stable_page(page);
1211 #ifdef CONFIG_FS_ENCRYPTION
1212 if (ext4_should_dioread_nolock(inode))
1213 ret = ext4_block_write_begin(page, pos, len,
1214 ext4_get_block_unwritten);
1216 ret = ext4_block_write_begin(page, pos, len,
1219 if (ext4_should_dioread_nolock(inode))
1220 ret = __block_write_begin(page, pos, len,
1221 ext4_get_block_unwritten);
1223 ret = __block_write_begin(page, pos, len, ext4_get_block);
1225 if (!ret && ext4_should_journal_data(inode)) {
1226 ret = ext4_walk_page_buffers(handle, inode,
1227 page_buffers(page), from, to, NULL,
1228 do_journal_get_write_access);
1232 bool extended = (pos + len > inode->i_size) &&
1233 !ext4_verity_in_progress(inode);
1237 * __block_write_begin may have instantiated a few blocks
1238 * outside i_size. Trim these off again. Don't need
1239 * i_size_read because we hold i_rwsem.
1241 * Add inode to orphan list in case we crash before
1244 if (extended && ext4_can_truncate(inode))
1245 ext4_orphan_add(handle, inode);
1247 ext4_journal_stop(handle);
1249 ext4_truncate_failed_write(inode);
1251 * If truncate failed early the inode might
1252 * still be on the orphan list; we need to
1253 * make sure the inode is removed from the
1254 * orphan list in that case.
1257 ext4_orphan_del(NULL, inode);
1260 if (ret == -ENOSPC &&
1261 ext4_should_retry_alloc(inode->i_sb, &retries))
1270 /* For write_end() in data=journal mode */
1271 static int write_end_fn(handle_t *handle, struct inode *inode,
1272 struct buffer_head *bh)
1275 if (!buffer_mapped(bh) || buffer_freed(bh))
1277 set_buffer_uptodate(bh);
1278 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1279 clear_buffer_meta(bh);
1280 clear_buffer_prio(bh);
1285 * We need to pick up the new inode size which generic_commit_write gave us
1286 * `file' can be NULL - eg, when called from page_symlink().
1288 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1289 * buffers are managed internally.
1291 static int ext4_write_end(struct file *file,
1292 struct address_space *mapping,
1293 loff_t pos, unsigned len, unsigned copied,
1294 struct page *page, void *fsdata)
1296 handle_t *handle = ext4_journal_current_handle();
1297 struct inode *inode = mapping->host;
1298 loff_t old_size = inode->i_size;
1300 int i_size_changed = 0;
1301 bool verity = ext4_verity_in_progress(inode);
1303 trace_ext4_write_end(inode, pos, len, copied);
1305 if (ext4_has_inline_data(inode))
1306 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1308 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1310 * it's important to update i_size while still holding page lock:
1311 * page writeout could otherwise come in and zero beyond i_size.
1313 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1314 * blocks are being written past EOF, so skip the i_size update.
1317 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1321 if (old_size < pos && !verity)
1322 pagecache_isize_extended(inode, old_size, pos);
1324 * Don't mark the inode dirty under page lock. First, it unnecessarily
1325 * makes the holding time of page lock longer. Second, it forces lock
1326 * ordering of page lock and transaction start for journaling
1330 ret = ext4_mark_inode_dirty(handle, inode);
1332 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1333 /* if we have allocated more blocks and copied
1334 * less. We will have blocks allocated outside
1335 * inode->i_size. So truncate them
1337 ext4_orphan_add(handle, inode);
1339 ret2 = ext4_journal_stop(handle);
1343 if (pos + len > inode->i_size && !verity) {
1344 ext4_truncate_failed_write(inode);
1346 * If truncate failed early the inode might still be
1347 * on the orphan list; we need to make sure the inode
1348 * is removed from the orphan list in that case.
1351 ext4_orphan_del(NULL, inode);
1354 return ret ? ret : copied;
1358 * This is a private version of page_zero_new_buffers() which doesn't
1359 * set the buffer to be dirty, since in data=journalled mode we need
1360 * to call ext4_handle_dirty_metadata() instead.
1362 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1363 struct inode *inode,
1365 unsigned from, unsigned to)
1367 unsigned int block_start = 0, block_end;
1368 struct buffer_head *head, *bh;
1370 bh = head = page_buffers(page);
1372 block_end = block_start + bh->b_size;
1373 if (buffer_new(bh)) {
1374 if (block_end > from && block_start < to) {
1375 if (!PageUptodate(page)) {
1376 unsigned start, size;
1378 start = max(from, block_start);
1379 size = min(to, block_end) - start;
1381 zero_user(page, start, size);
1382 write_end_fn(handle, inode, bh);
1384 clear_buffer_new(bh);
1387 block_start = block_end;
1388 bh = bh->b_this_page;
1389 } while (bh != head);
1392 static int ext4_journalled_write_end(struct file *file,
1393 struct address_space *mapping,
1394 loff_t pos, unsigned len, unsigned copied,
1395 struct page *page, void *fsdata)
1397 handle_t *handle = ext4_journal_current_handle();
1398 struct inode *inode = mapping->host;
1399 loff_t old_size = inode->i_size;
1403 int size_changed = 0;
1404 bool verity = ext4_verity_in_progress(inode);
1406 trace_ext4_journalled_write_end(inode, pos, len, copied);
1407 from = pos & (PAGE_SIZE - 1);
1410 BUG_ON(!ext4_handle_valid(handle));
1412 if (ext4_has_inline_data(inode))
1413 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1415 if (unlikely(copied < len) && !PageUptodate(page)) {
1417 ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
1419 if (unlikely(copied < len))
1420 ext4_journalled_zero_new_buffers(handle, inode, page,
1422 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
1423 from, from + copied, &partial,
1426 SetPageUptodate(page);
1429 size_changed = ext4_update_inode_size(inode, pos + copied);
1430 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1431 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1435 if (old_size < pos && !verity)
1436 pagecache_isize_extended(inode, old_size, pos);
1439 ret2 = ext4_mark_inode_dirty(handle, inode);
1444 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1445 /* if we have allocated more blocks and copied
1446 * less. We will have blocks allocated outside
1447 * inode->i_size. So truncate them
1449 ext4_orphan_add(handle, inode);
1451 ret2 = ext4_journal_stop(handle);
1454 if (pos + len > inode->i_size && !verity) {
1455 ext4_truncate_failed_write(inode);
1457 * If truncate failed early the inode might still be
1458 * on the orphan list; we need to make sure the inode
1459 * is removed from the orphan list in that case.
1462 ext4_orphan_del(NULL, inode);
1465 return ret ? ret : copied;
1469 * Reserve space for a single cluster
1471 static int ext4_da_reserve_space(struct inode *inode)
1473 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1474 struct ext4_inode_info *ei = EXT4_I(inode);
1478 * We will charge metadata quota at writeout time; this saves
1479 * us from metadata over-estimation, though we may go over by
1480 * a small amount in the end. Here we just reserve for data.
1482 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1486 spin_lock(&ei->i_block_reservation_lock);
1487 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1488 spin_unlock(&ei->i_block_reservation_lock);
1489 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1492 ei->i_reserved_data_blocks++;
1493 trace_ext4_da_reserve_space(inode);
1494 spin_unlock(&ei->i_block_reservation_lock);
1496 return 0; /* success */
1499 void ext4_da_release_space(struct inode *inode, int to_free)
1501 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1502 struct ext4_inode_info *ei = EXT4_I(inode);
1505 return; /* Nothing to release, exit */
1507 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1509 trace_ext4_da_release_space(inode, to_free);
1510 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1512 * if there aren't enough reserved blocks, then the
1513 * counter is messed up somewhere. Since this
1514 * function is called from invalidate page, it's
1515 * harmless to return without any action.
1517 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1518 "ino %lu, to_free %d with only %d reserved "
1519 "data blocks", inode->i_ino, to_free,
1520 ei->i_reserved_data_blocks);
1522 to_free = ei->i_reserved_data_blocks;
1524 ei->i_reserved_data_blocks -= to_free;
1526 /* update fs dirty data blocks counter */
1527 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1529 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1531 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1535 * Delayed allocation stuff
1538 struct mpage_da_data {
1539 struct inode *inode;
1540 struct writeback_control *wbc;
1542 pgoff_t first_page; /* The first page to write */
1543 pgoff_t next_page; /* Current page to examine */
1544 pgoff_t last_page; /* Last page to examine */
1546 * Extent to map - this can be after first_page because that can be
1547 * fully mapped. We somewhat abuse m_flags to store whether the extent
1548 * is delalloc or unwritten.
1550 struct ext4_map_blocks map;
1551 struct ext4_io_submit io_submit; /* IO submission data */
1552 unsigned int do_map:1;
1553 unsigned int scanned_until_end:1;
1556 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1561 struct folio_batch fbatch;
1562 struct inode *inode = mpd->inode;
1563 struct address_space *mapping = inode->i_mapping;
1565 /* This is necessary when next_page == 0. */
1566 if (mpd->first_page >= mpd->next_page)
1569 mpd->scanned_until_end = 0;
1570 index = mpd->first_page;
1571 end = mpd->next_page - 1;
1573 ext4_lblk_t start, last;
1574 start = index << (PAGE_SHIFT - inode->i_blkbits);
1575 last = end << (PAGE_SHIFT - inode->i_blkbits);
1578 * avoid racing with extent status tree scans made by
1579 * ext4_insert_delayed_block()
1581 down_write(&EXT4_I(inode)->i_data_sem);
1582 ext4_es_remove_extent(inode, start, last - start + 1);
1583 up_write(&EXT4_I(inode)->i_data_sem);
1586 folio_batch_init(&fbatch);
1587 while (index <= end) {
1588 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1591 for (i = 0; i < nr; i++) {
1592 struct folio *folio = fbatch.folios[i];
1594 if (folio->index < mpd->first_page)
1596 if (folio->index + folio_nr_pages(folio) - 1 > end)
1598 BUG_ON(!folio_test_locked(folio));
1599 BUG_ON(folio_test_writeback(folio));
1601 if (folio_mapped(folio))
1602 folio_clear_dirty_for_io(folio);
1603 block_invalidate_folio(folio, 0,
1605 folio_clear_uptodate(folio);
1607 folio_unlock(folio);
1609 folio_batch_release(&fbatch);
1613 static void ext4_print_free_blocks(struct inode *inode)
1615 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1616 struct super_block *sb = inode->i_sb;
1617 struct ext4_inode_info *ei = EXT4_I(inode);
1619 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1620 EXT4_C2B(EXT4_SB(inode->i_sb),
1621 ext4_count_free_clusters(sb)));
1622 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1623 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1624 (long long) EXT4_C2B(EXT4_SB(sb),
1625 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1626 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1627 (long long) EXT4_C2B(EXT4_SB(sb),
1628 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1629 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1630 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1631 ei->i_reserved_data_blocks);
1635 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1636 struct buffer_head *bh)
1638 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1642 * ext4_insert_delayed_block - adds a delayed block to the extents status
1643 * tree, incrementing the reserved cluster/block
1644 * count or making a pending reservation
1647 * @inode - file containing the newly added block
1648 * @lblk - logical block to be added
1650 * Returns 0 on success, negative error code on failure.
1652 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1654 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1656 bool allocated = false;
1657 bool reserved = false;
1660 * If the cluster containing lblk is shared with a delayed,
1661 * written, or unwritten extent in a bigalloc file system, it's
1662 * already been accounted for and does not need to be reserved.
1663 * A pending reservation must be made for the cluster if it's
1664 * shared with a written or unwritten extent and doesn't already
1665 * have one. Written and unwritten extents can be purged from the
1666 * extents status tree if the system is under memory pressure, so
1667 * it's necessary to examine the extent tree if a search of the
1668 * extents status tree doesn't get a match.
1670 if (sbi->s_cluster_ratio == 1) {
1671 ret = ext4_da_reserve_space(inode);
1672 if (ret != 0) /* ENOSPC */
1675 } else { /* bigalloc */
1676 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1677 if (!ext4_es_scan_clu(inode,
1678 &ext4_es_is_mapped, lblk)) {
1679 ret = ext4_clu_mapped(inode,
1680 EXT4_B2C(sbi, lblk));
1684 ret = ext4_da_reserve_space(inode);
1685 if (ret != 0) /* ENOSPC */
1697 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1698 if (ret && reserved)
1699 ext4_da_release_space(inode, 1);
1706 * This function is grabs code from the very beginning of
1707 * ext4_map_blocks, but assumes that the caller is from delayed write
1708 * time. This function looks up the requested blocks and sets the
1709 * buffer delay bit under the protection of i_data_sem.
1711 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1712 struct ext4_map_blocks *map,
1713 struct buffer_head *bh)
1715 struct extent_status es;
1717 sector_t invalid_block = ~((sector_t) 0xffff);
1718 #ifdef ES_AGGRESSIVE_TEST
1719 struct ext4_map_blocks orig_map;
1721 memcpy(&orig_map, map, sizeof(*map));
1724 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1728 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1729 (unsigned long) map->m_lblk);
1731 /* Lookup extent status tree firstly */
1732 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1733 if (ext4_es_is_hole(&es)) {
1735 down_read(&EXT4_I(inode)->i_data_sem);
1740 * Delayed extent could be allocated by fallocate.
1741 * So we need to check it.
1743 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1744 map_bh(bh, inode->i_sb, invalid_block);
1746 set_buffer_delay(bh);
1750 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1751 retval = es.es_len - (iblock - es.es_lblk);
1752 if (retval > map->m_len)
1753 retval = map->m_len;
1754 map->m_len = retval;
1755 if (ext4_es_is_written(&es))
1756 map->m_flags |= EXT4_MAP_MAPPED;
1757 else if (ext4_es_is_unwritten(&es))
1758 map->m_flags |= EXT4_MAP_UNWRITTEN;
1762 #ifdef ES_AGGRESSIVE_TEST
1763 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1769 * Try to see if we can get the block without requesting a new
1770 * file system block.
1772 down_read(&EXT4_I(inode)->i_data_sem);
1773 if (ext4_has_inline_data(inode))
1775 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1776 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1778 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1785 * XXX: __block_prepare_write() unmaps passed block,
1789 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1795 map_bh(bh, inode->i_sb, invalid_block);
1797 set_buffer_delay(bh);
1798 } else if (retval > 0) {
1800 unsigned int status;
1802 if (unlikely(retval != map->m_len)) {
1803 ext4_warning(inode->i_sb,
1804 "ES len assertion failed for inode "
1805 "%lu: retval %d != map->m_len %d",
1806 inode->i_ino, retval, map->m_len);
1810 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1811 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1812 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1813 map->m_pblk, status);
1819 up_read((&EXT4_I(inode)->i_data_sem));
1825 * This is a special get_block_t callback which is used by
1826 * ext4_da_write_begin(). It will either return mapped block or
1827 * reserve space for a single block.
1829 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1830 * We also have b_blocknr = -1 and b_bdev initialized properly
1832 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1833 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1834 * initialized properly.
1836 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1837 struct buffer_head *bh, int create)
1839 struct ext4_map_blocks map;
1842 BUG_ON(create == 0);
1843 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1845 map.m_lblk = iblock;
1849 * first, we need to know whether the block is allocated already
1850 * preallocated blocks are unmapped but should treated
1851 * the same as allocated blocks.
1853 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1857 map_bh(bh, inode->i_sb, map.m_pblk);
1858 ext4_update_bh_state(bh, map.m_flags);
1860 if (buffer_unwritten(bh)) {
1861 /* A delayed write to unwritten bh should be marked
1862 * new and mapped. Mapped ensures that we don't do
1863 * get_block multiple times when we write to the same
1864 * offset and new ensures that we do proper zero out
1865 * for partial write.
1868 set_buffer_mapped(bh);
1873 static int __ext4_journalled_writepage(struct page *page,
1876 struct address_space *mapping = page->mapping;
1877 struct inode *inode = mapping->host;
1878 handle_t *handle = NULL;
1879 int ret = 0, err = 0;
1880 int inline_data = ext4_has_inline_data(inode);
1881 struct buffer_head *inode_bh = NULL;
1884 ClearPageChecked(page);
1887 BUG_ON(page->index != 0);
1888 BUG_ON(len > ext4_get_max_inline_size(inode));
1889 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1890 if (inode_bh == NULL)
1894 * We need to release the page lock before we start the
1895 * journal, so grab a reference so the page won't disappear
1896 * out from under us.
1901 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1902 ext4_writepage_trans_blocks(inode));
1903 if (IS_ERR(handle)) {
1904 ret = PTR_ERR(handle);
1906 goto out_no_pagelock;
1908 BUG_ON(!ext4_handle_valid(handle));
1912 size = i_size_read(inode);
1913 if (page->mapping != mapping || page_offset(page) > size) {
1914 /* The page got truncated from under us */
1915 ext4_journal_stop(handle);
1921 ret = ext4_mark_inode_dirty(handle, inode);
1923 struct buffer_head *page_bufs = page_buffers(page);
1925 if (page->index == size >> PAGE_SHIFT)
1926 len = size & ~PAGE_MASK;
1930 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1931 NULL, do_journal_get_write_access);
1933 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1934 NULL, write_end_fn);
1938 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1941 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1942 err = ext4_journal_stop(handle);
1946 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1955 * Note that we don't need to start a transaction unless we're journaling data
1956 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1957 * need to file the inode to the transaction's list in ordered mode because if
1958 * we are writing back data added by write(), the inode is already there and if
1959 * we are writing back data modified via mmap(), no one guarantees in which
1960 * transaction the data will hit the disk. In case we are journaling data, we
1961 * cannot start transaction directly because transaction start ranks above page
1962 * lock so we have to do some magic.
1964 * This function can get called via...
1965 * - ext4_writepages after taking page lock (have journal handle)
1966 * - journal_submit_inode_data_buffers (no journal handle)
1967 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1968 * - grab_page_cache when doing write_begin (have journal handle)
1970 * We don't do any block allocation in this function. If we have page with
1971 * multiple blocks we need to write those buffer_heads that are mapped. This
1972 * is important for mmaped based write. So if we do with blocksize 1K
1973 * truncate(f, 1024);
1974 * a = mmap(f, 0, 4096);
1976 * truncate(f, 4096);
1977 * we have in the page first buffer_head mapped via page_mkwrite call back
1978 * but other buffer_heads would be unmapped but dirty (dirty done via the
1979 * do_wp_page). So writepage should write the first block. If we modify
1980 * the mmap area beyond 1024 we will again get a page_fault and the
1981 * page_mkwrite callback will do the block allocation and mark the
1982 * buffer_heads mapped.
1984 * We redirty the page if we have any buffer_heads that is either delay or
1985 * unwritten in the page.
1987 * We can get recursively called as show below.
1989 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1992 * But since we don't do any block allocation we should not deadlock.
1993 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1995 static int ext4_writepage(struct page *page,
1996 struct writeback_control *wbc)
1998 struct folio *folio = page_folio(page);
2002 struct buffer_head *page_bufs = NULL;
2003 struct inode *inode = page->mapping->host;
2004 struct ext4_io_submit io_submit;
2005 bool keep_towrite = false;
2007 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2008 folio_invalidate(folio, 0, folio_size(folio));
2009 folio_unlock(folio);
2013 trace_ext4_writepage(page);
2014 size = i_size_read(inode);
2015 if (page->index == size >> PAGE_SHIFT &&
2016 !ext4_verity_in_progress(inode))
2017 len = size & ~PAGE_MASK;
2021 /* Should never happen but for bugs in other kernel subsystems */
2022 if (!page_has_buffers(page)) {
2023 ext4_warning_inode(inode,
2024 "page %lu does not have buffers attached", page->index);
2025 ClearPageDirty(page);
2030 page_bufs = page_buffers(page);
2032 * We cannot do block allocation or other extent handling in this
2033 * function. If there are buffers needing that, we have to redirty
2034 * the page. But we may reach here when we do a journal commit via
2035 * journal_submit_inode_data_buffers() and in that case we must write
2036 * allocated buffers to achieve data=ordered mode guarantees.
2038 * Also, if there is only one buffer per page (the fs block
2039 * size == the page size), if one buffer needs block
2040 * allocation or needs to modify the extent tree to clear the
2041 * unwritten flag, we know that the page can't be written at
2042 * all, so we might as well refuse the write immediately.
2043 * Unfortunately if the block size != page size, we can't as
2044 * easily detect this case using ext4_walk_page_buffers(), but
2045 * for the extremely common case, this is an optimization that
2046 * skips a useless round trip through ext4_bio_write_page().
2048 if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2049 ext4_bh_delay_or_unwritten)) {
2050 redirty_page_for_writepage(wbc, page);
2051 if ((current->flags & PF_MEMALLOC) ||
2052 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2054 * For memory cleaning there's no point in writing only
2055 * some buffers. So just bail out. Warn if we came here
2056 * from direct reclaim.
2058 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2063 keep_towrite = true;
2066 if (PageChecked(page) && ext4_should_journal_data(inode))
2068 * It's mmapped pagecache. Add buffers and journal it. There
2069 * doesn't seem much point in redirtying the page here.
2071 return __ext4_journalled_writepage(page, len);
2073 ext4_io_submit_init(&io_submit, wbc);
2074 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2075 if (!io_submit.io_end) {
2076 redirty_page_for_writepage(wbc, page);
2080 ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite);
2081 ext4_io_submit(&io_submit);
2082 /* Drop io_end reference we got from init */
2083 ext4_put_io_end_defer(io_submit.io_end);
2087 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2093 BUG_ON(page->index != mpd->first_page);
2094 clear_page_dirty_for_io(page);
2096 * We have to be very careful here! Nothing protects writeback path
2097 * against i_size changes and the page can be writeably mapped into
2098 * page tables. So an application can be growing i_size and writing
2099 * data through mmap while writeback runs. clear_page_dirty_for_io()
2100 * write-protects our page in page tables and the page cannot get
2101 * written to again until we release page lock. So only after
2102 * clear_page_dirty_for_io() we are safe to sample i_size for
2103 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2104 * on the barrier provided by TestClearPageDirty in
2105 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2106 * after page tables are updated.
2108 size = i_size_read(mpd->inode);
2109 if (page->index == size >> PAGE_SHIFT &&
2110 !ext4_verity_in_progress(mpd->inode))
2111 len = size & ~PAGE_MASK;
2114 err = ext4_bio_write_page(&mpd->io_submit, page, len, false);
2116 mpd->wbc->nr_to_write--;
2122 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2125 * mballoc gives us at most this number of blocks...
2126 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2127 * The rest of mballoc seems to handle chunks up to full group size.
2129 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2132 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2134 * @mpd - extent of blocks
2135 * @lblk - logical number of the block in the file
2136 * @bh - buffer head we want to add to the extent
2138 * The function is used to collect contig. blocks in the same state. If the
2139 * buffer doesn't require mapping for writeback and we haven't started the
2140 * extent of buffers to map yet, the function returns 'true' immediately - the
2141 * caller can write the buffer right away. Otherwise the function returns true
2142 * if the block has been added to the extent, false if the block couldn't be
2145 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2146 struct buffer_head *bh)
2148 struct ext4_map_blocks *map = &mpd->map;
2150 /* Buffer that doesn't need mapping for writeback? */
2151 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2152 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2153 /* So far no extent to map => we write the buffer right away */
2154 if (map->m_len == 0)
2159 /* First block in the extent? */
2160 if (map->m_len == 0) {
2161 /* We cannot map unless handle is started... */
2166 map->m_flags = bh->b_state & BH_FLAGS;
2170 /* Don't go larger than mballoc is willing to allocate */
2171 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2174 /* Can we merge the block to our big extent? */
2175 if (lblk == map->m_lblk + map->m_len &&
2176 (bh->b_state & BH_FLAGS) == map->m_flags) {
2184 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2186 * @mpd - extent of blocks for mapping
2187 * @head - the first buffer in the page
2188 * @bh - buffer we should start processing from
2189 * @lblk - logical number of the block in the file corresponding to @bh
2191 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2192 * the page for IO if all buffers in this page were mapped and there's no
2193 * accumulated extent of buffers to map or add buffers in the page to the
2194 * extent of buffers to map. The function returns 1 if the caller can continue
2195 * by processing the next page, 0 if it should stop adding buffers to the
2196 * extent to map because we cannot extend it anymore. It can also return value
2197 * < 0 in case of error during IO submission.
2199 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2200 struct buffer_head *head,
2201 struct buffer_head *bh,
2204 struct inode *inode = mpd->inode;
2206 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2207 >> inode->i_blkbits;
2209 if (ext4_verity_in_progress(inode))
2210 blocks = EXT_MAX_BLOCKS;
2213 BUG_ON(buffer_locked(bh));
2215 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2216 /* Found extent to map? */
2219 /* Buffer needs mapping and handle is not started? */
2222 /* Everything mapped so far and we hit EOF */
2225 } while (lblk++, (bh = bh->b_this_page) != head);
2226 /* So far everything mapped? Submit the page for IO. */
2227 if (mpd->map.m_len == 0) {
2228 err = mpage_submit_page(mpd, head->b_page);
2232 if (lblk >= blocks) {
2233 mpd->scanned_until_end = 1;
2240 * mpage_process_page - update page buffers corresponding to changed extent and
2241 * may submit fully mapped page for IO
2243 * @mpd - description of extent to map, on return next extent to map
2244 * @m_lblk - logical block mapping.
2245 * @m_pblk - corresponding physical mapping.
2246 * @map_bh - determines on return whether this page requires any further
2248 * Scan given page buffers corresponding to changed extent and update buffer
2249 * state according to new extent state.
2250 * We map delalloc buffers to their physical location, clear unwritten bits.
2251 * If the given page is not fully mapped, we update @map to the next extent in
2252 * the given page that needs mapping & return @map_bh as true.
2254 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2255 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2258 struct buffer_head *head, *bh;
2259 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2260 ext4_lblk_t lblk = *m_lblk;
2261 ext4_fsblk_t pblock = *m_pblk;
2263 int blkbits = mpd->inode->i_blkbits;
2264 ssize_t io_end_size = 0;
2265 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2267 bh = head = page_buffers(page);
2269 if (lblk < mpd->map.m_lblk)
2271 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2273 * Buffer after end of mapped extent.
2274 * Find next buffer in the page to map.
2277 mpd->map.m_flags = 0;
2278 io_end_vec->size += io_end_size;
2280 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2283 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2284 io_end_vec = ext4_alloc_io_end_vec(io_end);
2285 if (IS_ERR(io_end_vec)) {
2286 err = PTR_ERR(io_end_vec);
2289 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2294 if (buffer_delay(bh)) {
2295 clear_buffer_delay(bh);
2296 bh->b_blocknr = pblock++;
2298 clear_buffer_unwritten(bh);
2299 io_end_size += (1 << blkbits);
2300 } while (lblk++, (bh = bh->b_this_page) != head);
2302 io_end_vec->size += io_end_size;
2311 * mpage_map_buffers - update buffers corresponding to changed extent and
2312 * submit fully mapped pages for IO
2314 * @mpd - description of extent to map, on return next extent to map
2316 * Scan buffers corresponding to changed extent (we expect corresponding pages
2317 * to be already locked) and update buffer state according to new extent state.
2318 * We map delalloc buffers to their physical location, clear unwritten bits,
2319 * and mark buffers as uninit when we perform writes to unwritten extents
2320 * and do extent conversion after IO is finished. If the last page is not fully
2321 * mapped, we update @map to the next extent in the last page that needs
2322 * mapping. Otherwise we submit the page for IO.
2324 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2326 struct folio_batch fbatch;
2328 struct inode *inode = mpd->inode;
2329 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2332 ext4_fsblk_t pblock;
2334 bool map_bh = false;
2336 start = mpd->map.m_lblk >> bpp_bits;
2337 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2338 lblk = start << bpp_bits;
2339 pblock = mpd->map.m_pblk;
2341 folio_batch_init(&fbatch);
2342 while (start <= end) {
2343 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2346 for (i = 0; i < nr; i++) {
2347 struct page *page = &fbatch.folios[i]->page;
2349 err = mpage_process_page(mpd, page, &lblk, &pblock,
2352 * If map_bh is true, means page may require further bh
2353 * mapping, or maybe the page was submitted for IO.
2354 * So we return to call further extent mapping.
2356 if (err < 0 || map_bh)
2358 /* Page fully mapped - let IO run! */
2359 err = mpage_submit_page(mpd, page);
2363 folio_batch_release(&fbatch);
2365 /* Extent fully mapped and matches with page boundary. We are done. */
2367 mpd->map.m_flags = 0;
2370 folio_batch_release(&fbatch);
2374 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2376 struct inode *inode = mpd->inode;
2377 struct ext4_map_blocks *map = &mpd->map;
2378 int get_blocks_flags;
2379 int err, dioread_nolock;
2381 trace_ext4_da_write_pages_extent(inode, map);
2383 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2384 * to convert an unwritten extent to be initialized (in the case
2385 * where we have written into one or more preallocated blocks). It is
2386 * possible that we're going to need more metadata blocks than
2387 * previously reserved. However we must not fail because we're in
2388 * writeback and there is nothing we can do about it so it might result
2389 * in data loss. So use reserved blocks to allocate metadata if
2392 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2393 * the blocks in question are delalloc blocks. This indicates
2394 * that the blocks and quotas has already been checked when
2395 * the data was copied into the page cache.
2397 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2398 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2399 EXT4_GET_BLOCKS_IO_SUBMIT;
2400 dioread_nolock = ext4_should_dioread_nolock(inode);
2402 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2403 if (map->m_flags & BIT(BH_Delay))
2404 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2406 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2409 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2410 if (!mpd->io_submit.io_end->handle &&
2411 ext4_handle_valid(handle)) {
2412 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2413 handle->h_rsv_handle = NULL;
2415 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2418 BUG_ON(map->m_len == 0);
2423 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2424 * mpd->len and submit pages underlying it for IO
2426 * @handle - handle for journal operations
2427 * @mpd - extent to map
2428 * @give_up_on_write - we set this to true iff there is a fatal error and there
2429 * is no hope of writing the data. The caller should discard
2430 * dirty pages to avoid infinite loops.
2432 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2433 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2434 * them to initialized or split the described range from larger unwritten
2435 * extent. Note that we need not map all the described range since allocation
2436 * can return less blocks or the range is covered by more unwritten extents. We
2437 * cannot map more because we are limited by reserved transaction credits. On
2438 * the other hand we always make sure that the last touched page is fully
2439 * mapped so that it can be written out (and thus forward progress is
2440 * guaranteed). After mapping we submit all mapped pages for IO.
2442 static int mpage_map_and_submit_extent(handle_t *handle,
2443 struct mpage_da_data *mpd,
2444 bool *give_up_on_write)
2446 struct inode *inode = mpd->inode;
2447 struct ext4_map_blocks *map = &mpd->map;
2451 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2452 struct ext4_io_end_vec *io_end_vec;
2454 io_end_vec = ext4_alloc_io_end_vec(io_end);
2455 if (IS_ERR(io_end_vec))
2456 return PTR_ERR(io_end_vec);
2457 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2459 err = mpage_map_one_extent(handle, mpd);
2461 struct super_block *sb = inode->i_sb;
2463 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2464 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2465 goto invalidate_dirty_pages;
2467 * Let the uper layers retry transient errors.
2468 * In the case of ENOSPC, if ext4_count_free_blocks()
2469 * is non-zero, a commit should free up blocks.
2471 if ((err == -ENOMEM) ||
2472 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2474 goto update_disksize;
2477 ext4_msg(sb, KERN_CRIT,
2478 "Delayed block allocation failed for "
2479 "inode %lu at logical offset %llu with"
2480 " max blocks %u with error %d",
2482 (unsigned long long)map->m_lblk,
2483 (unsigned)map->m_len, -err);
2484 ext4_msg(sb, KERN_CRIT,
2485 "This should not happen!! Data will "
2488 ext4_print_free_blocks(inode);
2489 invalidate_dirty_pages:
2490 *give_up_on_write = true;
2495 * Update buffer state, submit mapped pages, and get us new
2498 err = mpage_map_and_submit_buffers(mpd);
2500 goto update_disksize;
2501 } while (map->m_len);
2505 * Update on-disk size after IO is submitted. Races with
2506 * truncate are avoided by checking i_size under i_data_sem.
2508 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2509 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2513 down_write(&EXT4_I(inode)->i_data_sem);
2514 i_size = i_size_read(inode);
2515 if (disksize > i_size)
2517 if (disksize > EXT4_I(inode)->i_disksize)
2518 EXT4_I(inode)->i_disksize = disksize;
2519 up_write(&EXT4_I(inode)->i_data_sem);
2520 err2 = ext4_mark_inode_dirty(handle, inode);
2522 ext4_error_err(inode->i_sb, -err2,
2523 "Failed to mark inode %lu dirty",
2533 * Calculate the total number of credits to reserve for one writepages
2534 * iteration. This is called from ext4_writepages(). We map an extent of
2535 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2536 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2537 * bpp - 1 blocks in bpp different extents.
2539 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2541 int bpp = ext4_journal_blocks_per_page(inode);
2543 return ext4_meta_trans_blocks(inode,
2544 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2548 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2549 * and underlying extent to map
2551 * @mpd - where to look for pages
2553 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2554 * IO immediately. When we find a page which isn't mapped we start accumulating
2555 * extent of buffers underlying these pages that needs mapping (formed by
2556 * either delayed or unwritten buffers). We also lock the pages containing
2557 * these buffers. The extent found is returned in @mpd structure (starting at
2558 * mpd->lblk with length mpd->len blocks).
2560 * Note that this function can attach bios to one io_end structure which are
2561 * neither logically nor physically contiguous. Although it may seem as an
2562 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2563 * case as we need to track IO to all buffers underlying a page in one io_end.
2565 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2567 struct address_space *mapping = mpd->inode->i_mapping;
2568 struct pagevec pvec;
2569 unsigned int nr_pages;
2570 long left = mpd->wbc->nr_to_write;
2571 pgoff_t index = mpd->first_page;
2572 pgoff_t end = mpd->last_page;
2575 int blkbits = mpd->inode->i_blkbits;
2577 struct buffer_head *head;
2579 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2580 tag = PAGECACHE_TAG_TOWRITE;
2582 tag = PAGECACHE_TAG_DIRTY;
2584 pagevec_init(&pvec);
2586 mpd->next_page = index;
2587 while (index <= end) {
2588 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2593 for (i = 0; i < nr_pages; i++) {
2594 struct page *page = pvec.pages[i];
2597 * Accumulated enough dirty pages? This doesn't apply
2598 * to WB_SYNC_ALL mode. For integrity sync we have to
2599 * keep going because someone may be concurrently
2600 * dirtying pages, and we might have synced a lot of
2601 * newly appeared dirty pages, but have not synced all
2602 * of the old dirty pages.
2604 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2607 /* If we can't merge this page, we are done. */
2608 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2613 * If the page is no longer dirty, or its mapping no
2614 * longer corresponds to inode we are writing (which
2615 * means it has been truncated or invalidated), or the
2616 * page is already under writeback and we are not doing
2617 * a data integrity writeback, skip the page
2619 if (!PageDirty(page) ||
2620 (PageWriteback(page) &&
2621 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2622 unlikely(page->mapping != mapping)) {
2627 wait_on_page_writeback(page);
2628 BUG_ON(PageWriteback(page));
2631 * Should never happen but for buggy code in
2632 * other subsystems that call
2633 * set_page_dirty() without properly warning
2634 * the file system first. See [1] for more
2637 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2639 if (!page_has_buffers(page)) {
2640 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2641 ClearPageDirty(page);
2646 if (mpd->map.m_len == 0)
2647 mpd->first_page = page->index;
2648 mpd->next_page = page->index + 1;
2649 /* Add all dirty buffers to mpd */
2650 lblk = ((ext4_lblk_t)page->index) <<
2651 (PAGE_SHIFT - blkbits);
2652 head = page_buffers(page);
2653 err = mpage_process_page_bufs(mpd, head, head, lblk);
2659 pagevec_release(&pvec);
2662 mpd->scanned_until_end = 1;
2665 pagevec_release(&pvec);
2669 static int ext4_writepages(struct address_space *mapping,
2670 struct writeback_control *wbc)
2672 pgoff_t writeback_index = 0;
2673 long nr_to_write = wbc->nr_to_write;
2674 int range_whole = 0;
2676 handle_t *handle = NULL;
2677 struct mpage_da_data mpd;
2678 struct inode *inode = mapping->host;
2679 int needed_blocks, rsv_blocks = 0, ret = 0;
2680 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2681 struct blk_plug plug;
2682 bool give_up_on_write = false;
2684 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2687 percpu_down_read(&sbi->s_writepages_rwsem);
2688 trace_ext4_writepages(inode, wbc);
2691 * No pages to write? This is mainly a kludge to avoid starting
2692 * a transaction for special inodes like journal inode on last iput()
2693 * because that could violate lock ordering on umount
2695 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2696 goto out_writepages;
2698 if (ext4_should_journal_data(inode)) {
2699 ret = generic_writepages(mapping, wbc);
2700 goto out_writepages;
2704 * If the filesystem has aborted, it is read-only, so return
2705 * right away instead of dumping stack traces later on that
2706 * will obscure the real source of the problem. We test
2707 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2708 * the latter could be true if the filesystem is mounted
2709 * read-only, and in that case, ext4_writepages should
2710 * *never* be called, so if that ever happens, we would want
2713 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2714 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2716 goto out_writepages;
2720 * If we have inline data and arrive here, it means that
2721 * we will soon create the block for the 1st page, so
2722 * we'd better clear the inline data here.
2724 if (ext4_has_inline_data(inode)) {
2725 /* Just inode will be modified... */
2726 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2727 if (IS_ERR(handle)) {
2728 ret = PTR_ERR(handle);
2729 goto out_writepages;
2731 BUG_ON(ext4_test_inode_state(inode,
2732 EXT4_STATE_MAY_INLINE_DATA));
2733 ext4_destroy_inline_data(handle, inode);
2734 ext4_journal_stop(handle);
2737 if (ext4_should_dioread_nolock(inode)) {
2739 * We may need to convert up to one extent per block in
2740 * the page and we may dirty the inode.
2742 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2743 PAGE_SIZE >> inode->i_blkbits);
2746 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2749 if (wbc->range_cyclic) {
2750 writeback_index = mapping->writeback_index;
2751 if (writeback_index)
2753 mpd.first_page = writeback_index;
2756 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2757 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2762 ext4_io_submit_init(&mpd.io_submit, wbc);
2764 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2765 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2766 blk_start_plug(&plug);
2769 * First writeback pages that don't need mapping - we can avoid
2770 * starting a transaction unnecessarily and also avoid being blocked
2771 * in the block layer on device congestion while having transaction
2775 mpd.scanned_until_end = 0;
2776 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2777 if (!mpd.io_submit.io_end) {
2781 ret = mpage_prepare_extent_to_map(&mpd);
2782 /* Unlock pages we didn't use */
2783 mpage_release_unused_pages(&mpd, false);
2784 /* Submit prepared bio */
2785 ext4_io_submit(&mpd.io_submit);
2786 ext4_put_io_end_defer(mpd.io_submit.io_end);
2787 mpd.io_submit.io_end = NULL;
2791 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2792 /* For each extent of pages we use new io_end */
2793 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2794 if (!mpd.io_submit.io_end) {
2800 * We have two constraints: We find one extent to map and we
2801 * must always write out whole page (makes a difference when
2802 * blocksize < pagesize) so that we don't block on IO when we
2803 * try to write out the rest of the page. Journalled mode is
2804 * not supported by delalloc.
2806 BUG_ON(ext4_should_journal_data(inode));
2807 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2809 /* start a new transaction */
2810 handle = ext4_journal_start_with_reserve(inode,
2811 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2812 if (IS_ERR(handle)) {
2813 ret = PTR_ERR(handle);
2814 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2815 "%ld pages, ino %lu; err %d", __func__,
2816 wbc->nr_to_write, inode->i_ino, ret);
2817 /* Release allocated io_end */
2818 ext4_put_io_end(mpd.io_submit.io_end);
2819 mpd.io_submit.io_end = NULL;
2824 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2825 ret = mpage_prepare_extent_to_map(&mpd);
2826 if (!ret && mpd.map.m_len)
2827 ret = mpage_map_and_submit_extent(handle, &mpd,
2830 * Caution: If the handle is synchronous,
2831 * ext4_journal_stop() can wait for transaction commit
2832 * to finish which may depend on writeback of pages to
2833 * complete or on page lock to be released. In that
2834 * case, we have to wait until after we have
2835 * submitted all the IO, released page locks we hold,
2836 * and dropped io_end reference (for extent conversion
2837 * to be able to complete) before stopping the handle.
2839 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2840 ext4_journal_stop(handle);
2844 /* Unlock pages we didn't use */
2845 mpage_release_unused_pages(&mpd, give_up_on_write);
2846 /* Submit prepared bio */
2847 ext4_io_submit(&mpd.io_submit);
2850 * Drop our io_end reference we got from init. We have
2851 * to be careful and use deferred io_end finishing if
2852 * we are still holding the transaction as we can
2853 * release the last reference to io_end which may end
2854 * up doing unwritten extent conversion.
2857 ext4_put_io_end_defer(mpd.io_submit.io_end);
2858 ext4_journal_stop(handle);
2860 ext4_put_io_end(mpd.io_submit.io_end);
2861 mpd.io_submit.io_end = NULL;
2863 if (ret == -ENOSPC && sbi->s_journal) {
2865 * Commit the transaction which would
2866 * free blocks released in the transaction
2869 jbd2_journal_force_commit_nested(sbi->s_journal);
2873 /* Fatal error - ENOMEM, EIO... */
2878 blk_finish_plug(&plug);
2879 if (!ret && !cycled && wbc->nr_to_write > 0) {
2881 mpd.last_page = writeback_index - 1;
2887 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2889 * Set the writeback_index so that range_cyclic
2890 * mode will write it back later
2892 mapping->writeback_index = mpd.first_page;
2895 trace_ext4_writepages_result(inode, wbc, ret,
2896 nr_to_write - wbc->nr_to_write);
2897 percpu_up_read(&sbi->s_writepages_rwsem);
2901 static int ext4_dax_writepages(struct address_space *mapping,
2902 struct writeback_control *wbc)
2905 long nr_to_write = wbc->nr_to_write;
2906 struct inode *inode = mapping->host;
2907 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2909 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2912 percpu_down_read(&sbi->s_writepages_rwsem);
2913 trace_ext4_writepages(inode, wbc);
2915 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2916 trace_ext4_writepages_result(inode, wbc, ret,
2917 nr_to_write - wbc->nr_to_write);
2918 percpu_up_read(&sbi->s_writepages_rwsem);
2922 static int ext4_nonda_switch(struct super_block *sb)
2924 s64 free_clusters, dirty_clusters;
2925 struct ext4_sb_info *sbi = EXT4_SB(sb);
2928 * switch to non delalloc mode if we are running low
2929 * on free block. The free block accounting via percpu
2930 * counters can get slightly wrong with percpu_counter_batch getting
2931 * accumulated on each CPU without updating global counters
2932 * Delalloc need an accurate free block accounting. So switch
2933 * to non delalloc when we are near to error range.
2936 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2938 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2940 * Start pushing delalloc when 1/2 of free blocks are dirty.
2942 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2943 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2945 if (2 * free_clusters < 3 * dirty_clusters ||
2946 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2948 * free block count is less than 150% of dirty blocks
2949 * or free blocks is less than watermark
2956 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2957 loff_t pos, unsigned len,
2958 struct page **pagep, void **fsdata)
2960 int ret, retries = 0;
2963 struct inode *inode = mapping->host;
2965 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2968 index = pos >> PAGE_SHIFT;
2970 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2971 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2972 return ext4_write_begin(file, mapping, pos,
2973 len, pagep, fsdata);
2975 *fsdata = (void *)0;
2976 trace_ext4_da_write_begin(inode, pos, len);
2978 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2979 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2988 page = grab_cache_page_write_begin(mapping, index);
2992 /* In case writeback began while the page was unlocked */
2993 wait_for_stable_page(page);
2995 #ifdef CONFIG_FS_ENCRYPTION
2996 ret = ext4_block_write_begin(page, pos, len,
2997 ext4_da_get_block_prep);
2999 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3005 * block_write_begin may have instantiated a few blocks
3006 * outside i_size. Trim these off again. Don't need
3007 * i_size_read because we hold inode lock.
3009 if (pos + len > inode->i_size)
3010 ext4_truncate_failed_write(inode);
3012 if (ret == -ENOSPC &&
3013 ext4_should_retry_alloc(inode->i_sb, &retries))
3023 * Check if we should update i_disksize
3024 * when write to the end of file but not require block allocation
3026 static int ext4_da_should_update_i_disksize(struct page *page,
3027 unsigned long offset)
3029 struct buffer_head *bh;
3030 struct inode *inode = page->mapping->host;
3034 bh = page_buffers(page);
3035 idx = offset >> inode->i_blkbits;
3037 for (i = 0; i < idx; i++)
3038 bh = bh->b_this_page;
3040 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3045 static int ext4_da_write_end(struct file *file,
3046 struct address_space *mapping,
3047 loff_t pos, unsigned len, unsigned copied,
3048 struct page *page, void *fsdata)
3050 struct inode *inode = mapping->host;
3052 unsigned long start, end;
3053 int write_mode = (int)(unsigned long)fsdata;
3055 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3056 return ext4_write_end(file, mapping, pos,
3057 len, copied, page, fsdata);
3059 trace_ext4_da_write_end(inode, pos, len, copied);
3061 if (write_mode != CONVERT_INLINE_DATA &&
3062 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3063 ext4_has_inline_data(inode))
3064 return ext4_write_inline_data_end(inode, pos, len, copied, page);
3066 start = pos & (PAGE_SIZE - 1);
3067 end = start + copied - 1;
3070 * Since we are holding inode lock, we are sure i_disksize <=
3071 * i_size. We also know that if i_disksize < i_size, there are
3072 * delalloc writes pending in the range upto i_size. If the end of
3073 * the current write is <= i_size, there's no need to touch
3074 * i_disksize since writeback will push i_disksize upto i_size
3075 * eventually. If the end of the current write is > i_size and
3076 * inside an allocated block (ext4_da_should_update_i_disksize()
3077 * check), we need to update i_disksize here as neither
3078 * ext4_writepage() nor certain ext4_writepages() paths not
3079 * allocating blocks update i_disksize.
3081 * Note that we defer inode dirtying to generic_write_end() /
3082 * ext4_da_write_inline_data_end().
3084 new_i_size = pos + copied;
3085 if (copied && new_i_size > inode->i_size &&
3086 ext4_da_should_update_i_disksize(page, end))
3087 ext4_update_i_disksize(inode, new_i_size);
3089 return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3093 * Force all delayed allocation blocks to be allocated for a given inode.
3095 int ext4_alloc_da_blocks(struct inode *inode)
3097 trace_ext4_alloc_da_blocks(inode);
3099 if (!EXT4_I(inode)->i_reserved_data_blocks)
3103 * We do something simple for now. The filemap_flush() will
3104 * also start triggering a write of the data blocks, which is
3105 * not strictly speaking necessary (and for users of
3106 * laptop_mode, not even desirable). However, to do otherwise
3107 * would require replicating code paths in:
3109 * ext4_writepages() ->
3110 * write_cache_pages() ---> (via passed in callback function)
3111 * __mpage_da_writepage() -->
3112 * mpage_add_bh_to_extent()
3113 * mpage_da_map_blocks()
3115 * The problem is that write_cache_pages(), located in
3116 * mm/page-writeback.c, marks pages clean in preparation for
3117 * doing I/O, which is not desirable if we're not planning on
3120 * We could call write_cache_pages(), and then redirty all of
3121 * the pages by calling redirty_page_for_writepage() but that
3122 * would be ugly in the extreme. So instead we would need to
3123 * replicate parts of the code in the above functions,
3124 * simplifying them because we wouldn't actually intend to
3125 * write out the pages, but rather only collect contiguous
3126 * logical block extents, call the multi-block allocator, and
3127 * then update the buffer heads with the block allocations.
3129 * For now, though, we'll cheat by calling filemap_flush(),
3130 * which will map the blocks, and start the I/O, but not
3131 * actually wait for the I/O to complete.
3133 return filemap_flush(inode->i_mapping);
3137 * bmap() is special. It gets used by applications such as lilo and by
3138 * the swapper to find the on-disk block of a specific piece of data.
3140 * Naturally, this is dangerous if the block concerned is still in the
3141 * journal. If somebody makes a swapfile on an ext4 data-journaling
3142 * filesystem and enables swap, then they may get a nasty shock when the
3143 * data getting swapped to that swapfile suddenly gets overwritten by
3144 * the original zero's written out previously to the journal and
3145 * awaiting writeback in the kernel's buffer cache.
3147 * So, if we see any bmap calls here on a modified, data-journaled file,
3148 * take extra steps to flush any blocks which might be in the cache.
3150 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3152 struct inode *inode = mapping->host;
3157 inode_lock_shared(inode);
3159 * We can get here for an inline file via the FIBMAP ioctl
3161 if (ext4_has_inline_data(inode))
3164 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3165 test_opt(inode->i_sb, DELALLOC)) {
3167 * With delalloc we want to sync the file
3168 * so that we can make sure we allocate
3171 filemap_write_and_wait(mapping);
3174 if (EXT4_JOURNAL(inode) &&
3175 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3177 * This is a REALLY heavyweight approach, but the use of
3178 * bmap on dirty files is expected to be extremely rare:
3179 * only if we run lilo or swapon on a freshly made file
3180 * do we expect this to happen.
3182 * (bmap requires CAP_SYS_RAWIO so this does not
3183 * represent an unprivileged user DOS attack --- we'd be
3184 * in trouble if mortal users could trigger this path at
3187 * NB. EXT4_STATE_JDATA is not set on files other than
3188 * regular files. If somebody wants to bmap a directory
3189 * or symlink and gets confused because the buffer
3190 * hasn't yet been flushed to disk, they deserve
3191 * everything they get.
3194 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3195 journal = EXT4_JOURNAL(inode);
3196 jbd2_journal_lock_updates(journal);
3197 err = jbd2_journal_flush(journal, 0);
3198 jbd2_journal_unlock_updates(journal);
3204 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3207 inode_unlock_shared(inode);
3211 static int ext4_read_folio(struct file *file, struct folio *folio)
3213 struct page *page = &folio->page;
3215 struct inode *inode = page->mapping->host;
3217 trace_ext4_readpage(page);
3219 if (ext4_has_inline_data(inode))
3220 ret = ext4_readpage_inline(inode, page);
3223 return ext4_mpage_readpages(inode, NULL, page);
3228 static void ext4_readahead(struct readahead_control *rac)
3230 struct inode *inode = rac->mapping->host;
3232 /* If the file has inline data, no need to do readahead. */
3233 if (ext4_has_inline_data(inode))
3236 ext4_mpage_readpages(inode, rac, NULL);
3239 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3242 trace_ext4_invalidate_folio(folio, offset, length);
3244 /* No journalling happens on data buffers when this function is used */
3245 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3247 block_invalidate_folio(folio, offset, length);
3250 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3251 size_t offset, size_t length)
3253 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3255 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3258 * If it's a full truncate we just forget about the pending dirtying
3260 if (offset == 0 && length == folio_size(folio))
3261 folio_clear_checked(folio);
3263 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3266 /* Wrapper for aops... */
3267 static void ext4_journalled_invalidate_folio(struct folio *folio,
3271 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3274 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3276 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3278 trace_ext4_releasepage(&folio->page);
3280 /* Page has dirty journalled data -> cannot release */
3281 if (folio_test_checked(folio))
3284 return jbd2_journal_try_to_free_buffers(journal, folio);
3286 return try_to_free_buffers(folio);
3289 static bool ext4_inode_datasync_dirty(struct inode *inode)
3291 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3294 if (jbd2_transaction_committed(journal,
3295 EXT4_I(inode)->i_datasync_tid))
3297 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3298 return !list_empty(&EXT4_I(inode)->i_fc_list);
3302 /* Any metadata buffers to write? */
3303 if (!list_empty(&inode->i_mapping->private_list))
3305 return inode->i_state & I_DIRTY_DATASYNC;
3308 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3309 struct ext4_map_blocks *map, loff_t offset,
3310 loff_t length, unsigned int flags)
3312 u8 blkbits = inode->i_blkbits;
3315 * Writes that span EOF might trigger an I/O size update on completion,
3316 * so consider them to be dirty for the purpose of O_DSYNC, even if
3317 * there is no other metadata changes being made or are pending.
3320 if (ext4_inode_datasync_dirty(inode) ||
3321 offset + length > i_size_read(inode))
3322 iomap->flags |= IOMAP_F_DIRTY;
3324 if (map->m_flags & EXT4_MAP_NEW)
3325 iomap->flags |= IOMAP_F_NEW;
3327 if (flags & IOMAP_DAX)
3328 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3330 iomap->bdev = inode->i_sb->s_bdev;
3331 iomap->offset = (u64) map->m_lblk << blkbits;
3332 iomap->length = (u64) map->m_len << blkbits;
3334 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3335 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3336 iomap->flags |= IOMAP_F_MERGED;
3339 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3340 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3341 * set. In order for any allocated unwritten extents to be converted
3342 * into written extents correctly within the ->end_io() handler, we
3343 * need to ensure that the iomap->type is set appropriately. Hence, the
3344 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3347 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3348 iomap->type = IOMAP_UNWRITTEN;
3349 iomap->addr = (u64) map->m_pblk << blkbits;
3350 if (flags & IOMAP_DAX)
3351 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3352 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3353 iomap->type = IOMAP_MAPPED;
3354 iomap->addr = (u64) map->m_pblk << blkbits;
3355 if (flags & IOMAP_DAX)
3356 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3358 iomap->type = IOMAP_HOLE;
3359 iomap->addr = IOMAP_NULL_ADDR;
3363 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3367 u8 blkbits = inode->i_blkbits;
3368 int ret, dio_credits, m_flags = 0, retries = 0;
3371 * Trim the mapping request to the maximum value that we can map at
3372 * once for direct I/O.
3374 if (map->m_len > DIO_MAX_BLOCKS)
3375 map->m_len = DIO_MAX_BLOCKS;
3376 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3380 * Either we allocate blocks and then don't get an unwritten extent, so
3381 * in that case we have reserved enough credits. Or, the blocks are
3382 * already allocated and unwritten. In that case, the extent conversion
3383 * fits into the credits as well.
3385 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3387 return PTR_ERR(handle);
3390 * DAX and direct I/O are the only two operations that are currently
3391 * supported with IOMAP_WRITE.
3393 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3394 if (flags & IOMAP_DAX)
3395 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3397 * We use i_size instead of i_disksize here because delalloc writeback
3398 * can complete at any point during the I/O and subsequently push the
3399 * i_disksize out to i_size. This could be beyond where direct I/O is
3400 * happening and thus expose allocated blocks to direct I/O reads.
3402 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3403 m_flags = EXT4_GET_BLOCKS_CREATE;
3404 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3405 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3407 ret = ext4_map_blocks(handle, inode, map, m_flags);
3410 * We cannot fill holes in indirect tree based inodes as that could
3411 * expose stale data in the case of a crash. Use the magic error code
3412 * to fallback to buffered I/O.
3414 if (!m_flags && !ret)
3417 ext4_journal_stop(handle);
3418 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3425 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3426 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3429 struct ext4_map_blocks map;
3430 u8 blkbits = inode->i_blkbits;
3432 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3435 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3439 * Calculate the first and last logical blocks respectively.
3441 map.m_lblk = offset >> blkbits;
3442 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3443 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3445 if (flags & IOMAP_WRITE) {
3447 * We check here if the blocks are already allocated, then we
3448 * don't need to start a journal txn and we can directly return
3449 * the mapping information. This could boost performance
3450 * especially in multi-threaded overwrite requests.
3452 if (offset + length <= i_size_read(inode)) {
3453 ret = ext4_map_blocks(NULL, inode, &map, 0);
3454 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3457 ret = ext4_iomap_alloc(inode, &map, flags);
3459 ret = ext4_map_blocks(NULL, inode, &map, 0);
3466 * When inline encryption is enabled, sometimes I/O to an encrypted file
3467 * has to be broken up to guarantee DUN contiguity. Handle this by
3468 * limiting the length of the mapping returned.
3470 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3472 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3477 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3478 loff_t length, unsigned flags, struct iomap *iomap,
3479 struct iomap *srcmap)
3484 * Even for writes we don't need to allocate blocks, so just pretend
3485 * we are reading to save overhead of starting a transaction.
3487 flags &= ~IOMAP_WRITE;
3488 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3489 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3493 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3494 ssize_t written, unsigned flags, struct iomap *iomap)
3497 * Check to see whether an error occurred while writing out the data to
3498 * the allocated blocks. If so, return the magic error code so that we
3499 * fallback to buffered I/O and attempt to complete the remainder of
3500 * the I/O. Any blocks that may have been allocated in preparation for
3501 * the direct I/O will be reused during buffered I/O.
3503 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3509 const struct iomap_ops ext4_iomap_ops = {
3510 .iomap_begin = ext4_iomap_begin,
3511 .iomap_end = ext4_iomap_end,
3514 const struct iomap_ops ext4_iomap_overwrite_ops = {
3515 .iomap_begin = ext4_iomap_overwrite_begin,
3516 .iomap_end = ext4_iomap_end,
3519 static bool ext4_iomap_is_delalloc(struct inode *inode,
3520 struct ext4_map_blocks *map)
3522 struct extent_status es;
3523 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3525 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3526 map->m_lblk, end, &es);
3528 if (!es.es_len || es.es_lblk > end)
3531 if (es.es_lblk > map->m_lblk) {
3532 map->m_len = es.es_lblk - map->m_lblk;
3536 offset = map->m_lblk - es.es_lblk;
3537 map->m_len = es.es_len - offset;
3542 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3543 loff_t length, unsigned int flags,
3544 struct iomap *iomap, struct iomap *srcmap)
3547 bool delalloc = false;
3548 struct ext4_map_blocks map;
3549 u8 blkbits = inode->i_blkbits;
3551 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3554 if (ext4_has_inline_data(inode)) {
3555 ret = ext4_inline_data_iomap(inode, iomap);
3556 if (ret != -EAGAIN) {
3557 if (ret == 0 && offset >= iomap->length)
3564 * Calculate the first and last logical block respectively.
3566 map.m_lblk = offset >> blkbits;
3567 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3568 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3571 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3572 * So handle it here itself instead of querying ext4_map_blocks().
3573 * Since ext4_map_blocks() will warn about it and will return
3576 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3577 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3579 if (offset >= sbi->s_bitmap_maxbytes) {
3585 ret = ext4_map_blocks(NULL, inode, &map, 0);
3589 delalloc = ext4_iomap_is_delalloc(inode, &map);
3592 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3593 if (delalloc && iomap->type == IOMAP_HOLE)
3594 iomap->type = IOMAP_DELALLOC;
3599 const struct iomap_ops ext4_iomap_report_ops = {
3600 .iomap_begin = ext4_iomap_begin_report,
3604 * Whenever the folio is being dirtied, corresponding buffers should already
3605 * be attached to the transaction (we take care of this in ext4_page_mkwrite()
3606 * and ext4_write_begin()). However we cannot move buffers to dirty transaction
3607 * lists here because ->dirty_folio is called under VFS locks and the folio
3608 * is not necessarily locked.
3610 * We cannot just dirty the folio and leave attached buffers clean, because the
3611 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3612 * or jbddirty because all the journalling code will explode.
3614 * So what we do is to mark the folio "pending dirty" and next time writepage
3615 * is called, propagate that into the buffers appropriately.
3617 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3618 struct folio *folio)
3620 WARN_ON_ONCE(!folio_buffers(folio));
3621 folio_set_checked(folio);
3622 return filemap_dirty_folio(mapping, folio);
3625 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3627 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3628 WARN_ON_ONCE(!folio_buffers(folio));
3629 return block_dirty_folio(mapping, folio);
3632 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3633 struct file *file, sector_t *span)
3635 return iomap_swapfile_activate(sis, file, span,
3636 &ext4_iomap_report_ops);
3639 static const struct address_space_operations ext4_aops = {
3640 .read_folio = ext4_read_folio,
3641 .readahead = ext4_readahead,
3642 .writepage = ext4_writepage,
3643 .writepages = ext4_writepages,
3644 .write_begin = ext4_write_begin,
3645 .write_end = ext4_write_end,
3646 .dirty_folio = ext4_dirty_folio,
3648 .invalidate_folio = ext4_invalidate_folio,
3649 .release_folio = ext4_release_folio,
3650 .direct_IO = noop_direct_IO,
3651 .migrate_folio = buffer_migrate_folio,
3652 .is_partially_uptodate = block_is_partially_uptodate,
3653 .error_remove_page = generic_error_remove_page,
3654 .swap_activate = ext4_iomap_swap_activate,
3657 static const struct address_space_operations ext4_journalled_aops = {
3658 .read_folio = ext4_read_folio,
3659 .readahead = ext4_readahead,
3660 .writepage = ext4_writepage,
3661 .writepages = ext4_writepages,
3662 .write_begin = ext4_write_begin,
3663 .write_end = ext4_journalled_write_end,
3664 .dirty_folio = ext4_journalled_dirty_folio,
3666 .invalidate_folio = ext4_journalled_invalidate_folio,
3667 .release_folio = ext4_release_folio,
3668 .direct_IO = noop_direct_IO,
3669 .is_partially_uptodate = block_is_partially_uptodate,
3670 .error_remove_page = generic_error_remove_page,
3671 .swap_activate = ext4_iomap_swap_activate,
3674 static const struct address_space_operations ext4_da_aops = {
3675 .read_folio = ext4_read_folio,
3676 .readahead = ext4_readahead,
3677 .writepage = ext4_writepage,
3678 .writepages = ext4_writepages,
3679 .write_begin = ext4_da_write_begin,
3680 .write_end = ext4_da_write_end,
3681 .dirty_folio = ext4_dirty_folio,
3683 .invalidate_folio = ext4_invalidate_folio,
3684 .release_folio = ext4_release_folio,
3685 .direct_IO = noop_direct_IO,
3686 .migrate_folio = buffer_migrate_folio,
3687 .is_partially_uptodate = block_is_partially_uptodate,
3688 .error_remove_page = generic_error_remove_page,
3689 .swap_activate = ext4_iomap_swap_activate,
3692 static const struct address_space_operations ext4_dax_aops = {
3693 .writepages = ext4_dax_writepages,
3694 .direct_IO = noop_direct_IO,
3695 .dirty_folio = noop_dirty_folio,
3697 .swap_activate = ext4_iomap_swap_activate,
3700 void ext4_set_aops(struct inode *inode)
3702 switch (ext4_inode_journal_mode(inode)) {
3703 case EXT4_INODE_ORDERED_DATA_MODE:
3704 case EXT4_INODE_WRITEBACK_DATA_MODE:
3706 case EXT4_INODE_JOURNAL_DATA_MODE:
3707 inode->i_mapping->a_ops = &ext4_journalled_aops;
3713 inode->i_mapping->a_ops = &ext4_dax_aops;
3714 else if (test_opt(inode->i_sb, DELALLOC))
3715 inode->i_mapping->a_ops = &ext4_da_aops;
3717 inode->i_mapping->a_ops = &ext4_aops;
3720 static int __ext4_block_zero_page_range(handle_t *handle,
3721 struct address_space *mapping, loff_t from, loff_t length)
3723 ext4_fsblk_t index = from >> PAGE_SHIFT;
3724 unsigned offset = from & (PAGE_SIZE-1);
3725 unsigned blocksize, pos;
3727 struct inode *inode = mapping->host;
3728 struct buffer_head *bh;
3732 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3733 mapping_gfp_constraint(mapping, ~__GFP_FS));
3737 blocksize = inode->i_sb->s_blocksize;
3739 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3741 if (!page_has_buffers(page))
3742 create_empty_buffers(page, blocksize, 0);
3744 /* Find the buffer that contains "offset" */
3745 bh = page_buffers(page);
3747 while (offset >= pos) {
3748 bh = bh->b_this_page;
3752 if (buffer_freed(bh)) {
3753 BUFFER_TRACE(bh, "freed: skip");
3756 if (!buffer_mapped(bh)) {
3757 BUFFER_TRACE(bh, "unmapped");
3758 ext4_get_block(inode, iblock, bh, 0);
3759 /* unmapped? It's a hole - nothing to do */
3760 if (!buffer_mapped(bh)) {
3761 BUFFER_TRACE(bh, "still unmapped");
3766 /* Ok, it's mapped. Make sure it's up-to-date */
3767 if (PageUptodate(page))
3768 set_buffer_uptodate(bh);
3770 if (!buffer_uptodate(bh)) {
3771 err = ext4_read_bh_lock(bh, 0, true);
3774 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3775 /* We expect the key to be set. */
3776 BUG_ON(!fscrypt_has_encryption_key(inode));
3777 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3780 clear_buffer_uptodate(bh);
3785 if (ext4_should_journal_data(inode)) {
3786 BUFFER_TRACE(bh, "get write access");
3787 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3792 zero_user(page, offset, length);
3793 BUFFER_TRACE(bh, "zeroed end of block");
3795 if (ext4_should_journal_data(inode)) {
3796 err = ext4_handle_dirty_metadata(handle, inode, bh);
3799 mark_buffer_dirty(bh);
3800 if (ext4_should_order_data(inode))
3801 err = ext4_jbd2_inode_add_write(handle, inode, from,
3812 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3813 * starting from file offset 'from'. The range to be zero'd must
3814 * be contained with in one block. If the specified range exceeds
3815 * the end of the block it will be shortened to end of the block
3816 * that corresponds to 'from'
3818 static int ext4_block_zero_page_range(handle_t *handle,
3819 struct address_space *mapping, loff_t from, loff_t length)
3821 struct inode *inode = mapping->host;
3822 unsigned offset = from & (PAGE_SIZE-1);
3823 unsigned blocksize = inode->i_sb->s_blocksize;
3824 unsigned max = blocksize - (offset & (blocksize - 1));
3827 * correct length if it does not fall between
3828 * 'from' and the end of the block
3830 if (length > max || length < 0)
3833 if (IS_DAX(inode)) {
3834 return dax_zero_range(inode, from, length, NULL,
3837 return __ext4_block_zero_page_range(handle, mapping, from, length);
3841 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3842 * up to the end of the block which corresponds to `from'.
3843 * This required during truncate. We need to physically zero the tail end
3844 * of that block so it doesn't yield old data if the file is later grown.
3846 static int ext4_block_truncate_page(handle_t *handle,
3847 struct address_space *mapping, loff_t from)
3849 unsigned offset = from & (PAGE_SIZE-1);
3852 struct inode *inode = mapping->host;
3854 /* If we are processing an encrypted inode during orphan list handling */
3855 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3858 blocksize = inode->i_sb->s_blocksize;
3859 length = blocksize - (offset & (blocksize - 1));
3861 return ext4_block_zero_page_range(handle, mapping, from, length);
3864 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3865 loff_t lstart, loff_t length)
3867 struct super_block *sb = inode->i_sb;
3868 struct address_space *mapping = inode->i_mapping;
3869 unsigned partial_start, partial_end;
3870 ext4_fsblk_t start, end;
3871 loff_t byte_end = (lstart + length - 1);
3874 partial_start = lstart & (sb->s_blocksize - 1);
3875 partial_end = byte_end & (sb->s_blocksize - 1);
3877 start = lstart >> sb->s_blocksize_bits;
3878 end = byte_end >> sb->s_blocksize_bits;
3880 /* Handle partial zero within the single block */
3882 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3883 err = ext4_block_zero_page_range(handle, mapping,
3887 /* Handle partial zero out on the start of the range */
3888 if (partial_start) {
3889 err = ext4_block_zero_page_range(handle, mapping,
3890 lstart, sb->s_blocksize);
3894 /* Handle partial zero out on the end of the range */
3895 if (partial_end != sb->s_blocksize - 1)
3896 err = ext4_block_zero_page_range(handle, mapping,
3897 byte_end - partial_end,
3902 int ext4_can_truncate(struct inode *inode)
3904 if (S_ISREG(inode->i_mode))
3906 if (S_ISDIR(inode->i_mode))
3908 if (S_ISLNK(inode->i_mode))
3909 return !ext4_inode_is_fast_symlink(inode);
3914 * We have to make sure i_disksize gets properly updated before we truncate
3915 * page cache due to hole punching or zero range. Otherwise i_disksize update
3916 * can get lost as it may have been postponed to submission of writeback but
3917 * that will never happen after we truncate page cache.
3919 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3925 loff_t size = i_size_read(inode);
3927 WARN_ON(!inode_is_locked(inode));
3928 if (offset > size || offset + len < size)
3931 if (EXT4_I(inode)->i_disksize >= size)
3934 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3936 return PTR_ERR(handle);
3937 ext4_update_i_disksize(inode, size);
3938 ret = ext4_mark_inode_dirty(handle, inode);
3939 ext4_journal_stop(handle);
3944 static void ext4_wait_dax_page(struct inode *inode)
3946 filemap_invalidate_unlock(inode->i_mapping);
3948 filemap_invalidate_lock(inode->i_mapping);
3951 int ext4_break_layouts(struct inode *inode)
3956 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3960 page = dax_layout_busy_page(inode->i_mapping);
3964 error = ___wait_var_event(&page->_refcount,
3965 atomic_read(&page->_refcount) == 1,
3966 TASK_INTERRUPTIBLE, 0, 0,
3967 ext4_wait_dax_page(inode));
3968 } while (error == 0);
3974 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3975 * associated with the given offset and length
3977 * @inode: File inode
3978 * @offset: The offset where the hole will begin
3979 * @len: The length of the hole
3981 * Returns: 0 on success or negative on failure
3984 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3986 struct inode *inode = file_inode(file);
3987 struct super_block *sb = inode->i_sb;
3988 ext4_lblk_t first_block, stop_block;
3989 struct address_space *mapping = inode->i_mapping;
3990 loff_t first_block_offset, last_block_offset, max_length;
3991 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3993 unsigned int credits;
3994 int ret = 0, ret2 = 0;
3996 trace_ext4_punch_hole(inode, offset, length, 0);
3999 * Write out all dirty pages to avoid race conditions
4000 * Then release them.
4002 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4003 ret = filemap_write_and_wait_range(mapping, offset,
4004 offset + length - 1);
4011 /* No need to punch hole beyond i_size */
4012 if (offset >= inode->i_size)
4016 * If the hole extends beyond i_size, set the hole
4017 * to end after the page that contains i_size
4019 if (offset + length > inode->i_size) {
4020 length = inode->i_size +
4021 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4026 * For punch hole the length + offset needs to be within one block
4027 * before last range. Adjust the length if it goes beyond that limit.
4029 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4030 if (offset + length > max_length)
4031 length = max_length - offset;
4033 if (offset & (sb->s_blocksize - 1) ||
4034 (offset + length) & (sb->s_blocksize - 1)) {
4036 * Attach jinode to inode for jbd2 if we do any zeroing of
4039 ret = ext4_inode_attach_jinode(inode);
4045 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4046 inode_dio_wait(inode);
4048 ret = file_modified(file);
4053 * Prevent page faults from reinstantiating pages we have released from
4056 filemap_invalidate_lock(mapping);
4058 ret = ext4_break_layouts(inode);
4062 first_block_offset = round_up(offset, sb->s_blocksize);
4063 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4065 /* Now release the pages and zero block aligned part of pages*/
4066 if (last_block_offset > first_block_offset) {
4067 ret = ext4_update_disksize_before_punch(inode, offset, length);
4070 truncate_pagecache_range(inode, first_block_offset,
4074 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4075 credits = ext4_writepage_trans_blocks(inode);
4077 credits = ext4_blocks_for_truncate(inode);
4078 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4079 if (IS_ERR(handle)) {
4080 ret = PTR_ERR(handle);
4081 ext4_std_error(sb, ret);
4085 ret = ext4_zero_partial_blocks(handle, inode, offset,
4090 first_block = (offset + sb->s_blocksize - 1) >>
4091 EXT4_BLOCK_SIZE_BITS(sb);
4092 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4094 /* If there are blocks to remove, do it */
4095 if (stop_block > first_block) {
4097 down_write(&EXT4_I(inode)->i_data_sem);
4098 ext4_discard_preallocations(inode, 0);
4100 ret = ext4_es_remove_extent(inode, first_block,
4101 stop_block - first_block);
4103 up_write(&EXT4_I(inode)->i_data_sem);
4107 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4108 ret = ext4_ext_remove_space(inode, first_block,
4111 ret = ext4_ind_remove_space(handle, inode, first_block,
4114 up_write(&EXT4_I(inode)->i_data_sem);
4116 ext4_fc_track_range(handle, inode, first_block, stop_block);
4118 ext4_handle_sync(handle);
4120 inode->i_mtime = inode->i_ctime = current_time(inode);
4121 ret2 = ext4_mark_inode_dirty(handle, inode);
4125 ext4_update_inode_fsync_trans(handle, inode, 1);
4127 ext4_journal_stop(handle);
4129 filemap_invalidate_unlock(mapping);
4131 inode_unlock(inode);
4135 int ext4_inode_attach_jinode(struct inode *inode)
4137 struct ext4_inode_info *ei = EXT4_I(inode);
4138 struct jbd2_inode *jinode;
4140 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4143 jinode = jbd2_alloc_inode(GFP_KERNEL);
4144 spin_lock(&inode->i_lock);
4147 spin_unlock(&inode->i_lock);
4150 ei->jinode = jinode;
4151 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4154 spin_unlock(&inode->i_lock);
4155 if (unlikely(jinode != NULL))
4156 jbd2_free_inode(jinode);
4163 * We block out ext4_get_block() block instantiations across the entire
4164 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4165 * simultaneously on behalf of the same inode.
4167 * As we work through the truncate and commit bits of it to the journal there
4168 * is one core, guiding principle: the file's tree must always be consistent on
4169 * disk. We must be able to restart the truncate after a crash.
4171 * The file's tree may be transiently inconsistent in memory (although it
4172 * probably isn't), but whenever we close off and commit a journal transaction,
4173 * the contents of (the filesystem + the journal) must be consistent and
4174 * restartable. It's pretty simple, really: bottom up, right to left (although
4175 * left-to-right works OK too).
4177 * Note that at recovery time, journal replay occurs *before* the restart of
4178 * truncate against the orphan inode list.
4180 * The committed inode has the new, desired i_size (which is the same as
4181 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4182 * that this inode's truncate did not complete and it will again call
4183 * ext4_truncate() to have another go. So there will be instantiated blocks
4184 * to the right of the truncation point in a crashed ext4 filesystem. But
4185 * that's fine - as long as they are linked from the inode, the post-crash
4186 * ext4_truncate() run will find them and release them.
4188 int ext4_truncate(struct inode *inode)
4190 struct ext4_inode_info *ei = EXT4_I(inode);
4191 unsigned int credits;
4194 struct address_space *mapping = inode->i_mapping;
4197 * There is a possibility that we're either freeing the inode
4198 * or it's a completely new inode. In those cases we might not
4199 * have i_rwsem locked because it's not necessary.
4201 if (!(inode->i_state & (I_NEW|I_FREEING)))
4202 WARN_ON(!inode_is_locked(inode));
4203 trace_ext4_truncate_enter(inode);
4205 if (!ext4_can_truncate(inode))
4208 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4209 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4211 if (ext4_has_inline_data(inode)) {
4214 err = ext4_inline_data_truncate(inode, &has_inline);
4215 if (err || has_inline)
4219 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4220 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4221 if (ext4_inode_attach_jinode(inode) < 0)
4225 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4226 credits = ext4_writepage_trans_blocks(inode);
4228 credits = ext4_blocks_for_truncate(inode);
4230 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4231 if (IS_ERR(handle)) {
4232 err = PTR_ERR(handle);
4236 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4237 ext4_block_truncate_page(handle, mapping, inode->i_size);
4240 * We add the inode to the orphan list, so that if this
4241 * truncate spans multiple transactions, and we crash, we will
4242 * resume the truncate when the filesystem recovers. It also
4243 * marks the inode dirty, to catch the new size.
4245 * Implication: the file must always be in a sane, consistent
4246 * truncatable state while each transaction commits.
4248 err = ext4_orphan_add(handle, inode);
4252 down_write(&EXT4_I(inode)->i_data_sem);
4254 ext4_discard_preallocations(inode, 0);
4256 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4257 err = ext4_ext_truncate(handle, inode);
4259 ext4_ind_truncate(handle, inode);
4261 up_write(&ei->i_data_sem);
4266 ext4_handle_sync(handle);
4270 * If this was a simple ftruncate() and the file will remain alive,
4271 * then we need to clear up the orphan record which we created above.
4272 * However, if this was a real unlink then we were called by
4273 * ext4_evict_inode(), and we allow that function to clean up the
4274 * orphan info for us.
4277 ext4_orphan_del(handle, inode);
4279 inode->i_mtime = inode->i_ctime = current_time(inode);
4280 err2 = ext4_mark_inode_dirty(handle, inode);
4281 if (unlikely(err2 && !err))
4283 ext4_journal_stop(handle);
4286 trace_ext4_truncate_exit(inode);
4290 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4292 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4293 return inode_peek_iversion_raw(inode);
4295 return inode_peek_iversion(inode);
4298 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4299 struct ext4_inode_info *ei)
4301 struct inode *inode = &(ei->vfs_inode);
4302 u64 i_blocks = READ_ONCE(inode->i_blocks);
4303 struct super_block *sb = inode->i_sb;
4305 if (i_blocks <= ~0U) {
4307 * i_blocks can be represented in a 32 bit variable
4308 * as multiple of 512 bytes
4310 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4311 raw_inode->i_blocks_high = 0;
4312 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4317 * This should never happen since sb->s_maxbytes should not have
4318 * allowed this, sb->s_maxbytes was set according to the huge_file
4319 * feature in ext4_fill_super().
4321 if (!ext4_has_feature_huge_file(sb))
4322 return -EFSCORRUPTED;
4324 if (i_blocks <= 0xffffffffffffULL) {
4326 * i_blocks can be represented in a 48 bit variable
4327 * as multiple of 512 bytes
4329 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4330 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4331 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4333 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4334 /* i_block is stored in file system block size */
4335 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4336 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4337 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4342 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4344 struct ext4_inode_info *ei = EXT4_I(inode);
4351 err = ext4_inode_blocks_set(raw_inode, ei);
4353 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4354 i_uid = i_uid_read(inode);
4355 i_gid = i_gid_read(inode);
4356 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4357 if (!(test_opt(inode->i_sb, NO_UID32))) {
4358 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4359 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4361 * Fix up interoperability with old kernels. Otherwise,
4362 * old inodes get re-used with the upper 16 bits of the
4365 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4366 raw_inode->i_uid_high = 0;
4367 raw_inode->i_gid_high = 0;
4369 raw_inode->i_uid_high =
4370 cpu_to_le16(high_16_bits(i_uid));
4371 raw_inode->i_gid_high =
4372 cpu_to_le16(high_16_bits(i_gid));
4375 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4376 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4377 raw_inode->i_uid_high = 0;
4378 raw_inode->i_gid_high = 0;
4380 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4382 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4383 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4384 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4385 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4387 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4388 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4389 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4390 raw_inode->i_file_acl_high =
4391 cpu_to_le16(ei->i_file_acl >> 32);
4392 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4393 ext4_isize_set(raw_inode, ei->i_disksize);
4395 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4396 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4397 if (old_valid_dev(inode->i_rdev)) {
4398 raw_inode->i_block[0] =
4399 cpu_to_le32(old_encode_dev(inode->i_rdev));
4400 raw_inode->i_block[1] = 0;
4402 raw_inode->i_block[0] = 0;
4403 raw_inode->i_block[1] =
4404 cpu_to_le32(new_encode_dev(inode->i_rdev));
4405 raw_inode->i_block[2] = 0;
4407 } else if (!ext4_has_inline_data(inode)) {
4408 for (block = 0; block < EXT4_N_BLOCKS; block++)
4409 raw_inode->i_block[block] = ei->i_data[block];
4412 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4413 u64 ivers = ext4_inode_peek_iversion(inode);
4415 raw_inode->i_disk_version = cpu_to_le32(ivers);
4416 if (ei->i_extra_isize) {
4417 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4418 raw_inode->i_version_hi =
4419 cpu_to_le32(ivers >> 32);
4420 raw_inode->i_extra_isize =
4421 cpu_to_le16(ei->i_extra_isize);
4425 if (i_projid != EXT4_DEF_PROJID &&
4426 !ext4_has_feature_project(inode->i_sb))
4427 err = err ?: -EFSCORRUPTED;
4429 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4430 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4431 raw_inode->i_projid = cpu_to_le32(i_projid);
4433 ext4_inode_csum_set(inode, raw_inode, ei);
4438 * ext4_get_inode_loc returns with an extra refcount against the inode's
4439 * underlying buffer_head on success. If we pass 'inode' and it does not
4440 * have in-inode xattr, we have all inode data in memory that is needed
4441 * to recreate the on-disk version of this inode.
4443 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4444 struct inode *inode, struct ext4_iloc *iloc,
4445 ext4_fsblk_t *ret_block)
4447 struct ext4_group_desc *gdp;
4448 struct buffer_head *bh;
4450 struct blk_plug plug;
4451 int inodes_per_block, inode_offset;
4454 if (ino < EXT4_ROOT_INO ||
4455 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4456 return -EFSCORRUPTED;
4458 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4459 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4464 * Figure out the offset within the block group inode table
4466 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4467 inode_offset = ((ino - 1) %
4468 EXT4_INODES_PER_GROUP(sb));
4469 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4470 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4472 bh = sb_getblk(sb, block);
4475 if (ext4_buffer_uptodate(bh))
4479 if (ext4_buffer_uptodate(bh)) {
4480 /* Someone brought it uptodate while we waited */
4486 * If we have all information of the inode in memory and this
4487 * is the only valid inode in the block, we need not read the
4490 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4491 struct buffer_head *bitmap_bh;
4494 start = inode_offset & ~(inodes_per_block - 1);
4496 /* Is the inode bitmap in cache? */
4497 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4498 if (unlikely(!bitmap_bh))
4502 * If the inode bitmap isn't in cache then the
4503 * optimisation may end up performing two reads instead
4504 * of one, so skip it.
4506 if (!buffer_uptodate(bitmap_bh)) {
4510 for (i = start; i < start + inodes_per_block; i++) {
4511 if (i == inode_offset)
4513 if (ext4_test_bit(i, bitmap_bh->b_data))
4517 if (i == start + inodes_per_block) {
4518 struct ext4_inode *raw_inode =
4519 (struct ext4_inode *) (bh->b_data + iloc->offset);
4521 /* all other inodes are free, so skip I/O */
4522 memset(bh->b_data, 0, bh->b_size);
4523 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4524 ext4_fill_raw_inode(inode, raw_inode);
4525 set_buffer_uptodate(bh);
4533 * If we need to do any I/O, try to pre-readahead extra
4534 * blocks from the inode table.
4536 blk_start_plug(&plug);
4537 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4538 ext4_fsblk_t b, end, table;
4540 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4542 table = ext4_inode_table(sb, gdp);
4543 /* s_inode_readahead_blks is always a power of 2 */
4544 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4548 num = EXT4_INODES_PER_GROUP(sb);
4549 if (ext4_has_group_desc_csum(sb))
4550 num -= ext4_itable_unused_count(sb, gdp);
4551 table += num / inodes_per_block;
4555 ext4_sb_breadahead_unmovable(sb, b++);
4559 * There are other valid inodes in the buffer, this inode
4560 * has in-inode xattrs, or we don't have this inode in memory.
4561 * Read the block from disk.
4563 trace_ext4_load_inode(sb, ino);
4564 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4565 blk_finish_plug(&plug);
4567 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4568 if (!buffer_uptodate(bh)) {
4579 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4580 struct ext4_iloc *iloc)
4582 ext4_fsblk_t err_blk = 0;
4585 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4589 ext4_error_inode_block(inode, err_blk, EIO,
4590 "unable to read itable block");
4595 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4597 ext4_fsblk_t err_blk = 0;
4600 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4604 ext4_error_inode_block(inode, err_blk, EIO,
4605 "unable to read itable block");
4611 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4612 struct ext4_iloc *iloc)
4614 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4617 static bool ext4_should_enable_dax(struct inode *inode)
4619 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4621 if (test_opt2(inode->i_sb, DAX_NEVER))
4623 if (!S_ISREG(inode->i_mode))
4625 if (ext4_should_journal_data(inode))
4627 if (ext4_has_inline_data(inode))
4629 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4631 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4633 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4635 if (test_opt(inode->i_sb, DAX_ALWAYS))
4638 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4641 void ext4_set_inode_flags(struct inode *inode, bool init)
4643 unsigned int flags = EXT4_I(inode)->i_flags;
4644 unsigned int new_fl = 0;
4646 WARN_ON_ONCE(IS_DAX(inode) && init);
4648 if (flags & EXT4_SYNC_FL)
4650 if (flags & EXT4_APPEND_FL)
4652 if (flags & EXT4_IMMUTABLE_FL)
4653 new_fl |= S_IMMUTABLE;
4654 if (flags & EXT4_NOATIME_FL)
4655 new_fl |= S_NOATIME;
4656 if (flags & EXT4_DIRSYNC_FL)
4657 new_fl |= S_DIRSYNC;
4659 /* Because of the way inode_set_flags() works we must preserve S_DAX
4660 * here if already set. */
4661 new_fl |= (inode->i_flags & S_DAX);
4662 if (init && ext4_should_enable_dax(inode))
4665 if (flags & EXT4_ENCRYPT_FL)
4666 new_fl |= S_ENCRYPTED;
4667 if (flags & EXT4_CASEFOLD_FL)
4668 new_fl |= S_CASEFOLD;
4669 if (flags & EXT4_VERITY_FL)
4671 inode_set_flags(inode, new_fl,
4672 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4673 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4676 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4677 struct ext4_inode_info *ei)
4680 struct inode *inode = &(ei->vfs_inode);
4681 struct super_block *sb = inode->i_sb;
4683 if (ext4_has_feature_huge_file(sb)) {
4684 /* we are using combined 48 bit field */
4685 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4686 le32_to_cpu(raw_inode->i_blocks_lo);
4687 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4688 /* i_blocks represent file system block size */
4689 return i_blocks << (inode->i_blkbits - 9);
4694 return le32_to_cpu(raw_inode->i_blocks_lo);
4698 static inline int ext4_iget_extra_inode(struct inode *inode,
4699 struct ext4_inode *raw_inode,
4700 struct ext4_inode_info *ei)
4702 __le32 *magic = (void *)raw_inode +
4703 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4705 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4706 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4707 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4708 return ext4_find_inline_data_nolock(inode);
4710 EXT4_I(inode)->i_inline_off = 0;
4714 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4716 if (!ext4_has_feature_project(inode->i_sb))
4718 *projid = EXT4_I(inode)->i_projid;
4723 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4724 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4727 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4729 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4730 inode_set_iversion_raw(inode, val);
4732 inode_set_iversion_queried(inode, val);
4735 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4736 ext4_iget_flags flags, const char *function,
4739 struct ext4_iloc iloc;
4740 struct ext4_inode *raw_inode;
4741 struct ext4_inode_info *ei;
4742 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4743 struct inode *inode;
4744 journal_t *journal = EXT4_SB(sb)->s_journal;
4752 if ((!(flags & EXT4_IGET_SPECIAL) &&
4753 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4754 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4755 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4756 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4757 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4758 (ino < EXT4_ROOT_INO) ||
4759 (ino > le32_to_cpu(es->s_inodes_count))) {
4760 if (flags & EXT4_IGET_HANDLE)
4761 return ERR_PTR(-ESTALE);
4762 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4763 "inode #%lu: comm %s: iget: illegal inode #",
4764 ino, current->comm);
4765 return ERR_PTR(-EFSCORRUPTED);
4768 inode = iget_locked(sb, ino);
4770 return ERR_PTR(-ENOMEM);
4771 if (!(inode->i_state & I_NEW))
4777 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4780 raw_inode = ext4_raw_inode(&iloc);
4782 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4783 ext4_error_inode(inode, function, line, 0,
4784 "iget: root inode unallocated");
4785 ret = -EFSCORRUPTED;
4789 if ((flags & EXT4_IGET_HANDLE) &&
4790 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4795 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4796 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4797 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4798 EXT4_INODE_SIZE(inode->i_sb) ||
4799 (ei->i_extra_isize & 3)) {
4800 ext4_error_inode(inode, function, line, 0,
4801 "iget: bad extra_isize %u "
4804 EXT4_INODE_SIZE(inode->i_sb));
4805 ret = -EFSCORRUPTED;
4809 ei->i_extra_isize = 0;
4811 /* Precompute checksum seed for inode metadata */
4812 if (ext4_has_metadata_csum(sb)) {
4813 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4815 __le32 inum = cpu_to_le32(inode->i_ino);
4816 __le32 gen = raw_inode->i_generation;
4817 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4819 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4823 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4824 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4825 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4826 ext4_error_inode_err(inode, function, line, 0,
4827 EFSBADCRC, "iget: checksum invalid");
4832 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4833 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4834 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4835 if (ext4_has_feature_project(sb) &&
4836 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4837 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4838 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4840 i_projid = EXT4_DEF_PROJID;
4842 if (!(test_opt(inode->i_sb, NO_UID32))) {
4843 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4844 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4846 i_uid_write(inode, i_uid);
4847 i_gid_write(inode, i_gid);
4848 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4849 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4851 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4852 ei->i_inline_off = 0;
4853 ei->i_dir_start_lookup = 0;
4854 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4855 /* We now have enough fields to check if the inode was active or not.
4856 * This is needed because nfsd might try to access dead inodes
4857 * the test is that same one that e2fsck uses
4858 * NeilBrown 1999oct15
4860 if (inode->i_nlink == 0) {
4861 if ((inode->i_mode == 0 ||
4862 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4863 ino != EXT4_BOOT_LOADER_INO) {
4864 /* this inode is deleted */
4868 /* The only unlinked inodes we let through here have
4869 * valid i_mode and are being read by the orphan
4870 * recovery code: that's fine, we're about to complete
4871 * the process of deleting those.
4872 * OR it is the EXT4_BOOT_LOADER_INO which is
4873 * not initialized on a new filesystem. */
4875 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4876 ext4_set_inode_flags(inode, true);
4877 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4878 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4879 if (ext4_has_feature_64bit(sb))
4881 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4882 inode->i_size = ext4_isize(sb, raw_inode);
4883 if ((size = i_size_read(inode)) < 0) {
4884 ext4_error_inode(inode, function, line, 0,
4885 "iget: bad i_size value: %lld", size);
4886 ret = -EFSCORRUPTED;
4890 * If dir_index is not enabled but there's dir with INDEX flag set,
4891 * we'd normally treat htree data as empty space. But with metadata
4892 * checksumming that corrupts checksums so forbid that.
4894 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4895 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4896 ext4_error_inode(inode, function, line, 0,
4897 "iget: Dir with htree data on filesystem without dir_index feature.");
4898 ret = -EFSCORRUPTED;
4901 ei->i_disksize = inode->i_size;
4903 ei->i_reserved_quota = 0;
4905 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4906 ei->i_block_group = iloc.block_group;
4907 ei->i_last_alloc_group = ~0;
4909 * NOTE! The in-memory inode i_data array is in little-endian order
4910 * even on big-endian machines: we do NOT byteswap the block numbers!
4912 for (block = 0; block < EXT4_N_BLOCKS; block++)
4913 ei->i_data[block] = raw_inode->i_block[block];
4914 INIT_LIST_HEAD(&ei->i_orphan);
4915 ext4_fc_init_inode(&ei->vfs_inode);
4918 * Set transaction id's of transactions that have to be committed
4919 * to finish f[data]sync. We set them to currently running transaction
4920 * as we cannot be sure that the inode or some of its metadata isn't
4921 * part of the transaction - the inode could have been reclaimed and
4922 * now it is reread from disk.
4925 transaction_t *transaction;
4928 read_lock(&journal->j_state_lock);
4929 if (journal->j_running_transaction)
4930 transaction = journal->j_running_transaction;
4932 transaction = journal->j_committing_transaction;
4934 tid = transaction->t_tid;
4936 tid = journal->j_commit_sequence;
4937 read_unlock(&journal->j_state_lock);
4938 ei->i_sync_tid = tid;
4939 ei->i_datasync_tid = tid;
4942 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4943 if (ei->i_extra_isize == 0) {
4944 /* The extra space is currently unused. Use it. */
4945 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4946 ei->i_extra_isize = sizeof(struct ext4_inode) -
4947 EXT4_GOOD_OLD_INODE_SIZE;
4949 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4955 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4956 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4957 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4958 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4960 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4961 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4963 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4964 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4966 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4968 ext4_inode_set_iversion_queried(inode, ivers);
4972 if (ei->i_file_acl &&
4973 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4974 ext4_error_inode(inode, function, line, 0,
4975 "iget: bad extended attribute block %llu",
4977 ret = -EFSCORRUPTED;
4979 } else if (!ext4_has_inline_data(inode)) {
4980 /* validate the block references in the inode */
4981 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4982 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4983 (S_ISLNK(inode->i_mode) &&
4984 !ext4_inode_is_fast_symlink(inode)))) {
4985 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4986 ret = ext4_ext_check_inode(inode);
4988 ret = ext4_ind_check_inode(inode);
4994 if (S_ISREG(inode->i_mode)) {
4995 inode->i_op = &ext4_file_inode_operations;
4996 inode->i_fop = &ext4_file_operations;
4997 ext4_set_aops(inode);
4998 } else if (S_ISDIR(inode->i_mode)) {
4999 inode->i_op = &ext4_dir_inode_operations;
5000 inode->i_fop = &ext4_dir_operations;
5001 } else if (S_ISLNK(inode->i_mode)) {
5002 /* VFS does not allow setting these so must be corruption */
5003 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5004 ext4_error_inode(inode, function, line, 0,
5005 "iget: immutable or append flags "
5006 "not allowed on symlinks");
5007 ret = -EFSCORRUPTED;
5010 if (IS_ENCRYPTED(inode)) {
5011 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5012 } else if (ext4_inode_is_fast_symlink(inode)) {
5013 inode->i_link = (char *)ei->i_data;
5014 inode->i_op = &ext4_fast_symlink_inode_operations;
5015 nd_terminate_link(ei->i_data, inode->i_size,
5016 sizeof(ei->i_data) - 1);
5018 inode->i_op = &ext4_symlink_inode_operations;
5020 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5021 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5022 inode->i_op = &ext4_special_inode_operations;
5023 if (raw_inode->i_block[0])
5024 init_special_inode(inode, inode->i_mode,
5025 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5027 init_special_inode(inode, inode->i_mode,
5028 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5029 } else if (ino == EXT4_BOOT_LOADER_INO) {
5030 make_bad_inode(inode);
5032 ret = -EFSCORRUPTED;
5033 ext4_error_inode(inode, function, line, 0,
5034 "iget: bogus i_mode (%o)", inode->i_mode);
5037 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5038 ext4_error_inode(inode, function, line, 0,
5039 "casefold flag without casefold feature");
5042 unlock_new_inode(inode);
5048 return ERR_PTR(ret);
5051 static void __ext4_update_other_inode_time(struct super_block *sb,
5052 unsigned long orig_ino,
5054 struct ext4_inode *raw_inode)
5056 struct inode *inode;
5058 inode = find_inode_by_ino_rcu(sb, ino);
5062 if (!inode_is_dirtytime_only(inode))
5065 spin_lock(&inode->i_lock);
5066 if (inode_is_dirtytime_only(inode)) {
5067 struct ext4_inode_info *ei = EXT4_I(inode);
5069 inode->i_state &= ~I_DIRTY_TIME;
5070 spin_unlock(&inode->i_lock);
5072 spin_lock(&ei->i_raw_lock);
5073 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5074 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5075 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5076 ext4_inode_csum_set(inode, raw_inode, ei);
5077 spin_unlock(&ei->i_raw_lock);
5078 trace_ext4_other_inode_update_time(inode, orig_ino);
5081 spin_unlock(&inode->i_lock);
5085 * Opportunistically update the other time fields for other inodes in
5086 * the same inode table block.
5088 static void ext4_update_other_inodes_time(struct super_block *sb,
5089 unsigned long orig_ino, char *buf)
5092 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5093 int inode_size = EXT4_INODE_SIZE(sb);
5096 * Calculate the first inode in the inode table block. Inode
5097 * numbers are one-based. That is, the first inode in a block
5098 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5100 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5102 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5103 if (ino == orig_ino)
5105 __ext4_update_other_inode_time(sb, orig_ino, ino,
5106 (struct ext4_inode *)buf);
5112 * Post the struct inode info into an on-disk inode location in the
5113 * buffer-cache. This gobbles the caller's reference to the
5114 * buffer_head in the inode location struct.
5116 * The caller must have write access to iloc->bh.
5118 static int ext4_do_update_inode(handle_t *handle,
5119 struct inode *inode,
5120 struct ext4_iloc *iloc)
5122 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5123 struct ext4_inode_info *ei = EXT4_I(inode);
5124 struct buffer_head *bh = iloc->bh;
5125 struct super_block *sb = inode->i_sb;
5127 int need_datasync = 0, set_large_file = 0;
5129 spin_lock(&ei->i_raw_lock);
5132 * For fields not tracked in the in-memory inode, initialise them
5133 * to zero for new inodes.
5135 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5136 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5138 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5140 if (ei->i_disksize > 0x7fffffffULL) {
5141 if (!ext4_has_feature_large_file(sb) ||
5142 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5146 err = ext4_fill_raw_inode(inode, raw_inode);
5147 spin_unlock(&ei->i_raw_lock);
5149 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5153 if (inode->i_sb->s_flags & SB_LAZYTIME)
5154 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5157 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5158 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5161 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5162 if (set_large_file) {
5163 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5164 err = ext4_journal_get_write_access(handle, sb,
5169 lock_buffer(EXT4_SB(sb)->s_sbh);
5170 ext4_set_feature_large_file(sb);
5171 ext4_superblock_csum_set(sb);
5172 unlock_buffer(EXT4_SB(sb)->s_sbh);
5173 ext4_handle_sync(handle);
5174 err = ext4_handle_dirty_metadata(handle, NULL,
5175 EXT4_SB(sb)->s_sbh);
5177 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5179 ext4_std_error(inode->i_sb, err);
5186 * ext4_write_inode()
5188 * We are called from a few places:
5190 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5191 * Here, there will be no transaction running. We wait for any running
5192 * transaction to commit.
5194 * - Within flush work (sys_sync(), kupdate and such).
5195 * We wait on commit, if told to.
5197 * - Within iput_final() -> write_inode_now()
5198 * We wait on commit, if told to.
5200 * In all cases it is actually safe for us to return without doing anything,
5201 * because the inode has been copied into a raw inode buffer in
5202 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5205 * Note that we are absolutely dependent upon all inode dirtiers doing the
5206 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5207 * which we are interested.
5209 * It would be a bug for them to not do this. The code:
5211 * mark_inode_dirty(inode)
5213 * inode->i_size = expr;
5215 * is in error because write_inode() could occur while `stuff()' is running,
5216 * and the new i_size will be lost. Plus the inode will no longer be on the
5217 * superblock's dirty inode list.
5219 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5223 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5224 sb_rdonly(inode->i_sb))
5227 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5230 if (EXT4_SB(inode->i_sb)->s_journal) {
5231 if (ext4_journal_current_handle()) {
5232 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5238 * No need to force transaction in WB_SYNC_NONE mode. Also
5239 * ext4_sync_fs() will force the commit after everything is
5242 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5245 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5246 EXT4_I(inode)->i_sync_tid);
5248 struct ext4_iloc iloc;
5250 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5254 * sync(2) will flush the whole buffer cache. No need to do
5255 * it here separately for each inode.
5257 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5258 sync_dirty_buffer(iloc.bh);
5259 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5260 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5261 "IO error syncing inode");
5270 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5271 * buffers that are attached to a folio straddling i_size and are undergoing
5272 * commit. In that case we have to wait for commit to finish and try again.
5274 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5277 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5278 tid_t commit_tid = 0;
5281 offset = inode->i_size & (PAGE_SIZE - 1);
5283 * If the folio is fully truncated, we don't need to wait for any commit
5284 * (and we even should not as __ext4_journalled_invalidate_folio() may
5285 * strip all buffers from the folio but keep the folio dirty which can then
5286 * confuse e.g. concurrent ext4_writepage() seeing dirty folio without
5287 * buffers). Also we don't need to wait for any commit if all buffers in
5288 * the folio remain valid. This is most beneficial for the common case of
5289 * blocksize == PAGESIZE.
5291 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5294 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5295 inode->i_size >> PAGE_SHIFT);
5298 ret = __ext4_journalled_invalidate_folio(folio, offset,
5299 folio_size(folio) - offset);
5300 folio_unlock(folio);
5305 read_lock(&journal->j_state_lock);
5306 if (journal->j_committing_transaction)
5307 commit_tid = journal->j_committing_transaction->t_tid;
5308 read_unlock(&journal->j_state_lock);
5310 jbd2_log_wait_commit(journal, commit_tid);
5317 * Called from notify_change.
5319 * We want to trap VFS attempts to truncate the file as soon as
5320 * possible. In particular, we want to make sure that when the VFS
5321 * shrinks i_size, we put the inode on the orphan list and modify
5322 * i_disksize immediately, so that during the subsequent flushing of
5323 * dirty pages and freeing of disk blocks, we can guarantee that any
5324 * commit will leave the blocks being flushed in an unused state on
5325 * disk. (On recovery, the inode will get truncated and the blocks will
5326 * be freed, so we have a strong guarantee that no future commit will
5327 * leave these blocks visible to the user.)
5329 * Another thing we have to assure is that if we are in ordered mode
5330 * and inode is still attached to the committing transaction, we must
5331 * we start writeout of all the dirty pages which are being truncated.
5332 * This way we are sure that all the data written in the previous
5333 * transaction are already on disk (truncate waits for pages under
5336 * Called with inode->i_rwsem down.
5338 int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5341 struct inode *inode = d_inode(dentry);
5344 const unsigned int ia_valid = attr->ia_valid;
5346 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5349 if (unlikely(IS_IMMUTABLE(inode)))
5352 if (unlikely(IS_APPEND(inode) &&
5353 (ia_valid & (ATTR_MODE | ATTR_UID |
5354 ATTR_GID | ATTR_TIMES_SET))))
5357 error = setattr_prepare(mnt_userns, dentry, attr);
5361 error = fscrypt_prepare_setattr(dentry, attr);
5365 error = fsverity_prepare_setattr(dentry, attr);
5369 if (is_quota_modification(mnt_userns, inode, attr)) {
5370 error = dquot_initialize(inode);
5375 if (i_uid_needs_update(mnt_userns, attr, inode) ||
5376 i_gid_needs_update(mnt_userns, attr, inode)) {
5379 /* (user+group)*(old+new) structure, inode write (sb,
5380 * inode block, ? - but truncate inode update has it) */
5381 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5382 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5383 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5384 if (IS_ERR(handle)) {
5385 error = PTR_ERR(handle);
5389 /* dquot_transfer() calls back ext4_get_inode_usage() which
5390 * counts xattr inode references.
5392 down_read(&EXT4_I(inode)->xattr_sem);
5393 error = dquot_transfer(mnt_userns, inode, attr);
5394 up_read(&EXT4_I(inode)->xattr_sem);
5397 ext4_journal_stop(handle);
5400 /* Update corresponding info in inode so that everything is in
5401 * one transaction */
5402 i_uid_update(mnt_userns, attr, inode);
5403 i_gid_update(mnt_userns, attr, inode);
5404 error = ext4_mark_inode_dirty(handle, inode);
5405 ext4_journal_stop(handle);
5406 if (unlikely(error)) {
5411 if (attr->ia_valid & ATTR_SIZE) {
5413 loff_t oldsize = inode->i_size;
5414 loff_t old_disksize;
5415 int shrink = (attr->ia_size < inode->i_size);
5417 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5418 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5420 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5424 if (!S_ISREG(inode->i_mode)) {
5428 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5429 inode_inc_iversion(inode);
5432 if (ext4_should_order_data(inode)) {
5433 error = ext4_begin_ordered_truncate(inode,
5439 * Blocks are going to be removed from the inode. Wait
5440 * for dio in flight.
5442 inode_dio_wait(inode);
5445 filemap_invalidate_lock(inode->i_mapping);
5447 rc = ext4_break_layouts(inode);
5449 filemap_invalidate_unlock(inode->i_mapping);
5453 if (attr->ia_size != inode->i_size) {
5454 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5455 if (IS_ERR(handle)) {
5456 error = PTR_ERR(handle);
5459 if (ext4_handle_valid(handle) && shrink) {
5460 error = ext4_orphan_add(handle, inode);
5464 * Update c/mtime on truncate up, ext4_truncate() will
5465 * update c/mtime in shrink case below
5468 inode->i_mtime = current_time(inode);
5469 inode->i_ctime = inode->i_mtime;
5473 ext4_fc_track_range(handle, inode,
5474 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5475 inode->i_sb->s_blocksize_bits,
5476 EXT_MAX_BLOCKS - 1);
5478 ext4_fc_track_range(
5480 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5481 inode->i_sb->s_blocksize_bits,
5482 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5483 inode->i_sb->s_blocksize_bits);
5485 down_write(&EXT4_I(inode)->i_data_sem);
5486 old_disksize = EXT4_I(inode)->i_disksize;
5487 EXT4_I(inode)->i_disksize = attr->ia_size;
5488 rc = ext4_mark_inode_dirty(handle, inode);
5492 * We have to update i_size under i_data_sem together
5493 * with i_disksize to avoid races with writeback code
5494 * running ext4_wb_update_i_disksize().
5497 i_size_write(inode, attr->ia_size);
5499 EXT4_I(inode)->i_disksize = old_disksize;
5500 up_write(&EXT4_I(inode)->i_data_sem);
5501 ext4_journal_stop(handle);
5505 pagecache_isize_extended(inode, oldsize,
5507 } else if (ext4_should_journal_data(inode)) {
5508 ext4_wait_for_tail_page_commit(inode);
5513 * Truncate pagecache after we've waited for commit
5514 * in data=journal mode to make pages freeable.
5516 truncate_pagecache(inode, inode->i_size);
5518 * Call ext4_truncate() even if i_size didn't change to
5519 * truncate possible preallocated blocks.
5521 if (attr->ia_size <= oldsize) {
5522 rc = ext4_truncate(inode);
5527 filemap_invalidate_unlock(inode->i_mapping);
5531 setattr_copy(mnt_userns, inode, attr);
5532 mark_inode_dirty(inode);
5536 * If the call to ext4_truncate failed to get a transaction handle at
5537 * all, we need to clean up the in-core orphan list manually.
5539 if (orphan && inode->i_nlink)
5540 ext4_orphan_del(NULL, inode);
5542 if (!error && (ia_valid & ATTR_MODE))
5543 rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5547 ext4_std_error(inode->i_sb, error);
5553 int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5554 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5556 struct inode *inode = d_inode(path->dentry);
5557 struct ext4_inode *raw_inode;
5558 struct ext4_inode_info *ei = EXT4_I(inode);
5561 if ((request_mask & STATX_BTIME) &&
5562 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5563 stat->result_mask |= STATX_BTIME;
5564 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5565 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5568 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5569 if (flags & EXT4_APPEND_FL)
5570 stat->attributes |= STATX_ATTR_APPEND;
5571 if (flags & EXT4_COMPR_FL)
5572 stat->attributes |= STATX_ATTR_COMPRESSED;
5573 if (flags & EXT4_ENCRYPT_FL)
5574 stat->attributes |= STATX_ATTR_ENCRYPTED;
5575 if (flags & EXT4_IMMUTABLE_FL)
5576 stat->attributes |= STATX_ATTR_IMMUTABLE;
5577 if (flags & EXT4_NODUMP_FL)
5578 stat->attributes |= STATX_ATTR_NODUMP;
5579 if (flags & EXT4_VERITY_FL)
5580 stat->attributes |= STATX_ATTR_VERITY;
5582 stat->attributes_mask |= (STATX_ATTR_APPEND |
5583 STATX_ATTR_COMPRESSED |
5584 STATX_ATTR_ENCRYPTED |
5585 STATX_ATTR_IMMUTABLE |
5589 generic_fillattr(mnt_userns, inode, stat);
5593 int ext4_file_getattr(struct user_namespace *mnt_userns,
5594 const struct path *path, struct kstat *stat,
5595 u32 request_mask, unsigned int query_flags)
5597 struct inode *inode = d_inode(path->dentry);
5598 u64 delalloc_blocks;
5600 ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5603 * If there is inline data in the inode, the inode will normally not
5604 * have data blocks allocated (it may have an external xattr block).
5605 * Report at least one sector for such files, so tools like tar, rsync,
5606 * others don't incorrectly think the file is completely sparse.
5608 if (unlikely(ext4_has_inline_data(inode)))
5609 stat->blocks += (stat->size + 511) >> 9;
5612 * We can't update i_blocks if the block allocation is delayed
5613 * otherwise in the case of system crash before the real block
5614 * allocation is done, we will have i_blocks inconsistent with
5615 * on-disk file blocks.
5616 * We always keep i_blocks updated together with real
5617 * allocation. But to not confuse with user, stat
5618 * will return the blocks that include the delayed allocation
5619 * blocks for this file.
5621 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5622 EXT4_I(inode)->i_reserved_data_blocks);
5623 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5627 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5630 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5631 return ext4_ind_trans_blocks(inode, lblocks);
5632 return ext4_ext_index_trans_blocks(inode, pextents);
5636 * Account for index blocks, block groups bitmaps and block group
5637 * descriptor blocks if modify datablocks and index blocks
5638 * worse case, the indexs blocks spread over different block groups
5640 * If datablocks are discontiguous, they are possible to spread over
5641 * different block groups too. If they are contiguous, with flexbg,
5642 * they could still across block group boundary.
5644 * Also account for superblock, inode, quota and xattr blocks
5646 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5649 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5655 * How many index blocks need to touch to map @lblocks logical blocks
5656 * to @pextents physical extents?
5658 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5663 * Now let's see how many group bitmaps and group descriptors need
5666 groups = idxblocks + pextents;
5668 if (groups > ngroups)
5670 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5671 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5673 /* bitmaps and block group descriptor blocks */
5674 ret += groups + gdpblocks;
5676 /* Blocks for super block, inode, quota and xattr blocks */
5677 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5683 * Calculate the total number of credits to reserve to fit
5684 * the modification of a single pages into a single transaction,
5685 * which may include multiple chunks of block allocations.
5687 * This could be called via ext4_write_begin()
5689 * We need to consider the worse case, when
5690 * one new block per extent.
5692 int ext4_writepage_trans_blocks(struct inode *inode)
5694 int bpp = ext4_journal_blocks_per_page(inode);
5697 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5699 /* Account for data blocks for journalled mode */
5700 if (ext4_should_journal_data(inode))
5706 * Calculate the journal credits for a chunk of data modification.
5708 * This is called from DIO, fallocate or whoever calling
5709 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5711 * journal buffers for data blocks are not included here, as DIO
5712 * and fallocate do no need to journal data buffers.
5714 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5716 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5720 * The caller must have previously called ext4_reserve_inode_write().
5721 * Give this, we know that the caller already has write access to iloc->bh.
5723 int ext4_mark_iloc_dirty(handle_t *handle,
5724 struct inode *inode, struct ext4_iloc *iloc)
5728 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5732 ext4_fc_track_inode(handle, inode);
5734 if (IS_I_VERSION(inode))
5735 inode_inc_iversion(inode);
5737 /* the do_update_inode consumes one bh->b_count */
5740 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5741 err = ext4_do_update_inode(handle, inode, iloc);
5747 * On success, We end up with an outstanding reference count against
5748 * iloc->bh. This _must_ be cleaned up later.
5752 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5753 struct ext4_iloc *iloc)
5757 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5760 err = ext4_get_inode_loc(inode, iloc);
5762 BUFFER_TRACE(iloc->bh, "get_write_access");
5763 err = ext4_journal_get_write_access(handle, inode->i_sb,
5764 iloc->bh, EXT4_JTR_NONE);
5770 ext4_std_error(inode->i_sb, err);
5774 static int __ext4_expand_extra_isize(struct inode *inode,
5775 unsigned int new_extra_isize,
5776 struct ext4_iloc *iloc,
5777 handle_t *handle, int *no_expand)
5779 struct ext4_inode *raw_inode;
5780 struct ext4_xattr_ibody_header *header;
5781 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5782 struct ext4_inode_info *ei = EXT4_I(inode);
5785 /* this was checked at iget time, but double check for good measure */
5786 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5787 (ei->i_extra_isize & 3)) {
5788 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5790 EXT4_INODE_SIZE(inode->i_sb));
5791 return -EFSCORRUPTED;
5793 if ((new_extra_isize < ei->i_extra_isize) ||
5794 (new_extra_isize < 4) ||
5795 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5796 return -EINVAL; /* Should never happen */
5798 raw_inode = ext4_raw_inode(iloc);
5800 header = IHDR(inode, raw_inode);
5802 /* No extended attributes present */
5803 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5804 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5805 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5806 EXT4_I(inode)->i_extra_isize, 0,
5807 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5808 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5812 /* try to expand with EAs present */
5813 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5817 * Inode size expansion failed; don't try again
5826 * Expand an inode by new_extra_isize bytes.
5827 * Returns 0 on success or negative error number on failure.
5829 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5830 unsigned int new_extra_isize,
5831 struct ext4_iloc iloc,
5837 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5841 * In nojournal mode, we can immediately attempt to expand
5842 * the inode. When journaled, we first need to obtain extra
5843 * buffer credits since we may write into the EA block
5844 * with this same handle. If journal_extend fails, then it will
5845 * only result in a minor loss of functionality for that inode.
5846 * If this is felt to be critical, then e2fsck should be run to
5847 * force a large enough s_min_extra_isize.
5849 if (ext4_journal_extend(handle,
5850 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5853 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5856 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5857 handle, &no_expand);
5858 ext4_write_unlock_xattr(inode, &no_expand);
5863 int ext4_expand_extra_isize(struct inode *inode,
5864 unsigned int new_extra_isize,
5865 struct ext4_iloc *iloc)
5871 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5876 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5877 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5878 if (IS_ERR(handle)) {
5879 error = PTR_ERR(handle);
5884 ext4_write_lock_xattr(inode, &no_expand);
5886 BUFFER_TRACE(iloc->bh, "get_write_access");
5887 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5894 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5895 handle, &no_expand);
5897 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5902 ext4_write_unlock_xattr(inode, &no_expand);
5903 ext4_journal_stop(handle);
5908 * What we do here is to mark the in-core inode as clean with respect to inode
5909 * dirtiness (it may still be data-dirty).
5910 * This means that the in-core inode may be reaped by prune_icache
5911 * without having to perform any I/O. This is a very good thing,
5912 * because *any* task may call prune_icache - even ones which
5913 * have a transaction open against a different journal.
5915 * Is this cheating? Not really. Sure, we haven't written the
5916 * inode out, but prune_icache isn't a user-visible syncing function.
5917 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5918 * we start and wait on commits.
5920 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5921 const char *func, unsigned int line)
5923 struct ext4_iloc iloc;
5924 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5928 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5929 err = ext4_reserve_inode_write(handle, inode, &iloc);
5933 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5934 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5937 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5940 ext4_error_inode_err(inode, func, line, 0, err,
5941 "mark_inode_dirty error");
5946 * ext4_dirty_inode() is called from __mark_inode_dirty()
5948 * We're really interested in the case where a file is being extended.
5949 * i_size has been changed by generic_commit_write() and we thus need
5950 * to include the updated inode in the current transaction.
5952 * Also, dquot_alloc_block() will always dirty the inode when blocks
5953 * are allocated to the file.
5955 * If the inode is marked synchronous, we don't honour that here - doing
5956 * so would cause a commit on atime updates, which we don't bother doing.
5957 * We handle synchronous inodes at the highest possible level.
5959 void ext4_dirty_inode(struct inode *inode, int flags)
5963 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5966 ext4_mark_inode_dirty(handle, inode);
5967 ext4_journal_stop(handle);
5970 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5975 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5978 * We have to be very careful here: changing a data block's
5979 * journaling status dynamically is dangerous. If we write a
5980 * data block to the journal, change the status and then delete
5981 * that block, we risk forgetting to revoke the old log record
5982 * from the journal and so a subsequent replay can corrupt data.
5983 * So, first we make sure that the journal is empty and that
5984 * nobody is changing anything.
5987 journal = EXT4_JOURNAL(inode);
5990 if (is_journal_aborted(journal))
5993 /* Wait for all existing dio workers */
5994 inode_dio_wait(inode);
5997 * Before flushing the journal and switching inode's aops, we have
5998 * to flush all dirty data the inode has. There can be outstanding
5999 * delayed allocations, there can be unwritten extents created by
6000 * fallocate or buffered writes in dioread_nolock mode covered by
6001 * dirty data which can be converted only after flushing the dirty
6002 * data (and journalled aops don't know how to handle these cases).
6005 filemap_invalidate_lock(inode->i_mapping);
6006 err = filemap_write_and_wait(inode->i_mapping);
6008 filemap_invalidate_unlock(inode->i_mapping);
6013 percpu_down_write(&sbi->s_writepages_rwsem);
6014 jbd2_journal_lock_updates(journal);
6017 * OK, there are no updates running now, and all cached data is
6018 * synced to disk. We are now in a completely consistent state
6019 * which doesn't have anything in the journal, and we know that
6020 * no filesystem updates are running, so it is safe to modify
6021 * the inode's in-core data-journaling state flag now.
6025 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6027 err = jbd2_journal_flush(journal, 0);
6029 jbd2_journal_unlock_updates(journal);
6030 percpu_up_write(&sbi->s_writepages_rwsem);
6033 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6035 ext4_set_aops(inode);
6037 jbd2_journal_unlock_updates(journal);
6038 percpu_up_write(&sbi->s_writepages_rwsem);
6041 filemap_invalidate_unlock(inode->i_mapping);
6043 /* Finally we can mark the inode as dirty. */
6045 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6047 return PTR_ERR(handle);
6049 ext4_fc_mark_ineligible(inode->i_sb,
6050 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6051 err = ext4_mark_inode_dirty(handle, inode);
6052 ext4_handle_sync(handle);
6053 ext4_journal_stop(handle);
6054 ext4_std_error(inode->i_sb, err);
6059 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6060 struct buffer_head *bh)
6062 return !buffer_mapped(bh);
6065 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6067 struct vm_area_struct *vma = vmf->vma;
6068 struct page *page = vmf->page;
6073 struct file *file = vma->vm_file;
6074 struct inode *inode = file_inode(file);
6075 struct address_space *mapping = inode->i_mapping;
6077 get_block_t *get_block;
6080 if (unlikely(IS_IMMUTABLE(inode)))
6081 return VM_FAULT_SIGBUS;
6083 sb_start_pagefault(inode->i_sb);
6084 file_update_time(vma->vm_file);
6086 filemap_invalidate_lock_shared(mapping);
6088 err = ext4_convert_inline_data(inode);
6093 * On data journalling we skip straight to the transaction handle:
6094 * there's no delalloc; page truncated will be checked later; the
6095 * early return w/ all buffers mapped (calculates size/len) can't
6096 * be used; and there's no dioread_nolock, so only ext4_get_block.
6098 if (ext4_should_journal_data(inode))
6101 /* Delalloc case is easy... */
6102 if (test_opt(inode->i_sb, DELALLOC) &&
6103 !ext4_nonda_switch(inode->i_sb)) {
6105 err = block_page_mkwrite(vma, vmf,
6106 ext4_da_get_block_prep);
6107 } while (err == -ENOSPC &&
6108 ext4_should_retry_alloc(inode->i_sb, &retries));
6113 size = i_size_read(inode);
6114 /* Page got truncated from under us? */
6115 if (page->mapping != mapping || page_offset(page) > size) {
6117 ret = VM_FAULT_NOPAGE;
6121 if (page->index == size >> PAGE_SHIFT)
6122 len = size & ~PAGE_MASK;
6126 * Return if we have all the buffers mapped. This avoids the need to do
6127 * journal_start/journal_stop which can block and take a long time
6129 * This cannot be done for data journalling, as we have to add the
6130 * inode to the transaction's list to writeprotect pages on commit.
6132 if (page_has_buffers(page)) {
6133 if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6135 ext4_bh_unmapped)) {
6136 /* Wait so that we don't change page under IO */
6137 wait_for_stable_page(page);
6138 ret = VM_FAULT_LOCKED;
6143 /* OK, we need to fill the hole... */
6144 if (ext4_should_dioread_nolock(inode))
6145 get_block = ext4_get_block_unwritten;
6147 get_block = ext4_get_block;
6149 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6150 ext4_writepage_trans_blocks(inode));
6151 if (IS_ERR(handle)) {
6152 ret = VM_FAULT_SIGBUS;
6156 * Data journalling can't use block_page_mkwrite() because it
6157 * will set_buffer_dirty() before do_journal_get_write_access()
6158 * thus might hit warning messages for dirty metadata buffers.
6160 if (!ext4_should_journal_data(inode)) {
6161 err = block_page_mkwrite(vma, vmf, get_block);
6164 size = i_size_read(inode);
6165 /* Page got truncated from under us? */
6166 if (page->mapping != mapping || page_offset(page) > size) {
6167 ret = VM_FAULT_NOPAGE;
6171 if (page->index == size >> PAGE_SHIFT)
6172 len = size & ~PAGE_MASK;
6176 err = __block_write_begin(page, 0, len, ext4_get_block);
6178 ret = VM_FAULT_SIGBUS;
6179 if (ext4_walk_page_buffers(handle, inode,
6180 page_buffers(page), 0, len, NULL,
6181 do_journal_get_write_access))
6183 if (ext4_walk_page_buffers(handle, inode,
6184 page_buffers(page), 0, len, NULL,
6187 if (ext4_jbd2_inode_add_write(handle, inode,
6188 page_offset(page), len))
6190 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6195 ext4_journal_stop(handle);
6196 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6199 ret = block_page_mkwrite_return(err);
6201 filemap_invalidate_unlock_shared(mapping);
6202 sb_end_pagefault(inode->i_sb);
6206 ext4_journal_stop(handle);