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. And for inodes with dioread_nolock, unwritten
226 * extents converting worker could merge extents and also have dirtied
227 * the inode. Flush worker is ignoring it because of I_FREEING flag but
228 * we still need to remove the inode from the writeback lists.
230 if (!list_empty_careful(&inode->i_io_list))
231 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);
339 * Check out some where else accidentally dirty the evicting inode,
340 * which may probably cause inode use-after-free issues later.
342 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
344 if (!list_empty(&EXT4_I(inode)->i_fc_list))
345 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
346 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
350 qsize_t *ext4_get_reserved_space(struct inode *inode)
352 return &EXT4_I(inode)->i_reserved_quota;
357 * Called with i_data_sem down, which is important since we can call
358 * ext4_discard_preallocations() from here.
360 void ext4_da_update_reserve_space(struct inode *inode,
361 int used, int quota_claim)
363 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
364 struct ext4_inode_info *ei = EXT4_I(inode);
366 spin_lock(&ei->i_block_reservation_lock);
367 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
368 if (unlikely(used > ei->i_reserved_data_blocks)) {
369 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
370 "with only %d reserved data blocks",
371 __func__, inode->i_ino, used,
372 ei->i_reserved_data_blocks);
374 used = ei->i_reserved_data_blocks;
377 /* Update per-inode reservations */
378 ei->i_reserved_data_blocks -= used;
379 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
381 spin_unlock(&ei->i_block_reservation_lock);
383 /* Update quota subsystem for data blocks */
385 dquot_claim_block(inode, EXT4_C2B(sbi, used));
388 * We did fallocate with an offset that is already delayed
389 * allocated. So on delayed allocated writeback we should
390 * not re-claim the quota for fallocated blocks.
392 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
396 * If we have done all the pending block allocations and if
397 * there aren't any writers on the inode, we can discard the
398 * inode's preallocations.
400 if ((ei->i_reserved_data_blocks == 0) &&
401 !inode_is_open_for_write(inode))
402 ext4_discard_preallocations(inode, 0);
405 static int __check_block_validity(struct inode *inode, const char *func,
407 struct ext4_map_blocks *map)
409 if (ext4_has_feature_journal(inode->i_sb) &&
411 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
413 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock %llu "
416 "(length %d)", (unsigned long) map->m_lblk,
417 map->m_pblk, map->m_len);
418 return -EFSCORRUPTED;
423 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
428 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
429 return fscrypt_zeroout_range(inode, lblk, pblk, len);
431 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
438 #define check_block_validity(inode, map) \
439 __check_block_validity((inode), __func__, __LINE__, (map))
441 #ifdef ES_AGGRESSIVE_TEST
442 static void ext4_map_blocks_es_recheck(handle_t *handle,
444 struct ext4_map_blocks *es_map,
445 struct ext4_map_blocks *map,
452 * There is a race window that the result is not the same.
453 * e.g. xfstests #223 when dioread_nolock enables. The reason
454 * is that we lookup a block mapping in extent status tree with
455 * out taking i_data_sem. So at the time the unwritten extent
456 * could be converted.
458 down_read(&EXT4_I(inode)->i_data_sem);
459 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
460 retval = ext4_ext_map_blocks(handle, inode, map, 0);
462 retval = ext4_ind_map_blocks(handle, inode, map, 0);
464 up_read((&EXT4_I(inode)->i_data_sem));
467 * We don't check m_len because extent will be collpased in status
468 * tree. So the m_len might not equal.
470 if (es_map->m_lblk != map->m_lblk ||
471 es_map->m_flags != map->m_flags ||
472 es_map->m_pblk != map->m_pblk) {
473 printk("ES cache assertion failed for inode: %lu "
474 "es_cached ex [%d/%d/%llu/%x] != "
475 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
476 inode->i_ino, es_map->m_lblk, es_map->m_len,
477 es_map->m_pblk, es_map->m_flags, map->m_lblk,
478 map->m_len, map->m_pblk, map->m_flags,
482 #endif /* ES_AGGRESSIVE_TEST */
485 * The ext4_map_blocks() function tries to look up the requested blocks,
486 * and returns if the blocks are already mapped.
488 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
489 * and store the allocated blocks in the result buffer head and mark it
492 * If file type is extents based, it will call ext4_ext_map_blocks(),
493 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
496 * On success, it returns the number of blocks being mapped or allocated. if
497 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
498 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
500 * It returns 0 if plain look up failed (blocks have not been allocated), in
501 * that case, @map is returned as unmapped but we still do fill map->m_len to
502 * indicate the length of a hole starting at map->m_lblk.
504 * It returns the error in case of allocation failure.
506 int ext4_map_blocks(handle_t *handle, struct inode *inode,
507 struct ext4_map_blocks *map, int flags)
509 struct extent_status es;
512 #ifdef ES_AGGRESSIVE_TEST
513 struct ext4_map_blocks orig_map;
515 memcpy(&orig_map, map, sizeof(*map));
519 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
520 flags, map->m_len, (unsigned long) map->m_lblk);
523 * ext4_map_blocks returns an int, and m_len is an unsigned int
525 if (unlikely(map->m_len > INT_MAX))
526 map->m_len = INT_MAX;
528 /* We can handle the block number less than EXT_MAX_BLOCKS */
529 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
530 return -EFSCORRUPTED;
532 /* Lookup extent status tree firstly */
533 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
534 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
535 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
536 map->m_pblk = ext4_es_pblock(&es) +
537 map->m_lblk - es.es_lblk;
538 map->m_flags |= ext4_es_is_written(&es) ?
539 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
540 retval = es.es_len - (map->m_lblk - es.es_lblk);
541 if (retval > map->m_len)
544 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
546 retval = es.es_len - (map->m_lblk - es.es_lblk);
547 if (retval > map->m_len)
555 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
557 #ifdef ES_AGGRESSIVE_TEST
558 ext4_map_blocks_es_recheck(handle, inode, map,
564 * In the query cache no-wait mode, nothing we can do more if we
565 * cannot find extent in the cache.
567 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
571 * Try to see if we can get the block without requesting a new
574 down_read(&EXT4_I(inode)->i_data_sem);
575 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
576 retval = ext4_ext_map_blocks(handle, inode, map, 0);
578 retval = ext4_ind_map_blocks(handle, inode, map, 0);
583 if (unlikely(retval != map->m_len)) {
584 ext4_warning(inode->i_sb,
585 "ES len assertion failed for inode "
586 "%lu: retval %d != map->m_len %d",
587 inode->i_ino, retval, map->m_len);
591 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
592 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
593 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
594 !(status & EXTENT_STATUS_WRITTEN) &&
595 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
596 map->m_lblk + map->m_len - 1))
597 status |= EXTENT_STATUS_DELAYED;
598 ret = ext4_es_insert_extent(inode, map->m_lblk,
599 map->m_len, map->m_pblk, status);
603 up_read((&EXT4_I(inode)->i_data_sem));
606 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
607 ret = check_block_validity(inode, map);
612 /* If it is only a block(s) look up */
613 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
617 * Returns if the blocks have already allocated
619 * Note that if blocks have been preallocated
620 * ext4_ext_get_block() returns the create = 0
621 * with buffer head unmapped.
623 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
625 * If we need to convert extent to unwritten
626 * we continue and do the actual work in
627 * ext4_ext_map_blocks()
629 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
633 * Here we clear m_flags because after allocating an new extent,
634 * it will be set again.
636 map->m_flags &= ~EXT4_MAP_FLAGS;
639 * New blocks allocate and/or writing to unwritten extent
640 * will possibly result in updating i_data, so we take
641 * the write lock of i_data_sem, and call get_block()
642 * with create == 1 flag.
644 down_write(&EXT4_I(inode)->i_data_sem);
647 * We need to check for EXT4 here because migrate
648 * could have changed the inode type in between
650 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
651 retval = ext4_ext_map_blocks(handle, inode, map, flags);
653 retval = ext4_ind_map_blocks(handle, inode, map, flags);
655 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
657 * We allocated new blocks which will result in
658 * i_data's format changing. Force the migrate
659 * to fail by clearing migrate flags
661 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
665 * Update reserved blocks/metadata blocks after successful
666 * block allocation which had been deferred till now. We don't
667 * support fallocate for non extent files. So we can update
668 * reserve space here.
671 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
672 ext4_da_update_reserve_space(inode, retval, 1);
678 if (unlikely(retval != map->m_len)) {
679 ext4_warning(inode->i_sb,
680 "ES len assertion failed for inode "
681 "%lu: retval %d != map->m_len %d",
682 inode->i_ino, retval, map->m_len);
687 * We have to zeroout blocks before inserting them into extent
688 * status tree. Otherwise someone could look them up there and
689 * use them before they are really zeroed. We also have to
690 * unmap metadata before zeroing as otherwise writeback can
691 * overwrite zeros with stale data from block device.
693 if (flags & EXT4_GET_BLOCKS_ZERO &&
694 map->m_flags & EXT4_MAP_MAPPED &&
695 map->m_flags & EXT4_MAP_NEW) {
696 ret = ext4_issue_zeroout(inode, map->m_lblk,
697 map->m_pblk, map->m_len);
705 * If the extent has been zeroed out, we don't need to update
706 * extent status tree.
708 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
709 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
710 if (ext4_es_is_written(&es))
713 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
714 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
715 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
716 !(status & EXTENT_STATUS_WRITTEN) &&
717 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
718 map->m_lblk + map->m_len - 1))
719 status |= EXTENT_STATUS_DELAYED;
720 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
721 map->m_pblk, status);
729 up_write((&EXT4_I(inode)->i_data_sem));
730 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
731 ret = check_block_validity(inode, map);
736 * Inodes with freshly allocated blocks where contents will be
737 * visible after transaction commit must be on transaction's
740 if (map->m_flags & EXT4_MAP_NEW &&
741 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
742 !(flags & EXT4_GET_BLOCKS_ZERO) &&
743 !ext4_is_quota_file(inode) &&
744 ext4_should_order_data(inode)) {
746 (loff_t)map->m_lblk << inode->i_blkbits;
747 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
749 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
750 ret = ext4_jbd2_inode_add_wait(handle, inode,
753 ret = ext4_jbd2_inode_add_write(handle, inode,
759 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
760 map->m_flags & EXT4_MAP_MAPPED))
761 ext4_fc_track_range(handle, inode, map->m_lblk,
762 map->m_lblk + map->m_len - 1);
764 ext_debug(inode, "failed with err %d\n", retval);
769 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
770 * we have to be careful as someone else may be manipulating b_state as well.
772 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
774 unsigned long old_state;
775 unsigned long new_state;
777 flags &= EXT4_MAP_FLAGS;
779 /* Dummy buffer_head? Set non-atomically. */
781 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
785 * Someone else may be modifying b_state. Be careful! This is ugly but
786 * once we get rid of using bh as a container for mapping information
787 * to pass to / from get_block functions, this can go away.
790 old_state = READ_ONCE(bh->b_state);
791 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
793 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
796 static int _ext4_get_block(struct inode *inode, sector_t iblock,
797 struct buffer_head *bh, int flags)
799 struct ext4_map_blocks map;
802 if (ext4_has_inline_data(inode))
806 map.m_len = bh->b_size >> inode->i_blkbits;
808 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
811 map_bh(bh, inode->i_sb, map.m_pblk);
812 ext4_update_bh_state(bh, map.m_flags);
813 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
815 } else if (ret == 0) {
816 /* hole case, need to fill in bh->b_size */
817 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
822 int ext4_get_block(struct inode *inode, sector_t iblock,
823 struct buffer_head *bh, int create)
825 return _ext4_get_block(inode, iblock, bh,
826 create ? EXT4_GET_BLOCKS_CREATE : 0);
830 * Get block function used when preparing for buffered write if we require
831 * creating an unwritten extent if blocks haven't been allocated. The extent
832 * will be converted to written after the IO is complete.
834 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
835 struct buffer_head *bh_result, int create)
837 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
838 inode->i_ino, create);
839 return _ext4_get_block(inode, iblock, bh_result,
840 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
843 /* Maximum number of blocks we map for direct IO at once. */
844 #define DIO_MAX_BLOCKS 4096
847 * `handle' can be NULL if create is zero
849 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
850 ext4_lblk_t block, int map_flags)
852 struct ext4_map_blocks map;
853 struct buffer_head *bh;
854 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
855 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
858 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
859 || handle != NULL || create == 0);
860 ASSERT(create == 0 || !nowait);
864 err = ext4_map_blocks(handle, inode, &map, map_flags);
867 return create ? ERR_PTR(-ENOSPC) : NULL;
872 return sb_find_get_block(inode->i_sb, map.m_pblk);
874 bh = sb_getblk(inode->i_sb, map.m_pblk);
876 return ERR_PTR(-ENOMEM);
877 if (map.m_flags & EXT4_MAP_NEW) {
879 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
880 || (handle != NULL));
883 * Now that we do not always journal data, we should
884 * keep in mind whether this should always journal the
885 * new buffer as metadata. For now, regular file
886 * writes use ext4_get_block instead, so it's not a
890 BUFFER_TRACE(bh, "call get_create_access");
891 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
897 if (!buffer_uptodate(bh)) {
898 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
899 set_buffer_uptodate(bh);
902 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
903 err = ext4_handle_dirty_metadata(handle, inode, bh);
907 BUFFER_TRACE(bh, "not a new buffer");
914 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
915 ext4_lblk_t block, int map_flags)
917 struct buffer_head *bh;
920 bh = ext4_getblk(handle, inode, block, map_flags);
923 if (!bh || ext4_buffer_uptodate(bh))
926 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
934 /* Read a contiguous batch of blocks. */
935 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
936 bool wait, struct buffer_head **bhs)
940 for (i = 0; i < bh_count; i++) {
941 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
942 if (IS_ERR(bhs[i])) {
943 err = PTR_ERR(bhs[i]);
949 for (i = 0; i < bh_count; i++)
950 /* Note that NULL bhs[i] is valid because of holes. */
951 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
952 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
957 for (i = 0; i < bh_count; i++)
959 wait_on_buffer(bhs[i]);
961 for (i = 0; i < bh_count; i++) {
962 if (bhs[i] && !buffer_uptodate(bhs[i])) {
970 for (i = 0; i < bh_count; i++) {
977 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
978 struct buffer_head *head,
982 int (*fn)(handle_t *handle, struct inode *inode,
983 struct buffer_head *bh))
985 struct buffer_head *bh;
986 unsigned block_start, block_end;
987 unsigned blocksize = head->b_size;
989 struct buffer_head *next;
991 for (bh = head, block_start = 0;
992 ret == 0 && (bh != head || !block_start);
993 block_start = block_end, bh = next) {
994 next = bh->b_this_page;
995 block_end = block_start + blocksize;
996 if (block_end <= from || block_start >= to) {
997 if (partial && !buffer_uptodate(bh))
1001 err = (*fn)(handle, inode, bh);
1009 * To preserve ordering, it is essential that the hole instantiation and
1010 * the data write be encapsulated in a single transaction. We cannot
1011 * close off a transaction and start a new one between the ext4_get_block()
1012 * and the commit_write(). So doing the jbd2_journal_start at the start of
1013 * prepare_write() is the right place.
1015 * Also, this function can nest inside ext4_writepage(). In that case, we
1016 * *know* that ext4_writepage() has generated enough buffer credits to do the
1017 * whole page. So we won't block on the journal in that case, which is good,
1018 * because the caller may be PF_MEMALLOC.
1020 * By accident, ext4 can be reentered when a transaction is open via
1021 * quota file writes. If we were to commit the transaction while thus
1022 * reentered, there can be a deadlock - we would be holding a quota
1023 * lock, and the commit would never complete if another thread had a
1024 * transaction open and was blocking on the quota lock - a ranking
1027 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1028 * will _not_ run commit under these circumstances because handle->h_ref
1029 * is elevated. We'll still have enough credits for the tiny quotafile
1032 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1033 struct buffer_head *bh)
1035 int dirty = buffer_dirty(bh);
1038 if (!buffer_mapped(bh) || buffer_freed(bh))
1041 * __block_write_begin() could have dirtied some buffers. Clean
1042 * the dirty bit as jbd2_journal_get_write_access() could complain
1043 * otherwise about fs integrity issues. Setting of the dirty bit
1044 * by __block_write_begin() isn't a real problem here as we clear
1045 * the bit before releasing a page lock and thus writeback cannot
1046 * ever write the buffer.
1049 clear_buffer_dirty(bh);
1050 BUFFER_TRACE(bh, "get write access");
1051 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1054 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1058 #ifdef CONFIG_FS_ENCRYPTION
1059 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1060 get_block_t *get_block)
1062 unsigned from = pos & (PAGE_SIZE - 1);
1063 unsigned to = from + len;
1064 struct inode *inode = page->mapping->host;
1065 unsigned block_start, block_end;
1068 unsigned blocksize = inode->i_sb->s_blocksize;
1070 struct buffer_head *bh, *head, *wait[2];
1074 BUG_ON(!PageLocked(page));
1075 BUG_ON(from > PAGE_SIZE);
1076 BUG_ON(to > PAGE_SIZE);
1079 if (!page_has_buffers(page))
1080 create_empty_buffers(page, blocksize, 0);
1081 head = page_buffers(page);
1082 bbits = ilog2(blocksize);
1083 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1085 for (bh = head, block_start = 0; bh != head || !block_start;
1086 block++, block_start = block_end, bh = bh->b_this_page) {
1087 block_end = block_start + blocksize;
1088 if (block_end <= from || block_start >= to) {
1089 if (PageUptodate(page)) {
1090 set_buffer_uptodate(bh);
1095 clear_buffer_new(bh);
1096 if (!buffer_mapped(bh)) {
1097 WARN_ON(bh->b_size != blocksize);
1098 err = get_block(inode, block, bh, 1);
1101 if (buffer_new(bh)) {
1102 if (PageUptodate(page)) {
1103 clear_buffer_new(bh);
1104 set_buffer_uptodate(bh);
1105 mark_buffer_dirty(bh);
1108 if (block_end > to || block_start < from)
1109 zero_user_segments(page, to, block_end,
1114 if (PageUptodate(page)) {
1115 set_buffer_uptodate(bh);
1118 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1119 !buffer_unwritten(bh) &&
1120 (block_start < from || block_end > to)) {
1121 ext4_read_bh_lock(bh, 0, false);
1122 wait[nr_wait++] = bh;
1126 * If we issued read requests, let them complete.
1128 for (i = 0; i < nr_wait; i++) {
1129 wait_on_buffer(wait[i]);
1130 if (!buffer_uptodate(wait[i]))
1133 if (unlikely(err)) {
1134 page_zero_new_buffers(page, from, to);
1135 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1136 for (i = 0; i < nr_wait; i++) {
1139 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1140 bh_offset(wait[i]));
1142 clear_buffer_uptodate(wait[i]);
1152 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1153 loff_t pos, unsigned len,
1154 struct page **pagep, void **fsdata)
1156 struct inode *inode = mapping->host;
1157 int ret, needed_blocks;
1164 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1167 trace_ext4_write_begin(inode, pos, len);
1169 * Reserve one block more for addition to orphan list in case
1170 * we allocate blocks but write fails for some reason
1172 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1173 index = pos >> PAGE_SHIFT;
1174 from = pos & (PAGE_SIZE - 1);
1177 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1178 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1187 * grab_cache_page_write_begin() can take a long time if the
1188 * system is thrashing due to memory pressure, or if the page
1189 * is being written back. So grab it first before we start
1190 * the transaction handle. This also allows us to allocate
1191 * the page (if needed) without using GFP_NOFS.
1194 page = grab_cache_page_write_begin(mapping, index);
1198 * The same as page allocation, we prealloc buffer heads before
1199 * starting the handle.
1201 if (!page_has_buffers(page))
1202 create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1207 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1208 if (IS_ERR(handle)) {
1210 return PTR_ERR(handle);
1214 if (page->mapping != mapping) {
1215 /* The page got truncated from under us */
1218 ext4_journal_stop(handle);
1221 /* In case writeback began while the page was unlocked */
1222 wait_for_stable_page(page);
1224 #ifdef CONFIG_FS_ENCRYPTION
1225 if (ext4_should_dioread_nolock(inode))
1226 ret = ext4_block_write_begin(page, pos, len,
1227 ext4_get_block_unwritten);
1229 ret = ext4_block_write_begin(page, pos, len,
1232 if (ext4_should_dioread_nolock(inode))
1233 ret = __block_write_begin(page, pos, len,
1234 ext4_get_block_unwritten);
1236 ret = __block_write_begin(page, pos, len, ext4_get_block);
1238 if (!ret && ext4_should_journal_data(inode)) {
1239 ret = ext4_walk_page_buffers(handle, inode,
1240 page_buffers(page), from, to, NULL,
1241 do_journal_get_write_access);
1245 bool extended = (pos + len > inode->i_size) &&
1246 !ext4_verity_in_progress(inode);
1250 * __block_write_begin may have instantiated a few blocks
1251 * outside i_size. Trim these off again. Don't need
1252 * i_size_read because we hold i_rwsem.
1254 * Add inode to orphan list in case we crash before
1257 if (extended && ext4_can_truncate(inode))
1258 ext4_orphan_add(handle, inode);
1260 ext4_journal_stop(handle);
1262 ext4_truncate_failed_write(inode);
1264 * If truncate failed early the inode might
1265 * still be on the orphan list; we need to
1266 * make sure the inode is removed from the
1267 * orphan list in that case.
1270 ext4_orphan_del(NULL, inode);
1273 if (ret == -ENOSPC &&
1274 ext4_should_retry_alloc(inode->i_sb, &retries))
1283 /* For write_end() in data=journal mode */
1284 static int write_end_fn(handle_t *handle, struct inode *inode,
1285 struct buffer_head *bh)
1288 if (!buffer_mapped(bh) || buffer_freed(bh))
1290 set_buffer_uptodate(bh);
1291 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1292 clear_buffer_meta(bh);
1293 clear_buffer_prio(bh);
1298 * We need to pick up the new inode size which generic_commit_write gave us
1299 * `file' can be NULL - eg, when called from page_symlink().
1301 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1302 * buffers are managed internally.
1304 static int ext4_write_end(struct file *file,
1305 struct address_space *mapping,
1306 loff_t pos, unsigned len, unsigned copied,
1307 struct page *page, void *fsdata)
1309 handle_t *handle = ext4_journal_current_handle();
1310 struct inode *inode = mapping->host;
1311 loff_t old_size = inode->i_size;
1313 int i_size_changed = 0;
1314 bool verity = ext4_verity_in_progress(inode);
1316 trace_ext4_write_end(inode, pos, len, copied);
1318 if (ext4_has_inline_data(inode))
1319 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1321 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1323 * it's important to update i_size while still holding page lock:
1324 * page writeout could otherwise come in and zero beyond i_size.
1326 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1327 * blocks are being written past EOF, so skip the i_size update.
1330 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1334 if (old_size < pos && !verity)
1335 pagecache_isize_extended(inode, old_size, pos);
1337 * Don't mark the inode dirty under page lock. First, it unnecessarily
1338 * makes the holding time of page lock longer. Second, it forces lock
1339 * ordering of page lock and transaction start for journaling
1343 ret = ext4_mark_inode_dirty(handle, inode);
1345 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1346 /* if we have allocated more blocks and copied
1347 * less. We will have blocks allocated outside
1348 * inode->i_size. So truncate them
1350 ext4_orphan_add(handle, inode);
1352 ret2 = ext4_journal_stop(handle);
1356 if (pos + len > inode->i_size && !verity) {
1357 ext4_truncate_failed_write(inode);
1359 * If truncate failed early the inode might still be
1360 * on the orphan list; we need to make sure the inode
1361 * is removed from the orphan list in that case.
1364 ext4_orphan_del(NULL, inode);
1367 return ret ? ret : copied;
1371 * This is a private version of page_zero_new_buffers() which doesn't
1372 * set the buffer to be dirty, since in data=journalled mode we need
1373 * to call ext4_handle_dirty_metadata() instead.
1375 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1376 struct inode *inode,
1378 unsigned from, unsigned to)
1380 unsigned int block_start = 0, block_end;
1381 struct buffer_head *head, *bh;
1383 bh = head = page_buffers(page);
1385 block_end = block_start + bh->b_size;
1386 if (buffer_new(bh)) {
1387 if (block_end > from && block_start < to) {
1388 if (!PageUptodate(page)) {
1389 unsigned start, size;
1391 start = max(from, block_start);
1392 size = min(to, block_end) - start;
1394 zero_user(page, start, size);
1395 write_end_fn(handle, inode, bh);
1397 clear_buffer_new(bh);
1400 block_start = block_end;
1401 bh = bh->b_this_page;
1402 } while (bh != head);
1405 static int ext4_journalled_write_end(struct file *file,
1406 struct address_space *mapping,
1407 loff_t pos, unsigned len, unsigned copied,
1408 struct page *page, void *fsdata)
1410 handle_t *handle = ext4_journal_current_handle();
1411 struct inode *inode = mapping->host;
1412 loff_t old_size = inode->i_size;
1416 int size_changed = 0;
1417 bool verity = ext4_verity_in_progress(inode);
1419 trace_ext4_journalled_write_end(inode, pos, len, copied);
1420 from = pos & (PAGE_SIZE - 1);
1423 BUG_ON(!ext4_handle_valid(handle));
1425 if (ext4_has_inline_data(inode))
1426 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1428 if (unlikely(copied < len) && !PageUptodate(page)) {
1430 ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
1432 if (unlikely(copied < len))
1433 ext4_journalled_zero_new_buffers(handle, inode, page,
1435 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
1436 from, from + copied, &partial,
1439 SetPageUptodate(page);
1442 size_changed = ext4_update_inode_size(inode, pos + copied);
1443 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1444 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1448 if (old_size < pos && !verity)
1449 pagecache_isize_extended(inode, old_size, pos);
1452 ret2 = ext4_mark_inode_dirty(handle, inode);
1457 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1458 /* if we have allocated more blocks and copied
1459 * less. We will have blocks allocated outside
1460 * inode->i_size. So truncate them
1462 ext4_orphan_add(handle, inode);
1464 ret2 = ext4_journal_stop(handle);
1467 if (pos + len > inode->i_size && !verity) {
1468 ext4_truncate_failed_write(inode);
1470 * If truncate failed early the inode might still be
1471 * on the orphan list; we need to make sure the inode
1472 * is removed from the orphan list in that case.
1475 ext4_orphan_del(NULL, inode);
1478 return ret ? ret : copied;
1482 * Reserve space for a single cluster
1484 static int ext4_da_reserve_space(struct inode *inode)
1486 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1487 struct ext4_inode_info *ei = EXT4_I(inode);
1491 * We will charge metadata quota at writeout time; this saves
1492 * us from metadata over-estimation, though we may go over by
1493 * a small amount in the end. Here we just reserve for data.
1495 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1499 spin_lock(&ei->i_block_reservation_lock);
1500 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1501 spin_unlock(&ei->i_block_reservation_lock);
1502 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1505 ei->i_reserved_data_blocks++;
1506 trace_ext4_da_reserve_space(inode);
1507 spin_unlock(&ei->i_block_reservation_lock);
1509 return 0; /* success */
1512 void ext4_da_release_space(struct inode *inode, int to_free)
1514 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1515 struct ext4_inode_info *ei = EXT4_I(inode);
1518 return; /* Nothing to release, exit */
1520 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1522 trace_ext4_da_release_space(inode, to_free);
1523 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1525 * if there aren't enough reserved blocks, then the
1526 * counter is messed up somewhere. Since this
1527 * function is called from invalidate page, it's
1528 * harmless to return without any action.
1530 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1531 "ino %lu, to_free %d with only %d reserved "
1532 "data blocks", inode->i_ino, to_free,
1533 ei->i_reserved_data_blocks);
1535 to_free = ei->i_reserved_data_blocks;
1537 ei->i_reserved_data_blocks -= to_free;
1539 /* update fs dirty data blocks counter */
1540 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1542 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1544 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1548 * Delayed allocation stuff
1551 struct mpage_da_data {
1552 struct inode *inode;
1553 struct writeback_control *wbc;
1555 pgoff_t first_page; /* The first page to write */
1556 pgoff_t next_page; /* Current page to examine */
1557 pgoff_t last_page; /* Last page to examine */
1559 * Extent to map - this can be after first_page because that can be
1560 * fully mapped. We somewhat abuse m_flags to store whether the extent
1561 * is delalloc or unwritten.
1563 struct ext4_map_blocks map;
1564 struct ext4_io_submit io_submit; /* IO submission data */
1565 unsigned int do_map:1;
1566 unsigned int scanned_until_end:1;
1569 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1574 struct folio_batch fbatch;
1575 struct inode *inode = mpd->inode;
1576 struct address_space *mapping = inode->i_mapping;
1578 /* This is necessary when next_page == 0. */
1579 if (mpd->first_page >= mpd->next_page)
1582 mpd->scanned_until_end = 0;
1583 index = mpd->first_page;
1584 end = mpd->next_page - 1;
1586 ext4_lblk_t start, last;
1587 start = index << (PAGE_SHIFT - inode->i_blkbits);
1588 last = end << (PAGE_SHIFT - inode->i_blkbits);
1591 * avoid racing with extent status tree scans made by
1592 * ext4_insert_delayed_block()
1594 down_write(&EXT4_I(inode)->i_data_sem);
1595 ext4_es_remove_extent(inode, start, last - start + 1);
1596 up_write(&EXT4_I(inode)->i_data_sem);
1599 folio_batch_init(&fbatch);
1600 while (index <= end) {
1601 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1604 for (i = 0; i < nr; i++) {
1605 struct folio *folio = fbatch.folios[i];
1607 if (folio->index < mpd->first_page)
1609 if (folio->index + folio_nr_pages(folio) - 1 > end)
1611 BUG_ON(!folio_test_locked(folio));
1612 BUG_ON(folio_test_writeback(folio));
1614 if (folio_mapped(folio))
1615 folio_clear_dirty_for_io(folio);
1616 block_invalidate_folio(folio, 0,
1618 folio_clear_uptodate(folio);
1620 folio_unlock(folio);
1622 folio_batch_release(&fbatch);
1626 static void ext4_print_free_blocks(struct inode *inode)
1628 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1629 struct super_block *sb = inode->i_sb;
1630 struct ext4_inode_info *ei = EXT4_I(inode);
1632 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1633 EXT4_C2B(EXT4_SB(inode->i_sb),
1634 ext4_count_free_clusters(sb)));
1635 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1636 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1637 (long long) EXT4_C2B(EXT4_SB(sb),
1638 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1639 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1640 (long long) EXT4_C2B(EXT4_SB(sb),
1641 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1642 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1643 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1644 ei->i_reserved_data_blocks);
1648 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1649 struct buffer_head *bh)
1651 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1655 * ext4_insert_delayed_block - adds a delayed block to the extents status
1656 * tree, incrementing the reserved cluster/block
1657 * count or making a pending reservation
1660 * @inode - file containing the newly added block
1661 * @lblk - logical block to be added
1663 * Returns 0 on success, negative error code on failure.
1665 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1667 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1669 bool allocated = false;
1670 bool reserved = false;
1673 * If the cluster containing lblk is shared with a delayed,
1674 * written, or unwritten extent in a bigalloc file system, it's
1675 * already been accounted for and does not need to be reserved.
1676 * A pending reservation must be made for the cluster if it's
1677 * shared with a written or unwritten extent and doesn't already
1678 * have one. Written and unwritten extents can be purged from the
1679 * extents status tree if the system is under memory pressure, so
1680 * it's necessary to examine the extent tree if a search of the
1681 * extents status tree doesn't get a match.
1683 if (sbi->s_cluster_ratio == 1) {
1684 ret = ext4_da_reserve_space(inode);
1685 if (ret != 0) /* ENOSPC */
1688 } else { /* bigalloc */
1689 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1690 if (!ext4_es_scan_clu(inode,
1691 &ext4_es_is_mapped, lblk)) {
1692 ret = ext4_clu_mapped(inode,
1693 EXT4_B2C(sbi, lblk));
1697 ret = ext4_da_reserve_space(inode);
1698 if (ret != 0) /* ENOSPC */
1710 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1711 if (ret && reserved)
1712 ext4_da_release_space(inode, 1);
1719 * This function is grabs code from the very beginning of
1720 * ext4_map_blocks, but assumes that the caller is from delayed write
1721 * time. This function looks up the requested blocks and sets the
1722 * buffer delay bit under the protection of i_data_sem.
1724 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1725 struct ext4_map_blocks *map,
1726 struct buffer_head *bh)
1728 struct extent_status es;
1730 sector_t invalid_block = ~((sector_t) 0xffff);
1731 #ifdef ES_AGGRESSIVE_TEST
1732 struct ext4_map_blocks orig_map;
1734 memcpy(&orig_map, map, sizeof(*map));
1737 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1741 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1742 (unsigned long) map->m_lblk);
1744 /* Lookup extent status tree firstly */
1745 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1746 if (ext4_es_is_hole(&es)) {
1748 down_read(&EXT4_I(inode)->i_data_sem);
1753 * Delayed extent could be allocated by fallocate.
1754 * So we need to check it.
1756 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1757 map_bh(bh, inode->i_sb, invalid_block);
1759 set_buffer_delay(bh);
1763 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1764 retval = es.es_len - (iblock - es.es_lblk);
1765 if (retval > map->m_len)
1766 retval = map->m_len;
1767 map->m_len = retval;
1768 if (ext4_es_is_written(&es))
1769 map->m_flags |= EXT4_MAP_MAPPED;
1770 else if (ext4_es_is_unwritten(&es))
1771 map->m_flags |= EXT4_MAP_UNWRITTEN;
1775 #ifdef ES_AGGRESSIVE_TEST
1776 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1782 * Try to see if we can get the block without requesting a new
1783 * file system block.
1785 down_read(&EXT4_I(inode)->i_data_sem);
1786 if (ext4_has_inline_data(inode))
1788 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1789 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1791 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1798 * XXX: __block_prepare_write() unmaps passed block,
1802 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1808 map_bh(bh, inode->i_sb, invalid_block);
1810 set_buffer_delay(bh);
1811 } else if (retval > 0) {
1813 unsigned int status;
1815 if (unlikely(retval != map->m_len)) {
1816 ext4_warning(inode->i_sb,
1817 "ES len assertion failed for inode "
1818 "%lu: retval %d != map->m_len %d",
1819 inode->i_ino, retval, map->m_len);
1823 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1824 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1825 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1826 map->m_pblk, status);
1832 up_read((&EXT4_I(inode)->i_data_sem));
1838 * This is a special get_block_t callback which is used by
1839 * ext4_da_write_begin(). It will either return mapped block or
1840 * reserve space for a single block.
1842 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1843 * We also have b_blocknr = -1 and b_bdev initialized properly
1845 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1846 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1847 * initialized properly.
1849 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1850 struct buffer_head *bh, int create)
1852 struct ext4_map_blocks map;
1855 BUG_ON(create == 0);
1856 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1858 map.m_lblk = iblock;
1862 * first, we need to know whether the block is allocated already
1863 * preallocated blocks are unmapped but should treated
1864 * the same as allocated blocks.
1866 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1870 map_bh(bh, inode->i_sb, map.m_pblk);
1871 ext4_update_bh_state(bh, map.m_flags);
1873 if (buffer_unwritten(bh)) {
1874 /* A delayed write to unwritten bh should be marked
1875 * new and mapped. Mapped ensures that we don't do
1876 * get_block multiple times when we write to the same
1877 * offset and new ensures that we do proper zero out
1878 * for partial write.
1881 set_buffer_mapped(bh);
1886 static int __ext4_journalled_writepage(struct page *page,
1889 struct address_space *mapping = page->mapping;
1890 struct inode *inode = mapping->host;
1891 handle_t *handle = NULL;
1892 int ret = 0, err = 0;
1893 int inline_data = ext4_has_inline_data(inode);
1894 struct buffer_head *inode_bh = NULL;
1897 ClearPageChecked(page);
1900 BUG_ON(page->index != 0);
1901 BUG_ON(len > ext4_get_max_inline_size(inode));
1902 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1903 if (inode_bh == NULL)
1907 * We need to release the page lock before we start the
1908 * journal, so grab a reference so the page won't disappear
1909 * out from under us.
1914 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1915 ext4_writepage_trans_blocks(inode));
1916 if (IS_ERR(handle)) {
1917 ret = PTR_ERR(handle);
1919 goto out_no_pagelock;
1921 BUG_ON(!ext4_handle_valid(handle));
1925 size = i_size_read(inode);
1926 if (page->mapping != mapping || page_offset(page) > size) {
1927 /* The page got truncated from under us */
1928 ext4_journal_stop(handle);
1934 ret = ext4_mark_inode_dirty(handle, inode);
1936 struct buffer_head *page_bufs = page_buffers(page);
1938 if (page->index == size >> PAGE_SHIFT)
1939 len = size & ~PAGE_MASK;
1943 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1944 NULL, do_journal_get_write_access);
1946 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1947 NULL, write_end_fn);
1951 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1954 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1955 err = ext4_journal_stop(handle);
1959 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1968 * Note that we don't need to start a transaction unless we're journaling data
1969 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1970 * need to file the inode to the transaction's list in ordered mode because if
1971 * we are writing back data added by write(), the inode is already there and if
1972 * we are writing back data modified via mmap(), no one guarantees in which
1973 * transaction the data will hit the disk. In case we are journaling data, we
1974 * cannot start transaction directly because transaction start ranks above page
1975 * lock so we have to do some magic.
1977 * This function can get called via...
1978 * - ext4_writepages after taking page lock (have journal handle)
1979 * - journal_submit_inode_data_buffers (no journal handle)
1980 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1981 * - grab_page_cache when doing write_begin (have journal handle)
1983 * We don't do any block allocation in this function. If we have page with
1984 * multiple blocks we need to write those buffer_heads that are mapped. This
1985 * is important for mmaped based write. So if we do with blocksize 1K
1986 * truncate(f, 1024);
1987 * a = mmap(f, 0, 4096);
1989 * truncate(f, 4096);
1990 * we have in the page first buffer_head mapped via page_mkwrite call back
1991 * but other buffer_heads would be unmapped but dirty (dirty done via the
1992 * do_wp_page). So writepage should write the first block. If we modify
1993 * the mmap area beyond 1024 we will again get a page_fault and the
1994 * page_mkwrite callback will do the block allocation and mark the
1995 * buffer_heads mapped.
1997 * We redirty the page if we have any buffer_heads that is either delay or
1998 * unwritten in the page.
2000 * We can get recursively called as show below.
2002 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2005 * But since we don't do any block allocation we should not deadlock.
2006 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2008 static int ext4_writepage(struct page *page,
2009 struct writeback_control *wbc)
2011 struct folio *folio = page_folio(page);
2015 struct buffer_head *page_bufs = NULL;
2016 struct inode *inode = page->mapping->host;
2017 struct ext4_io_submit io_submit;
2018 bool keep_towrite = false;
2020 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2021 folio_invalidate(folio, 0, folio_size(folio));
2022 folio_unlock(folio);
2026 trace_ext4_writepage(page);
2027 size = i_size_read(inode);
2028 if (page->index == size >> PAGE_SHIFT &&
2029 !ext4_verity_in_progress(inode))
2030 len = size & ~PAGE_MASK;
2034 /* Should never happen but for bugs in other kernel subsystems */
2035 if (!page_has_buffers(page)) {
2036 ext4_warning_inode(inode,
2037 "page %lu does not have buffers attached", page->index);
2038 ClearPageDirty(page);
2043 page_bufs = page_buffers(page);
2045 * We cannot do block allocation or other extent handling in this
2046 * function. If there are buffers needing that, we have to redirty
2047 * the page. But we may reach here when we do a journal commit via
2048 * journal_submit_inode_data_buffers() and in that case we must write
2049 * allocated buffers to achieve data=ordered mode guarantees.
2051 * Also, if there is only one buffer per page (the fs block
2052 * size == the page size), if one buffer needs block
2053 * allocation or needs to modify the extent tree to clear the
2054 * unwritten flag, we know that the page can't be written at
2055 * all, so we might as well refuse the write immediately.
2056 * Unfortunately if the block size != page size, we can't as
2057 * easily detect this case using ext4_walk_page_buffers(), but
2058 * for the extremely common case, this is an optimization that
2059 * skips a useless round trip through ext4_bio_write_page().
2061 if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2062 ext4_bh_delay_or_unwritten)) {
2063 redirty_page_for_writepage(wbc, page);
2064 if ((current->flags & PF_MEMALLOC) ||
2065 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2067 * For memory cleaning there's no point in writing only
2068 * some buffers. So just bail out. Warn if we came here
2069 * from direct reclaim.
2071 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2076 keep_towrite = true;
2079 if (PageChecked(page) && ext4_should_journal_data(inode))
2081 * It's mmapped pagecache. Add buffers and journal it. There
2082 * doesn't seem much point in redirtying the page here.
2084 return __ext4_journalled_writepage(page, len);
2086 ext4_io_submit_init(&io_submit, wbc);
2087 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2088 if (!io_submit.io_end) {
2089 redirty_page_for_writepage(wbc, page);
2093 ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite);
2094 ext4_io_submit(&io_submit);
2095 /* Drop io_end reference we got from init */
2096 ext4_put_io_end_defer(io_submit.io_end);
2100 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2106 BUG_ON(page->index != mpd->first_page);
2107 clear_page_dirty_for_io(page);
2109 * We have to be very careful here! Nothing protects writeback path
2110 * against i_size changes and the page can be writeably mapped into
2111 * page tables. So an application can be growing i_size and writing
2112 * data through mmap while writeback runs. clear_page_dirty_for_io()
2113 * write-protects our page in page tables and the page cannot get
2114 * written to again until we release page lock. So only after
2115 * clear_page_dirty_for_io() we are safe to sample i_size for
2116 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2117 * on the barrier provided by TestClearPageDirty in
2118 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2119 * after page tables are updated.
2121 size = i_size_read(mpd->inode);
2122 if (page->index == size >> PAGE_SHIFT &&
2123 !ext4_verity_in_progress(mpd->inode))
2124 len = size & ~PAGE_MASK;
2127 err = ext4_bio_write_page(&mpd->io_submit, page, len, false);
2129 mpd->wbc->nr_to_write--;
2135 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2138 * mballoc gives us at most this number of blocks...
2139 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2140 * The rest of mballoc seems to handle chunks up to full group size.
2142 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2145 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2147 * @mpd - extent of blocks
2148 * @lblk - logical number of the block in the file
2149 * @bh - buffer head we want to add to the extent
2151 * The function is used to collect contig. blocks in the same state. If the
2152 * buffer doesn't require mapping for writeback and we haven't started the
2153 * extent of buffers to map yet, the function returns 'true' immediately - the
2154 * caller can write the buffer right away. Otherwise the function returns true
2155 * if the block has been added to the extent, false if the block couldn't be
2158 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2159 struct buffer_head *bh)
2161 struct ext4_map_blocks *map = &mpd->map;
2163 /* Buffer that doesn't need mapping for writeback? */
2164 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2165 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2166 /* So far no extent to map => we write the buffer right away */
2167 if (map->m_len == 0)
2172 /* First block in the extent? */
2173 if (map->m_len == 0) {
2174 /* We cannot map unless handle is started... */
2179 map->m_flags = bh->b_state & BH_FLAGS;
2183 /* Don't go larger than mballoc is willing to allocate */
2184 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2187 /* Can we merge the block to our big extent? */
2188 if (lblk == map->m_lblk + map->m_len &&
2189 (bh->b_state & BH_FLAGS) == map->m_flags) {
2197 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2199 * @mpd - extent of blocks for mapping
2200 * @head - the first buffer in the page
2201 * @bh - buffer we should start processing from
2202 * @lblk - logical number of the block in the file corresponding to @bh
2204 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2205 * the page for IO if all buffers in this page were mapped and there's no
2206 * accumulated extent of buffers to map or add buffers in the page to the
2207 * extent of buffers to map. The function returns 1 if the caller can continue
2208 * by processing the next page, 0 if it should stop adding buffers to the
2209 * extent to map because we cannot extend it anymore. It can also return value
2210 * < 0 in case of error during IO submission.
2212 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2213 struct buffer_head *head,
2214 struct buffer_head *bh,
2217 struct inode *inode = mpd->inode;
2219 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2220 >> inode->i_blkbits;
2222 if (ext4_verity_in_progress(inode))
2223 blocks = EXT_MAX_BLOCKS;
2226 BUG_ON(buffer_locked(bh));
2228 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2229 /* Found extent to map? */
2232 /* Buffer needs mapping and handle is not started? */
2235 /* Everything mapped so far and we hit EOF */
2238 } while (lblk++, (bh = bh->b_this_page) != head);
2239 /* So far everything mapped? Submit the page for IO. */
2240 if (mpd->map.m_len == 0) {
2241 err = mpage_submit_page(mpd, head->b_page);
2245 if (lblk >= blocks) {
2246 mpd->scanned_until_end = 1;
2253 * mpage_process_page - update page buffers corresponding to changed extent and
2254 * may submit fully mapped page for IO
2256 * @mpd - description of extent to map, on return next extent to map
2257 * @m_lblk - logical block mapping.
2258 * @m_pblk - corresponding physical mapping.
2259 * @map_bh - determines on return whether this page requires any further
2261 * Scan given page buffers corresponding to changed extent and update buffer
2262 * state according to new extent state.
2263 * We map delalloc buffers to their physical location, clear unwritten bits.
2264 * If the given page is not fully mapped, we update @map to the next extent in
2265 * the given page that needs mapping & return @map_bh as true.
2267 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2268 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2271 struct buffer_head *head, *bh;
2272 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2273 ext4_lblk_t lblk = *m_lblk;
2274 ext4_fsblk_t pblock = *m_pblk;
2276 int blkbits = mpd->inode->i_blkbits;
2277 ssize_t io_end_size = 0;
2278 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2280 bh = head = page_buffers(page);
2282 if (lblk < mpd->map.m_lblk)
2284 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2286 * Buffer after end of mapped extent.
2287 * Find next buffer in the page to map.
2290 mpd->map.m_flags = 0;
2291 io_end_vec->size += io_end_size;
2293 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2296 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2297 io_end_vec = ext4_alloc_io_end_vec(io_end);
2298 if (IS_ERR(io_end_vec)) {
2299 err = PTR_ERR(io_end_vec);
2302 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2307 if (buffer_delay(bh)) {
2308 clear_buffer_delay(bh);
2309 bh->b_blocknr = pblock++;
2311 clear_buffer_unwritten(bh);
2312 io_end_size += (1 << blkbits);
2313 } while (lblk++, (bh = bh->b_this_page) != head);
2315 io_end_vec->size += io_end_size;
2324 * mpage_map_buffers - update buffers corresponding to changed extent and
2325 * submit fully mapped pages for IO
2327 * @mpd - description of extent to map, on return next extent to map
2329 * Scan buffers corresponding to changed extent (we expect corresponding pages
2330 * to be already locked) and update buffer state according to new extent state.
2331 * We map delalloc buffers to their physical location, clear unwritten bits,
2332 * and mark buffers as uninit when we perform writes to unwritten extents
2333 * and do extent conversion after IO is finished. If the last page is not fully
2334 * mapped, we update @map to the next extent in the last page that needs
2335 * mapping. Otherwise we submit the page for IO.
2337 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2339 struct folio_batch fbatch;
2341 struct inode *inode = mpd->inode;
2342 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2345 ext4_fsblk_t pblock;
2347 bool map_bh = false;
2349 start = mpd->map.m_lblk >> bpp_bits;
2350 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2351 lblk = start << bpp_bits;
2352 pblock = mpd->map.m_pblk;
2354 folio_batch_init(&fbatch);
2355 while (start <= end) {
2356 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2359 for (i = 0; i < nr; i++) {
2360 struct page *page = &fbatch.folios[i]->page;
2362 err = mpage_process_page(mpd, page, &lblk, &pblock,
2365 * If map_bh is true, means page may require further bh
2366 * mapping, or maybe the page was submitted for IO.
2367 * So we return to call further extent mapping.
2369 if (err < 0 || map_bh)
2371 /* Page fully mapped - let IO run! */
2372 err = mpage_submit_page(mpd, page);
2376 folio_batch_release(&fbatch);
2378 /* Extent fully mapped and matches with page boundary. We are done. */
2380 mpd->map.m_flags = 0;
2383 folio_batch_release(&fbatch);
2387 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2389 struct inode *inode = mpd->inode;
2390 struct ext4_map_blocks *map = &mpd->map;
2391 int get_blocks_flags;
2392 int err, dioread_nolock;
2394 trace_ext4_da_write_pages_extent(inode, map);
2396 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2397 * to convert an unwritten extent to be initialized (in the case
2398 * where we have written into one or more preallocated blocks). It is
2399 * possible that we're going to need more metadata blocks than
2400 * previously reserved. However we must not fail because we're in
2401 * writeback and there is nothing we can do about it so it might result
2402 * in data loss. So use reserved blocks to allocate metadata if
2405 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2406 * the blocks in question are delalloc blocks. This indicates
2407 * that the blocks and quotas has already been checked when
2408 * the data was copied into the page cache.
2410 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2411 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2412 EXT4_GET_BLOCKS_IO_SUBMIT;
2413 dioread_nolock = ext4_should_dioread_nolock(inode);
2415 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2416 if (map->m_flags & BIT(BH_Delay))
2417 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2419 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2422 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2423 if (!mpd->io_submit.io_end->handle &&
2424 ext4_handle_valid(handle)) {
2425 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2426 handle->h_rsv_handle = NULL;
2428 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2431 BUG_ON(map->m_len == 0);
2436 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2437 * mpd->len and submit pages underlying it for IO
2439 * @handle - handle for journal operations
2440 * @mpd - extent to map
2441 * @give_up_on_write - we set this to true iff there is a fatal error and there
2442 * is no hope of writing the data. The caller should discard
2443 * dirty pages to avoid infinite loops.
2445 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2446 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2447 * them to initialized or split the described range from larger unwritten
2448 * extent. Note that we need not map all the described range since allocation
2449 * can return less blocks or the range is covered by more unwritten extents. We
2450 * cannot map more because we are limited by reserved transaction credits. On
2451 * the other hand we always make sure that the last touched page is fully
2452 * mapped so that it can be written out (and thus forward progress is
2453 * guaranteed). After mapping we submit all mapped pages for IO.
2455 static int mpage_map_and_submit_extent(handle_t *handle,
2456 struct mpage_da_data *mpd,
2457 bool *give_up_on_write)
2459 struct inode *inode = mpd->inode;
2460 struct ext4_map_blocks *map = &mpd->map;
2464 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2465 struct ext4_io_end_vec *io_end_vec;
2467 io_end_vec = ext4_alloc_io_end_vec(io_end);
2468 if (IS_ERR(io_end_vec))
2469 return PTR_ERR(io_end_vec);
2470 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2472 err = mpage_map_one_extent(handle, mpd);
2474 struct super_block *sb = inode->i_sb;
2476 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2477 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2478 goto invalidate_dirty_pages;
2480 * Let the uper layers retry transient errors.
2481 * In the case of ENOSPC, if ext4_count_free_blocks()
2482 * is non-zero, a commit should free up blocks.
2484 if ((err == -ENOMEM) ||
2485 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2487 goto update_disksize;
2490 ext4_msg(sb, KERN_CRIT,
2491 "Delayed block allocation failed for "
2492 "inode %lu at logical offset %llu with"
2493 " max blocks %u with error %d",
2495 (unsigned long long)map->m_lblk,
2496 (unsigned)map->m_len, -err);
2497 ext4_msg(sb, KERN_CRIT,
2498 "This should not happen!! Data will "
2501 ext4_print_free_blocks(inode);
2502 invalidate_dirty_pages:
2503 *give_up_on_write = true;
2508 * Update buffer state, submit mapped pages, and get us new
2511 err = mpage_map_and_submit_buffers(mpd);
2513 goto update_disksize;
2514 } while (map->m_len);
2518 * Update on-disk size after IO is submitted. Races with
2519 * truncate are avoided by checking i_size under i_data_sem.
2521 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2522 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2526 down_write(&EXT4_I(inode)->i_data_sem);
2527 i_size = i_size_read(inode);
2528 if (disksize > i_size)
2530 if (disksize > EXT4_I(inode)->i_disksize)
2531 EXT4_I(inode)->i_disksize = disksize;
2532 up_write(&EXT4_I(inode)->i_data_sem);
2533 err2 = ext4_mark_inode_dirty(handle, inode);
2535 ext4_error_err(inode->i_sb, -err2,
2536 "Failed to mark inode %lu dirty",
2546 * Calculate the total number of credits to reserve for one writepages
2547 * iteration. This is called from ext4_writepages(). We map an extent of
2548 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2549 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2550 * bpp - 1 blocks in bpp different extents.
2552 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2554 int bpp = ext4_journal_blocks_per_page(inode);
2556 return ext4_meta_trans_blocks(inode,
2557 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2561 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2562 * and underlying extent to map
2564 * @mpd - where to look for pages
2566 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2567 * IO immediately. When we find a page which isn't mapped we start accumulating
2568 * extent of buffers underlying these pages that needs mapping (formed by
2569 * either delayed or unwritten buffers). We also lock the pages containing
2570 * these buffers. The extent found is returned in @mpd structure (starting at
2571 * mpd->lblk with length mpd->len blocks).
2573 * Note that this function can attach bios to one io_end structure which are
2574 * neither logically nor physically contiguous. Although it may seem as an
2575 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2576 * case as we need to track IO to all buffers underlying a page in one io_end.
2578 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2580 struct address_space *mapping = mpd->inode->i_mapping;
2581 struct pagevec pvec;
2582 unsigned int nr_pages;
2583 long left = mpd->wbc->nr_to_write;
2584 pgoff_t index = mpd->first_page;
2585 pgoff_t end = mpd->last_page;
2588 int blkbits = mpd->inode->i_blkbits;
2590 struct buffer_head *head;
2592 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2593 tag = PAGECACHE_TAG_TOWRITE;
2595 tag = PAGECACHE_TAG_DIRTY;
2597 pagevec_init(&pvec);
2599 mpd->next_page = index;
2600 while (index <= end) {
2601 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2606 for (i = 0; i < nr_pages; i++) {
2607 struct page *page = pvec.pages[i];
2610 * Accumulated enough dirty pages? This doesn't apply
2611 * to WB_SYNC_ALL mode. For integrity sync we have to
2612 * keep going because someone may be concurrently
2613 * dirtying pages, and we might have synced a lot of
2614 * newly appeared dirty pages, but have not synced all
2615 * of the old dirty pages.
2617 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2620 /* If we can't merge this page, we are done. */
2621 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2626 * If the page is no longer dirty, or its mapping no
2627 * longer corresponds to inode we are writing (which
2628 * means it has been truncated or invalidated), or the
2629 * page is already under writeback and we are not doing
2630 * a data integrity writeback, skip the page
2632 if (!PageDirty(page) ||
2633 (PageWriteback(page) &&
2634 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2635 unlikely(page->mapping != mapping)) {
2640 wait_on_page_writeback(page);
2641 BUG_ON(PageWriteback(page));
2644 * Should never happen but for buggy code in
2645 * other subsystems that call
2646 * set_page_dirty() without properly warning
2647 * the file system first. See [1] for more
2650 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2652 if (!page_has_buffers(page)) {
2653 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2654 ClearPageDirty(page);
2659 if (mpd->map.m_len == 0)
2660 mpd->first_page = page->index;
2661 mpd->next_page = page->index + 1;
2662 /* Add all dirty buffers to mpd */
2663 lblk = ((ext4_lblk_t)page->index) <<
2664 (PAGE_SHIFT - blkbits);
2665 head = page_buffers(page);
2666 err = mpage_process_page_bufs(mpd, head, head, lblk);
2672 pagevec_release(&pvec);
2675 mpd->scanned_until_end = 1;
2678 pagevec_release(&pvec);
2682 static int ext4_writepages(struct address_space *mapping,
2683 struct writeback_control *wbc)
2685 pgoff_t writeback_index = 0;
2686 long nr_to_write = wbc->nr_to_write;
2687 int range_whole = 0;
2689 handle_t *handle = NULL;
2690 struct mpage_da_data mpd;
2691 struct inode *inode = mapping->host;
2692 int needed_blocks, rsv_blocks = 0, ret = 0;
2693 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2694 struct blk_plug plug;
2695 bool give_up_on_write = false;
2697 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2700 percpu_down_read(&sbi->s_writepages_rwsem);
2701 trace_ext4_writepages(inode, wbc);
2704 * No pages to write? This is mainly a kludge to avoid starting
2705 * a transaction for special inodes like journal inode on last iput()
2706 * because that could violate lock ordering on umount
2708 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2709 goto out_writepages;
2711 if (ext4_should_journal_data(inode)) {
2712 ret = generic_writepages(mapping, wbc);
2713 goto out_writepages;
2717 * If the filesystem has aborted, it is read-only, so return
2718 * right away instead of dumping stack traces later on that
2719 * will obscure the real source of the problem. We test
2720 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2721 * the latter could be true if the filesystem is mounted
2722 * read-only, and in that case, ext4_writepages should
2723 * *never* be called, so if that ever happens, we would want
2726 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2727 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2729 goto out_writepages;
2733 * If we have inline data and arrive here, it means that
2734 * we will soon create the block for the 1st page, so
2735 * we'd better clear the inline data here.
2737 if (ext4_has_inline_data(inode)) {
2738 /* Just inode will be modified... */
2739 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2740 if (IS_ERR(handle)) {
2741 ret = PTR_ERR(handle);
2742 goto out_writepages;
2744 BUG_ON(ext4_test_inode_state(inode,
2745 EXT4_STATE_MAY_INLINE_DATA));
2746 ext4_destroy_inline_data(handle, inode);
2747 ext4_journal_stop(handle);
2750 if (ext4_should_dioread_nolock(inode)) {
2752 * We may need to convert up to one extent per block in
2753 * the page and we may dirty the inode.
2755 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2756 PAGE_SIZE >> inode->i_blkbits);
2759 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2762 if (wbc->range_cyclic) {
2763 writeback_index = mapping->writeback_index;
2764 if (writeback_index)
2766 mpd.first_page = writeback_index;
2769 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2770 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2775 ext4_io_submit_init(&mpd.io_submit, wbc);
2777 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2778 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2779 blk_start_plug(&plug);
2782 * First writeback pages that don't need mapping - we can avoid
2783 * starting a transaction unnecessarily and also avoid being blocked
2784 * in the block layer on device congestion while having transaction
2788 mpd.scanned_until_end = 0;
2789 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2790 if (!mpd.io_submit.io_end) {
2794 ret = mpage_prepare_extent_to_map(&mpd);
2795 /* Unlock pages we didn't use */
2796 mpage_release_unused_pages(&mpd, false);
2797 /* Submit prepared bio */
2798 ext4_io_submit(&mpd.io_submit);
2799 ext4_put_io_end_defer(mpd.io_submit.io_end);
2800 mpd.io_submit.io_end = NULL;
2804 while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2805 /* For each extent of pages we use new io_end */
2806 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2807 if (!mpd.io_submit.io_end) {
2813 * We have two constraints: We find one extent to map and we
2814 * must always write out whole page (makes a difference when
2815 * blocksize < pagesize) so that we don't block on IO when we
2816 * try to write out the rest of the page. Journalled mode is
2817 * not supported by delalloc.
2819 BUG_ON(ext4_should_journal_data(inode));
2820 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2822 /* start a new transaction */
2823 handle = ext4_journal_start_with_reserve(inode,
2824 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2825 if (IS_ERR(handle)) {
2826 ret = PTR_ERR(handle);
2827 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2828 "%ld pages, ino %lu; err %d", __func__,
2829 wbc->nr_to_write, inode->i_ino, ret);
2830 /* Release allocated io_end */
2831 ext4_put_io_end(mpd.io_submit.io_end);
2832 mpd.io_submit.io_end = NULL;
2837 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2838 ret = mpage_prepare_extent_to_map(&mpd);
2839 if (!ret && mpd.map.m_len)
2840 ret = mpage_map_and_submit_extent(handle, &mpd,
2843 * Caution: If the handle is synchronous,
2844 * ext4_journal_stop() can wait for transaction commit
2845 * to finish which may depend on writeback of pages to
2846 * complete or on page lock to be released. In that
2847 * case, we have to wait until after we have
2848 * submitted all the IO, released page locks we hold,
2849 * and dropped io_end reference (for extent conversion
2850 * to be able to complete) before stopping the handle.
2852 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2853 ext4_journal_stop(handle);
2857 /* Unlock pages we didn't use */
2858 mpage_release_unused_pages(&mpd, give_up_on_write);
2859 /* Submit prepared bio */
2860 ext4_io_submit(&mpd.io_submit);
2863 * Drop our io_end reference we got from init. We have
2864 * to be careful and use deferred io_end finishing if
2865 * we are still holding the transaction as we can
2866 * release the last reference to io_end which may end
2867 * up doing unwritten extent conversion.
2870 ext4_put_io_end_defer(mpd.io_submit.io_end);
2871 ext4_journal_stop(handle);
2873 ext4_put_io_end(mpd.io_submit.io_end);
2874 mpd.io_submit.io_end = NULL;
2876 if (ret == -ENOSPC && sbi->s_journal) {
2878 * Commit the transaction which would
2879 * free blocks released in the transaction
2882 jbd2_journal_force_commit_nested(sbi->s_journal);
2886 /* Fatal error - ENOMEM, EIO... */
2891 blk_finish_plug(&plug);
2892 if (!ret && !cycled && wbc->nr_to_write > 0) {
2894 mpd.last_page = writeback_index - 1;
2900 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2902 * Set the writeback_index so that range_cyclic
2903 * mode will write it back later
2905 mapping->writeback_index = mpd.first_page;
2908 trace_ext4_writepages_result(inode, wbc, ret,
2909 nr_to_write - wbc->nr_to_write);
2910 percpu_up_read(&sbi->s_writepages_rwsem);
2914 static int ext4_dax_writepages(struct address_space *mapping,
2915 struct writeback_control *wbc)
2918 long nr_to_write = wbc->nr_to_write;
2919 struct inode *inode = mapping->host;
2920 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2922 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2925 percpu_down_read(&sbi->s_writepages_rwsem);
2926 trace_ext4_writepages(inode, wbc);
2928 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2929 trace_ext4_writepages_result(inode, wbc, ret,
2930 nr_to_write - wbc->nr_to_write);
2931 percpu_up_read(&sbi->s_writepages_rwsem);
2935 static int ext4_nonda_switch(struct super_block *sb)
2937 s64 free_clusters, dirty_clusters;
2938 struct ext4_sb_info *sbi = EXT4_SB(sb);
2941 * switch to non delalloc mode if we are running low
2942 * on free block. The free block accounting via percpu
2943 * counters can get slightly wrong with percpu_counter_batch getting
2944 * accumulated on each CPU without updating global counters
2945 * Delalloc need an accurate free block accounting. So switch
2946 * to non delalloc when we are near to error range.
2949 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2951 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2953 * Start pushing delalloc when 1/2 of free blocks are dirty.
2955 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2956 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2958 if (2 * free_clusters < 3 * dirty_clusters ||
2959 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2961 * free block count is less than 150% of dirty blocks
2962 * or free blocks is less than watermark
2969 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2970 loff_t pos, unsigned len,
2971 struct page **pagep, void **fsdata)
2973 int ret, retries = 0;
2976 struct inode *inode = mapping->host;
2978 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2981 index = pos >> PAGE_SHIFT;
2983 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2984 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2985 return ext4_write_begin(file, mapping, pos,
2986 len, pagep, fsdata);
2988 *fsdata = (void *)0;
2989 trace_ext4_da_write_begin(inode, pos, len);
2991 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2992 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
3001 page = grab_cache_page_write_begin(mapping, index);
3005 /* In case writeback began while the page was unlocked */
3006 wait_for_stable_page(page);
3008 #ifdef CONFIG_FS_ENCRYPTION
3009 ret = ext4_block_write_begin(page, pos, len,
3010 ext4_da_get_block_prep);
3012 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3018 * block_write_begin may have instantiated a few blocks
3019 * outside i_size. Trim these off again. Don't need
3020 * i_size_read because we hold inode lock.
3022 if (pos + len > inode->i_size)
3023 ext4_truncate_failed_write(inode);
3025 if (ret == -ENOSPC &&
3026 ext4_should_retry_alloc(inode->i_sb, &retries))
3036 * Check if we should update i_disksize
3037 * when write to the end of file but not require block allocation
3039 static int ext4_da_should_update_i_disksize(struct page *page,
3040 unsigned long offset)
3042 struct buffer_head *bh;
3043 struct inode *inode = page->mapping->host;
3047 bh = page_buffers(page);
3048 idx = offset >> inode->i_blkbits;
3050 for (i = 0; i < idx; i++)
3051 bh = bh->b_this_page;
3053 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3058 static int ext4_da_write_end(struct file *file,
3059 struct address_space *mapping,
3060 loff_t pos, unsigned len, unsigned copied,
3061 struct page *page, void *fsdata)
3063 struct inode *inode = mapping->host;
3065 unsigned long start, end;
3066 int write_mode = (int)(unsigned long)fsdata;
3068 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3069 return ext4_write_end(file, mapping, pos,
3070 len, copied, page, fsdata);
3072 trace_ext4_da_write_end(inode, pos, len, copied);
3074 if (write_mode != CONVERT_INLINE_DATA &&
3075 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3076 ext4_has_inline_data(inode))
3077 return ext4_write_inline_data_end(inode, pos, len, copied, page);
3079 start = pos & (PAGE_SIZE - 1);
3080 end = start + copied - 1;
3083 * Since we are holding inode lock, we are sure i_disksize <=
3084 * i_size. We also know that if i_disksize < i_size, there are
3085 * delalloc writes pending in the range upto i_size. If the end of
3086 * the current write is <= i_size, there's no need to touch
3087 * i_disksize since writeback will push i_disksize upto i_size
3088 * eventually. If the end of the current write is > i_size and
3089 * inside an allocated block (ext4_da_should_update_i_disksize()
3090 * check), we need to update i_disksize here as neither
3091 * ext4_writepage() nor certain ext4_writepages() paths not
3092 * allocating blocks update i_disksize.
3094 * Note that we defer inode dirtying to generic_write_end() /
3095 * ext4_da_write_inline_data_end().
3097 new_i_size = pos + copied;
3098 if (copied && new_i_size > inode->i_size &&
3099 ext4_da_should_update_i_disksize(page, end))
3100 ext4_update_i_disksize(inode, new_i_size);
3102 return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3106 * Force all delayed allocation blocks to be allocated for a given inode.
3108 int ext4_alloc_da_blocks(struct inode *inode)
3110 trace_ext4_alloc_da_blocks(inode);
3112 if (!EXT4_I(inode)->i_reserved_data_blocks)
3116 * We do something simple for now. The filemap_flush() will
3117 * also start triggering a write of the data blocks, which is
3118 * not strictly speaking necessary (and for users of
3119 * laptop_mode, not even desirable). However, to do otherwise
3120 * would require replicating code paths in:
3122 * ext4_writepages() ->
3123 * write_cache_pages() ---> (via passed in callback function)
3124 * __mpage_da_writepage() -->
3125 * mpage_add_bh_to_extent()
3126 * mpage_da_map_blocks()
3128 * The problem is that write_cache_pages(), located in
3129 * mm/page-writeback.c, marks pages clean in preparation for
3130 * doing I/O, which is not desirable if we're not planning on
3133 * We could call write_cache_pages(), and then redirty all of
3134 * the pages by calling redirty_page_for_writepage() but that
3135 * would be ugly in the extreme. So instead we would need to
3136 * replicate parts of the code in the above functions,
3137 * simplifying them because we wouldn't actually intend to
3138 * write out the pages, but rather only collect contiguous
3139 * logical block extents, call the multi-block allocator, and
3140 * then update the buffer heads with the block allocations.
3142 * For now, though, we'll cheat by calling filemap_flush(),
3143 * which will map the blocks, and start the I/O, but not
3144 * actually wait for the I/O to complete.
3146 return filemap_flush(inode->i_mapping);
3150 * bmap() is special. It gets used by applications such as lilo and by
3151 * the swapper to find the on-disk block of a specific piece of data.
3153 * Naturally, this is dangerous if the block concerned is still in the
3154 * journal. If somebody makes a swapfile on an ext4 data-journaling
3155 * filesystem and enables swap, then they may get a nasty shock when the
3156 * data getting swapped to that swapfile suddenly gets overwritten by
3157 * the original zero's written out previously to the journal and
3158 * awaiting writeback in the kernel's buffer cache.
3160 * So, if we see any bmap calls here on a modified, data-journaled file,
3161 * take extra steps to flush any blocks which might be in the cache.
3163 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3165 struct inode *inode = mapping->host;
3170 inode_lock_shared(inode);
3172 * We can get here for an inline file via the FIBMAP ioctl
3174 if (ext4_has_inline_data(inode))
3177 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3178 test_opt(inode->i_sb, DELALLOC)) {
3180 * With delalloc we want to sync the file
3181 * so that we can make sure we allocate
3184 filemap_write_and_wait(mapping);
3187 if (EXT4_JOURNAL(inode) &&
3188 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3190 * This is a REALLY heavyweight approach, but the use of
3191 * bmap on dirty files is expected to be extremely rare:
3192 * only if we run lilo or swapon on a freshly made file
3193 * do we expect this to happen.
3195 * (bmap requires CAP_SYS_RAWIO so this does not
3196 * represent an unprivileged user DOS attack --- we'd be
3197 * in trouble if mortal users could trigger this path at
3200 * NB. EXT4_STATE_JDATA is not set on files other than
3201 * regular files. If somebody wants to bmap a directory
3202 * or symlink and gets confused because the buffer
3203 * hasn't yet been flushed to disk, they deserve
3204 * everything they get.
3207 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3208 journal = EXT4_JOURNAL(inode);
3209 jbd2_journal_lock_updates(journal);
3210 err = jbd2_journal_flush(journal, 0);
3211 jbd2_journal_unlock_updates(journal);
3217 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3220 inode_unlock_shared(inode);
3224 static int ext4_read_folio(struct file *file, struct folio *folio)
3226 struct page *page = &folio->page;
3228 struct inode *inode = page->mapping->host;
3230 trace_ext4_readpage(page);
3232 if (ext4_has_inline_data(inode))
3233 ret = ext4_readpage_inline(inode, page);
3236 return ext4_mpage_readpages(inode, NULL, page);
3241 static void ext4_readahead(struct readahead_control *rac)
3243 struct inode *inode = rac->mapping->host;
3245 /* If the file has inline data, no need to do readahead. */
3246 if (ext4_has_inline_data(inode))
3249 ext4_mpage_readpages(inode, rac, NULL);
3252 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3255 trace_ext4_invalidate_folio(folio, offset, length);
3257 /* No journalling happens on data buffers when this function is used */
3258 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3260 block_invalidate_folio(folio, offset, length);
3263 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3264 size_t offset, size_t length)
3266 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3268 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3271 * If it's a full truncate we just forget about the pending dirtying
3273 if (offset == 0 && length == folio_size(folio))
3274 folio_clear_checked(folio);
3276 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3279 /* Wrapper for aops... */
3280 static void ext4_journalled_invalidate_folio(struct folio *folio,
3284 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3287 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3289 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3291 trace_ext4_releasepage(&folio->page);
3293 /* Page has dirty journalled data -> cannot release */
3294 if (folio_test_checked(folio))
3297 return jbd2_journal_try_to_free_buffers(journal, folio);
3299 return try_to_free_buffers(folio);
3302 static bool ext4_inode_datasync_dirty(struct inode *inode)
3304 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3307 if (jbd2_transaction_committed(journal,
3308 EXT4_I(inode)->i_datasync_tid))
3310 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3311 return !list_empty(&EXT4_I(inode)->i_fc_list);
3315 /* Any metadata buffers to write? */
3316 if (!list_empty(&inode->i_mapping->private_list))
3318 return inode->i_state & I_DIRTY_DATASYNC;
3321 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3322 struct ext4_map_blocks *map, loff_t offset,
3323 loff_t length, unsigned int flags)
3325 u8 blkbits = inode->i_blkbits;
3328 * Writes that span EOF might trigger an I/O size update on completion,
3329 * so consider them to be dirty for the purpose of O_DSYNC, even if
3330 * there is no other metadata changes being made or are pending.
3333 if (ext4_inode_datasync_dirty(inode) ||
3334 offset + length > i_size_read(inode))
3335 iomap->flags |= IOMAP_F_DIRTY;
3337 if (map->m_flags & EXT4_MAP_NEW)
3338 iomap->flags |= IOMAP_F_NEW;
3340 if (flags & IOMAP_DAX)
3341 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3343 iomap->bdev = inode->i_sb->s_bdev;
3344 iomap->offset = (u64) map->m_lblk << blkbits;
3345 iomap->length = (u64) map->m_len << blkbits;
3347 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3348 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3349 iomap->flags |= IOMAP_F_MERGED;
3352 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3353 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3354 * set. In order for any allocated unwritten extents to be converted
3355 * into written extents correctly within the ->end_io() handler, we
3356 * need to ensure that the iomap->type is set appropriately. Hence, the
3357 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3360 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3361 iomap->type = IOMAP_UNWRITTEN;
3362 iomap->addr = (u64) map->m_pblk << blkbits;
3363 if (flags & IOMAP_DAX)
3364 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3365 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3366 iomap->type = IOMAP_MAPPED;
3367 iomap->addr = (u64) map->m_pblk << blkbits;
3368 if (flags & IOMAP_DAX)
3369 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3371 iomap->type = IOMAP_HOLE;
3372 iomap->addr = IOMAP_NULL_ADDR;
3376 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3380 u8 blkbits = inode->i_blkbits;
3381 int ret, dio_credits, m_flags = 0, retries = 0;
3384 * Trim the mapping request to the maximum value that we can map at
3385 * once for direct I/O.
3387 if (map->m_len > DIO_MAX_BLOCKS)
3388 map->m_len = DIO_MAX_BLOCKS;
3389 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3393 * Either we allocate blocks and then don't get an unwritten extent, so
3394 * in that case we have reserved enough credits. Or, the blocks are
3395 * already allocated and unwritten. In that case, the extent conversion
3396 * fits into the credits as well.
3398 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3400 return PTR_ERR(handle);
3403 * DAX and direct I/O are the only two operations that are currently
3404 * supported with IOMAP_WRITE.
3406 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3407 if (flags & IOMAP_DAX)
3408 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3410 * We use i_size instead of i_disksize here because delalloc writeback
3411 * can complete at any point during the I/O and subsequently push the
3412 * i_disksize out to i_size. This could be beyond where direct I/O is
3413 * happening and thus expose allocated blocks to direct I/O reads.
3415 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3416 m_flags = EXT4_GET_BLOCKS_CREATE;
3417 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3418 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3420 ret = ext4_map_blocks(handle, inode, map, m_flags);
3423 * We cannot fill holes in indirect tree based inodes as that could
3424 * expose stale data in the case of a crash. Use the magic error code
3425 * to fallback to buffered I/O.
3427 if (!m_flags && !ret)
3430 ext4_journal_stop(handle);
3431 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3438 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3439 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3442 struct ext4_map_blocks map;
3443 u8 blkbits = inode->i_blkbits;
3445 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3448 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3452 * Calculate the first and last logical blocks respectively.
3454 map.m_lblk = offset >> blkbits;
3455 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3456 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3458 if (flags & IOMAP_WRITE) {
3460 * We check here if the blocks are already allocated, then we
3461 * don't need to start a journal txn and we can directly return
3462 * the mapping information. This could boost performance
3463 * especially in multi-threaded overwrite requests.
3465 if (offset + length <= i_size_read(inode)) {
3466 ret = ext4_map_blocks(NULL, inode, &map, 0);
3467 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3470 ret = ext4_iomap_alloc(inode, &map, flags);
3472 ret = ext4_map_blocks(NULL, inode, &map, 0);
3479 * When inline encryption is enabled, sometimes I/O to an encrypted file
3480 * has to be broken up to guarantee DUN contiguity. Handle this by
3481 * limiting the length of the mapping returned.
3483 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3485 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3490 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3491 loff_t length, unsigned flags, struct iomap *iomap,
3492 struct iomap *srcmap)
3497 * Even for writes we don't need to allocate blocks, so just pretend
3498 * we are reading to save overhead of starting a transaction.
3500 flags &= ~IOMAP_WRITE;
3501 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3502 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3506 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3507 ssize_t written, unsigned flags, struct iomap *iomap)
3510 * Check to see whether an error occurred while writing out the data to
3511 * the allocated blocks. If so, return the magic error code so that we
3512 * fallback to buffered I/O and attempt to complete the remainder of
3513 * the I/O. Any blocks that may have been allocated in preparation for
3514 * the direct I/O will be reused during buffered I/O.
3516 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3522 const struct iomap_ops ext4_iomap_ops = {
3523 .iomap_begin = ext4_iomap_begin,
3524 .iomap_end = ext4_iomap_end,
3527 const struct iomap_ops ext4_iomap_overwrite_ops = {
3528 .iomap_begin = ext4_iomap_overwrite_begin,
3529 .iomap_end = ext4_iomap_end,
3532 static bool ext4_iomap_is_delalloc(struct inode *inode,
3533 struct ext4_map_blocks *map)
3535 struct extent_status es;
3536 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3538 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3539 map->m_lblk, end, &es);
3541 if (!es.es_len || es.es_lblk > end)
3544 if (es.es_lblk > map->m_lblk) {
3545 map->m_len = es.es_lblk - map->m_lblk;
3549 offset = map->m_lblk - es.es_lblk;
3550 map->m_len = es.es_len - offset;
3555 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3556 loff_t length, unsigned int flags,
3557 struct iomap *iomap, struct iomap *srcmap)
3560 bool delalloc = false;
3561 struct ext4_map_blocks map;
3562 u8 blkbits = inode->i_blkbits;
3564 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3567 if (ext4_has_inline_data(inode)) {
3568 ret = ext4_inline_data_iomap(inode, iomap);
3569 if (ret != -EAGAIN) {
3570 if (ret == 0 && offset >= iomap->length)
3577 * Calculate the first and last logical block respectively.
3579 map.m_lblk = offset >> blkbits;
3580 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3581 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3584 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3585 * So handle it here itself instead of querying ext4_map_blocks().
3586 * Since ext4_map_blocks() will warn about it and will return
3589 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3590 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3592 if (offset >= sbi->s_bitmap_maxbytes) {
3598 ret = ext4_map_blocks(NULL, inode, &map, 0);
3602 delalloc = ext4_iomap_is_delalloc(inode, &map);
3605 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3606 if (delalloc && iomap->type == IOMAP_HOLE)
3607 iomap->type = IOMAP_DELALLOC;
3612 const struct iomap_ops ext4_iomap_report_ops = {
3613 .iomap_begin = ext4_iomap_begin_report,
3617 * Whenever the folio is being dirtied, corresponding buffers should already
3618 * be attached to the transaction (we take care of this in ext4_page_mkwrite()
3619 * and ext4_write_begin()). However we cannot move buffers to dirty transaction
3620 * lists here because ->dirty_folio is called under VFS locks and the folio
3621 * is not necessarily locked.
3623 * We cannot just dirty the folio and leave attached buffers clean, because the
3624 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3625 * or jbddirty because all the journalling code will explode.
3627 * So what we do is to mark the folio "pending dirty" and next time writepage
3628 * is called, propagate that into the buffers appropriately.
3630 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3631 struct folio *folio)
3633 WARN_ON_ONCE(!folio_buffers(folio));
3634 folio_set_checked(folio);
3635 return filemap_dirty_folio(mapping, folio);
3638 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3640 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3641 WARN_ON_ONCE(!folio_buffers(folio));
3642 return block_dirty_folio(mapping, folio);
3645 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3646 struct file *file, sector_t *span)
3648 return iomap_swapfile_activate(sis, file, span,
3649 &ext4_iomap_report_ops);
3652 static const struct address_space_operations ext4_aops = {
3653 .read_folio = ext4_read_folio,
3654 .readahead = ext4_readahead,
3655 .writepage = ext4_writepage,
3656 .writepages = ext4_writepages,
3657 .write_begin = ext4_write_begin,
3658 .write_end = ext4_write_end,
3659 .dirty_folio = ext4_dirty_folio,
3661 .invalidate_folio = ext4_invalidate_folio,
3662 .release_folio = ext4_release_folio,
3663 .direct_IO = noop_direct_IO,
3664 .migrate_folio = buffer_migrate_folio,
3665 .is_partially_uptodate = block_is_partially_uptodate,
3666 .error_remove_page = generic_error_remove_page,
3667 .swap_activate = ext4_iomap_swap_activate,
3670 static const struct address_space_operations ext4_journalled_aops = {
3671 .read_folio = ext4_read_folio,
3672 .readahead = ext4_readahead,
3673 .writepage = ext4_writepage,
3674 .writepages = ext4_writepages,
3675 .write_begin = ext4_write_begin,
3676 .write_end = ext4_journalled_write_end,
3677 .dirty_folio = ext4_journalled_dirty_folio,
3679 .invalidate_folio = ext4_journalled_invalidate_folio,
3680 .release_folio = ext4_release_folio,
3681 .direct_IO = noop_direct_IO,
3682 .is_partially_uptodate = block_is_partially_uptodate,
3683 .error_remove_page = generic_error_remove_page,
3684 .swap_activate = ext4_iomap_swap_activate,
3687 static const struct address_space_operations ext4_da_aops = {
3688 .read_folio = ext4_read_folio,
3689 .readahead = ext4_readahead,
3690 .writepage = ext4_writepage,
3691 .writepages = ext4_writepages,
3692 .write_begin = ext4_da_write_begin,
3693 .write_end = ext4_da_write_end,
3694 .dirty_folio = ext4_dirty_folio,
3696 .invalidate_folio = ext4_invalidate_folio,
3697 .release_folio = ext4_release_folio,
3698 .direct_IO = noop_direct_IO,
3699 .migrate_folio = buffer_migrate_folio,
3700 .is_partially_uptodate = block_is_partially_uptodate,
3701 .error_remove_page = generic_error_remove_page,
3702 .swap_activate = ext4_iomap_swap_activate,
3705 static const struct address_space_operations ext4_dax_aops = {
3706 .writepages = ext4_dax_writepages,
3707 .direct_IO = noop_direct_IO,
3708 .dirty_folio = noop_dirty_folio,
3710 .swap_activate = ext4_iomap_swap_activate,
3713 void ext4_set_aops(struct inode *inode)
3715 switch (ext4_inode_journal_mode(inode)) {
3716 case EXT4_INODE_ORDERED_DATA_MODE:
3717 case EXT4_INODE_WRITEBACK_DATA_MODE:
3719 case EXT4_INODE_JOURNAL_DATA_MODE:
3720 inode->i_mapping->a_ops = &ext4_journalled_aops;
3726 inode->i_mapping->a_ops = &ext4_dax_aops;
3727 else if (test_opt(inode->i_sb, DELALLOC))
3728 inode->i_mapping->a_ops = &ext4_da_aops;
3730 inode->i_mapping->a_ops = &ext4_aops;
3733 static int __ext4_block_zero_page_range(handle_t *handle,
3734 struct address_space *mapping, loff_t from, loff_t length)
3736 ext4_fsblk_t index = from >> PAGE_SHIFT;
3737 unsigned offset = from & (PAGE_SIZE-1);
3738 unsigned blocksize, pos;
3740 struct inode *inode = mapping->host;
3741 struct buffer_head *bh;
3745 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3746 mapping_gfp_constraint(mapping, ~__GFP_FS));
3750 blocksize = inode->i_sb->s_blocksize;
3752 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3754 if (!page_has_buffers(page))
3755 create_empty_buffers(page, blocksize, 0);
3757 /* Find the buffer that contains "offset" */
3758 bh = page_buffers(page);
3760 while (offset >= pos) {
3761 bh = bh->b_this_page;
3765 if (buffer_freed(bh)) {
3766 BUFFER_TRACE(bh, "freed: skip");
3769 if (!buffer_mapped(bh)) {
3770 BUFFER_TRACE(bh, "unmapped");
3771 ext4_get_block(inode, iblock, bh, 0);
3772 /* unmapped? It's a hole - nothing to do */
3773 if (!buffer_mapped(bh)) {
3774 BUFFER_TRACE(bh, "still unmapped");
3779 /* Ok, it's mapped. Make sure it's up-to-date */
3780 if (PageUptodate(page))
3781 set_buffer_uptodate(bh);
3783 if (!buffer_uptodate(bh)) {
3784 err = ext4_read_bh_lock(bh, 0, true);
3787 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3788 /* We expect the key to be set. */
3789 BUG_ON(!fscrypt_has_encryption_key(inode));
3790 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3793 clear_buffer_uptodate(bh);
3798 if (ext4_should_journal_data(inode)) {
3799 BUFFER_TRACE(bh, "get write access");
3800 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3805 zero_user(page, offset, length);
3806 BUFFER_TRACE(bh, "zeroed end of block");
3808 if (ext4_should_journal_data(inode)) {
3809 err = ext4_handle_dirty_metadata(handle, inode, bh);
3812 mark_buffer_dirty(bh);
3813 if (ext4_should_order_data(inode))
3814 err = ext4_jbd2_inode_add_write(handle, inode, from,
3825 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3826 * starting from file offset 'from'. The range to be zero'd must
3827 * be contained with in one block. If the specified range exceeds
3828 * the end of the block it will be shortened to end of the block
3829 * that corresponds to 'from'
3831 static int ext4_block_zero_page_range(handle_t *handle,
3832 struct address_space *mapping, loff_t from, loff_t length)
3834 struct inode *inode = mapping->host;
3835 unsigned offset = from & (PAGE_SIZE-1);
3836 unsigned blocksize = inode->i_sb->s_blocksize;
3837 unsigned max = blocksize - (offset & (blocksize - 1));
3840 * correct length if it does not fall between
3841 * 'from' and the end of the block
3843 if (length > max || length < 0)
3846 if (IS_DAX(inode)) {
3847 return dax_zero_range(inode, from, length, NULL,
3850 return __ext4_block_zero_page_range(handle, mapping, from, length);
3854 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3855 * up to the end of the block which corresponds to `from'.
3856 * This required during truncate. We need to physically zero the tail end
3857 * of that block so it doesn't yield old data if the file is later grown.
3859 static int ext4_block_truncate_page(handle_t *handle,
3860 struct address_space *mapping, loff_t from)
3862 unsigned offset = from & (PAGE_SIZE-1);
3865 struct inode *inode = mapping->host;
3867 /* If we are processing an encrypted inode during orphan list handling */
3868 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3871 blocksize = inode->i_sb->s_blocksize;
3872 length = blocksize - (offset & (blocksize - 1));
3874 return ext4_block_zero_page_range(handle, mapping, from, length);
3877 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3878 loff_t lstart, loff_t length)
3880 struct super_block *sb = inode->i_sb;
3881 struct address_space *mapping = inode->i_mapping;
3882 unsigned partial_start, partial_end;
3883 ext4_fsblk_t start, end;
3884 loff_t byte_end = (lstart + length - 1);
3887 partial_start = lstart & (sb->s_blocksize - 1);
3888 partial_end = byte_end & (sb->s_blocksize - 1);
3890 start = lstart >> sb->s_blocksize_bits;
3891 end = byte_end >> sb->s_blocksize_bits;
3893 /* Handle partial zero within the single block */
3895 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3896 err = ext4_block_zero_page_range(handle, mapping,
3900 /* Handle partial zero out on the start of the range */
3901 if (partial_start) {
3902 err = ext4_block_zero_page_range(handle, mapping,
3903 lstart, sb->s_blocksize);
3907 /* Handle partial zero out on the end of the range */
3908 if (partial_end != sb->s_blocksize - 1)
3909 err = ext4_block_zero_page_range(handle, mapping,
3910 byte_end - partial_end,
3915 int ext4_can_truncate(struct inode *inode)
3917 if (S_ISREG(inode->i_mode))
3919 if (S_ISDIR(inode->i_mode))
3921 if (S_ISLNK(inode->i_mode))
3922 return !ext4_inode_is_fast_symlink(inode);
3927 * We have to make sure i_disksize gets properly updated before we truncate
3928 * page cache due to hole punching or zero range. Otherwise i_disksize update
3929 * can get lost as it may have been postponed to submission of writeback but
3930 * that will never happen after we truncate page cache.
3932 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3938 loff_t size = i_size_read(inode);
3940 WARN_ON(!inode_is_locked(inode));
3941 if (offset > size || offset + len < size)
3944 if (EXT4_I(inode)->i_disksize >= size)
3947 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3949 return PTR_ERR(handle);
3950 ext4_update_i_disksize(inode, size);
3951 ret = ext4_mark_inode_dirty(handle, inode);
3952 ext4_journal_stop(handle);
3957 static void ext4_wait_dax_page(struct inode *inode)
3959 filemap_invalidate_unlock(inode->i_mapping);
3961 filemap_invalidate_lock(inode->i_mapping);
3964 int ext4_break_layouts(struct inode *inode)
3969 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3973 page = dax_layout_busy_page(inode->i_mapping);
3977 error = ___wait_var_event(&page->_refcount,
3978 atomic_read(&page->_refcount) == 1,
3979 TASK_INTERRUPTIBLE, 0, 0,
3980 ext4_wait_dax_page(inode));
3981 } while (error == 0);
3987 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3988 * associated with the given offset and length
3990 * @inode: File inode
3991 * @offset: The offset where the hole will begin
3992 * @len: The length of the hole
3994 * Returns: 0 on success or negative on failure
3997 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3999 struct inode *inode = file_inode(file);
4000 struct super_block *sb = inode->i_sb;
4001 ext4_lblk_t first_block, stop_block;
4002 struct address_space *mapping = inode->i_mapping;
4003 loff_t first_block_offset, last_block_offset, max_length;
4004 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4006 unsigned int credits;
4007 int ret = 0, ret2 = 0;
4009 trace_ext4_punch_hole(inode, offset, length, 0);
4012 * Write out all dirty pages to avoid race conditions
4013 * Then release them.
4015 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4016 ret = filemap_write_and_wait_range(mapping, offset,
4017 offset + length - 1);
4024 /* No need to punch hole beyond i_size */
4025 if (offset >= inode->i_size)
4029 * If the hole extends beyond i_size, set the hole
4030 * to end after the page that contains i_size
4032 if (offset + length > inode->i_size) {
4033 length = inode->i_size +
4034 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4039 * For punch hole the length + offset needs to be within one block
4040 * before last range. Adjust the length if it goes beyond that limit.
4042 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4043 if (offset + length > max_length)
4044 length = max_length - offset;
4046 if (offset & (sb->s_blocksize - 1) ||
4047 (offset + length) & (sb->s_blocksize - 1)) {
4049 * Attach jinode to inode for jbd2 if we do any zeroing of
4052 ret = ext4_inode_attach_jinode(inode);
4058 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4059 inode_dio_wait(inode);
4061 ret = file_modified(file);
4066 * Prevent page faults from reinstantiating pages we have released from
4069 filemap_invalidate_lock(mapping);
4071 ret = ext4_break_layouts(inode);
4075 first_block_offset = round_up(offset, sb->s_blocksize);
4076 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4078 /* Now release the pages and zero block aligned part of pages*/
4079 if (last_block_offset > first_block_offset) {
4080 ret = ext4_update_disksize_before_punch(inode, offset, length);
4083 truncate_pagecache_range(inode, first_block_offset,
4087 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4088 credits = ext4_writepage_trans_blocks(inode);
4090 credits = ext4_blocks_for_truncate(inode);
4091 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4092 if (IS_ERR(handle)) {
4093 ret = PTR_ERR(handle);
4094 ext4_std_error(sb, ret);
4098 ret = ext4_zero_partial_blocks(handle, inode, offset,
4103 first_block = (offset + sb->s_blocksize - 1) >>
4104 EXT4_BLOCK_SIZE_BITS(sb);
4105 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4107 /* If there are blocks to remove, do it */
4108 if (stop_block > first_block) {
4110 down_write(&EXT4_I(inode)->i_data_sem);
4111 ext4_discard_preallocations(inode, 0);
4113 ret = ext4_es_remove_extent(inode, first_block,
4114 stop_block - first_block);
4116 up_write(&EXT4_I(inode)->i_data_sem);
4120 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4121 ret = ext4_ext_remove_space(inode, first_block,
4124 ret = ext4_ind_remove_space(handle, inode, first_block,
4127 up_write(&EXT4_I(inode)->i_data_sem);
4129 ext4_fc_track_range(handle, inode, first_block, stop_block);
4131 ext4_handle_sync(handle);
4133 inode->i_mtime = inode->i_ctime = current_time(inode);
4134 ret2 = ext4_mark_inode_dirty(handle, inode);
4138 ext4_update_inode_fsync_trans(handle, inode, 1);
4140 ext4_journal_stop(handle);
4142 filemap_invalidate_unlock(mapping);
4144 inode_unlock(inode);
4148 int ext4_inode_attach_jinode(struct inode *inode)
4150 struct ext4_inode_info *ei = EXT4_I(inode);
4151 struct jbd2_inode *jinode;
4153 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4156 jinode = jbd2_alloc_inode(GFP_KERNEL);
4157 spin_lock(&inode->i_lock);
4160 spin_unlock(&inode->i_lock);
4163 ei->jinode = jinode;
4164 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4167 spin_unlock(&inode->i_lock);
4168 if (unlikely(jinode != NULL))
4169 jbd2_free_inode(jinode);
4176 * We block out ext4_get_block() block instantiations across the entire
4177 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4178 * simultaneously on behalf of the same inode.
4180 * As we work through the truncate and commit bits of it to the journal there
4181 * is one core, guiding principle: the file's tree must always be consistent on
4182 * disk. We must be able to restart the truncate after a crash.
4184 * The file's tree may be transiently inconsistent in memory (although it
4185 * probably isn't), but whenever we close off and commit a journal transaction,
4186 * the contents of (the filesystem + the journal) must be consistent and
4187 * restartable. It's pretty simple, really: bottom up, right to left (although
4188 * left-to-right works OK too).
4190 * Note that at recovery time, journal replay occurs *before* the restart of
4191 * truncate against the orphan inode list.
4193 * The committed inode has the new, desired i_size (which is the same as
4194 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4195 * that this inode's truncate did not complete and it will again call
4196 * ext4_truncate() to have another go. So there will be instantiated blocks
4197 * to the right of the truncation point in a crashed ext4 filesystem. But
4198 * that's fine - as long as they are linked from the inode, the post-crash
4199 * ext4_truncate() run will find them and release them.
4201 int ext4_truncate(struct inode *inode)
4203 struct ext4_inode_info *ei = EXT4_I(inode);
4204 unsigned int credits;
4207 struct address_space *mapping = inode->i_mapping;
4210 * There is a possibility that we're either freeing the inode
4211 * or it's a completely new inode. In those cases we might not
4212 * have i_rwsem locked because it's not necessary.
4214 if (!(inode->i_state & (I_NEW|I_FREEING)))
4215 WARN_ON(!inode_is_locked(inode));
4216 trace_ext4_truncate_enter(inode);
4218 if (!ext4_can_truncate(inode))
4221 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4222 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4224 if (ext4_has_inline_data(inode)) {
4227 err = ext4_inline_data_truncate(inode, &has_inline);
4228 if (err || has_inline)
4232 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4233 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4234 err = ext4_inode_attach_jinode(inode);
4239 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4240 credits = ext4_writepage_trans_blocks(inode);
4242 credits = ext4_blocks_for_truncate(inode);
4244 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4245 if (IS_ERR(handle)) {
4246 err = PTR_ERR(handle);
4250 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4251 ext4_block_truncate_page(handle, mapping, inode->i_size);
4254 * We add the inode to the orphan list, so that if this
4255 * truncate spans multiple transactions, and we crash, we will
4256 * resume the truncate when the filesystem recovers. It also
4257 * marks the inode dirty, to catch the new size.
4259 * Implication: the file must always be in a sane, consistent
4260 * truncatable state while each transaction commits.
4262 err = ext4_orphan_add(handle, inode);
4266 down_write(&EXT4_I(inode)->i_data_sem);
4268 ext4_discard_preallocations(inode, 0);
4270 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4271 err = ext4_ext_truncate(handle, inode);
4273 ext4_ind_truncate(handle, inode);
4275 up_write(&ei->i_data_sem);
4280 ext4_handle_sync(handle);
4284 * If this was a simple ftruncate() and the file will remain alive,
4285 * then we need to clear up the orphan record which we created above.
4286 * However, if this was a real unlink then we were called by
4287 * ext4_evict_inode(), and we allow that function to clean up the
4288 * orphan info for us.
4291 ext4_orphan_del(handle, inode);
4293 inode->i_mtime = inode->i_ctime = current_time(inode);
4294 err2 = ext4_mark_inode_dirty(handle, inode);
4295 if (unlikely(err2 && !err))
4297 ext4_journal_stop(handle);
4300 trace_ext4_truncate_exit(inode);
4304 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4306 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4307 return inode_peek_iversion_raw(inode);
4309 return inode_peek_iversion(inode);
4312 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4313 struct ext4_inode_info *ei)
4315 struct inode *inode = &(ei->vfs_inode);
4316 u64 i_blocks = READ_ONCE(inode->i_blocks);
4317 struct super_block *sb = inode->i_sb;
4319 if (i_blocks <= ~0U) {
4321 * i_blocks can be represented in a 32 bit variable
4322 * as multiple of 512 bytes
4324 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4325 raw_inode->i_blocks_high = 0;
4326 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4331 * This should never happen since sb->s_maxbytes should not have
4332 * allowed this, sb->s_maxbytes was set according to the huge_file
4333 * feature in ext4_fill_super().
4335 if (!ext4_has_feature_huge_file(sb))
4336 return -EFSCORRUPTED;
4338 if (i_blocks <= 0xffffffffffffULL) {
4340 * i_blocks can be represented in a 48 bit variable
4341 * as multiple of 512 bytes
4343 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4344 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4345 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4347 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4348 /* i_block is stored in file system block size */
4349 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4350 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4351 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4356 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4358 struct ext4_inode_info *ei = EXT4_I(inode);
4365 err = ext4_inode_blocks_set(raw_inode, ei);
4367 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4368 i_uid = i_uid_read(inode);
4369 i_gid = i_gid_read(inode);
4370 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4371 if (!(test_opt(inode->i_sb, NO_UID32))) {
4372 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4373 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4375 * Fix up interoperability with old kernels. Otherwise,
4376 * old inodes get re-used with the upper 16 bits of the
4379 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4380 raw_inode->i_uid_high = 0;
4381 raw_inode->i_gid_high = 0;
4383 raw_inode->i_uid_high =
4384 cpu_to_le16(high_16_bits(i_uid));
4385 raw_inode->i_gid_high =
4386 cpu_to_le16(high_16_bits(i_gid));
4389 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4390 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4391 raw_inode->i_uid_high = 0;
4392 raw_inode->i_gid_high = 0;
4394 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4396 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4397 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4398 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4399 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4401 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4402 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4403 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4404 raw_inode->i_file_acl_high =
4405 cpu_to_le16(ei->i_file_acl >> 32);
4406 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4407 ext4_isize_set(raw_inode, ei->i_disksize);
4409 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4410 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4411 if (old_valid_dev(inode->i_rdev)) {
4412 raw_inode->i_block[0] =
4413 cpu_to_le32(old_encode_dev(inode->i_rdev));
4414 raw_inode->i_block[1] = 0;
4416 raw_inode->i_block[0] = 0;
4417 raw_inode->i_block[1] =
4418 cpu_to_le32(new_encode_dev(inode->i_rdev));
4419 raw_inode->i_block[2] = 0;
4421 } else if (!ext4_has_inline_data(inode)) {
4422 for (block = 0; block < EXT4_N_BLOCKS; block++)
4423 raw_inode->i_block[block] = ei->i_data[block];
4426 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4427 u64 ivers = ext4_inode_peek_iversion(inode);
4429 raw_inode->i_disk_version = cpu_to_le32(ivers);
4430 if (ei->i_extra_isize) {
4431 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4432 raw_inode->i_version_hi =
4433 cpu_to_le32(ivers >> 32);
4434 raw_inode->i_extra_isize =
4435 cpu_to_le16(ei->i_extra_isize);
4439 if (i_projid != EXT4_DEF_PROJID &&
4440 !ext4_has_feature_project(inode->i_sb))
4441 err = err ?: -EFSCORRUPTED;
4443 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4444 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4445 raw_inode->i_projid = cpu_to_le32(i_projid);
4447 ext4_inode_csum_set(inode, raw_inode, ei);
4452 * ext4_get_inode_loc returns with an extra refcount against the inode's
4453 * underlying buffer_head on success. If we pass 'inode' and it does not
4454 * have in-inode xattr, we have all inode data in memory that is needed
4455 * to recreate the on-disk version of this inode.
4457 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4458 struct inode *inode, struct ext4_iloc *iloc,
4459 ext4_fsblk_t *ret_block)
4461 struct ext4_group_desc *gdp;
4462 struct buffer_head *bh;
4464 struct blk_plug plug;
4465 int inodes_per_block, inode_offset;
4468 if (ino < EXT4_ROOT_INO ||
4469 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4470 return -EFSCORRUPTED;
4472 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4473 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4478 * Figure out the offset within the block group inode table
4480 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4481 inode_offset = ((ino - 1) %
4482 EXT4_INODES_PER_GROUP(sb));
4483 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4485 block = ext4_inode_table(sb, gdp);
4486 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4487 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4488 ext4_error(sb, "Invalid inode table block %llu in "
4489 "block_group %u", block, iloc->block_group);
4490 return -EFSCORRUPTED;
4492 block += (inode_offset / inodes_per_block);
4494 bh = sb_getblk(sb, block);
4497 if (ext4_buffer_uptodate(bh))
4501 if (ext4_buffer_uptodate(bh)) {
4502 /* Someone brought it uptodate while we waited */
4508 * If we have all information of the inode in memory and this
4509 * is the only valid inode in the block, we need not read the
4512 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4513 struct buffer_head *bitmap_bh;
4516 start = inode_offset & ~(inodes_per_block - 1);
4518 /* Is the inode bitmap in cache? */
4519 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4520 if (unlikely(!bitmap_bh))
4524 * If the inode bitmap isn't in cache then the
4525 * optimisation may end up performing two reads instead
4526 * of one, so skip it.
4528 if (!buffer_uptodate(bitmap_bh)) {
4532 for (i = start; i < start + inodes_per_block; i++) {
4533 if (i == inode_offset)
4535 if (ext4_test_bit(i, bitmap_bh->b_data))
4539 if (i == start + inodes_per_block) {
4540 struct ext4_inode *raw_inode =
4541 (struct ext4_inode *) (bh->b_data + iloc->offset);
4543 /* all other inodes are free, so skip I/O */
4544 memset(bh->b_data, 0, bh->b_size);
4545 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4546 ext4_fill_raw_inode(inode, raw_inode);
4547 set_buffer_uptodate(bh);
4555 * If we need to do any I/O, try to pre-readahead extra
4556 * blocks from the inode table.
4558 blk_start_plug(&plug);
4559 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4560 ext4_fsblk_t b, end, table;
4562 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4564 table = ext4_inode_table(sb, gdp);
4565 /* s_inode_readahead_blks is always a power of 2 */
4566 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4570 num = EXT4_INODES_PER_GROUP(sb);
4571 if (ext4_has_group_desc_csum(sb))
4572 num -= ext4_itable_unused_count(sb, gdp);
4573 table += num / inodes_per_block;
4577 ext4_sb_breadahead_unmovable(sb, b++);
4581 * There are other valid inodes in the buffer, this inode
4582 * has in-inode xattrs, or we don't have this inode in memory.
4583 * Read the block from disk.
4585 trace_ext4_load_inode(sb, ino);
4586 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4587 blk_finish_plug(&plug);
4589 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4590 if (!buffer_uptodate(bh)) {
4601 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4602 struct ext4_iloc *iloc)
4604 ext4_fsblk_t err_blk = 0;
4607 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4611 ext4_error_inode_block(inode, err_blk, EIO,
4612 "unable to read itable block");
4617 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4619 ext4_fsblk_t err_blk = 0;
4622 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4626 ext4_error_inode_block(inode, err_blk, EIO,
4627 "unable to read itable block");
4633 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4634 struct ext4_iloc *iloc)
4636 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4639 static bool ext4_should_enable_dax(struct inode *inode)
4641 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4643 if (test_opt2(inode->i_sb, DAX_NEVER))
4645 if (!S_ISREG(inode->i_mode))
4647 if (ext4_should_journal_data(inode))
4649 if (ext4_has_inline_data(inode))
4651 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4653 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4655 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4657 if (test_opt(inode->i_sb, DAX_ALWAYS))
4660 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4663 void ext4_set_inode_flags(struct inode *inode, bool init)
4665 unsigned int flags = EXT4_I(inode)->i_flags;
4666 unsigned int new_fl = 0;
4668 WARN_ON_ONCE(IS_DAX(inode) && init);
4670 if (flags & EXT4_SYNC_FL)
4672 if (flags & EXT4_APPEND_FL)
4674 if (flags & EXT4_IMMUTABLE_FL)
4675 new_fl |= S_IMMUTABLE;
4676 if (flags & EXT4_NOATIME_FL)
4677 new_fl |= S_NOATIME;
4678 if (flags & EXT4_DIRSYNC_FL)
4679 new_fl |= S_DIRSYNC;
4681 /* Because of the way inode_set_flags() works we must preserve S_DAX
4682 * here if already set. */
4683 new_fl |= (inode->i_flags & S_DAX);
4684 if (init && ext4_should_enable_dax(inode))
4687 if (flags & EXT4_ENCRYPT_FL)
4688 new_fl |= S_ENCRYPTED;
4689 if (flags & EXT4_CASEFOLD_FL)
4690 new_fl |= S_CASEFOLD;
4691 if (flags & EXT4_VERITY_FL)
4693 inode_set_flags(inode, new_fl,
4694 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4695 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4698 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4699 struct ext4_inode_info *ei)
4702 struct inode *inode = &(ei->vfs_inode);
4703 struct super_block *sb = inode->i_sb;
4705 if (ext4_has_feature_huge_file(sb)) {
4706 /* we are using combined 48 bit field */
4707 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4708 le32_to_cpu(raw_inode->i_blocks_lo);
4709 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4710 /* i_blocks represent file system block size */
4711 return i_blocks << (inode->i_blkbits - 9);
4716 return le32_to_cpu(raw_inode->i_blocks_lo);
4720 static inline int ext4_iget_extra_inode(struct inode *inode,
4721 struct ext4_inode *raw_inode,
4722 struct ext4_inode_info *ei)
4724 __le32 *magic = (void *)raw_inode +
4725 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4727 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4728 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4729 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4730 return ext4_find_inline_data_nolock(inode);
4732 EXT4_I(inode)->i_inline_off = 0;
4736 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4738 if (!ext4_has_feature_project(inode->i_sb))
4740 *projid = EXT4_I(inode)->i_projid;
4745 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4746 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4749 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4751 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4752 inode_set_iversion_raw(inode, val);
4754 inode_set_iversion_queried(inode, val);
4757 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4758 ext4_iget_flags flags, const char *function,
4761 struct ext4_iloc iloc;
4762 struct ext4_inode *raw_inode;
4763 struct ext4_inode_info *ei;
4764 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4765 struct inode *inode;
4766 journal_t *journal = EXT4_SB(sb)->s_journal;
4774 if ((!(flags & EXT4_IGET_SPECIAL) &&
4775 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4776 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4777 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4778 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4779 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4780 (ino < EXT4_ROOT_INO) ||
4781 (ino > le32_to_cpu(es->s_inodes_count))) {
4782 if (flags & EXT4_IGET_HANDLE)
4783 return ERR_PTR(-ESTALE);
4784 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4785 "inode #%lu: comm %s: iget: illegal inode #",
4786 ino, current->comm);
4787 return ERR_PTR(-EFSCORRUPTED);
4790 inode = iget_locked(sb, ino);
4792 return ERR_PTR(-ENOMEM);
4793 if (!(inode->i_state & I_NEW))
4799 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4802 raw_inode = ext4_raw_inode(&iloc);
4804 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4805 ext4_error_inode(inode, function, line, 0,
4806 "iget: root inode unallocated");
4807 ret = -EFSCORRUPTED;
4811 if ((flags & EXT4_IGET_HANDLE) &&
4812 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4817 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4818 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4819 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4820 EXT4_INODE_SIZE(inode->i_sb) ||
4821 (ei->i_extra_isize & 3)) {
4822 ext4_error_inode(inode, function, line, 0,
4823 "iget: bad extra_isize %u "
4826 EXT4_INODE_SIZE(inode->i_sb));
4827 ret = -EFSCORRUPTED;
4831 ei->i_extra_isize = 0;
4833 /* Precompute checksum seed for inode metadata */
4834 if (ext4_has_metadata_csum(sb)) {
4835 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4837 __le32 inum = cpu_to_le32(inode->i_ino);
4838 __le32 gen = raw_inode->i_generation;
4839 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4841 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4845 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4846 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4847 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4848 ext4_error_inode_err(inode, function, line, 0,
4849 EFSBADCRC, "iget: checksum invalid");
4854 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4855 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4856 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4857 if (ext4_has_feature_project(sb) &&
4858 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4859 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4860 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4862 i_projid = EXT4_DEF_PROJID;
4864 if (!(test_opt(inode->i_sb, NO_UID32))) {
4865 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4866 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4868 i_uid_write(inode, i_uid);
4869 i_gid_write(inode, i_gid);
4870 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4871 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4873 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4874 ei->i_inline_off = 0;
4875 ei->i_dir_start_lookup = 0;
4876 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4877 /* We now have enough fields to check if the inode was active or not.
4878 * This is needed because nfsd might try to access dead inodes
4879 * the test is that same one that e2fsck uses
4880 * NeilBrown 1999oct15
4882 if (inode->i_nlink == 0) {
4883 if ((inode->i_mode == 0 ||
4884 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4885 ino != EXT4_BOOT_LOADER_INO) {
4886 /* this inode is deleted */
4890 /* The only unlinked inodes we let through here have
4891 * valid i_mode and are being read by the orphan
4892 * recovery code: that's fine, we're about to complete
4893 * the process of deleting those.
4894 * OR it is the EXT4_BOOT_LOADER_INO which is
4895 * not initialized on a new filesystem. */
4897 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4898 ext4_set_inode_flags(inode, true);
4899 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4900 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4901 if (ext4_has_feature_64bit(sb))
4903 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4904 inode->i_size = ext4_isize(sb, raw_inode);
4905 if ((size = i_size_read(inode)) < 0) {
4906 ext4_error_inode(inode, function, line, 0,
4907 "iget: bad i_size value: %lld", size);
4908 ret = -EFSCORRUPTED;
4912 * If dir_index is not enabled but there's dir with INDEX flag set,
4913 * we'd normally treat htree data as empty space. But with metadata
4914 * checksumming that corrupts checksums so forbid that.
4916 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4917 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4918 ext4_error_inode(inode, function, line, 0,
4919 "iget: Dir with htree data on filesystem without dir_index feature.");
4920 ret = -EFSCORRUPTED;
4923 ei->i_disksize = inode->i_size;
4925 ei->i_reserved_quota = 0;
4927 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4928 ei->i_block_group = iloc.block_group;
4929 ei->i_last_alloc_group = ~0;
4931 * NOTE! The in-memory inode i_data array is in little-endian order
4932 * even on big-endian machines: we do NOT byteswap the block numbers!
4934 for (block = 0; block < EXT4_N_BLOCKS; block++)
4935 ei->i_data[block] = raw_inode->i_block[block];
4936 INIT_LIST_HEAD(&ei->i_orphan);
4937 ext4_fc_init_inode(&ei->vfs_inode);
4940 * Set transaction id's of transactions that have to be committed
4941 * to finish f[data]sync. We set them to currently running transaction
4942 * as we cannot be sure that the inode or some of its metadata isn't
4943 * part of the transaction - the inode could have been reclaimed and
4944 * now it is reread from disk.
4947 transaction_t *transaction;
4950 read_lock(&journal->j_state_lock);
4951 if (journal->j_running_transaction)
4952 transaction = journal->j_running_transaction;
4954 transaction = journal->j_committing_transaction;
4956 tid = transaction->t_tid;
4958 tid = journal->j_commit_sequence;
4959 read_unlock(&journal->j_state_lock);
4960 ei->i_sync_tid = tid;
4961 ei->i_datasync_tid = tid;
4964 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4965 if (ei->i_extra_isize == 0) {
4966 /* The extra space is currently unused. Use it. */
4967 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4968 ei->i_extra_isize = sizeof(struct ext4_inode) -
4969 EXT4_GOOD_OLD_INODE_SIZE;
4971 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4977 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4978 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4979 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4980 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4982 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4983 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4985 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4986 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4988 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4990 ext4_inode_set_iversion_queried(inode, ivers);
4994 if (ei->i_file_acl &&
4995 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4996 ext4_error_inode(inode, function, line, 0,
4997 "iget: bad extended attribute block %llu",
4999 ret = -EFSCORRUPTED;
5001 } else if (!ext4_has_inline_data(inode)) {
5002 /* validate the block references in the inode */
5003 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
5004 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5005 (S_ISLNK(inode->i_mode) &&
5006 !ext4_inode_is_fast_symlink(inode)))) {
5007 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5008 ret = ext4_ext_check_inode(inode);
5010 ret = ext4_ind_check_inode(inode);
5016 if (S_ISREG(inode->i_mode)) {
5017 inode->i_op = &ext4_file_inode_operations;
5018 inode->i_fop = &ext4_file_operations;
5019 ext4_set_aops(inode);
5020 } else if (S_ISDIR(inode->i_mode)) {
5021 inode->i_op = &ext4_dir_inode_operations;
5022 inode->i_fop = &ext4_dir_operations;
5023 } else if (S_ISLNK(inode->i_mode)) {
5024 /* VFS does not allow setting these so must be corruption */
5025 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5026 ext4_error_inode(inode, function, line, 0,
5027 "iget: immutable or append flags "
5028 "not allowed on symlinks");
5029 ret = -EFSCORRUPTED;
5032 if (IS_ENCRYPTED(inode)) {
5033 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5034 } else if (ext4_inode_is_fast_symlink(inode)) {
5035 inode->i_link = (char *)ei->i_data;
5036 inode->i_op = &ext4_fast_symlink_inode_operations;
5037 nd_terminate_link(ei->i_data, inode->i_size,
5038 sizeof(ei->i_data) - 1);
5040 inode->i_op = &ext4_symlink_inode_operations;
5042 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5043 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5044 inode->i_op = &ext4_special_inode_operations;
5045 if (raw_inode->i_block[0])
5046 init_special_inode(inode, inode->i_mode,
5047 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5049 init_special_inode(inode, inode->i_mode,
5050 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5051 } else if (ino == EXT4_BOOT_LOADER_INO) {
5052 make_bad_inode(inode);
5054 ret = -EFSCORRUPTED;
5055 ext4_error_inode(inode, function, line, 0,
5056 "iget: bogus i_mode (%o)", inode->i_mode);
5059 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5060 ext4_error_inode(inode, function, line, 0,
5061 "casefold flag without casefold feature");
5062 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
5063 ext4_error_inode(inode, function, line, 0,
5064 "bad inode without EXT4_IGET_BAD flag");
5070 unlock_new_inode(inode);
5076 return ERR_PTR(ret);
5079 static void __ext4_update_other_inode_time(struct super_block *sb,
5080 unsigned long orig_ino,
5082 struct ext4_inode *raw_inode)
5084 struct inode *inode;
5086 inode = find_inode_by_ino_rcu(sb, ino);
5090 if (!inode_is_dirtytime_only(inode))
5093 spin_lock(&inode->i_lock);
5094 if (inode_is_dirtytime_only(inode)) {
5095 struct ext4_inode_info *ei = EXT4_I(inode);
5097 inode->i_state &= ~I_DIRTY_TIME;
5098 spin_unlock(&inode->i_lock);
5100 spin_lock(&ei->i_raw_lock);
5101 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5102 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5103 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5104 ext4_inode_csum_set(inode, raw_inode, ei);
5105 spin_unlock(&ei->i_raw_lock);
5106 trace_ext4_other_inode_update_time(inode, orig_ino);
5109 spin_unlock(&inode->i_lock);
5113 * Opportunistically update the other time fields for other inodes in
5114 * the same inode table block.
5116 static void ext4_update_other_inodes_time(struct super_block *sb,
5117 unsigned long orig_ino, char *buf)
5120 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5121 int inode_size = EXT4_INODE_SIZE(sb);
5124 * Calculate the first inode in the inode table block. Inode
5125 * numbers are one-based. That is, the first inode in a block
5126 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5128 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5130 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5131 if (ino == orig_ino)
5133 __ext4_update_other_inode_time(sb, orig_ino, ino,
5134 (struct ext4_inode *)buf);
5140 * Post the struct inode info into an on-disk inode location in the
5141 * buffer-cache. This gobbles the caller's reference to the
5142 * buffer_head in the inode location struct.
5144 * The caller must have write access to iloc->bh.
5146 static int ext4_do_update_inode(handle_t *handle,
5147 struct inode *inode,
5148 struct ext4_iloc *iloc)
5150 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5151 struct ext4_inode_info *ei = EXT4_I(inode);
5152 struct buffer_head *bh = iloc->bh;
5153 struct super_block *sb = inode->i_sb;
5155 int need_datasync = 0, set_large_file = 0;
5157 spin_lock(&ei->i_raw_lock);
5160 * For fields not tracked in the in-memory inode, initialise them
5161 * to zero for new inodes.
5163 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5164 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5166 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5168 if (ei->i_disksize > 0x7fffffffULL) {
5169 if (!ext4_has_feature_large_file(sb) ||
5170 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5174 err = ext4_fill_raw_inode(inode, raw_inode);
5175 spin_unlock(&ei->i_raw_lock);
5177 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5181 if (inode->i_sb->s_flags & SB_LAZYTIME)
5182 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5185 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5186 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5189 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5190 if (set_large_file) {
5191 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5192 err = ext4_journal_get_write_access(handle, sb,
5197 lock_buffer(EXT4_SB(sb)->s_sbh);
5198 ext4_set_feature_large_file(sb);
5199 ext4_superblock_csum_set(sb);
5200 unlock_buffer(EXT4_SB(sb)->s_sbh);
5201 ext4_handle_sync(handle);
5202 err = ext4_handle_dirty_metadata(handle, NULL,
5203 EXT4_SB(sb)->s_sbh);
5205 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5207 ext4_std_error(inode->i_sb, err);
5214 * ext4_write_inode()
5216 * We are called from a few places:
5218 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5219 * Here, there will be no transaction running. We wait for any running
5220 * transaction to commit.
5222 * - Within flush work (sys_sync(), kupdate and such).
5223 * We wait on commit, if told to.
5225 * - Within iput_final() -> write_inode_now()
5226 * We wait on commit, if told to.
5228 * In all cases it is actually safe for us to return without doing anything,
5229 * because the inode has been copied into a raw inode buffer in
5230 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5233 * Note that we are absolutely dependent upon all inode dirtiers doing the
5234 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5235 * which we are interested.
5237 * It would be a bug for them to not do this. The code:
5239 * mark_inode_dirty(inode)
5241 * inode->i_size = expr;
5243 * is in error because write_inode() could occur while `stuff()' is running,
5244 * and the new i_size will be lost. Plus the inode will no longer be on the
5245 * superblock's dirty inode list.
5247 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5251 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5252 sb_rdonly(inode->i_sb))
5255 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5258 if (EXT4_SB(inode->i_sb)->s_journal) {
5259 if (ext4_journal_current_handle()) {
5260 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5266 * No need to force transaction in WB_SYNC_NONE mode. Also
5267 * ext4_sync_fs() will force the commit after everything is
5270 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5273 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5274 EXT4_I(inode)->i_sync_tid);
5276 struct ext4_iloc iloc;
5278 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5282 * sync(2) will flush the whole buffer cache. No need to do
5283 * it here separately for each inode.
5285 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5286 sync_dirty_buffer(iloc.bh);
5287 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5288 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5289 "IO error syncing inode");
5298 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5299 * buffers that are attached to a folio straddling i_size and are undergoing
5300 * commit. In that case we have to wait for commit to finish and try again.
5302 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5305 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5306 tid_t commit_tid = 0;
5309 offset = inode->i_size & (PAGE_SIZE - 1);
5311 * If the folio is fully truncated, we don't need to wait for any commit
5312 * (and we even should not as __ext4_journalled_invalidate_folio() may
5313 * strip all buffers from the folio but keep the folio dirty which can then
5314 * confuse e.g. concurrent ext4_writepage() seeing dirty folio without
5315 * buffers). Also we don't need to wait for any commit if all buffers in
5316 * the folio remain valid. This is most beneficial for the common case of
5317 * blocksize == PAGESIZE.
5319 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5322 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5323 inode->i_size >> PAGE_SHIFT);
5326 ret = __ext4_journalled_invalidate_folio(folio, offset,
5327 folio_size(folio) - offset);
5328 folio_unlock(folio);
5333 read_lock(&journal->j_state_lock);
5334 if (journal->j_committing_transaction)
5335 commit_tid = journal->j_committing_transaction->t_tid;
5336 read_unlock(&journal->j_state_lock);
5338 jbd2_log_wait_commit(journal, commit_tid);
5345 * Called from notify_change.
5347 * We want to trap VFS attempts to truncate the file as soon as
5348 * possible. In particular, we want to make sure that when the VFS
5349 * shrinks i_size, we put the inode on the orphan list and modify
5350 * i_disksize immediately, so that during the subsequent flushing of
5351 * dirty pages and freeing of disk blocks, we can guarantee that any
5352 * commit will leave the blocks being flushed in an unused state on
5353 * disk. (On recovery, the inode will get truncated and the blocks will
5354 * be freed, so we have a strong guarantee that no future commit will
5355 * leave these blocks visible to the user.)
5357 * Another thing we have to assure is that if we are in ordered mode
5358 * and inode is still attached to the committing transaction, we must
5359 * we start writeout of all the dirty pages which are being truncated.
5360 * This way we are sure that all the data written in the previous
5361 * transaction are already on disk (truncate waits for pages under
5364 * Called with inode->i_rwsem down.
5366 int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5369 struct inode *inode = d_inode(dentry);
5372 const unsigned int ia_valid = attr->ia_valid;
5373 bool inc_ivers = true;
5375 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5378 if (unlikely(IS_IMMUTABLE(inode)))
5381 if (unlikely(IS_APPEND(inode) &&
5382 (ia_valid & (ATTR_MODE | ATTR_UID |
5383 ATTR_GID | ATTR_TIMES_SET))))
5386 error = setattr_prepare(mnt_userns, dentry, attr);
5390 error = fscrypt_prepare_setattr(dentry, attr);
5394 error = fsverity_prepare_setattr(dentry, attr);
5398 if (is_quota_modification(mnt_userns, inode, attr)) {
5399 error = dquot_initialize(inode);
5404 if (i_uid_needs_update(mnt_userns, attr, inode) ||
5405 i_gid_needs_update(mnt_userns, attr, inode)) {
5408 /* (user+group)*(old+new) structure, inode write (sb,
5409 * inode block, ? - but truncate inode update has it) */
5410 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5411 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5412 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5413 if (IS_ERR(handle)) {
5414 error = PTR_ERR(handle);
5418 /* dquot_transfer() calls back ext4_get_inode_usage() which
5419 * counts xattr inode references.
5421 down_read(&EXT4_I(inode)->xattr_sem);
5422 error = dquot_transfer(mnt_userns, inode, attr);
5423 up_read(&EXT4_I(inode)->xattr_sem);
5426 ext4_journal_stop(handle);
5429 /* Update corresponding info in inode so that everything is in
5430 * one transaction */
5431 i_uid_update(mnt_userns, attr, inode);
5432 i_gid_update(mnt_userns, attr, inode);
5433 error = ext4_mark_inode_dirty(handle, inode);
5434 ext4_journal_stop(handle);
5435 if (unlikely(error)) {
5440 if (attr->ia_valid & ATTR_SIZE) {
5442 loff_t oldsize = inode->i_size;
5443 loff_t old_disksize;
5444 int shrink = (attr->ia_size < inode->i_size);
5446 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5447 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5449 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5453 if (!S_ISREG(inode->i_mode)) {
5457 if (attr->ia_size == inode->i_size)
5461 if (ext4_should_order_data(inode)) {
5462 error = ext4_begin_ordered_truncate(inode,
5468 * Blocks are going to be removed from the inode. Wait
5469 * for dio in flight.
5471 inode_dio_wait(inode);
5474 filemap_invalidate_lock(inode->i_mapping);
5476 rc = ext4_break_layouts(inode);
5478 filemap_invalidate_unlock(inode->i_mapping);
5482 if (attr->ia_size != inode->i_size) {
5483 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5484 if (IS_ERR(handle)) {
5485 error = PTR_ERR(handle);
5488 if (ext4_handle_valid(handle) && shrink) {
5489 error = ext4_orphan_add(handle, inode);
5493 * Update c/mtime on truncate up, ext4_truncate() will
5494 * update c/mtime in shrink case below
5497 inode->i_mtime = current_time(inode);
5498 inode->i_ctime = inode->i_mtime;
5502 ext4_fc_track_range(handle, inode,
5503 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5504 inode->i_sb->s_blocksize_bits,
5505 EXT_MAX_BLOCKS - 1);
5507 ext4_fc_track_range(
5509 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5510 inode->i_sb->s_blocksize_bits,
5511 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5512 inode->i_sb->s_blocksize_bits);
5514 down_write(&EXT4_I(inode)->i_data_sem);
5515 old_disksize = EXT4_I(inode)->i_disksize;
5516 EXT4_I(inode)->i_disksize = attr->ia_size;
5517 rc = ext4_mark_inode_dirty(handle, inode);
5521 * We have to update i_size under i_data_sem together
5522 * with i_disksize to avoid races with writeback code
5523 * running ext4_wb_update_i_disksize().
5526 i_size_write(inode, attr->ia_size);
5528 EXT4_I(inode)->i_disksize = old_disksize;
5529 up_write(&EXT4_I(inode)->i_data_sem);
5530 ext4_journal_stop(handle);
5534 pagecache_isize_extended(inode, oldsize,
5536 } else if (ext4_should_journal_data(inode)) {
5537 ext4_wait_for_tail_page_commit(inode);
5542 * Truncate pagecache after we've waited for commit
5543 * in data=journal mode to make pages freeable.
5545 truncate_pagecache(inode, inode->i_size);
5547 * Call ext4_truncate() even if i_size didn't change to
5548 * truncate possible preallocated blocks.
5550 if (attr->ia_size <= oldsize) {
5551 rc = ext4_truncate(inode);
5556 filemap_invalidate_unlock(inode->i_mapping);
5561 inode_inc_iversion(inode);
5562 setattr_copy(mnt_userns, inode, attr);
5563 mark_inode_dirty(inode);
5567 * If the call to ext4_truncate failed to get a transaction handle at
5568 * all, we need to clean up the in-core orphan list manually.
5570 if (orphan && inode->i_nlink)
5571 ext4_orphan_del(NULL, inode);
5573 if (!error && (ia_valid & ATTR_MODE))
5574 rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5578 ext4_std_error(inode->i_sb, error);
5584 u32 ext4_dio_alignment(struct inode *inode)
5586 if (fsverity_active(inode))
5588 if (ext4_should_journal_data(inode))
5590 if (ext4_has_inline_data(inode))
5592 if (IS_ENCRYPTED(inode)) {
5593 if (!fscrypt_dio_supported(inode))
5595 return i_blocksize(inode);
5597 return 1; /* use the iomap defaults */
5600 int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5601 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5603 struct inode *inode = d_inode(path->dentry);
5604 struct ext4_inode *raw_inode;
5605 struct ext4_inode_info *ei = EXT4_I(inode);
5608 if ((request_mask & STATX_BTIME) &&
5609 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5610 stat->result_mask |= STATX_BTIME;
5611 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5612 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5616 * Return the DIO alignment restrictions if requested. We only return
5617 * this information when requested, since on encrypted files it might
5618 * take a fair bit of work to get if the file wasn't opened recently.
5620 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5621 u32 dio_align = ext4_dio_alignment(inode);
5623 stat->result_mask |= STATX_DIOALIGN;
5624 if (dio_align == 1) {
5625 struct block_device *bdev = inode->i_sb->s_bdev;
5627 /* iomap defaults */
5628 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5629 stat->dio_offset_align = bdev_logical_block_size(bdev);
5631 stat->dio_mem_align = dio_align;
5632 stat->dio_offset_align = dio_align;
5636 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5637 if (flags & EXT4_APPEND_FL)
5638 stat->attributes |= STATX_ATTR_APPEND;
5639 if (flags & EXT4_COMPR_FL)
5640 stat->attributes |= STATX_ATTR_COMPRESSED;
5641 if (flags & EXT4_ENCRYPT_FL)
5642 stat->attributes |= STATX_ATTR_ENCRYPTED;
5643 if (flags & EXT4_IMMUTABLE_FL)
5644 stat->attributes |= STATX_ATTR_IMMUTABLE;
5645 if (flags & EXT4_NODUMP_FL)
5646 stat->attributes |= STATX_ATTR_NODUMP;
5647 if (flags & EXT4_VERITY_FL)
5648 stat->attributes |= STATX_ATTR_VERITY;
5650 stat->attributes_mask |= (STATX_ATTR_APPEND |
5651 STATX_ATTR_COMPRESSED |
5652 STATX_ATTR_ENCRYPTED |
5653 STATX_ATTR_IMMUTABLE |
5657 generic_fillattr(mnt_userns, inode, stat);
5661 int ext4_file_getattr(struct user_namespace *mnt_userns,
5662 const struct path *path, struct kstat *stat,
5663 u32 request_mask, unsigned int query_flags)
5665 struct inode *inode = d_inode(path->dentry);
5666 u64 delalloc_blocks;
5668 ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5671 * If there is inline data in the inode, the inode will normally not
5672 * have data blocks allocated (it may have an external xattr block).
5673 * Report at least one sector for such files, so tools like tar, rsync,
5674 * others don't incorrectly think the file is completely sparse.
5676 if (unlikely(ext4_has_inline_data(inode)))
5677 stat->blocks += (stat->size + 511) >> 9;
5680 * We can't update i_blocks if the block allocation is delayed
5681 * otherwise in the case of system crash before the real block
5682 * allocation is done, we will have i_blocks inconsistent with
5683 * on-disk file blocks.
5684 * We always keep i_blocks updated together with real
5685 * allocation. But to not confuse with user, stat
5686 * will return the blocks that include the delayed allocation
5687 * blocks for this file.
5689 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5690 EXT4_I(inode)->i_reserved_data_blocks);
5691 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5695 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5698 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5699 return ext4_ind_trans_blocks(inode, lblocks);
5700 return ext4_ext_index_trans_blocks(inode, pextents);
5704 * Account for index blocks, block groups bitmaps and block group
5705 * descriptor blocks if modify datablocks and index blocks
5706 * worse case, the indexs blocks spread over different block groups
5708 * If datablocks are discontiguous, they are possible to spread over
5709 * different block groups too. If they are contiguous, with flexbg,
5710 * they could still across block group boundary.
5712 * Also account for superblock, inode, quota and xattr blocks
5714 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5717 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5723 * How many index blocks need to touch to map @lblocks logical blocks
5724 * to @pextents physical extents?
5726 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5731 * Now let's see how many group bitmaps and group descriptors need
5734 groups = idxblocks + pextents;
5736 if (groups > ngroups)
5738 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5739 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5741 /* bitmaps and block group descriptor blocks */
5742 ret += groups + gdpblocks;
5744 /* Blocks for super block, inode, quota and xattr blocks */
5745 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5751 * Calculate the total number of credits to reserve to fit
5752 * the modification of a single pages into a single transaction,
5753 * which may include multiple chunks of block allocations.
5755 * This could be called via ext4_write_begin()
5757 * We need to consider the worse case, when
5758 * one new block per extent.
5760 int ext4_writepage_trans_blocks(struct inode *inode)
5762 int bpp = ext4_journal_blocks_per_page(inode);
5765 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5767 /* Account for data blocks for journalled mode */
5768 if (ext4_should_journal_data(inode))
5774 * Calculate the journal credits for a chunk of data modification.
5776 * This is called from DIO, fallocate or whoever calling
5777 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5779 * journal buffers for data blocks are not included here, as DIO
5780 * and fallocate do no need to journal data buffers.
5782 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5784 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5788 * The caller must have previously called ext4_reserve_inode_write().
5789 * Give this, we know that the caller already has write access to iloc->bh.
5791 int ext4_mark_iloc_dirty(handle_t *handle,
5792 struct inode *inode, struct ext4_iloc *iloc)
5796 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5800 ext4_fc_track_inode(handle, inode);
5802 /* the do_update_inode consumes one bh->b_count */
5805 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5806 err = ext4_do_update_inode(handle, inode, iloc);
5812 * On success, We end up with an outstanding reference count against
5813 * iloc->bh. This _must_ be cleaned up later.
5817 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5818 struct ext4_iloc *iloc)
5822 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5825 err = ext4_get_inode_loc(inode, iloc);
5827 BUFFER_TRACE(iloc->bh, "get_write_access");
5828 err = ext4_journal_get_write_access(handle, inode->i_sb,
5829 iloc->bh, EXT4_JTR_NONE);
5835 ext4_std_error(inode->i_sb, err);
5839 static int __ext4_expand_extra_isize(struct inode *inode,
5840 unsigned int new_extra_isize,
5841 struct ext4_iloc *iloc,
5842 handle_t *handle, int *no_expand)
5844 struct ext4_inode *raw_inode;
5845 struct ext4_xattr_ibody_header *header;
5846 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5847 struct ext4_inode_info *ei = EXT4_I(inode);
5850 /* this was checked at iget time, but double check for good measure */
5851 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5852 (ei->i_extra_isize & 3)) {
5853 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5855 EXT4_INODE_SIZE(inode->i_sb));
5856 return -EFSCORRUPTED;
5858 if ((new_extra_isize < ei->i_extra_isize) ||
5859 (new_extra_isize < 4) ||
5860 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5861 return -EINVAL; /* Should never happen */
5863 raw_inode = ext4_raw_inode(iloc);
5865 header = IHDR(inode, raw_inode);
5867 /* No extended attributes present */
5868 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5869 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5870 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5871 EXT4_I(inode)->i_extra_isize, 0,
5872 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5873 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5877 /* try to expand with EAs present */
5878 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5882 * Inode size expansion failed; don't try again
5891 * Expand an inode by new_extra_isize bytes.
5892 * Returns 0 on success or negative error number on failure.
5894 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5895 unsigned int new_extra_isize,
5896 struct ext4_iloc iloc,
5902 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5906 * In nojournal mode, we can immediately attempt to expand
5907 * the inode. When journaled, we first need to obtain extra
5908 * buffer credits since we may write into the EA block
5909 * with this same handle. If journal_extend fails, then it will
5910 * only result in a minor loss of functionality for that inode.
5911 * If this is felt to be critical, then e2fsck should be run to
5912 * force a large enough s_min_extra_isize.
5914 if (ext4_journal_extend(handle,
5915 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5918 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5921 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5922 handle, &no_expand);
5923 ext4_write_unlock_xattr(inode, &no_expand);
5928 int ext4_expand_extra_isize(struct inode *inode,
5929 unsigned int new_extra_isize,
5930 struct ext4_iloc *iloc)
5936 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5941 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5942 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5943 if (IS_ERR(handle)) {
5944 error = PTR_ERR(handle);
5949 ext4_write_lock_xattr(inode, &no_expand);
5951 BUFFER_TRACE(iloc->bh, "get_write_access");
5952 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5959 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5960 handle, &no_expand);
5962 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5967 ext4_write_unlock_xattr(inode, &no_expand);
5968 ext4_journal_stop(handle);
5973 * What we do here is to mark the in-core inode as clean with respect to inode
5974 * dirtiness (it may still be data-dirty).
5975 * This means that the in-core inode may be reaped by prune_icache
5976 * without having to perform any I/O. This is a very good thing,
5977 * because *any* task may call prune_icache - even ones which
5978 * have a transaction open against a different journal.
5980 * Is this cheating? Not really. Sure, we haven't written the
5981 * inode out, but prune_icache isn't a user-visible syncing function.
5982 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5983 * we start and wait on commits.
5985 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5986 const char *func, unsigned int line)
5988 struct ext4_iloc iloc;
5989 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5993 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5994 err = ext4_reserve_inode_write(handle, inode, &iloc);
5998 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5999 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6002 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6005 ext4_error_inode_err(inode, func, line, 0, err,
6006 "mark_inode_dirty error");
6011 * ext4_dirty_inode() is called from __mark_inode_dirty()
6013 * We're really interested in the case where a file is being extended.
6014 * i_size has been changed by generic_commit_write() and we thus need
6015 * to include the updated inode in the current transaction.
6017 * Also, dquot_alloc_block() will always dirty the inode when blocks
6018 * are allocated to the file.
6020 * If the inode is marked synchronous, we don't honour that here - doing
6021 * so would cause a commit on atime updates, which we don't bother doing.
6022 * We handle synchronous inodes at the highest possible level.
6024 void ext4_dirty_inode(struct inode *inode, int flags)
6028 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6031 ext4_mark_inode_dirty(handle, inode);
6032 ext4_journal_stop(handle);
6035 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6040 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6043 * We have to be very careful here: changing a data block's
6044 * journaling status dynamically is dangerous. If we write a
6045 * data block to the journal, change the status and then delete
6046 * that block, we risk forgetting to revoke the old log record
6047 * from the journal and so a subsequent replay can corrupt data.
6048 * So, first we make sure that the journal is empty and that
6049 * nobody is changing anything.
6052 journal = EXT4_JOURNAL(inode);
6055 if (is_journal_aborted(journal))
6058 /* Wait for all existing dio workers */
6059 inode_dio_wait(inode);
6062 * Before flushing the journal and switching inode's aops, we have
6063 * to flush all dirty data the inode has. There can be outstanding
6064 * delayed allocations, there can be unwritten extents created by
6065 * fallocate or buffered writes in dioread_nolock mode covered by
6066 * dirty data which can be converted only after flushing the dirty
6067 * data (and journalled aops don't know how to handle these cases).
6070 filemap_invalidate_lock(inode->i_mapping);
6071 err = filemap_write_and_wait(inode->i_mapping);
6073 filemap_invalidate_unlock(inode->i_mapping);
6078 percpu_down_write(&sbi->s_writepages_rwsem);
6079 jbd2_journal_lock_updates(journal);
6082 * OK, there are no updates running now, and all cached data is
6083 * synced to disk. We are now in a completely consistent state
6084 * which doesn't have anything in the journal, and we know that
6085 * no filesystem updates are running, so it is safe to modify
6086 * the inode's in-core data-journaling state flag now.
6090 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6092 err = jbd2_journal_flush(journal, 0);
6094 jbd2_journal_unlock_updates(journal);
6095 percpu_up_write(&sbi->s_writepages_rwsem);
6098 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6100 ext4_set_aops(inode);
6102 jbd2_journal_unlock_updates(journal);
6103 percpu_up_write(&sbi->s_writepages_rwsem);
6106 filemap_invalidate_unlock(inode->i_mapping);
6108 /* Finally we can mark the inode as dirty. */
6110 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6112 return PTR_ERR(handle);
6114 ext4_fc_mark_ineligible(inode->i_sb,
6115 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6116 err = ext4_mark_inode_dirty(handle, inode);
6117 ext4_handle_sync(handle);
6118 ext4_journal_stop(handle);
6119 ext4_std_error(inode->i_sb, err);
6124 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6125 struct buffer_head *bh)
6127 return !buffer_mapped(bh);
6130 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6132 struct vm_area_struct *vma = vmf->vma;
6133 struct page *page = vmf->page;
6138 struct file *file = vma->vm_file;
6139 struct inode *inode = file_inode(file);
6140 struct address_space *mapping = inode->i_mapping;
6142 get_block_t *get_block;
6145 if (unlikely(IS_IMMUTABLE(inode)))
6146 return VM_FAULT_SIGBUS;
6148 sb_start_pagefault(inode->i_sb);
6149 file_update_time(vma->vm_file);
6151 filemap_invalidate_lock_shared(mapping);
6153 err = ext4_convert_inline_data(inode);
6158 * On data journalling we skip straight to the transaction handle:
6159 * there's no delalloc; page truncated will be checked later; the
6160 * early return w/ all buffers mapped (calculates size/len) can't
6161 * be used; and there's no dioread_nolock, so only ext4_get_block.
6163 if (ext4_should_journal_data(inode))
6166 /* Delalloc case is easy... */
6167 if (test_opt(inode->i_sb, DELALLOC) &&
6168 !ext4_nonda_switch(inode->i_sb)) {
6170 err = block_page_mkwrite(vma, vmf,
6171 ext4_da_get_block_prep);
6172 } while (err == -ENOSPC &&
6173 ext4_should_retry_alloc(inode->i_sb, &retries));
6178 size = i_size_read(inode);
6179 /* Page got truncated from under us? */
6180 if (page->mapping != mapping || page_offset(page) > size) {
6182 ret = VM_FAULT_NOPAGE;
6186 if (page->index == size >> PAGE_SHIFT)
6187 len = size & ~PAGE_MASK;
6191 * Return if we have all the buffers mapped. This avoids the need to do
6192 * journal_start/journal_stop which can block and take a long time
6194 * This cannot be done for data journalling, as we have to add the
6195 * inode to the transaction's list to writeprotect pages on commit.
6197 if (page_has_buffers(page)) {
6198 if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6200 ext4_bh_unmapped)) {
6201 /* Wait so that we don't change page under IO */
6202 wait_for_stable_page(page);
6203 ret = VM_FAULT_LOCKED;
6208 /* OK, we need to fill the hole... */
6209 if (ext4_should_dioread_nolock(inode))
6210 get_block = ext4_get_block_unwritten;
6212 get_block = ext4_get_block;
6214 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6215 ext4_writepage_trans_blocks(inode));
6216 if (IS_ERR(handle)) {
6217 ret = VM_FAULT_SIGBUS;
6221 * Data journalling can't use block_page_mkwrite() because it
6222 * will set_buffer_dirty() before do_journal_get_write_access()
6223 * thus might hit warning messages for dirty metadata buffers.
6225 if (!ext4_should_journal_data(inode)) {
6226 err = block_page_mkwrite(vma, vmf, get_block);
6229 size = i_size_read(inode);
6230 /* Page got truncated from under us? */
6231 if (page->mapping != mapping || page_offset(page) > size) {
6232 ret = VM_FAULT_NOPAGE;
6236 if (page->index == size >> PAGE_SHIFT)
6237 len = size & ~PAGE_MASK;
6241 err = __block_write_begin(page, 0, len, ext4_get_block);
6243 ret = VM_FAULT_SIGBUS;
6244 if (ext4_walk_page_buffers(handle, inode,
6245 page_buffers(page), 0, len, NULL,
6246 do_journal_get_write_access))
6248 if (ext4_walk_page_buffers(handle, inode,
6249 page_buffers(page), 0, len, NULL,
6252 if (ext4_jbd2_inode_add_write(handle, inode,
6253 page_offset(page), len))
6255 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6260 ext4_journal_stop(handle);
6261 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6264 ret = block_page_mkwrite_return(err);
6266 filemap_invalidate_unlock_shared(mapping);
6267 sb_end_pagefault(inode->i_sb);
6271 ext4_journal_stop(handle);