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 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1320 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1322 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1324 * it's important to update i_size while still holding page lock:
1325 * page writeout could otherwise come in and zero beyond i_size.
1327 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1328 * blocks are being written past EOF, so skip the i_size update.
1331 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1335 if (old_size < pos && !verity)
1336 pagecache_isize_extended(inode, old_size, pos);
1338 * Don't mark the inode dirty under page lock. First, it unnecessarily
1339 * makes the holding time of page lock longer. Second, it forces lock
1340 * ordering of page lock and transaction start for journaling
1344 ret = ext4_mark_inode_dirty(handle, inode);
1346 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1347 /* if we have allocated more blocks and copied
1348 * less. We will have blocks allocated outside
1349 * inode->i_size. So truncate them
1351 ext4_orphan_add(handle, inode);
1353 ret2 = ext4_journal_stop(handle);
1357 if (pos + len > inode->i_size && !verity) {
1358 ext4_truncate_failed_write(inode);
1360 * If truncate failed early the inode might still be
1361 * on the orphan list; we need to make sure the inode
1362 * is removed from the orphan list in that case.
1365 ext4_orphan_del(NULL, inode);
1368 return ret ? ret : copied;
1372 * This is a private version of page_zero_new_buffers() which doesn't
1373 * set the buffer to be dirty, since in data=journalled mode we need
1374 * to call ext4_handle_dirty_metadata() instead.
1376 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1377 struct inode *inode,
1379 unsigned from, unsigned to)
1381 unsigned int block_start = 0, block_end;
1382 struct buffer_head *head, *bh;
1384 bh = head = page_buffers(page);
1386 block_end = block_start + bh->b_size;
1387 if (buffer_new(bh)) {
1388 if (block_end > from && block_start < to) {
1389 if (!PageUptodate(page)) {
1390 unsigned start, size;
1392 start = max(from, block_start);
1393 size = min(to, block_end) - start;
1395 zero_user(page, start, size);
1396 write_end_fn(handle, inode, bh);
1398 clear_buffer_new(bh);
1401 block_start = block_end;
1402 bh = bh->b_this_page;
1403 } while (bh != head);
1406 static int ext4_journalled_write_end(struct file *file,
1407 struct address_space *mapping,
1408 loff_t pos, unsigned len, unsigned copied,
1409 struct page *page, void *fsdata)
1411 handle_t *handle = ext4_journal_current_handle();
1412 struct inode *inode = mapping->host;
1413 loff_t old_size = inode->i_size;
1417 int size_changed = 0;
1418 bool verity = ext4_verity_in_progress(inode);
1420 trace_ext4_journalled_write_end(inode, pos, len, copied);
1421 from = pos & (PAGE_SIZE - 1);
1424 BUG_ON(!ext4_handle_valid(handle));
1426 if (ext4_has_inline_data(inode))
1427 return ext4_write_inline_data_end(inode, pos, len, copied, page);
1429 if (unlikely(copied < len) && !PageUptodate(page)) {
1431 ext4_journalled_zero_new_buffers(handle, inode, page, from, to);
1433 if (unlikely(copied < len))
1434 ext4_journalled_zero_new_buffers(handle, inode, page,
1436 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
1437 from, from + copied, &partial,
1440 SetPageUptodate(page);
1443 size_changed = ext4_update_inode_size(inode, pos + copied);
1444 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1445 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1449 if (old_size < pos && !verity)
1450 pagecache_isize_extended(inode, old_size, pos);
1453 ret2 = ext4_mark_inode_dirty(handle, inode);
1458 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1459 /* if we have allocated more blocks and copied
1460 * less. We will have blocks allocated outside
1461 * inode->i_size. So truncate them
1463 ext4_orphan_add(handle, inode);
1465 ret2 = ext4_journal_stop(handle);
1468 if (pos + len > inode->i_size && !verity) {
1469 ext4_truncate_failed_write(inode);
1471 * If truncate failed early the inode might still be
1472 * on the orphan list; we need to make sure the inode
1473 * is removed from the orphan list in that case.
1476 ext4_orphan_del(NULL, inode);
1479 return ret ? ret : copied;
1483 * Reserve space for a single cluster
1485 static int ext4_da_reserve_space(struct inode *inode)
1487 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1488 struct ext4_inode_info *ei = EXT4_I(inode);
1492 * We will charge metadata quota at writeout time; this saves
1493 * us from metadata over-estimation, though we may go over by
1494 * a small amount in the end. Here we just reserve for data.
1496 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1500 spin_lock(&ei->i_block_reservation_lock);
1501 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1502 spin_unlock(&ei->i_block_reservation_lock);
1503 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1506 ei->i_reserved_data_blocks++;
1507 trace_ext4_da_reserve_space(inode);
1508 spin_unlock(&ei->i_block_reservation_lock);
1510 return 0; /* success */
1513 void ext4_da_release_space(struct inode *inode, int to_free)
1515 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1516 struct ext4_inode_info *ei = EXT4_I(inode);
1519 return; /* Nothing to release, exit */
1521 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1523 trace_ext4_da_release_space(inode, to_free);
1524 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1526 * if there aren't enough reserved blocks, then the
1527 * counter is messed up somewhere. Since this
1528 * function is called from invalidate page, it's
1529 * harmless to return without any action.
1531 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1532 "ino %lu, to_free %d with only %d reserved "
1533 "data blocks", inode->i_ino, to_free,
1534 ei->i_reserved_data_blocks);
1536 to_free = ei->i_reserved_data_blocks;
1538 ei->i_reserved_data_blocks -= to_free;
1540 /* update fs dirty data blocks counter */
1541 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1543 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1545 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1549 * Delayed allocation stuff
1552 struct mpage_da_data {
1553 /* These are input fields for ext4_do_writepages() */
1554 struct inode *inode;
1555 struct writeback_control *wbc;
1556 unsigned int can_map:1; /* Can writepages call map blocks? */
1558 /* These are internal state of ext4_do_writepages() */
1559 pgoff_t first_page; /* The first page to write */
1560 pgoff_t next_page; /* Current page to examine */
1561 pgoff_t last_page; /* Last page to examine */
1563 * Extent to map - this can be after first_page because that can be
1564 * fully mapped. We somewhat abuse m_flags to store whether the extent
1565 * is delalloc or unwritten.
1567 struct ext4_map_blocks map;
1568 struct ext4_io_submit io_submit; /* IO submission data */
1569 unsigned int do_map:1;
1570 unsigned int scanned_until_end:1;
1573 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1578 struct folio_batch fbatch;
1579 struct inode *inode = mpd->inode;
1580 struct address_space *mapping = inode->i_mapping;
1582 /* This is necessary when next_page == 0. */
1583 if (mpd->first_page >= mpd->next_page)
1586 mpd->scanned_until_end = 0;
1587 index = mpd->first_page;
1588 end = mpd->next_page - 1;
1590 ext4_lblk_t start, last;
1591 start = index << (PAGE_SHIFT - inode->i_blkbits);
1592 last = end << (PAGE_SHIFT - inode->i_blkbits);
1595 * avoid racing with extent status tree scans made by
1596 * ext4_insert_delayed_block()
1598 down_write(&EXT4_I(inode)->i_data_sem);
1599 ext4_es_remove_extent(inode, start, last - start + 1);
1600 up_write(&EXT4_I(inode)->i_data_sem);
1603 folio_batch_init(&fbatch);
1604 while (index <= end) {
1605 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1608 for (i = 0; i < nr; i++) {
1609 struct folio *folio = fbatch.folios[i];
1611 if (folio->index < mpd->first_page)
1613 if (folio->index + folio_nr_pages(folio) - 1 > end)
1615 BUG_ON(!folio_test_locked(folio));
1616 BUG_ON(folio_test_writeback(folio));
1618 if (folio_mapped(folio))
1619 folio_clear_dirty_for_io(folio);
1620 block_invalidate_folio(folio, 0,
1622 folio_clear_uptodate(folio);
1624 folio_unlock(folio);
1626 folio_batch_release(&fbatch);
1630 static void ext4_print_free_blocks(struct inode *inode)
1632 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1633 struct super_block *sb = inode->i_sb;
1634 struct ext4_inode_info *ei = EXT4_I(inode);
1636 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1637 EXT4_C2B(EXT4_SB(inode->i_sb),
1638 ext4_count_free_clusters(sb)));
1639 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1640 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1641 (long long) EXT4_C2B(EXT4_SB(sb),
1642 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1643 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1644 (long long) EXT4_C2B(EXT4_SB(sb),
1645 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1646 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1647 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1648 ei->i_reserved_data_blocks);
1652 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode,
1653 struct buffer_head *bh)
1655 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1659 * ext4_insert_delayed_block - adds a delayed block to the extents status
1660 * tree, incrementing the reserved cluster/block
1661 * count or making a pending reservation
1664 * @inode - file containing the newly added block
1665 * @lblk - logical block to be added
1667 * Returns 0 on success, negative error code on failure.
1669 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1671 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1673 bool allocated = false;
1674 bool reserved = false;
1677 * If the cluster containing lblk is shared with a delayed,
1678 * written, or unwritten extent in a bigalloc file system, it's
1679 * already been accounted for and does not need to be reserved.
1680 * A pending reservation must be made for the cluster if it's
1681 * shared with a written or unwritten extent and doesn't already
1682 * have one. Written and unwritten extents can be purged from the
1683 * extents status tree if the system is under memory pressure, so
1684 * it's necessary to examine the extent tree if a search of the
1685 * extents status tree doesn't get a match.
1687 if (sbi->s_cluster_ratio == 1) {
1688 ret = ext4_da_reserve_space(inode);
1689 if (ret != 0) /* ENOSPC */
1692 } else { /* bigalloc */
1693 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1694 if (!ext4_es_scan_clu(inode,
1695 &ext4_es_is_mapped, lblk)) {
1696 ret = ext4_clu_mapped(inode,
1697 EXT4_B2C(sbi, lblk));
1701 ret = ext4_da_reserve_space(inode);
1702 if (ret != 0) /* ENOSPC */
1714 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1715 if (ret && reserved)
1716 ext4_da_release_space(inode, 1);
1723 * This function is grabs code from the very beginning of
1724 * ext4_map_blocks, but assumes that the caller is from delayed write
1725 * time. This function looks up the requested blocks and sets the
1726 * buffer delay bit under the protection of i_data_sem.
1728 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1729 struct ext4_map_blocks *map,
1730 struct buffer_head *bh)
1732 struct extent_status es;
1734 sector_t invalid_block = ~((sector_t) 0xffff);
1735 #ifdef ES_AGGRESSIVE_TEST
1736 struct ext4_map_blocks orig_map;
1738 memcpy(&orig_map, map, sizeof(*map));
1741 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1745 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1746 (unsigned long) map->m_lblk);
1748 /* Lookup extent status tree firstly */
1749 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1750 if (ext4_es_is_hole(&es)) {
1752 down_read(&EXT4_I(inode)->i_data_sem);
1757 * Delayed extent could be allocated by fallocate.
1758 * So we need to check it.
1760 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1761 map_bh(bh, inode->i_sb, invalid_block);
1763 set_buffer_delay(bh);
1767 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1768 retval = es.es_len - (iblock - es.es_lblk);
1769 if (retval > map->m_len)
1770 retval = map->m_len;
1771 map->m_len = retval;
1772 if (ext4_es_is_written(&es))
1773 map->m_flags |= EXT4_MAP_MAPPED;
1774 else if (ext4_es_is_unwritten(&es))
1775 map->m_flags |= EXT4_MAP_UNWRITTEN;
1779 #ifdef ES_AGGRESSIVE_TEST
1780 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1786 * Try to see if we can get the block without requesting a new
1787 * file system block.
1789 down_read(&EXT4_I(inode)->i_data_sem);
1790 if (ext4_has_inline_data(inode))
1792 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1793 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1795 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1802 * XXX: __block_prepare_write() unmaps passed block,
1806 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1812 map_bh(bh, inode->i_sb, invalid_block);
1814 set_buffer_delay(bh);
1815 } else if (retval > 0) {
1817 unsigned int status;
1819 if (unlikely(retval != map->m_len)) {
1820 ext4_warning(inode->i_sb,
1821 "ES len assertion failed for inode "
1822 "%lu: retval %d != map->m_len %d",
1823 inode->i_ino, retval, map->m_len);
1827 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1828 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1829 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1830 map->m_pblk, status);
1836 up_read((&EXT4_I(inode)->i_data_sem));
1842 * This is a special get_block_t callback which is used by
1843 * ext4_da_write_begin(). It will either return mapped block or
1844 * reserve space for a single block.
1846 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1847 * We also have b_blocknr = -1 and b_bdev initialized properly
1849 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1850 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1851 * initialized properly.
1853 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1854 struct buffer_head *bh, int create)
1856 struct ext4_map_blocks map;
1859 BUG_ON(create == 0);
1860 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1862 map.m_lblk = iblock;
1866 * first, we need to know whether the block is allocated already
1867 * preallocated blocks are unmapped but should treated
1868 * the same as allocated blocks.
1870 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1874 map_bh(bh, inode->i_sb, map.m_pblk);
1875 ext4_update_bh_state(bh, map.m_flags);
1877 if (buffer_unwritten(bh)) {
1878 /* A delayed write to unwritten bh should be marked
1879 * new and mapped. Mapped ensures that we don't do
1880 * get_block multiple times when we write to the same
1881 * offset and new ensures that we do proper zero out
1882 * for partial write.
1885 set_buffer_mapped(bh);
1890 static int __ext4_journalled_writepage(struct page *page,
1893 struct address_space *mapping = page->mapping;
1894 struct inode *inode = mapping->host;
1895 handle_t *handle = NULL;
1896 int ret = 0, err = 0;
1897 int inline_data = ext4_has_inline_data(inode);
1898 struct buffer_head *inode_bh = NULL;
1901 ClearPageChecked(page);
1904 BUG_ON(page->index != 0);
1905 BUG_ON(len > ext4_get_max_inline_size(inode));
1906 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1907 if (inode_bh == NULL)
1911 * We need to release the page lock before we start the
1912 * journal, so grab a reference so the page won't disappear
1913 * out from under us.
1918 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1919 ext4_writepage_trans_blocks(inode));
1920 if (IS_ERR(handle)) {
1921 ret = PTR_ERR(handle);
1923 goto out_no_pagelock;
1925 BUG_ON(!ext4_handle_valid(handle));
1929 size = i_size_read(inode);
1930 if (page->mapping != mapping || page_offset(page) > size) {
1931 /* The page got truncated from under us */
1932 ext4_journal_stop(handle);
1938 ret = ext4_mark_inode_dirty(handle, inode);
1940 struct buffer_head *page_bufs = page_buffers(page);
1942 if (page->index == size >> PAGE_SHIFT)
1943 len = size & ~PAGE_MASK;
1947 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1948 NULL, do_journal_get_write_access);
1950 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
1951 NULL, write_end_fn);
1955 err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1958 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1959 err = ext4_journal_stop(handle);
1963 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1972 * Note that we don't need to start a transaction unless we're journaling data
1973 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1974 * need to file the inode to the transaction's list in ordered mode because if
1975 * we are writing back data added by write(), the inode is already there and if
1976 * we are writing back data modified via mmap(), no one guarantees in which
1977 * transaction the data will hit the disk. In case we are journaling data, we
1978 * cannot start transaction directly because transaction start ranks above page
1979 * lock so we have to do some magic.
1981 * This function can get called via...
1982 * - ext4_writepages after taking page lock (have journal handle)
1983 * - journal_submit_inode_data_buffers (no journal handle)
1984 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1985 * - grab_page_cache when doing write_begin (have journal handle)
1987 * We don't do any block allocation in this function. If we have page with
1988 * multiple blocks we need to write those buffer_heads that are mapped. This
1989 * is important for mmaped based write. So if we do with blocksize 1K
1990 * truncate(f, 1024);
1991 * a = mmap(f, 0, 4096);
1993 * truncate(f, 4096);
1994 * we have in the page first buffer_head mapped via page_mkwrite call back
1995 * but other buffer_heads would be unmapped but dirty (dirty done via the
1996 * do_wp_page). So writepage should write the first block. If we modify
1997 * the mmap area beyond 1024 we will again get a page_fault and the
1998 * page_mkwrite callback will do the block allocation and mark the
1999 * buffer_heads mapped.
2001 * We redirty the page if we have any buffer_heads that is either delay or
2002 * unwritten in the page.
2004 * We can get recursively called as show below.
2006 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2009 * But since we don't do any block allocation we should not deadlock.
2010 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2012 static int ext4_writepage(struct page *page,
2013 struct writeback_control *wbc)
2015 struct folio *folio = page_folio(page);
2019 struct buffer_head *page_bufs = NULL;
2020 struct inode *inode = page->mapping->host;
2021 struct ext4_io_submit io_submit;
2023 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2024 folio_invalidate(folio, 0, folio_size(folio));
2025 folio_unlock(folio);
2029 trace_ext4_writepage(page);
2030 size = i_size_read(inode);
2031 if (page->index == size >> PAGE_SHIFT &&
2032 !ext4_verity_in_progress(inode))
2033 len = size & ~PAGE_MASK;
2037 /* Should never happen but for bugs in other kernel subsystems */
2038 if (!page_has_buffers(page)) {
2039 ext4_warning_inode(inode,
2040 "page %lu does not have buffers attached", page->index);
2041 ClearPageDirty(page);
2046 page_bufs = page_buffers(page);
2048 * We cannot do block allocation or other extent handling in this
2049 * function. If there are buffers needing that, we have to redirty
2050 * the page. But we may reach here when we do a journal commit via
2051 * journal_submit_inode_data_buffers() and in that case we must write
2052 * allocated buffers to achieve data=ordered mode guarantees.
2054 * Also, if there is only one buffer per page (the fs block
2055 * size == the page size), if one buffer needs block
2056 * allocation or needs to modify the extent tree to clear the
2057 * unwritten flag, we know that the page can't be written at
2058 * all, so we might as well refuse the write immediately.
2059 * Unfortunately if the block size != page size, we can't as
2060 * easily detect this case using ext4_walk_page_buffers(), but
2061 * for the extremely common case, this is an optimization that
2062 * skips a useless round trip through ext4_bio_write_page().
2064 if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL,
2065 ext4_bh_delay_or_unwritten)) {
2066 redirty_page_for_writepage(wbc, page);
2067 if ((current->flags & PF_MEMALLOC) ||
2068 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2070 * For memory cleaning there's no point in writing only
2071 * some buffers. So just bail out. Warn if we came here
2072 * from direct reclaim.
2074 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2081 if (PageChecked(page) && ext4_should_journal_data(inode))
2083 * It's mmapped pagecache. Add buffers and journal it. There
2084 * doesn't seem much point in redirtying the page here.
2086 return __ext4_journalled_writepage(page, len);
2088 ext4_io_submit_init(&io_submit, wbc);
2089 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2090 if (!io_submit.io_end) {
2091 redirty_page_for_writepage(wbc, page);
2095 ret = ext4_bio_write_page(&io_submit, page, len);
2096 ext4_io_submit(&io_submit);
2097 /* Drop io_end reference we got from init */
2098 ext4_put_io_end_defer(io_submit.io_end);
2102 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2108 BUG_ON(page->index != mpd->first_page);
2109 clear_page_dirty_for_io(page);
2111 * We have to be very careful here! Nothing protects writeback path
2112 * against i_size changes and the page can be writeably mapped into
2113 * page tables. So an application can be growing i_size and writing
2114 * data through mmap while writeback runs. clear_page_dirty_for_io()
2115 * write-protects our page in page tables and the page cannot get
2116 * written to again until we release page lock. So only after
2117 * clear_page_dirty_for_io() we are safe to sample i_size for
2118 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2119 * on the barrier provided by TestClearPageDirty in
2120 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2121 * after page tables are updated.
2123 size = i_size_read(mpd->inode);
2124 if (page->index == size >> PAGE_SHIFT &&
2125 !ext4_verity_in_progress(mpd->inode))
2126 len = size & ~PAGE_MASK;
2129 err = ext4_bio_write_page(&mpd->io_submit, page, len);
2131 mpd->wbc->nr_to_write--;
2137 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2140 * mballoc gives us at most this number of blocks...
2141 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2142 * The rest of mballoc seems to handle chunks up to full group size.
2144 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2147 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2149 * @mpd - extent of blocks
2150 * @lblk - logical number of the block in the file
2151 * @bh - buffer head we want to add to the extent
2153 * The function is used to collect contig. blocks in the same state. If the
2154 * buffer doesn't require mapping for writeback and we haven't started the
2155 * extent of buffers to map yet, the function returns 'true' immediately - the
2156 * caller can write the buffer right away. Otherwise the function returns true
2157 * if the block has been added to the extent, false if the block couldn't be
2160 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2161 struct buffer_head *bh)
2163 struct ext4_map_blocks *map = &mpd->map;
2165 /* Buffer that doesn't need mapping for writeback? */
2166 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2167 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2168 /* So far no extent to map => we write the buffer right away */
2169 if (map->m_len == 0)
2174 /* First block in the extent? */
2175 if (map->m_len == 0) {
2176 /* We cannot map unless handle is started... */
2181 map->m_flags = bh->b_state & BH_FLAGS;
2185 /* Don't go larger than mballoc is willing to allocate */
2186 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2189 /* Can we merge the block to our big extent? */
2190 if (lblk == map->m_lblk + map->m_len &&
2191 (bh->b_state & BH_FLAGS) == map->m_flags) {
2199 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2201 * @mpd - extent of blocks for mapping
2202 * @head - the first buffer in the page
2203 * @bh - buffer we should start processing from
2204 * @lblk - logical number of the block in the file corresponding to @bh
2206 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2207 * the page for IO if all buffers in this page were mapped and there's no
2208 * accumulated extent of buffers to map or add buffers in the page to the
2209 * extent of buffers to map. The function returns 1 if the caller can continue
2210 * by processing the next page, 0 if it should stop adding buffers to the
2211 * extent to map because we cannot extend it anymore. It can also return value
2212 * < 0 in case of error during IO submission.
2214 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2215 struct buffer_head *head,
2216 struct buffer_head *bh,
2219 struct inode *inode = mpd->inode;
2221 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2222 >> inode->i_blkbits;
2224 if (ext4_verity_in_progress(inode))
2225 blocks = EXT_MAX_BLOCKS;
2228 BUG_ON(buffer_locked(bh));
2230 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2231 /* Found extent to map? */
2234 /* Buffer needs mapping and handle is not started? */
2237 /* Everything mapped so far and we hit EOF */
2240 } while (lblk++, (bh = bh->b_this_page) != head);
2241 /* So far everything mapped? Submit the page for IO. */
2242 if (mpd->map.m_len == 0) {
2243 err = mpage_submit_page(mpd, head->b_page);
2247 if (lblk >= blocks) {
2248 mpd->scanned_until_end = 1;
2255 * mpage_process_page - update page buffers corresponding to changed extent and
2256 * may submit fully mapped page for IO
2258 * @mpd - description of extent to map, on return next extent to map
2259 * @m_lblk - logical block mapping.
2260 * @m_pblk - corresponding physical mapping.
2261 * @map_bh - determines on return whether this page requires any further
2263 * Scan given page buffers corresponding to changed extent and update buffer
2264 * state according to new extent state.
2265 * We map delalloc buffers to their physical location, clear unwritten bits.
2266 * If the given page is not fully mapped, we update @map to the next extent in
2267 * the given page that needs mapping & return @map_bh as true.
2269 static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2270 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2273 struct buffer_head *head, *bh;
2274 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2275 ext4_lblk_t lblk = *m_lblk;
2276 ext4_fsblk_t pblock = *m_pblk;
2278 int blkbits = mpd->inode->i_blkbits;
2279 ssize_t io_end_size = 0;
2280 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2282 bh = head = page_buffers(page);
2284 if (lblk < mpd->map.m_lblk)
2286 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2288 * Buffer after end of mapped extent.
2289 * Find next buffer in the page to map.
2292 mpd->map.m_flags = 0;
2293 io_end_vec->size += io_end_size;
2295 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2298 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2299 io_end_vec = ext4_alloc_io_end_vec(io_end);
2300 if (IS_ERR(io_end_vec)) {
2301 err = PTR_ERR(io_end_vec);
2304 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2309 if (buffer_delay(bh)) {
2310 clear_buffer_delay(bh);
2311 bh->b_blocknr = pblock++;
2313 clear_buffer_unwritten(bh);
2314 io_end_size += (1 << blkbits);
2315 } while (lblk++, (bh = bh->b_this_page) != head);
2317 io_end_vec->size += io_end_size;
2326 * mpage_map_buffers - update buffers corresponding to changed extent and
2327 * submit fully mapped pages for IO
2329 * @mpd - description of extent to map, on return next extent to map
2331 * Scan buffers corresponding to changed extent (we expect corresponding pages
2332 * to be already locked) and update buffer state according to new extent state.
2333 * We map delalloc buffers to their physical location, clear unwritten bits,
2334 * and mark buffers as uninit when we perform writes to unwritten extents
2335 * and do extent conversion after IO is finished. If the last page is not fully
2336 * mapped, we update @map to the next extent in the last page that needs
2337 * mapping. Otherwise we submit the page for IO.
2339 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2341 struct folio_batch fbatch;
2343 struct inode *inode = mpd->inode;
2344 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2347 ext4_fsblk_t pblock;
2349 bool map_bh = false;
2351 start = mpd->map.m_lblk >> bpp_bits;
2352 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2353 lblk = start << bpp_bits;
2354 pblock = mpd->map.m_pblk;
2356 folio_batch_init(&fbatch);
2357 while (start <= end) {
2358 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2361 for (i = 0; i < nr; i++) {
2362 struct page *page = &fbatch.folios[i]->page;
2364 err = mpage_process_page(mpd, page, &lblk, &pblock,
2367 * If map_bh is true, means page may require further bh
2368 * mapping, or maybe the page was submitted for IO.
2369 * So we return to call further extent mapping.
2371 if (err < 0 || map_bh)
2373 /* Page fully mapped - let IO run! */
2374 err = mpage_submit_page(mpd, page);
2378 folio_batch_release(&fbatch);
2380 /* Extent fully mapped and matches with page boundary. We are done. */
2382 mpd->map.m_flags = 0;
2385 folio_batch_release(&fbatch);
2389 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2391 struct inode *inode = mpd->inode;
2392 struct ext4_map_blocks *map = &mpd->map;
2393 int get_blocks_flags;
2394 int err, dioread_nolock;
2396 trace_ext4_da_write_pages_extent(inode, map);
2398 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2399 * to convert an unwritten extent to be initialized (in the case
2400 * where we have written into one or more preallocated blocks). It is
2401 * possible that we're going to need more metadata blocks than
2402 * previously reserved. However we must not fail because we're in
2403 * writeback and there is nothing we can do about it so it might result
2404 * in data loss. So use reserved blocks to allocate metadata if
2407 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2408 * the blocks in question are delalloc blocks. This indicates
2409 * that the blocks and quotas has already been checked when
2410 * the data was copied into the page cache.
2412 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2413 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2414 EXT4_GET_BLOCKS_IO_SUBMIT;
2415 dioread_nolock = ext4_should_dioread_nolock(inode);
2417 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2418 if (map->m_flags & BIT(BH_Delay))
2419 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2421 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2424 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2425 if (!mpd->io_submit.io_end->handle &&
2426 ext4_handle_valid(handle)) {
2427 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2428 handle->h_rsv_handle = NULL;
2430 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2433 BUG_ON(map->m_len == 0);
2438 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2439 * mpd->len and submit pages underlying it for IO
2441 * @handle - handle for journal operations
2442 * @mpd - extent to map
2443 * @give_up_on_write - we set this to true iff there is a fatal error and there
2444 * is no hope of writing the data. The caller should discard
2445 * dirty pages to avoid infinite loops.
2447 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2448 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2449 * them to initialized or split the described range from larger unwritten
2450 * extent. Note that we need not map all the described range since allocation
2451 * can return less blocks or the range is covered by more unwritten extents. We
2452 * cannot map more because we are limited by reserved transaction credits. On
2453 * the other hand we always make sure that the last touched page is fully
2454 * mapped so that it can be written out (and thus forward progress is
2455 * guaranteed). After mapping we submit all mapped pages for IO.
2457 static int mpage_map_and_submit_extent(handle_t *handle,
2458 struct mpage_da_data *mpd,
2459 bool *give_up_on_write)
2461 struct inode *inode = mpd->inode;
2462 struct ext4_map_blocks *map = &mpd->map;
2466 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2467 struct ext4_io_end_vec *io_end_vec;
2469 io_end_vec = ext4_alloc_io_end_vec(io_end);
2470 if (IS_ERR(io_end_vec))
2471 return PTR_ERR(io_end_vec);
2472 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2474 err = mpage_map_one_extent(handle, mpd);
2476 struct super_block *sb = inode->i_sb;
2478 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2479 ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2480 goto invalidate_dirty_pages;
2482 * Let the uper layers retry transient errors.
2483 * In the case of ENOSPC, if ext4_count_free_blocks()
2484 * is non-zero, a commit should free up blocks.
2486 if ((err == -ENOMEM) ||
2487 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2489 goto update_disksize;
2492 ext4_msg(sb, KERN_CRIT,
2493 "Delayed block allocation failed for "
2494 "inode %lu at logical offset %llu with"
2495 " max blocks %u with error %d",
2497 (unsigned long long)map->m_lblk,
2498 (unsigned)map->m_len, -err);
2499 ext4_msg(sb, KERN_CRIT,
2500 "This should not happen!! Data will "
2503 ext4_print_free_blocks(inode);
2504 invalidate_dirty_pages:
2505 *give_up_on_write = true;
2510 * Update buffer state, submit mapped pages, and get us new
2513 err = mpage_map_and_submit_buffers(mpd);
2515 goto update_disksize;
2516 } while (map->m_len);
2520 * Update on-disk size after IO is submitted. Races with
2521 * truncate are avoided by checking i_size under i_data_sem.
2523 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2524 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2528 down_write(&EXT4_I(inode)->i_data_sem);
2529 i_size = i_size_read(inode);
2530 if (disksize > i_size)
2532 if (disksize > EXT4_I(inode)->i_disksize)
2533 EXT4_I(inode)->i_disksize = disksize;
2534 up_write(&EXT4_I(inode)->i_data_sem);
2535 err2 = ext4_mark_inode_dirty(handle, inode);
2537 ext4_error_err(inode->i_sb, -err2,
2538 "Failed to mark inode %lu dirty",
2548 * Calculate the total number of credits to reserve for one writepages
2549 * iteration. This is called from ext4_writepages(). We map an extent of
2550 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2551 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2552 * bpp - 1 blocks in bpp different extents.
2554 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2556 int bpp = ext4_journal_blocks_per_page(inode);
2558 return ext4_meta_trans_blocks(inode,
2559 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2562 /* Return true if the page needs to be written as part of transaction commit */
2563 static bool ext4_page_nomap_can_writeout(struct page *page)
2565 struct buffer_head *bh, *head;
2567 bh = head = page_buffers(page);
2569 if (buffer_dirty(bh) && buffer_mapped(bh) && !buffer_delay(bh))
2571 } while ((bh = bh->b_this_page) != head);
2576 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2577 * needing mapping, submit mapped pages
2579 * @mpd - where to look for pages
2581 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2582 * IO immediately. If we cannot map blocks, we submit just already mapped
2583 * buffers in the page for IO and keep page dirty. When we can map blocks and
2584 * we find a page which isn't mapped we start accumulating extent of buffers
2585 * underlying these pages that needs mapping (formed by either delayed or
2586 * unwritten buffers). We also lock the pages containing these buffers. The
2587 * extent found is returned in @mpd structure (starting at mpd->lblk with
2588 * length mpd->len blocks).
2590 * Note that this function can attach bios to one io_end structure which are
2591 * neither logically nor physically contiguous. Although it may seem as an
2592 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2593 * case as we need to track IO to all buffers underlying a page in one io_end.
2595 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2597 struct address_space *mapping = mpd->inode->i_mapping;
2598 struct pagevec pvec;
2599 unsigned int nr_pages;
2600 long left = mpd->wbc->nr_to_write;
2601 pgoff_t index = mpd->first_page;
2602 pgoff_t end = mpd->last_page;
2605 int blkbits = mpd->inode->i_blkbits;
2607 struct buffer_head *head;
2609 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2610 tag = PAGECACHE_TAG_TOWRITE;
2612 tag = PAGECACHE_TAG_DIRTY;
2614 pagevec_init(&pvec);
2616 mpd->next_page = index;
2617 while (index <= end) {
2618 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2623 for (i = 0; i < nr_pages; i++) {
2624 struct page *page = pvec.pages[i];
2627 * Accumulated enough dirty pages? This doesn't apply
2628 * to WB_SYNC_ALL mode. For integrity sync we have to
2629 * keep going because someone may be concurrently
2630 * dirtying pages, and we might have synced a lot of
2631 * newly appeared dirty pages, but have not synced all
2632 * of the old dirty pages.
2634 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2637 /* If we can't merge this page, we are done. */
2638 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2643 * If the page is no longer dirty, or its mapping no
2644 * longer corresponds to inode we are writing (which
2645 * means it has been truncated or invalidated), or the
2646 * page is already under writeback and we are not doing
2647 * a data integrity writeback, skip the page
2649 if (!PageDirty(page) ||
2650 (PageWriteback(page) &&
2651 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2652 unlikely(page->mapping != mapping)) {
2657 wait_on_page_writeback(page);
2658 BUG_ON(PageWriteback(page));
2661 * Should never happen but for buggy code in
2662 * other subsystems that call
2663 * set_page_dirty() without properly warning
2664 * the file system first. See [1] for more
2667 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2669 if (!page_has_buffers(page)) {
2670 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2671 ClearPageDirty(page);
2676 if (mpd->map.m_len == 0)
2677 mpd->first_page = page->index;
2678 mpd->next_page = page->index + 1;
2680 * Writeout for transaction commit where we cannot
2681 * modify metadata is simple. Just submit the page.
2683 if (!mpd->can_map) {
2684 if (ext4_page_nomap_can_writeout(page)) {
2685 err = mpage_submit_page(mpd, page);
2693 /* Add all dirty buffers to mpd */
2694 lblk = ((ext4_lblk_t)page->index) <<
2695 (PAGE_SHIFT - blkbits);
2696 head = page_buffers(page);
2697 err = mpage_process_page_bufs(mpd, head, head,
2705 pagevec_release(&pvec);
2708 mpd->scanned_until_end = 1;
2711 pagevec_release(&pvec);
2715 static int ext4_writepage_cb(struct page *page, struct writeback_control *wbc,
2718 return ext4_writepage(page, wbc);
2721 static int ext4_do_writepages(struct mpage_da_data *mpd)
2723 struct writeback_control *wbc = mpd->wbc;
2724 pgoff_t writeback_index = 0;
2725 long nr_to_write = wbc->nr_to_write;
2726 int range_whole = 0;
2728 handle_t *handle = NULL;
2729 struct inode *inode = mpd->inode;
2730 struct address_space *mapping = inode->i_mapping;
2731 int needed_blocks, rsv_blocks = 0, ret = 0;
2732 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2733 struct blk_plug plug;
2734 bool give_up_on_write = false;
2736 trace_ext4_writepages(inode, wbc);
2739 * No pages to write? This is mainly a kludge to avoid starting
2740 * a transaction for special inodes like journal inode on last iput()
2741 * because that could violate lock ordering on umount
2743 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2744 goto out_writepages;
2746 if (ext4_should_journal_data(inode)) {
2747 blk_start_plug(&plug);
2748 ret = write_cache_pages(mapping, wbc, ext4_writepage_cb, NULL);
2749 blk_finish_plug(&plug);
2750 goto out_writepages;
2754 * If the filesystem has aborted, it is read-only, so return
2755 * right away instead of dumping stack traces later on that
2756 * will obscure the real source of the problem. We test
2757 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2758 * the latter could be true if the filesystem is mounted
2759 * read-only, and in that case, ext4_writepages should
2760 * *never* be called, so if that ever happens, we would want
2763 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2764 ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2766 goto out_writepages;
2770 * If we have inline data and arrive here, it means that
2771 * we will soon create the block for the 1st page, so
2772 * we'd better clear the inline data here.
2774 if (ext4_has_inline_data(inode)) {
2775 /* Just inode will be modified... */
2776 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2777 if (IS_ERR(handle)) {
2778 ret = PTR_ERR(handle);
2779 goto out_writepages;
2781 BUG_ON(ext4_test_inode_state(inode,
2782 EXT4_STATE_MAY_INLINE_DATA));
2783 ext4_destroy_inline_data(handle, inode);
2784 ext4_journal_stop(handle);
2787 if (ext4_should_dioread_nolock(inode)) {
2789 * We may need to convert up to one extent per block in
2790 * the page and we may dirty the inode.
2792 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2793 PAGE_SIZE >> inode->i_blkbits);
2796 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2799 if (wbc->range_cyclic) {
2800 writeback_index = mapping->writeback_index;
2801 if (writeback_index)
2803 mpd->first_page = writeback_index;
2804 mpd->last_page = -1;
2806 mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2807 mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2810 ext4_io_submit_init(&mpd->io_submit, wbc);
2812 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2813 tag_pages_for_writeback(mapping, mpd->first_page,
2815 blk_start_plug(&plug);
2818 * First writeback pages that don't need mapping - we can avoid
2819 * starting a transaction unnecessarily and also avoid being blocked
2820 * in the block layer on device congestion while having transaction
2824 mpd->scanned_until_end = 0;
2825 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2826 if (!mpd->io_submit.io_end) {
2830 ret = mpage_prepare_extent_to_map(mpd);
2831 /* Unlock pages we didn't use */
2832 mpage_release_unused_pages(mpd, false);
2833 /* Submit prepared bio */
2834 ext4_io_submit(&mpd->io_submit);
2835 ext4_put_io_end_defer(mpd->io_submit.io_end);
2836 mpd->io_submit.io_end = NULL;
2840 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2841 /* For each extent of pages we use new io_end */
2842 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2843 if (!mpd->io_submit.io_end) {
2848 WARN_ON_ONCE(!mpd->can_map);
2850 * We have two constraints: We find one extent to map and we
2851 * must always write out whole page (makes a difference when
2852 * blocksize < pagesize) so that we don't block on IO when we
2853 * try to write out the rest of the page. Journalled mode is
2854 * not supported by delalloc.
2856 BUG_ON(ext4_should_journal_data(inode));
2857 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2859 /* start a new transaction */
2860 handle = ext4_journal_start_with_reserve(inode,
2861 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2862 if (IS_ERR(handle)) {
2863 ret = PTR_ERR(handle);
2864 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2865 "%ld pages, ino %lu; err %d", __func__,
2866 wbc->nr_to_write, inode->i_ino, ret);
2867 /* Release allocated io_end */
2868 ext4_put_io_end(mpd->io_submit.io_end);
2869 mpd->io_submit.io_end = NULL;
2874 trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2875 ret = mpage_prepare_extent_to_map(mpd);
2876 if (!ret && mpd->map.m_len)
2877 ret = mpage_map_and_submit_extent(handle, mpd,
2880 * Caution: If the handle is synchronous,
2881 * ext4_journal_stop() can wait for transaction commit
2882 * to finish which may depend on writeback of pages to
2883 * complete or on page lock to be released. In that
2884 * case, we have to wait until after we have
2885 * submitted all the IO, released page locks we hold,
2886 * and dropped io_end reference (for extent conversion
2887 * to be able to complete) before stopping the handle.
2889 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2890 ext4_journal_stop(handle);
2894 /* Unlock pages we didn't use */
2895 mpage_release_unused_pages(mpd, give_up_on_write);
2896 /* Submit prepared bio */
2897 ext4_io_submit(&mpd->io_submit);
2900 * Drop our io_end reference we got from init. We have
2901 * to be careful and use deferred io_end finishing if
2902 * we are still holding the transaction as we can
2903 * release the last reference to io_end which may end
2904 * up doing unwritten extent conversion.
2907 ext4_put_io_end_defer(mpd->io_submit.io_end);
2908 ext4_journal_stop(handle);
2910 ext4_put_io_end(mpd->io_submit.io_end);
2911 mpd->io_submit.io_end = NULL;
2913 if (ret == -ENOSPC && sbi->s_journal) {
2915 * Commit the transaction which would
2916 * free blocks released in the transaction
2919 jbd2_journal_force_commit_nested(sbi->s_journal);
2923 /* Fatal error - ENOMEM, EIO... */
2928 blk_finish_plug(&plug);
2929 if (!ret && !cycled && wbc->nr_to_write > 0) {
2931 mpd->last_page = writeback_index - 1;
2932 mpd->first_page = 0;
2937 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2939 * Set the writeback_index so that range_cyclic
2940 * mode will write it back later
2942 mapping->writeback_index = mpd->first_page;
2945 trace_ext4_writepages_result(inode, wbc, ret,
2946 nr_to_write - wbc->nr_to_write);
2950 static int ext4_writepages(struct address_space *mapping,
2951 struct writeback_control *wbc)
2953 struct super_block *sb = mapping->host->i_sb;
2954 struct mpage_da_data mpd = {
2955 .inode = mapping->host,
2961 if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
2964 percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
2965 ret = ext4_do_writepages(&mpd);
2966 percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
2971 int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2973 struct writeback_control wbc = {
2974 .sync_mode = WB_SYNC_ALL,
2975 .nr_to_write = LONG_MAX,
2976 .range_start = jinode->i_dirty_start,
2977 .range_end = jinode->i_dirty_end,
2979 struct mpage_da_data mpd = {
2980 .inode = jinode->i_vfs_inode,
2984 return ext4_do_writepages(&mpd);
2987 static int ext4_dax_writepages(struct address_space *mapping,
2988 struct writeback_control *wbc)
2991 long nr_to_write = wbc->nr_to_write;
2992 struct inode *inode = mapping->host;
2993 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2995 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2998 percpu_down_read(&sbi->s_writepages_rwsem);
2999 trace_ext4_writepages(inode, wbc);
3001 ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
3002 trace_ext4_writepages_result(inode, wbc, ret,
3003 nr_to_write - wbc->nr_to_write);
3004 percpu_up_read(&sbi->s_writepages_rwsem);
3008 static int ext4_nonda_switch(struct super_block *sb)
3010 s64 free_clusters, dirty_clusters;
3011 struct ext4_sb_info *sbi = EXT4_SB(sb);
3014 * switch to non delalloc mode if we are running low
3015 * on free block. The free block accounting via percpu
3016 * counters can get slightly wrong with percpu_counter_batch getting
3017 * accumulated on each CPU without updating global counters
3018 * Delalloc need an accurate free block accounting. So switch
3019 * to non delalloc when we are near to error range.
3022 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3024 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3026 * Start pushing delalloc when 1/2 of free blocks are dirty.
3028 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3029 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3031 if (2 * free_clusters < 3 * dirty_clusters ||
3032 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3034 * free block count is less than 150% of dirty blocks
3035 * or free blocks is less than watermark
3042 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3043 loff_t pos, unsigned len,
3044 struct page **pagep, void **fsdata)
3046 int ret, retries = 0;
3049 struct inode *inode = mapping->host;
3051 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3054 index = pos >> PAGE_SHIFT;
3056 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
3057 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3058 return ext4_write_begin(file, mapping, pos,
3059 len, pagep, fsdata);
3061 *fsdata = (void *)0;
3062 trace_ext4_da_write_begin(inode, pos, len);
3064 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3065 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
3074 page = grab_cache_page_write_begin(mapping, index);
3078 /* In case writeback began while the page was unlocked */
3079 wait_for_stable_page(page);
3081 #ifdef CONFIG_FS_ENCRYPTION
3082 ret = ext4_block_write_begin(page, pos, len,
3083 ext4_da_get_block_prep);
3085 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3091 * block_write_begin may have instantiated a few blocks
3092 * outside i_size. Trim these off again. Don't need
3093 * i_size_read because we hold inode lock.
3095 if (pos + len > inode->i_size)
3096 ext4_truncate_failed_write(inode);
3098 if (ret == -ENOSPC &&
3099 ext4_should_retry_alloc(inode->i_sb, &retries))
3109 * Check if we should update i_disksize
3110 * when write to the end of file but not require block allocation
3112 static int ext4_da_should_update_i_disksize(struct page *page,
3113 unsigned long offset)
3115 struct buffer_head *bh;
3116 struct inode *inode = page->mapping->host;
3120 bh = page_buffers(page);
3121 idx = offset >> inode->i_blkbits;
3123 for (i = 0; i < idx; i++)
3124 bh = bh->b_this_page;
3126 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3131 static int ext4_da_write_end(struct file *file,
3132 struct address_space *mapping,
3133 loff_t pos, unsigned len, unsigned copied,
3134 struct page *page, void *fsdata)
3136 struct inode *inode = mapping->host;
3138 unsigned long start, end;
3139 int write_mode = (int)(unsigned long)fsdata;
3141 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3142 return ext4_write_end(file, mapping, pos,
3143 len, copied, page, fsdata);
3145 trace_ext4_da_write_end(inode, pos, len, copied);
3147 if (write_mode != CONVERT_INLINE_DATA &&
3148 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3149 ext4_has_inline_data(inode))
3150 return ext4_write_inline_data_end(inode, pos, len, copied, page);
3152 start = pos & (PAGE_SIZE - 1);
3153 end = start + copied - 1;
3156 * Since we are holding inode lock, we are sure i_disksize <=
3157 * i_size. We also know that if i_disksize < i_size, there are
3158 * delalloc writes pending in the range upto i_size. If the end of
3159 * the current write is <= i_size, there's no need to touch
3160 * i_disksize since writeback will push i_disksize upto i_size
3161 * eventually. If the end of the current write is > i_size and
3162 * inside an allocated block (ext4_da_should_update_i_disksize()
3163 * check), we need to update i_disksize here as neither
3164 * ext4_writepage() nor certain ext4_writepages() paths not
3165 * allocating blocks update i_disksize.
3167 * Note that we defer inode dirtying to generic_write_end() /
3168 * ext4_da_write_inline_data_end().
3170 new_i_size = pos + copied;
3171 if (copied && new_i_size > inode->i_size &&
3172 ext4_da_should_update_i_disksize(page, end))
3173 ext4_update_i_disksize(inode, new_i_size);
3175 return generic_write_end(file, mapping, pos, len, copied, page, fsdata);
3179 * Force all delayed allocation blocks to be allocated for a given inode.
3181 int ext4_alloc_da_blocks(struct inode *inode)
3183 trace_ext4_alloc_da_blocks(inode);
3185 if (!EXT4_I(inode)->i_reserved_data_blocks)
3189 * We do something simple for now. The filemap_flush() will
3190 * also start triggering a write of the data blocks, which is
3191 * not strictly speaking necessary (and for users of
3192 * laptop_mode, not even desirable). However, to do otherwise
3193 * would require replicating code paths in:
3195 * ext4_writepages() ->
3196 * write_cache_pages() ---> (via passed in callback function)
3197 * __mpage_da_writepage() -->
3198 * mpage_add_bh_to_extent()
3199 * mpage_da_map_blocks()
3201 * The problem is that write_cache_pages(), located in
3202 * mm/page-writeback.c, marks pages clean in preparation for
3203 * doing I/O, which is not desirable if we're not planning on
3206 * We could call write_cache_pages(), and then redirty all of
3207 * the pages by calling redirty_page_for_writepage() but that
3208 * would be ugly in the extreme. So instead we would need to
3209 * replicate parts of the code in the above functions,
3210 * simplifying them because we wouldn't actually intend to
3211 * write out the pages, but rather only collect contiguous
3212 * logical block extents, call the multi-block allocator, and
3213 * then update the buffer heads with the block allocations.
3215 * For now, though, we'll cheat by calling filemap_flush(),
3216 * which will map the blocks, and start the I/O, but not
3217 * actually wait for the I/O to complete.
3219 return filemap_flush(inode->i_mapping);
3223 * bmap() is special. It gets used by applications such as lilo and by
3224 * the swapper to find the on-disk block of a specific piece of data.
3226 * Naturally, this is dangerous if the block concerned is still in the
3227 * journal. If somebody makes a swapfile on an ext4 data-journaling
3228 * filesystem and enables swap, then they may get a nasty shock when the
3229 * data getting swapped to that swapfile suddenly gets overwritten by
3230 * the original zero's written out previously to the journal and
3231 * awaiting writeback in the kernel's buffer cache.
3233 * So, if we see any bmap calls here on a modified, data-journaled file,
3234 * take extra steps to flush any blocks which might be in the cache.
3236 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3238 struct inode *inode = mapping->host;
3243 inode_lock_shared(inode);
3245 * We can get here for an inline file via the FIBMAP ioctl
3247 if (ext4_has_inline_data(inode))
3250 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3251 test_opt(inode->i_sb, DELALLOC)) {
3253 * With delalloc we want to sync the file
3254 * so that we can make sure we allocate
3257 filemap_write_and_wait(mapping);
3260 if (EXT4_JOURNAL(inode) &&
3261 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3263 * This is a REALLY heavyweight approach, but the use of
3264 * bmap on dirty files is expected to be extremely rare:
3265 * only if we run lilo or swapon on a freshly made file
3266 * do we expect this to happen.
3268 * (bmap requires CAP_SYS_RAWIO so this does not
3269 * represent an unprivileged user DOS attack --- we'd be
3270 * in trouble if mortal users could trigger this path at
3273 * NB. EXT4_STATE_JDATA is not set on files other than
3274 * regular files. If somebody wants to bmap a directory
3275 * or symlink and gets confused because the buffer
3276 * hasn't yet been flushed to disk, they deserve
3277 * everything they get.
3280 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3281 journal = EXT4_JOURNAL(inode);
3282 jbd2_journal_lock_updates(journal);
3283 err = jbd2_journal_flush(journal, 0);
3284 jbd2_journal_unlock_updates(journal);
3290 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3293 inode_unlock_shared(inode);
3297 static int ext4_read_folio(struct file *file, struct folio *folio)
3299 struct page *page = &folio->page;
3301 struct inode *inode = page->mapping->host;
3303 trace_ext4_readpage(page);
3305 if (ext4_has_inline_data(inode))
3306 ret = ext4_readpage_inline(inode, page);
3309 return ext4_mpage_readpages(inode, NULL, page);
3314 static void ext4_readahead(struct readahead_control *rac)
3316 struct inode *inode = rac->mapping->host;
3318 /* If the file has inline data, no need to do readahead. */
3319 if (ext4_has_inline_data(inode))
3322 ext4_mpage_readpages(inode, rac, NULL);
3325 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3328 trace_ext4_invalidate_folio(folio, offset, length);
3330 /* No journalling happens on data buffers when this function is used */
3331 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3333 block_invalidate_folio(folio, offset, length);
3336 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3337 size_t offset, size_t length)
3339 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3341 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3344 * If it's a full truncate we just forget about the pending dirtying
3346 if (offset == 0 && length == folio_size(folio))
3347 folio_clear_checked(folio);
3349 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3352 /* Wrapper for aops... */
3353 static void ext4_journalled_invalidate_folio(struct folio *folio,
3357 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3360 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3362 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3364 trace_ext4_releasepage(&folio->page);
3366 /* Page has dirty journalled data -> cannot release */
3367 if (folio_test_checked(folio))
3370 return jbd2_journal_try_to_free_buffers(journal, folio);
3372 return try_to_free_buffers(folio);
3375 static bool ext4_inode_datasync_dirty(struct inode *inode)
3377 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3380 if (jbd2_transaction_committed(journal,
3381 EXT4_I(inode)->i_datasync_tid))
3383 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3384 return !list_empty(&EXT4_I(inode)->i_fc_list);
3388 /* Any metadata buffers to write? */
3389 if (!list_empty(&inode->i_mapping->private_list))
3391 return inode->i_state & I_DIRTY_DATASYNC;
3394 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3395 struct ext4_map_blocks *map, loff_t offset,
3396 loff_t length, unsigned int flags)
3398 u8 blkbits = inode->i_blkbits;
3401 * Writes that span EOF might trigger an I/O size update on completion,
3402 * so consider them to be dirty for the purpose of O_DSYNC, even if
3403 * there is no other metadata changes being made or are pending.
3406 if (ext4_inode_datasync_dirty(inode) ||
3407 offset + length > i_size_read(inode))
3408 iomap->flags |= IOMAP_F_DIRTY;
3410 if (map->m_flags & EXT4_MAP_NEW)
3411 iomap->flags |= IOMAP_F_NEW;
3413 if (flags & IOMAP_DAX)
3414 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3416 iomap->bdev = inode->i_sb->s_bdev;
3417 iomap->offset = (u64) map->m_lblk << blkbits;
3418 iomap->length = (u64) map->m_len << blkbits;
3420 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3421 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3422 iomap->flags |= IOMAP_F_MERGED;
3425 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3426 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3427 * set. In order for any allocated unwritten extents to be converted
3428 * into written extents correctly within the ->end_io() handler, we
3429 * need to ensure that the iomap->type is set appropriately. Hence, the
3430 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3433 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3434 iomap->type = IOMAP_UNWRITTEN;
3435 iomap->addr = (u64) map->m_pblk << blkbits;
3436 if (flags & IOMAP_DAX)
3437 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3438 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3439 iomap->type = IOMAP_MAPPED;
3440 iomap->addr = (u64) map->m_pblk << blkbits;
3441 if (flags & IOMAP_DAX)
3442 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3444 iomap->type = IOMAP_HOLE;
3445 iomap->addr = IOMAP_NULL_ADDR;
3449 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3453 u8 blkbits = inode->i_blkbits;
3454 int ret, dio_credits, m_flags = 0, retries = 0;
3457 * Trim the mapping request to the maximum value that we can map at
3458 * once for direct I/O.
3460 if (map->m_len > DIO_MAX_BLOCKS)
3461 map->m_len = DIO_MAX_BLOCKS;
3462 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3466 * Either we allocate blocks and then don't get an unwritten extent, so
3467 * in that case we have reserved enough credits. Or, the blocks are
3468 * already allocated and unwritten. In that case, the extent conversion
3469 * fits into the credits as well.
3471 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3473 return PTR_ERR(handle);
3476 * DAX and direct I/O are the only two operations that are currently
3477 * supported with IOMAP_WRITE.
3479 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3480 if (flags & IOMAP_DAX)
3481 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3483 * We use i_size instead of i_disksize here because delalloc writeback
3484 * can complete at any point during the I/O and subsequently push the
3485 * i_disksize out to i_size. This could be beyond where direct I/O is
3486 * happening and thus expose allocated blocks to direct I/O reads.
3488 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3489 m_flags = EXT4_GET_BLOCKS_CREATE;
3490 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3491 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3493 ret = ext4_map_blocks(handle, inode, map, m_flags);
3496 * We cannot fill holes in indirect tree based inodes as that could
3497 * expose stale data in the case of a crash. Use the magic error code
3498 * to fallback to buffered I/O.
3500 if (!m_flags && !ret)
3503 ext4_journal_stop(handle);
3504 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3511 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3512 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3515 struct ext4_map_blocks map;
3516 u8 blkbits = inode->i_blkbits;
3518 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3521 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3525 * Calculate the first and last logical blocks respectively.
3527 map.m_lblk = offset >> blkbits;
3528 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3529 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3531 if (flags & IOMAP_WRITE) {
3533 * We check here if the blocks are already allocated, then we
3534 * don't need to start a journal txn and we can directly return
3535 * the mapping information. This could boost performance
3536 * especially in multi-threaded overwrite requests.
3538 if (offset + length <= i_size_read(inode)) {
3539 ret = ext4_map_blocks(NULL, inode, &map, 0);
3540 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3543 ret = ext4_iomap_alloc(inode, &map, flags);
3545 ret = ext4_map_blocks(NULL, inode, &map, 0);
3552 * When inline encryption is enabled, sometimes I/O to an encrypted file
3553 * has to be broken up to guarantee DUN contiguity. Handle this by
3554 * limiting the length of the mapping returned.
3556 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3558 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3563 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3564 loff_t length, unsigned flags, struct iomap *iomap,
3565 struct iomap *srcmap)
3570 * Even for writes we don't need to allocate blocks, so just pretend
3571 * we are reading to save overhead of starting a transaction.
3573 flags &= ~IOMAP_WRITE;
3574 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3575 WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3579 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3580 ssize_t written, unsigned flags, struct iomap *iomap)
3583 * Check to see whether an error occurred while writing out the data to
3584 * the allocated blocks. If so, return the magic error code so that we
3585 * fallback to buffered I/O and attempt to complete the remainder of
3586 * the I/O. Any blocks that may have been allocated in preparation for
3587 * the direct I/O will be reused during buffered I/O.
3589 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3595 const struct iomap_ops ext4_iomap_ops = {
3596 .iomap_begin = ext4_iomap_begin,
3597 .iomap_end = ext4_iomap_end,
3600 const struct iomap_ops ext4_iomap_overwrite_ops = {
3601 .iomap_begin = ext4_iomap_overwrite_begin,
3602 .iomap_end = ext4_iomap_end,
3605 static bool ext4_iomap_is_delalloc(struct inode *inode,
3606 struct ext4_map_blocks *map)
3608 struct extent_status es;
3609 ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3611 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3612 map->m_lblk, end, &es);
3614 if (!es.es_len || es.es_lblk > end)
3617 if (es.es_lblk > map->m_lblk) {
3618 map->m_len = es.es_lblk - map->m_lblk;
3622 offset = map->m_lblk - es.es_lblk;
3623 map->m_len = es.es_len - offset;
3628 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3629 loff_t length, unsigned int flags,
3630 struct iomap *iomap, struct iomap *srcmap)
3633 bool delalloc = false;
3634 struct ext4_map_blocks map;
3635 u8 blkbits = inode->i_blkbits;
3637 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3640 if (ext4_has_inline_data(inode)) {
3641 ret = ext4_inline_data_iomap(inode, iomap);
3642 if (ret != -EAGAIN) {
3643 if (ret == 0 && offset >= iomap->length)
3650 * Calculate the first and last logical block respectively.
3652 map.m_lblk = offset >> blkbits;
3653 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3654 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3657 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3658 * So handle it here itself instead of querying ext4_map_blocks().
3659 * Since ext4_map_blocks() will warn about it and will return
3662 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3663 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3665 if (offset >= sbi->s_bitmap_maxbytes) {
3671 ret = ext4_map_blocks(NULL, inode, &map, 0);
3675 delalloc = ext4_iomap_is_delalloc(inode, &map);
3678 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3679 if (delalloc && iomap->type == IOMAP_HOLE)
3680 iomap->type = IOMAP_DELALLOC;
3685 const struct iomap_ops ext4_iomap_report_ops = {
3686 .iomap_begin = ext4_iomap_begin_report,
3690 * Whenever the folio is being dirtied, corresponding buffers should already
3691 * be attached to the transaction (we take care of this in ext4_page_mkwrite()
3692 * and ext4_write_begin()). However we cannot move buffers to dirty transaction
3693 * lists here because ->dirty_folio is called under VFS locks and the folio
3694 * is not necessarily locked.
3696 * We cannot just dirty the folio and leave attached buffers clean, because the
3697 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3698 * or jbddirty because all the journalling code will explode.
3700 * So what we do is to mark the folio "pending dirty" and next time writepage
3701 * is called, propagate that into the buffers appropriately.
3703 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3704 struct folio *folio)
3706 WARN_ON_ONCE(!folio_buffers(folio));
3707 folio_set_checked(folio);
3708 return filemap_dirty_folio(mapping, folio);
3711 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3713 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3714 WARN_ON_ONCE(!folio_buffers(folio));
3715 return block_dirty_folio(mapping, folio);
3718 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3719 struct file *file, sector_t *span)
3721 return iomap_swapfile_activate(sis, file, span,
3722 &ext4_iomap_report_ops);
3725 static const struct address_space_operations ext4_aops = {
3726 .read_folio = ext4_read_folio,
3727 .readahead = ext4_readahead,
3728 .writepages = ext4_writepages,
3729 .write_begin = ext4_write_begin,
3730 .write_end = ext4_write_end,
3731 .dirty_folio = ext4_dirty_folio,
3733 .invalidate_folio = ext4_invalidate_folio,
3734 .release_folio = ext4_release_folio,
3735 .direct_IO = noop_direct_IO,
3736 .migrate_folio = buffer_migrate_folio,
3737 .is_partially_uptodate = block_is_partially_uptodate,
3738 .error_remove_page = generic_error_remove_page,
3739 .swap_activate = ext4_iomap_swap_activate,
3742 static const struct address_space_operations ext4_journalled_aops = {
3743 .read_folio = ext4_read_folio,
3744 .readahead = ext4_readahead,
3745 .writepages = ext4_writepages,
3746 .write_begin = ext4_write_begin,
3747 .write_end = ext4_journalled_write_end,
3748 .dirty_folio = ext4_journalled_dirty_folio,
3750 .invalidate_folio = ext4_journalled_invalidate_folio,
3751 .release_folio = ext4_release_folio,
3752 .direct_IO = noop_direct_IO,
3753 .migrate_folio = buffer_migrate_folio_norefs,
3754 .is_partially_uptodate = block_is_partially_uptodate,
3755 .error_remove_page = generic_error_remove_page,
3756 .swap_activate = ext4_iomap_swap_activate,
3759 static const struct address_space_operations ext4_da_aops = {
3760 .read_folio = ext4_read_folio,
3761 .readahead = ext4_readahead,
3762 .writepages = ext4_writepages,
3763 .write_begin = ext4_da_write_begin,
3764 .write_end = ext4_da_write_end,
3765 .dirty_folio = ext4_dirty_folio,
3767 .invalidate_folio = ext4_invalidate_folio,
3768 .release_folio = ext4_release_folio,
3769 .direct_IO = noop_direct_IO,
3770 .migrate_folio = buffer_migrate_folio,
3771 .is_partially_uptodate = block_is_partially_uptodate,
3772 .error_remove_page = generic_error_remove_page,
3773 .swap_activate = ext4_iomap_swap_activate,
3776 static const struct address_space_operations ext4_dax_aops = {
3777 .writepages = ext4_dax_writepages,
3778 .direct_IO = noop_direct_IO,
3779 .dirty_folio = noop_dirty_folio,
3781 .swap_activate = ext4_iomap_swap_activate,
3784 void ext4_set_aops(struct inode *inode)
3786 switch (ext4_inode_journal_mode(inode)) {
3787 case EXT4_INODE_ORDERED_DATA_MODE:
3788 case EXT4_INODE_WRITEBACK_DATA_MODE:
3790 case EXT4_INODE_JOURNAL_DATA_MODE:
3791 inode->i_mapping->a_ops = &ext4_journalled_aops;
3797 inode->i_mapping->a_ops = &ext4_dax_aops;
3798 else if (test_opt(inode->i_sb, DELALLOC))
3799 inode->i_mapping->a_ops = &ext4_da_aops;
3801 inode->i_mapping->a_ops = &ext4_aops;
3804 static int __ext4_block_zero_page_range(handle_t *handle,
3805 struct address_space *mapping, loff_t from, loff_t length)
3807 ext4_fsblk_t index = from >> PAGE_SHIFT;
3808 unsigned offset = from & (PAGE_SIZE-1);
3809 unsigned blocksize, pos;
3811 struct inode *inode = mapping->host;
3812 struct buffer_head *bh;
3816 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3817 mapping_gfp_constraint(mapping, ~__GFP_FS));
3821 blocksize = inode->i_sb->s_blocksize;
3823 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3825 if (!page_has_buffers(page))
3826 create_empty_buffers(page, blocksize, 0);
3828 /* Find the buffer that contains "offset" */
3829 bh = page_buffers(page);
3831 while (offset >= pos) {
3832 bh = bh->b_this_page;
3836 if (buffer_freed(bh)) {
3837 BUFFER_TRACE(bh, "freed: skip");
3840 if (!buffer_mapped(bh)) {
3841 BUFFER_TRACE(bh, "unmapped");
3842 ext4_get_block(inode, iblock, bh, 0);
3843 /* unmapped? It's a hole - nothing to do */
3844 if (!buffer_mapped(bh)) {
3845 BUFFER_TRACE(bh, "still unmapped");
3850 /* Ok, it's mapped. Make sure it's up-to-date */
3851 if (PageUptodate(page))
3852 set_buffer_uptodate(bh);
3854 if (!buffer_uptodate(bh)) {
3855 err = ext4_read_bh_lock(bh, 0, true);
3858 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3859 /* We expect the key to be set. */
3860 BUG_ON(!fscrypt_has_encryption_key(inode));
3861 err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3864 clear_buffer_uptodate(bh);
3869 if (ext4_should_journal_data(inode)) {
3870 BUFFER_TRACE(bh, "get write access");
3871 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3876 zero_user(page, offset, length);
3877 BUFFER_TRACE(bh, "zeroed end of block");
3879 if (ext4_should_journal_data(inode)) {
3880 err = ext4_handle_dirty_metadata(handle, inode, bh);
3883 mark_buffer_dirty(bh);
3884 if (ext4_should_order_data(inode))
3885 err = ext4_jbd2_inode_add_write(handle, inode, from,
3896 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3897 * starting from file offset 'from'. The range to be zero'd must
3898 * be contained with in one block. If the specified range exceeds
3899 * the end of the block it will be shortened to end of the block
3900 * that corresponds to 'from'
3902 static int ext4_block_zero_page_range(handle_t *handle,
3903 struct address_space *mapping, loff_t from, loff_t length)
3905 struct inode *inode = mapping->host;
3906 unsigned offset = from & (PAGE_SIZE-1);
3907 unsigned blocksize = inode->i_sb->s_blocksize;
3908 unsigned max = blocksize - (offset & (blocksize - 1));
3911 * correct length if it does not fall between
3912 * 'from' and the end of the block
3914 if (length > max || length < 0)
3917 if (IS_DAX(inode)) {
3918 return dax_zero_range(inode, from, length, NULL,
3921 return __ext4_block_zero_page_range(handle, mapping, from, length);
3925 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3926 * up to the end of the block which corresponds to `from'.
3927 * This required during truncate. We need to physically zero the tail end
3928 * of that block so it doesn't yield old data if the file is later grown.
3930 static int ext4_block_truncate_page(handle_t *handle,
3931 struct address_space *mapping, loff_t from)
3933 unsigned offset = from & (PAGE_SIZE-1);
3936 struct inode *inode = mapping->host;
3938 /* If we are processing an encrypted inode during orphan list handling */
3939 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3942 blocksize = inode->i_sb->s_blocksize;
3943 length = blocksize - (offset & (blocksize - 1));
3945 return ext4_block_zero_page_range(handle, mapping, from, length);
3948 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3949 loff_t lstart, loff_t length)
3951 struct super_block *sb = inode->i_sb;
3952 struct address_space *mapping = inode->i_mapping;
3953 unsigned partial_start, partial_end;
3954 ext4_fsblk_t start, end;
3955 loff_t byte_end = (lstart + length - 1);
3958 partial_start = lstart & (sb->s_blocksize - 1);
3959 partial_end = byte_end & (sb->s_blocksize - 1);
3961 start = lstart >> sb->s_blocksize_bits;
3962 end = byte_end >> sb->s_blocksize_bits;
3964 /* Handle partial zero within the single block */
3966 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3967 err = ext4_block_zero_page_range(handle, mapping,
3971 /* Handle partial zero out on the start of the range */
3972 if (partial_start) {
3973 err = ext4_block_zero_page_range(handle, mapping,
3974 lstart, sb->s_blocksize);
3978 /* Handle partial zero out on the end of the range */
3979 if (partial_end != sb->s_blocksize - 1)
3980 err = ext4_block_zero_page_range(handle, mapping,
3981 byte_end - partial_end,
3986 int ext4_can_truncate(struct inode *inode)
3988 if (S_ISREG(inode->i_mode))
3990 if (S_ISDIR(inode->i_mode))
3992 if (S_ISLNK(inode->i_mode))
3993 return !ext4_inode_is_fast_symlink(inode);
3998 * We have to make sure i_disksize gets properly updated before we truncate
3999 * page cache due to hole punching or zero range. Otherwise i_disksize update
4000 * can get lost as it may have been postponed to submission of writeback but
4001 * that will never happen after we truncate page cache.
4003 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4009 loff_t size = i_size_read(inode);
4011 WARN_ON(!inode_is_locked(inode));
4012 if (offset > size || offset + len < size)
4015 if (EXT4_I(inode)->i_disksize >= size)
4018 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4020 return PTR_ERR(handle);
4021 ext4_update_i_disksize(inode, size);
4022 ret = ext4_mark_inode_dirty(handle, inode);
4023 ext4_journal_stop(handle);
4028 static void ext4_wait_dax_page(struct inode *inode)
4030 filemap_invalidate_unlock(inode->i_mapping);
4032 filemap_invalidate_lock(inode->i_mapping);
4035 int ext4_break_layouts(struct inode *inode)
4040 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
4044 page = dax_layout_busy_page(inode->i_mapping);
4048 error = ___wait_var_event(&page->_refcount,
4049 atomic_read(&page->_refcount) == 1,
4050 TASK_INTERRUPTIBLE, 0, 0,
4051 ext4_wait_dax_page(inode));
4052 } while (error == 0);
4058 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4059 * associated with the given offset and length
4061 * @inode: File inode
4062 * @offset: The offset where the hole will begin
4063 * @len: The length of the hole
4065 * Returns: 0 on success or negative on failure
4068 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4070 struct inode *inode = file_inode(file);
4071 struct super_block *sb = inode->i_sb;
4072 ext4_lblk_t first_block, stop_block;
4073 struct address_space *mapping = inode->i_mapping;
4074 loff_t first_block_offset, last_block_offset, max_length;
4075 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4077 unsigned int credits;
4078 int ret = 0, ret2 = 0;
4080 trace_ext4_punch_hole(inode, offset, length, 0);
4083 * Write out all dirty pages to avoid race conditions
4084 * Then release them.
4086 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4087 ret = filemap_write_and_wait_range(mapping, offset,
4088 offset + length - 1);
4095 /* No need to punch hole beyond i_size */
4096 if (offset >= inode->i_size)
4100 * If the hole extends beyond i_size, set the hole
4101 * to end after the page that contains i_size
4103 if (offset + length > inode->i_size) {
4104 length = inode->i_size +
4105 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4110 * For punch hole the length + offset needs to be within one block
4111 * before last range. Adjust the length if it goes beyond that limit.
4113 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4114 if (offset + length > max_length)
4115 length = max_length - offset;
4117 if (offset & (sb->s_blocksize - 1) ||
4118 (offset + length) & (sb->s_blocksize - 1)) {
4120 * Attach jinode to inode for jbd2 if we do any zeroing of
4123 ret = ext4_inode_attach_jinode(inode);
4129 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4130 inode_dio_wait(inode);
4132 ret = file_modified(file);
4137 * Prevent page faults from reinstantiating pages we have released from
4140 filemap_invalidate_lock(mapping);
4142 ret = ext4_break_layouts(inode);
4146 first_block_offset = round_up(offset, sb->s_blocksize);
4147 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4149 /* Now release the pages and zero block aligned part of pages*/
4150 if (last_block_offset > first_block_offset) {
4151 ret = ext4_update_disksize_before_punch(inode, offset, length);
4154 truncate_pagecache_range(inode, first_block_offset,
4158 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4159 credits = ext4_writepage_trans_blocks(inode);
4161 credits = ext4_blocks_for_truncate(inode);
4162 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4163 if (IS_ERR(handle)) {
4164 ret = PTR_ERR(handle);
4165 ext4_std_error(sb, ret);
4169 ret = ext4_zero_partial_blocks(handle, inode, offset,
4174 first_block = (offset + sb->s_blocksize - 1) >>
4175 EXT4_BLOCK_SIZE_BITS(sb);
4176 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4178 /* If there are blocks to remove, do it */
4179 if (stop_block > first_block) {
4181 down_write(&EXT4_I(inode)->i_data_sem);
4182 ext4_discard_preallocations(inode, 0);
4184 ret = ext4_es_remove_extent(inode, first_block,
4185 stop_block - first_block);
4187 up_write(&EXT4_I(inode)->i_data_sem);
4191 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4192 ret = ext4_ext_remove_space(inode, first_block,
4195 ret = ext4_ind_remove_space(handle, inode, first_block,
4198 up_write(&EXT4_I(inode)->i_data_sem);
4200 ext4_fc_track_range(handle, inode, first_block, stop_block);
4202 ext4_handle_sync(handle);
4204 inode->i_mtime = inode->i_ctime = current_time(inode);
4205 ret2 = ext4_mark_inode_dirty(handle, inode);
4209 ext4_update_inode_fsync_trans(handle, inode, 1);
4211 ext4_journal_stop(handle);
4213 filemap_invalidate_unlock(mapping);
4215 inode_unlock(inode);
4219 int ext4_inode_attach_jinode(struct inode *inode)
4221 struct ext4_inode_info *ei = EXT4_I(inode);
4222 struct jbd2_inode *jinode;
4224 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4227 jinode = jbd2_alloc_inode(GFP_KERNEL);
4228 spin_lock(&inode->i_lock);
4231 spin_unlock(&inode->i_lock);
4234 ei->jinode = jinode;
4235 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4238 spin_unlock(&inode->i_lock);
4239 if (unlikely(jinode != NULL))
4240 jbd2_free_inode(jinode);
4247 * We block out ext4_get_block() block instantiations across the entire
4248 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4249 * simultaneously on behalf of the same inode.
4251 * As we work through the truncate and commit bits of it to the journal there
4252 * is one core, guiding principle: the file's tree must always be consistent on
4253 * disk. We must be able to restart the truncate after a crash.
4255 * The file's tree may be transiently inconsistent in memory (although it
4256 * probably isn't), but whenever we close off and commit a journal transaction,
4257 * the contents of (the filesystem + the journal) must be consistent and
4258 * restartable. It's pretty simple, really: bottom up, right to left (although
4259 * left-to-right works OK too).
4261 * Note that at recovery time, journal replay occurs *before* the restart of
4262 * truncate against the orphan inode list.
4264 * The committed inode has the new, desired i_size (which is the same as
4265 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4266 * that this inode's truncate did not complete and it will again call
4267 * ext4_truncate() to have another go. So there will be instantiated blocks
4268 * to the right of the truncation point in a crashed ext4 filesystem. But
4269 * that's fine - as long as they are linked from the inode, the post-crash
4270 * ext4_truncate() run will find them and release them.
4272 int ext4_truncate(struct inode *inode)
4274 struct ext4_inode_info *ei = EXT4_I(inode);
4275 unsigned int credits;
4278 struct address_space *mapping = inode->i_mapping;
4281 * There is a possibility that we're either freeing the inode
4282 * or it's a completely new inode. In those cases we might not
4283 * have i_rwsem locked because it's not necessary.
4285 if (!(inode->i_state & (I_NEW|I_FREEING)))
4286 WARN_ON(!inode_is_locked(inode));
4287 trace_ext4_truncate_enter(inode);
4289 if (!ext4_can_truncate(inode))
4292 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4293 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4295 if (ext4_has_inline_data(inode)) {
4298 err = ext4_inline_data_truncate(inode, &has_inline);
4299 if (err || has_inline)
4303 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4304 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4305 err = ext4_inode_attach_jinode(inode);
4310 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4311 credits = ext4_writepage_trans_blocks(inode);
4313 credits = ext4_blocks_for_truncate(inode);
4315 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4316 if (IS_ERR(handle)) {
4317 err = PTR_ERR(handle);
4321 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4322 ext4_block_truncate_page(handle, mapping, inode->i_size);
4325 * We add the inode to the orphan list, so that if this
4326 * truncate spans multiple transactions, and we crash, we will
4327 * resume the truncate when the filesystem recovers. It also
4328 * marks the inode dirty, to catch the new size.
4330 * Implication: the file must always be in a sane, consistent
4331 * truncatable state while each transaction commits.
4333 err = ext4_orphan_add(handle, inode);
4337 down_write(&EXT4_I(inode)->i_data_sem);
4339 ext4_discard_preallocations(inode, 0);
4341 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4342 err = ext4_ext_truncate(handle, inode);
4344 ext4_ind_truncate(handle, inode);
4346 up_write(&ei->i_data_sem);
4351 ext4_handle_sync(handle);
4355 * If this was a simple ftruncate() and the file will remain alive,
4356 * then we need to clear up the orphan record which we created above.
4357 * However, if this was a real unlink then we were called by
4358 * ext4_evict_inode(), and we allow that function to clean up the
4359 * orphan info for us.
4362 ext4_orphan_del(handle, inode);
4364 inode->i_mtime = inode->i_ctime = current_time(inode);
4365 err2 = ext4_mark_inode_dirty(handle, inode);
4366 if (unlikely(err2 && !err))
4368 ext4_journal_stop(handle);
4371 trace_ext4_truncate_exit(inode);
4375 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4377 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4378 return inode_peek_iversion_raw(inode);
4380 return inode_peek_iversion(inode);
4383 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4384 struct ext4_inode_info *ei)
4386 struct inode *inode = &(ei->vfs_inode);
4387 u64 i_blocks = READ_ONCE(inode->i_blocks);
4388 struct super_block *sb = inode->i_sb;
4390 if (i_blocks <= ~0U) {
4392 * i_blocks can be represented in a 32 bit variable
4393 * as multiple of 512 bytes
4395 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4396 raw_inode->i_blocks_high = 0;
4397 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4402 * This should never happen since sb->s_maxbytes should not have
4403 * allowed this, sb->s_maxbytes was set according to the huge_file
4404 * feature in ext4_fill_super().
4406 if (!ext4_has_feature_huge_file(sb))
4407 return -EFSCORRUPTED;
4409 if (i_blocks <= 0xffffffffffffULL) {
4411 * i_blocks can be represented in a 48 bit variable
4412 * as multiple of 512 bytes
4414 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4415 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4416 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4418 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4419 /* i_block is stored in file system block size */
4420 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4421 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4422 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4427 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4429 struct ext4_inode_info *ei = EXT4_I(inode);
4436 err = ext4_inode_blocks_set(raw_inode, ei);
4438 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4439 i_uid = i_uid_read(inode);
4440 i_gid = i_gid_read(inode);
4441 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4442 if (!(test_opt(inode->i_sb, NO_UID32))) {
4443 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4444 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4446 * Fix up interoperability with old kernels. Otherwise,
4447 * old inodes get re-used with the upper 16 bits of the
4450 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4451 raw_inode->i_uid_high = 0;
4452 raw_inode->i_gid_high = 0;
4454 raw_inode->i_uid_high =
4455 cpu_to_le16(high_16_bits(i_uid));
4456 raw_inode->i_gid_high =
4457 cpu_to_le16(high_16_bits(i_gid));
4460 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4461 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4462 raw_inode->i_uid_high = 0;
4463 raw_inode->i_gid_high = 0;
4465 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4467 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4468 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4469 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4470 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4472 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4473 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4474 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4475 raw_inode->i_file_acl_high =
4476 cpu_to_le16(ei->i_file_acl >> 32);
4477 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4478 ext4_isize_set(raw_inode, ei->i_disksize);
4480 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4481 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4482 if (old_valid_dev(inode->i_rdev)) {
4483 raw_inode->i_block[0] =
4484 cpu_to_le32(old_encode_dev(inode->i_rdev));
4485 raw_inode->i_block[1] = 0;
4487 raw_inode->i_block[0] = 0;
4488 raw_inode->i_block[1] =
4489 cpu_to_le32(new_encode_dev(inode->i_rdev));
4490 raw_inode->i_block[2] = 0;
4492 } else if (!ext4_has_inline_data(inode)) {
4493 for (block = 0; block < EXT4_N_BLOCKS; block++)
4494 raw_inode->i_block[block] = ei->i_data[block];
4497 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4498 u64 ivers = ext4_inode_peek_iversion(inode);
4500 raw_inode->i_disk_version = cpu_to_le32(ivers);
4501 if (ei->i_extra_isize) {
4502 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4503 raw_inode->i_version_hi =
4504 cpu_to_le32(ivers >> 32);
4505 raw_inode->i_extra_isize =
4506 cpu_to_le16(ei->i_extra_isize);
4510 if (i_projid != EXT4_DEF_PROJID &&
4511 !ext4_has_feature_project(inode->i_sb))
4512 err = err ?: -EFSCORRUPTED;
4514 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4515 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4516 raw_inode->i_projid = cpu_to_le32(i_projid);
4518 ext4_inode_csum_set(inode, raw_inode, ei);
4523 * ext4_get_inode_loc returns with an extra refcount against the inode's
4524 * underlying buffer_head on success. If we pass 'inode' and it does not
4525 * have in-inode xattr, we have all inode data in memory that is needed
4526 * to recreate the on-disk version of this inode.
4528 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4529 struct inode *inode, struct ext4_iloc *iloc,
4530 ext4_fsblk_t *ret_block)
4532 struct ext4_group_desc *gdp;
4533 struct buffer_head *bh;
4535 struct blk_plug plug;
4536 int inodes_per_block, inode_offset;
4539 if (ino < EXT4_ROOT_INO ||
4540 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4541 return -EFSCORRUPTED;
4543 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4544 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4549 * Figure out the offset within the block group inode table
4551 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4552 inode_offset = ((ino - 1) %
4553 EXT4_INODES_PER_GROUP(sb));
4554 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4556 block = ext4_inode_table(sb, gdp);
4557 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4558 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4559 ext4_error(sb, "Invalid inode table block %llu in "
4560 "block_group %u", block, iloc->block_group);
4561 return -EFSCORRUPTED;
4563 block += (inode_offset / inodes_per_block);
4565 bh = sb_getblk(sb, block);
4568 if (ext4_buffer_uptodate(bh))
4572 if (ext4_buffer_uptodate(bh)) {
4573 /* Someone brought it uptodate while we waited */
4579 * If we have all information of the inode in memory and this
4580 * is the only valid inode in the block, we need not read the
4583 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4584 struct buffer_head *bitmap_bh;
4587 start = inode_offset & ~(inodes_per_block - 1);
4589 /* Is the inode bitmap in cache? */
4590 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4591 if (unlikely(!bitmap_bh))
4595 * If the inode bitmap isn't in cache then the
4596 * optimisation may end up performing two reads instead
4597 * of one, so skip it.
4599 if (!buffer_uptodate(bitmap_bh)) {
4603 for (i = start; i < start + inodes_per_block; i++) {
4604 if (i == inode_offset)
4606 if (ext4_test_bit(i, bitmap_bh->b_data))
4610 if (i == start + inodes_per_block) {
4611 struct ext4_inode *raw_inode =
4612 (struct ext4_inode *) (bh->b_data + iloc->offset);
4614 /* all other inodes are free, so skip I/O */
4615 memset(bh->b_data, 0, bh->b_size);
4616 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4617 ext4_fill_raw_inode(inode, raw_inode);
4618 set_buffer_uptodate(bh);
4626 * If we need to do any I/O, try to pre-readahead extra
4627 * blocks from the inode table.
4629 blk_start_plug(&plug);
4630 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4631 ext4_fsblk_t b, end, table;
4633 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4635 table = ext4_inode_table(sb, gdp);
4636 /* s_inode_readahead_blks is always a power of 2 */
4637 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4641 num = EXT4_INODES_PER_GROUP(sb);
4642 if (ext4_has_group_desc_csum(sb))
4643 num -= ext4_itable_unused_count(sb, gdp);
4644 table += num / inodes_per_block;
4648 ext4_sb_breadahead_unmovable(sb, b++);
4652 * There are other valid inodes in the buffer, this inode
4653 * has in-inode xattrs, or we don't have this inode in memory.
4654 * Read the block from disk.
4656 trace_ext4_load_inode(sb, ino);
4657 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4658 blk_finish_plug(&plug);
4660 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4661 if (!buffer_uptodate(bh)) {
4672 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4673 struct ext4_iloc *iloc)
4675 ext4_fsblk_t err_blk = 0;
4678 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4682 ext4_error_inode_block(inode, err_blk, EIO,
4683 "unable to read itable block");
4688 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4690 ext4_fsblk_t err_blk = 0;
4693 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4697 ext4_error_inode_block(inode, err_blk, EIO,
4698 "unable to read itable block");
4704 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4705 struct ext4_iloc *iloc)
4707 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4710 static bool ext4_should_enable_dax(struct inode *inode)
4712 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4714 if (test_opt2(inode->i_sb, DAX_NEVER))
4716 if (!S_ISREG(inode->i_mode))
4718 if (ext4_should_journal_data(inode))
4720 if (ext4_has_inline_data(inode))
4722 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4724 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4726 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4728 if (test_opt(inode->i_sb, DAX_ALWAYS))
4731 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4734 void ext4_set_inode_flags(struct inode *inode, bool init)
4736 unsigned int flags = EXT4_I(inode)->i_flags;
4737 unsigned int new_fl = 0;
4739 WARN_ON_ONCE(IS_DAX(inode) && init);
4741 if (flags & EXT4_SYNC_FL)
4743 if (flags & EXT4_APPEND_FL)
4745 if (flags & EXT4_IMMUTABLE_FL)
4746 new_fl |= S_IMMUTABLE;
4747 if (flags & EXT4_NOATIME_FL)
4748 new_fl |= S_NOATIME;
4749 if (flags & EXT4_DIRSYNC_FL)
4750 new_fl |= S_DIRSYNC;
4752 /* Because of the way inode_set_flags() works we must preserve S_DAX
4753 * here if already set. */
4754 new_fl |= (inode->i_flags & S_DAX);
4755 if (init && ext4_should_enable_dax(inode))
4758 if (flags & EXT4_ENCRYPT_FL)
4759 new_fl |= S_ENCRYPTED;
4760 if (flags & EXT4_CASEFOLD_FL)
4761 new_fl |= S_CASEFOLD;
4762 if (flags & EXT4_VERITY_FL)
4764 inode_set_flags(inode, new_fl,
4765 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4766 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4769 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4770 struct ext4_inode_info *ei)
4773 struct inode *inode = &(ei->vfs_inode);
4774 struct super_block *sb = inode->i_sb;
4776 if (ext4_has_feature_huge_file(sb)) {
4777 /* we are using combined 48 bit field */
4778 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4779 le32_to_cpu(raw_inode->i_blocks_lo);
4780 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4781 /* i_blocks represent file system block size */
4782 return i_blocks << (inode->i_blkbits - 9);
4787 return le32_to_cpu(raw_inode->i_blocks_lo);
4791 static inline int ext4_iget_extra_inode(struct inode *inode,
4792 struct ext4_inode *raw_inode,
4793 struct ext4_inode_info *ei)
4795 __le32 *magic = (void *)raw_inode +
4796 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4798 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4799 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4800 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4801 return ext4_find_inline_data_nolock(inode);
4803 EXT4_I(inode)->i_inline_off = 0;
4807 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4809 if (!ext4_has_feature_project(inode->i_sb))
4811 *projid = EXT4_I(inode)->i_projid;
4816 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4817 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4820 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4822 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4823 inode_set_iversion_raw(inode, val);
4825 inode_set_iversion_queried(inode, val);
4828 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4829 ext4_iget_flags flags, const char *function,
4832 struct ext4_iloc iloc;
4833 struct ext4_inode *raw_inode;
4834 struct ext4_inode_info *ei;
4835 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4836 struct inode *inode;
4837 journal_t *journal = EXT4_SB(sb)->s_journal;
4845 if ((!(flags & EXT4_IGET_SPECIAL) &&
4846 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4847 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4848 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4849 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4850 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4851 (ino < EXT4_ROOT_INO) ||
4852 (ino > le32_to_cpu(es->s_inodes_count))) {
4853 if (flags & EXT4_IGET_HANDLE)
4854 return ERR_PTR(-ESTALE);
4855 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4856 "inode #%lu: comm %s: iget: illegal inode #",
4857 ino, current->comm);
4858 return ERR_PTR(-EFSCORRUPTED);
4861 inode = iget_locked(sb, ino);
4863 return ERR_PTR(-ENOMEM);
4864 if (!(inode->i_state & I_NEW))
4870 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4873 raw_inode = ext4_raw_inode(&iloc);
4875 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4876 ext4_error_inode(inode, function, line, 0,
4877 "iget: root inode unallocated");
4878 ret = -EFSCORRUPTED;
4882 if ((flags & EXT4_IGET_HANDLE) &&
4883 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4888 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4889 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4890 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4891 EXT4_INODE_SIZE(inode->i_sb) ||
4892 (ei->i_extra_isize & 3)) {
4893 ext4_error_inode(inode, function, line, 0,
4894 "iget: bad extra_isize %u "
4897 EXT4_INODE_SIZE(inode->i_sb));
4898 ret = -EFSCORRUPTED;
4902 ei->i_extra_isize = 0;
4904 /* Precompute checksum seed for inode metadata */
4905 if (ext4_has_metadata_csum(sb)) {
4906 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4908 __le32 inum = cpu_to_le32(inode->i_ino);
4909 __le32 gen = raw_inode->i_generation;
4910 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4912 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4916 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4917 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4918 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4919 ext4_error_inode_err(inode, function, line, 0,
4920 EFSBADCRC, "iget: checksum invalid");
4925 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4926 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4927 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4928 if (ext4_has_feature_project(sb) &&
4929 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4930 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4931 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4933 i_projid = EXT4_DEF_PROJID;
4935 if (!(test_opt(inode->i_sb, NO_UID32))) {
4936 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4937 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4939 i_uid_write(inode, i_uid);
4940 i_gid_write(inode, i_gid);
4941 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4942 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4944 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4945 ei->i_inline_off = 0;
4946 ei->i_dir_start_lookup = 0;
4947 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4948 /* We now have enough fields to check if the inode was active or not.
4949 * This is needed because nfsd might try to access dead inodes
4950 * the test is that same one that e2fsck uses
4951 * NeilBrown 1999oct15
4953 if (inode->i_nlink == 0) {
4954 if ((inode->i_mode == 0 ||
4955 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4956 ino != EXT4_BOOT_LOADER_INO) {
4957 /* this inode is deleted */
4961 /* The only unlinked inodes we let through here have
4962 * valid i_mode and are being read by the orphan
4963 * recovery code: that's fine, we're about to complete
4964 * the process of deleting those.
4965 * OR it is the EXT4_BOOT_LOADER_INO which is
4966 * not initialized on a new filesystem. */
4968 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4969 ext4_set_inode_flags(inode, true);
4970 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4971 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4972 if (ext4_has_feature_64bit(sb))
4974 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4975 inode->i_size = ext4_isize(sb, raw_inode);
4976 if ((size = i_size_read(inode)) < 0) {
4977 ext4_error_inode(inode, function, line, 0,
4978 "iget: bad i_size value: %lld", size);
4979 ret = -EFSCORRUPTED;
4983 * If dir_index is not enabled but there's dir with INDEX flag set,
4984 * we'd normally treat htree data as empty space. But with metadata
4985 * checksumming that corrupts checksums so forbid that.
4987 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4988 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4989 ext4_error_inode(inode, function, line, 0,
4990 "iget: Dir with htree data on filesystem without dir_index feature.");
4991 ret = -EFSCORRUPTED;
4994 ei->i_disksize = inode->i_size;
4996 ei->i_reserved_quota = 0;
4998 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4999 ei->i_block_group = iloc.block_group;
5000 ei->i_last_alloc_group = ~0;
5002 * NOTE! The in-memory inode i_data array is in little-endian order
5003 * even on big-endian machines: we do NOT byteswap the block numbers!
5005 for (block = 0; block < EXT4_N_BLOCKS; block++)
5006 ei->i_data[block] = raw_inode->i_block[block];
5007 INIT_LIST_HEAD(&ei->i_orphan);
5008 ext4_fc_init_inode(&ei->vfs_inode);
5011 * Set transaction id's of transactions that have to be committed
5012 * to finish f[data]sync. We set them to currently running transaction
5013 * as we cannot be sure that the inode or some of its metadata isn't
5014 * part of the transaction - the inode could have been reclaimed and
5015 * now it is reread from disk.
5018 transaction_t *transaction;
5021 read_lock(&journal->j_state_lock);
5022 if (journal->j_running_transaction)
5023 transaction = journal->j_running_transaction;
5025 transaction = journal->j_committing_transaction;
5027 tid = transaction->t_tid;
5029 tid = journal->j_commit_sequence;
5030 read_unlock(&journal->j_state_lock);
5031 ei->i_sync_tid = tid;
5032 ei->i_datasync_tid = tid;
5035 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5036 if (ei->i_extra_isize == 0) {
5037 /* The extra space is currently unused. Use it. */
5038 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5039 ei->i_extra_isize = sizeof(struct ext4_inode) -
5040 EXT4_GOOD_OLD_INODE_SIZE;
5042 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5048 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5049 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5050 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5051 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5053 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5054 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5056 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5057 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5059 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5061 ext4_inode_set_iversion_queried(inode, ivers);
5065 if (ei->i_file_acl &&
5066 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
5067 ext4_error_inode(inode, function, line, 0,
5068 "iget: bad extended attribute block %llu",
5070 ret = -EFSCORRUPTED;
5072 } else if (!ext4_has_inline_data(inode)) {
5073 /* validate the block references in the inode */
5074 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
5075 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5076 (S_ISLNK(inode->i_mode) &&
5077 !ext4_inode_is_fast_symlink(inode)))) {
5078 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5079 ret = ext4_ext_check_inode(inode);
5081 ret = ext4_ind_check_inode(inode);
5087 if (S_ISREG(inode->i_mode)) {
5088 inode->i_op = &ext4_file_inode_operations;
5089 inode->i_fop = &ext4_file_operations;
5090 ext4_set_aops(inode);
5091 } else if (S_ISDIR(inode->i_mode)) {
5092 inode->i_op = &ext4_dir_inode_operations;
5093 inode->i_fop = &ext4_dir_operations;
5094 } else if (S_ISLNK(inode->i_mode)) {
5095 /* VFS does not allow setting these so must be corruption */
5096 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5097 ext4_error_inode(inode, function, line, 0,
5098 "iget: immutable or append flags "
5099 "not allowed on symlinks");
5100 ret = -EFSCORRUPTED;
5103 if (IS_ENCRYPTED(inode)) {
5104 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5105 } else if (ext4_inode_is_fast_symlink(inode)) {
5106 inode->i_link = (char *)ei->i_data;
5107 inode->i_op = &ext4_fast_symlink_inode_operations;
5108 nd_terminate_link(ei->i_data, inode->i_size,
5109 sizeof(ei->i_data) - 1);
5111 inode->i_op = &ext4_symlink_inode_operations;
5113 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5114 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5115 inode->i_op = &ext4_special_inode_operations;
5116 if (raw_inode->i_block[0])
5117 init_special_inode(inode, inode->i_mode,
5118 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5120 init_special_inode(inode, inode->i_mode,
5121 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5122 } else if (ino == EXT4_BOOT_LOADER_INO) {
5123 make_bad_inode(inode);
5125 ret = -EFSCORRUPTED;
5126 ext4_error_inode(inode, function, line, 0,
5127 "iget: bogus i_mode (%o)", inode->i_mode);
5130 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5131 ext4_error_inode(inode, function, line, 0,
5132 "casefold flag without casefold feature");
5133 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
5134 ext4_error_inode(inode, function, line, 0,
5135 "bad inode without EXT4_IGET_BAD flag");
5141 unlock_new_inode(inode);
5147 return ERR_PTR(ret);
5150 static void __ext4_update_other_inode_time(struct super_block *sb,
5151 unsigned long orig_ino,
5153 struct ext4_inode *raw_inode)
5155 struct inode *inode;
5157 inode = find_inode_by_ino_rcu(sb, ino);
5161 if (!inode_is_dirtytime_only(inode))
5164 spin_lock(&inode->i_lock);
5165 if (inode_is_dirtytime_only(inode)) {
5166 struct ext4_inode_info *ei = EXT4_I(inode);
5168 inode->i_state &= ~I_DIRTY_TIME;
5169 spin_unlock(&inode->i_lock);
5171 spin_lock(&ei->i_raw_lock);
5172 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5173 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5174 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5175 ext4_inode_csum_set(inode, raw_inode, ei);
5176 spin_unlock(&ei->i_raw_lock);
5177 trace_ext4_other_inode_update_time(inode, orig_ino);
5180 spin_unlock(&inode->i_lock);
5184 * Opportunistically update the other time fields for other inodes in
5185 * the same inode table block.
5187 static void ext4_update_other_inodes_time(struct super_block *sb,
5188 unsigned long orig_ino, char *buf)
5191 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5192 int inode_size = EXT4_INODE_SIZE(sb);
5195 * Calculate the first inode in the inode table block. Inode
5196 * numbers are one-based. That is, the first inode in a block
5197 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5199 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5201 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5202 if (ino == orig_ino)
5204 __ext4_update_other_inode_time(sb, orig_ino, ino,
5205 (struct ext4_inode *)buf);
5211 * Post the struct inode info into an on-disk inode location in the
5212 * buffer-cache. This gobbles the caller's reference to the
5213 * buffer_head in the inode location struct.
5215 * The caller must have write access to iloc->bh.
5217 static int ext4_do_update_inode(handle_t *handle,
5218 struct inode *inode,
5219 struct ext4_iloc *iloc)
5221 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5222 struct ext4_inode_info *ei = EXT4_I(inode);
5223 struct buffer_head *bh = iloc->bh;
5224 struct super_block *sb = inode->i_sb;
5226 int need_datasync = 0, set_large_file = 0;
5228 spin_lock(&ei->i_raw_lock);
5231 * For fields not tracked in the in-memory inode, initialise them
5232 * to zero for new inodes.
5234 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5235 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5237 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5239 if (ei->i_disksize > 0x7fffffffULL) {
5240 if (!ext4_has_feature_large_file(sb) ||
5241 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5245 err = ext4_fill_raw_inode(inode, raw_inode);
5246 spin_unlock(&ei->i_raw_lock);
5248 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5252 if (inode->i_sb->s_flags & SB_LAZYTIME)
5253 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5256 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5257 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5260 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5261 if (set_large_file) {
5262 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5263 err = ext4_journal_get_write_access(handle, sb,
5268 lock_buffer(EXT4_SB(sb)->s_sbh);
5269 ext4_set_feature_large_file(sb);
5270 ext4_superblock_csum_set(sb);
5271 unlock_buffer(EXT4_SB(sb)->s_sbh);
5272 ext4_handle_sync(handle);
5273 err = ext4_handle_dirty_metadata(handle, NULL,
5274 EXT4_SB(sb)->s_sbh);
5276 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5278 ext4_std_error(inode->i_sb, err);
5285 * ext4_write_inode()
5287 * We are called from a few places:
5289 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5290 * Here, there will be no transaction running. We wait for any running
5291 * transaction to commit.
5293 * - Within flush work (sys_sync(), kupdate and such).
5294 * We wait on commit, if told to.
5296 * - Within iput_final() -> write_inode_now()
5297 * We wait on commit, if told to.
5299 * In all cases it is actually safe for us to return without doing anything,
5300 * because the inode has been copied into a raw inode buffer in
5301 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5304 * Note that we are absolutely dependent upon all inode dirtiers doing the
5305 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5306 * which we are interested.
5308 * It would be a bug for them to not do this. The code:
5310 * mark_inode_dirty(inode)
5312 * inode->i_size = expr;
5314 * is in error because write_inode() could occur while `stuff()' is running,
5315 * and the new i_size will be lost. Plus the inode will no longer be on the
5316 * superblock's dirty inode list.
5318 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5322 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5323 sb_rdonly(inode->i_sb))
5326 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5329 if (EXT4_SB(inode->i_sb)->s_journal) {
5330 if (ext4_journal_current_handle()) {
5331 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5337 * No need to force transaction in WB_SYNC_NONE mode. Also
5338 * ext4_sync_fs() will force the commit after everything is
5341 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5344 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5345 EXT4_I(inode)->i_sync_tid);
5347 struct ext4_iloc iloc;
5349 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5353 * sync(2) will flush the whole buffer cache. No need to do
5354 * it here separately for each inode.
5356 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5357 sync_dirty_buffer(iloc.bh);
5358 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5359 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5360 "IO error syncing inode");
5369 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5370 * buffers that are attached to a folio straddling i_size and are undergoing
5371 * commit. In that case we have to wait for commit to finish and try again.
5373 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5376 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5377 tid_t commit_tid = 0;
5380 offset = inode->i_size & (PAGE_SIZE - 1);
5382 * If the folio is fully truncated, we don't need to wait for any commit
5383 * (and we even should not as __ext4_journalled_invalidate_folio() may
5384 * strip all buffers from the folio but keep the folio dirty which can then
5385 * confuse e.g. concurrent ext4_writepage() seeing dirty folio without
5386 * buffers). Also we don't need to wait for any commit if all buffers in
5387 * the folio remain valid. This is most beneficial for the common case of
5388 * blocksize == PAGESIZE.
5390 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5393 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5394 inode->i_size >> PAGE_SHIFT);
5397 ret = __ext4_journalled_invalidate_folio(folio, offset,
5398 folio_size(folio) - offset);
5399 folio_unlock(folio);
5404 read_lock(&journal->j_state_lock);
5405 if (journal->j_committing_transaction)
5406 commit_tid = journal->j_committing_transaction->t_tid;
5407 read_unlock(&journal->j_state_lock);
5409 jbd2_log_wait_commit(journal, commit_tid);
5416 * Called from notify_change.
5418 * We want to trap VFS attempts to truncate the file as soon as
5419 * possible. In particular, we want to make sure that when the VFS
5420 * shrinks i_size, we put the inode on the orphan list and modify
5421 * i_disksize immediately, so that during the subsequent flushing of
5422 * dirty pages and freeing of disk blocks, we can guarantee that any
5423 * commit will leave the blocks being flushed in an unused state on
5424 * disk. (On recovery, the inode will get truncated and the blocks will
5425 * be freed, so we have a strong guarantee that no future commit will
5426 * leave these blocks visible to the user.)
5428 * Another thing we have to assure is that if we are in ordered mode
5429 * and inode is still attached to the committing transaction, we must
5430 * we start writeout of all the dirty pages which are being truncated.
5431 * This way we are sure that all the data written in the previous
5432 * transaction are already on disk (truncate waits for pages under
5435 * Called with inode->i_rwsem down.
5437 int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5440 struct inode *inode = d_inode(dentry);
5443 const unsigned int ia_valid = attr->ia_valid;
5444 bool inc_ivers = true;
5446 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5449 if (unlikely(IS_IMMUTABLE(inode)))
5452 if (unlikely(IS_APPEND(inode) &&
5453 (ia_valid & (ATTR_MODE | ATTR_UID |
5454 ATTR_GID | ATTR_TIMES_SET))))
5457 error = setattr_prepare(mnt_userns, dentry, attr);
5461 error = fscrypt_prepare_setattr(dentry, attr);
5465 error = fsverity_prepare_setattr(dentry, attr);
5469 if (is_quota_modification(mnt_userns, inode, attr)) {
5470 error = dquot_initialize(inode);
5475 if (i_uid_needs_update(mnt_userns, attr, inode) ||
5476 i_gid_needs_update(mnt_userns, attr, inode)) {
5479 /* (user+group)*(old+new) structure, inode write (sb,
5480 * inode block, ? - but truncate inode update has it) */
5481 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5482 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5483 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5484 if (IS_ERR(handle)) {
5485 error = PTR_ERR(handle);
5489 /* dquot_transfer() calls back ext4_get_inode_usage() which
5490 * counts xattr inode references.
5492 down_read(&EXT4_I(inode)->xattr_sem);
5493 error = dquot_transfer(mnt_userns, inode, attr);
5494 up_read(&EXT4_I(inode)->xattr_sem);
5497 ext4_journal_stop(handle);
5500 /* Update corresponding info in inode so that everything is in
5501 * one transaction */
5502 i_uid_update(mnt_userns, attr, inode);
5503 i_gid_update(mnt_userns, attr, inode);
5504 error = ext4_mark_inode_dirty(handle, inode);
5505 ext4_journal_stop(handle);
5506 if (unlikely(error)) {
5511 if (attr->ia_valid & ATTR_SIZE) {
5513 loff_t oldsize = inode->i_size;
5514 loff_t old_disksize;
5515 int shrink = (attr->ia_size < inode->i_size);
5517 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5518 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5520 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5524 if (!S_ISREG(inode->i_mode)) {
5528 if (attr->ia_size == inode->i_size)
5532 if (ext4_should_order_data(inode)) {
5533 error = ext4_begin_ordered_truncate(inode,
5539 * Blocks are going to be removed from the inode. Wait
5540 * for dio in flight.
5542 inode_dio_wait(inode);
5545 filemap_invalidate_lock(inode->i_mapping);
5547 rc = ext4_break_layouts(inode);
5549 filemap_invalidate_unlock(inode->i_mapping);
5553 if (attr->ia_size != inode->i_size) {
5554 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5555 if (IS_ERR(handle)) {
5556 error = PTR_ERR(handle);
5559 if (ext4_handle_valid(handle) && shrink) {
5560 error = ext4_orphan_add(handle, inode);
5564 * Update c/mtime on truncate up, ext4_truncate() will
5565 * update c/mtime in shrink case below
5568 inode->i_mtime = current_time(inode);
5569 inode->i_ctime = inode->i_mtime;
5573 ext4_fc_track_range(handle, inode,
5574 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5575 inode->i_sb->s_blocksize_bits,
5576 EXT_MAX_BLOCKS - 1);
5578 ext4_fc_track_range(
5580 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5581 inode->i_sb->s_blocksize_bits,
5582 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5583 inode->i_sb->s_blocksize_bits);
5585 down_write(&EXT4_I(inode)->i_data_sem);
5586 old_disksize = EXT4_I(inode)->i_disksize;
5587 EXT4_I(inode)->i_disksize = attr->ia_size;
5588 rc = ext4_mark_inode_dirty(handle, inode);
5592 * We have to update i_size under i_data_sem together
5593 * with i_disksize to avoid races with writeback code
5594 * running ext4_wb_update_i_disksize().
5597 i_size_write(inode, attr->ia_size);
5599 EXT4_I(inode)->i_disksize = old_disksize;
5600 up_write(&EXT4_I(inode)->i_data_sem);
5601 ext4_journal_stop(handle);
5605 pagecache_isize_extended(inode, oldsize,
5607 } else if (ext4_should_journal_data(inode)) {
5608 ext4_wait_for_tail_page_commit(inode);
5613 * Truncate pagecache after we've waited for commit
5614 * in data=journal mode to make pages freeable.
5616 truncate_pagecache(inode, inode->i_size);
5618 * Call ext4_truncate() even if i_size didn't change to
5619 * truncate possible preallocated blocks.
5621 if (attr->ia_size <= oldsize) {
5622 rc = ext4_truncate(inode);
5627 filemap_invalidate_unlock(inode->i_mapping);
5632 inode_inc_iversion(inode);
5633 setattr_copy(mnt_userns, inode, attr);
5634 mark_inode_dirty(inode);
5638 * If the call to ext4_truncate failed to get a transaction handle at
5639 * all, we need to clean up the in-core orphan list manually.
5641 if (orphan && inode->i_nlink)
5642 ext4_orphan_del(NULL, inode);
5644 if (!error && (ia_valid & ATTR_MODE))
5645 rc = posix_acl_chmod(mnt_userns, dentry, inode->i_mode);
5649 ext4_std_error(inode->i_sb, error);
5655 u32 ext4_dio_alignment(struct inode *inode)
5657 if (fsverity_active(inode))
5659 if (ext4_should_journal_data(inode))
5661 if (ext4_has_inline_data(inode))
5663 if (IS_ENCRYPTED(inode)) {
5664 if (!fscrypt_dio_supported(inode))
5666 return i_blocksize(inode);
5668 return 1; /* use the iomap defaults */
5671 int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5672 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5674 struct inode *inode = d_inode(path->dentry);
5675 struct ext4_inode *raw_inode;
5676 struct ext4_inode_info *ei = EXT4_I(inode);
5679 if ((request_mask & STATX_BTIME) &&
5680 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5681 stat->result_mask |= STATX_BTIME;
5682 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5683 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5687 * Return the DIO alignment restrictions if requested. We only return
5688 * this information when requested, since on encrypted files it might
5689 * take a fair bit of work to get if the file wasn't opened recently.
5691 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5692 u32 dio_align = ext4_dio_alignment(inode);
5694 stat->result_mask |= STATX_DIOALIGN;
5695 if (dio_align == 1) {
5696 struct block_device *bdev = inode->i_sb->s_bdev;
5698 /* iomap defaults */
5699 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5700 stat->dio_offset_align = bdev_logical_block_size(bdev);
5702 stat->dio_mem_align = dio_align;
5703 stat->dio_offset_align = dio_align;
5707 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5708 if (flags & EXT4_APPEND_FL)
5709 stat->attributes |= STATX_ATTR_APPEND;
5710 if (flags & EXT4_COMPR_FL)
5711 stat->attributes |= STATX_ATTR_COMPRESSED;
5712 if (flags & EXT4_ENCRYPT_FL)
5713 stat->attributes |= STATX_ATTR_ENCRYPTED;
5714 if (flags & EXT4_IMMUTABLE_FL)
5715 stat->attributes |= STATX_ATTR_IMMUTABLE;
5716 if (flags & EXT4_NODUMP_FL)
5717 stat->attributes |= STATX_ATTR_NODUMP;
5718 if (flags & EXT4_VERITY_FL)
5719 stat->attributes |= STATX_ATTR_VERITY;
5721 stat->attributes_mask |= (STATX_ATTR_APPEND |
5722 STATX_ATTR_COMPRESSED |
5723 STATX_ATTR_ENCRYPTED |
5724 STATX_ATTR_IMMUTABLE |
5728 generic_fillattr(mnt_userns, inode, stat);
5732 int ext4_file_getattr(struct user_namespace *mnt_userns,
5733 const struct path *path, struct kstat *stat,
5734 u32 request_mask, unsigned int query_flags)
5736 struct inode *inode = d_inode(path->dentry);
5737 u64 delalloc_blocks;
5739 ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5742 * If there is inline data in the inode, the inode will normally not
5743 * have data blocks allocated (it may have an external xattr block).
5744 * Report at least one sector for such files, so tools like tar, rsync,
5745 * others don't incorrectly think the file is completely sparse.
5747 if (unlikely(ext4_has_inline_data(inode)))
5748 stat->blocks += (stat->size + 511) >> 9;
5751 * We can't update i_blocks if the block allocation is delayed
5752 * otherwise in the case of system crash before the real block
5753 * allocation is done, we will have i_blocks inconsistent with
5754 * on-disk file blocks.
5755 * We always keep i_blocks updated together with real
5756 * allocation. But to not confuse with user, stat
5757 * will return the blocks that include the delayed allocation
5758 * blocks for this file.
5760 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5761 EXT4_I(inode)->i_reserved_data_blocks);
5762 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5766 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5769 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5770 return ext4_ind_trans_blocks(inode, lblocks);
5771 return ext4_ext_index_trans_blocks(inode, pextents);
5775 * Account for index blocks, block groups bitmaps and block group
5776 * descriptor blocks if modify datablocks and index blocks
5777 * worse case, the indexs blocks spread over different block groups
5779 * If datablocks are discontiguous, they are possible to spread over
5780 * different block groups too. If they are contiguous, with flexbg,
5781 * they could still across block group boundary.
5783 * Also account for superblock, inode, quota and xattr blocks
5785 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5788 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5794 * How many index blocks need to touch to map @lblocks logical blocks
5795 * to @pextents physical extents?
5797 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5802 * Now let's see how many group bitmaps and group descriptors need
5805 groups = idxblocks + pextents;
5807 if (groups > ngroups)
5809 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5810 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5812 /* bitmaps and block group descriptor blocks */
5813 ret += groups + gdpblocks;
5815 /* Blocks for super block, inode, quota and xattr blocks */
5816 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5822 * Calculate the total number of credits to reserve to fit
5823 * the modification of a single pages into a single transaction,
5824 * which may include multiple chunks of block allocations.
5826 * This could be called via ext4_write_begin()
5828 * We need to consider the worse case, when
5829 * one new block per extent.
5831 int ext4_writepage_trans_blocks(struct inode *inode)
5833 int bpp = ext4_journal_blocks_per_page(inode);
5836 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5838 /* Account for data blocks for journalled mode */
5839 if (ext4_should_journal_data(inode))
5845 * Calculate the journal credits for a chunk of data modification.
5847 * This is called from DIO, fallocate or whoever calling
5848 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5850 * journal buffers for data blocks are not included here, as DIO
5851 * and fallocate do no need to journal data buffers.
5853 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5855 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5859 * The caller must have previously called ext4_reserve_inode_write().
5860 * Give this, we know that the caller already has write access to iloc->bh.
5862 int ext4_mark_iloc_dirty(handle_t *handle,
5863 struct inode *inode, struct ext4_iloc *iloc)
5867 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5871 ext4_fc_track_inode(handle, inode);
5873 /* the do_update_inode consumes one bh->b_count */
5876 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5877 err = ext4_do_update_inode(handle, inode, iloc);
5883 * On success, We end up with an outstanding reference count against
5884 * iloc->bh. This _must_ be cleaned up later.
5888 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5889 struct ext4_iloc *iloc)
5893 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5896 err = ext4_get_inode_loc(inode, iloc);
5898 BUFFER_TRACE(iloc->bh, "get_write_access");
5899 err = ext4_journal_get_write_access(handle, inode->i_sb,
5900 iloc->bh, EXT4_JTR_NONE);
5906 ext4_std_error(inode->i_sb, err);
5910 static int __ext4_expand_extra_isize(struct inode *inode,
5911 unsigned int new_extra_isize,
5912 struct ext4_iloc *iloc,
5913 handle_t *handle, int *no_expand)
5915 struct ext4_inode *raw_inode;
5916 struct ext4_xattr_ibody_header *header;
5917 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5918 struct ext4_inode_info *ei = EXT4_I(inode);
5921 /* this was checked at iget time, but double check for good measure */
5922 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5923 (ei->i_extra_isize & 3)) {
5924 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5926 EXT4_INODE_SIZE(inode->i_sb));
5927 return -EFSCORRUPTED;
5929 if ((new_extra_isize < ei->i_extra_isize) ||
5930 (new_extra_isize < 4) ||
5931 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5932 return -EINVAL; /* Should never happen */
5934 raw_inode = ext4_raw_inode(iloc);
5936 header = IHDR(inode, raw_inode);
5938 /* No extended attributes present */
5939 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5940 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5941 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5942 EXT4_I(inode)->i_extra_isize, 0,
5943 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5944 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5949 * We may need to allocate external xattr block so we need quotas
5950 * initialized. Here we can be called with various locks held so we
5951 * cannot affort to initialize quotas ourselves. So just bail.
5953 if (dquot_initialize_needed(inode))
5956 /* try to expand with EAs present */
5957 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5961 * Inode size expansion failed; don't try again
5970 * Expand an inode by new_extra_isize bytes.
5971 * Returns 0 on success or negative error number on failure.
5973 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5974 unsigned int new_extra_isize,
5975 struct ext4_iloc iloc,
5981 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5985 * In nojournal mode, we can immediately attempt to expand
5986 * the inode. When journaled, we first need to obtain extra
5987 * buffer credits since we may write into the EA block
5988 * with this same handle. If journal_extend fails, then it will
5989 * only result in a minor loss of functionality for that inode.
5990 * If this is felt to be critical, then e2fsck should be run to
5991 * force a large enough s_min_extra_isize.
5993 if (ext4_journal_extend(handle,
5994 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5997 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
6000 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
6001 handle, &no_expand);
6002 ext4_write_unlock_xattr(inode, &no_expand);
6007 int ext4_expand_extra_isize(struct inode *inode,
6008 unsigned int new_extra_isize,
6009 struct ext4_iloc *iloc)
6015 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6020 handle = ext4_journal_start(inode, EXT4_HT_INODE,
6021 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6022 if (IS_ERR(handle)) {
6023 error = PTR_ERR(handle);
6028 ext4_write_lock_xattr(inode, &no_expand);
6030 BUFFER_TRACE(iloc->bh, "get_write_access");
6031 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
6038 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6039 handle, &no_expand);
6041 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6046 ext4_write_unlock_xattr(inode, &no_expand);
6047 ext4_journal_stop(handle);
6052 * What we do here is to mark the in-core inode as clean with respect to inode
6053 * dirtiness (it may still be data-dirty).
6054 * This means that the in-core inode may be reaped by prune_icache
6055 * without having to perform any I/O. This is a very good thing,
6056 * because *any* task may call prune_icache - even ones which
6057 * have a transaction open against a different journal.
6059 * Is this cheating? Not really. Sure, we haven't written the
6060 * inode out, but prune_icache isn't a user-visible syncing function.
6061 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6062 * we start and wait on commits.
6064 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
6065 const char *func, unsigned int line)
6067 struct ext4_iloc iloc;
6068 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6072 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6073 err = ext4_reserve_inode_write(handle, inode, &iloc);
6077 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6078 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6081 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6084 ext4_error_inode_err(inode, func, line, 0, err,
6085 "mark_inode_dirty error");
6090 * ext4_dirty_inode() is called from __mark_inode_dirty()
6092 * We're really interested in the case where a file is being extended.
6093 * i_size has been changed by generic_commit_write() and we thus need
6094 * to include the updated inode in the current transaction.
6096 * Also, dquot_alloc_block() will always dirty the inode when blocks
6097 * are allocated to the file.
6099 * If the inode is marked synchronous, we don't honour that here - doing
6100 * so would cause a commit on atime updates, which we don't bother doing.
6101 * We handle synchronous inodes at the highest possible level.
6103 void ext4_dirty_inode(struct inode *inode, int flags)
6107 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6110 ext4_mark_inode_dirty(handle, inode);
6111 ext4_journal_stop(handle);
6114 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6119 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6122 * We have to be very careful here: changing a data block's
6123 * journaling status dynamically is dangerous. If we write a
6124 * data block to the journal, change the status and then delete
6125 * that block, we risk forgetting to revoke the old log record
6126 * from the journal and so a subsequent replay can corrupt data.
6127 * So, first we make sure that the journal is empty and that
6128 * nobody is changing anything.
6131 journal = EXT4_JOURNAL(inode);
6134 if (is_journal_aborted(journal))
6137 /* Wait for all existing dio workers */
6138 inode_dio_wait(inode);
6141 * Before flushing the journal and switching inode's aops, we have
6142 * to flush all dirty data the inode has. There can be outstanding
6143 * delayed allocations, there can be unwritten extents created by
6144 * fallocate or buffered writes in dioread_nolock mode covered by
6145 * dirty data which can be converted only after flushing the dirty
6146 * data (and journalled aops don't know how to handle these cases).
6149 filemap_invalidate_lock(inode->i_mapping);
6150 err = filemap_write_and_wait(inode->i_mapping);
6152 filemap_invalidate_unlock(inode->i_mapping);
6157 percpu_down_write(&sbi->s_writepages_rwsem);
6158 jbd2_journal_lock_updates(journal);
6161 * OK, there are no updates running now, and all cached data is
6162 * synced to disk. We are now in a completely consistent state
6163 * which doesn't have anything in the journal, and we know that
6164 * no filesystem updates are running, so it is safe to modify
6165 * the inode's in-core data-journaling state flag now.
6169 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6171 err = jbd2_journal_flush(journal, 0);
6173 jbd2_journal_unlock_updates(journal);
6174 percpu_up_write(&sbi->s_writepages_rwsem);
6177 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6179 ext4_set_aops(inode);
6181 jbd2_journal_unlock_updates(journal);
6182 percpu_up_write(&sbi->s_writepages_rwsem);
6185 filemap_invalidate_unlock(inode->i_mapping);
6187 /* Finally we can mark the inode as dirty. */
6189 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6191 return PTR_ERR(handle);
6193 ext4_fc_mark_ineligible(inode->i_sb,
6194 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6195 err = ext4_mark_inode_dirty(handle, inode);
6196 ext4_handle_sync(handle);
6197 ext4_journal_stop(handle);
6198 ext4_std_error(inode->i_sb, err);
6203 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6204 struct buffer_head *bh)
6206 return !buffer_mapped(bh);
6209 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6211 struct vm_area_struct *vma = vmf->vma;
6212 struct page *page = vmf->page;
6217 struct file *file = vma->vm_file;
6218 struct inode *inode = file_inode(file);
6219 struct address_space *mapping = inode->i_mapping;
6221 get_block_t *get_block;
6224 if (unlikely(IS_IMMUTABLE(inode)))
6225 return VM_FAULT_SIGBUS;
6227 sb_start_pagefault(inode->i_sb);
6228 file_update_time(vma->vm_file);
6230 filemap_invalidate_lock_shared(mapping);
6232 err = ext4_convert_inline_data(inode);
6237 * On data journalling we skip straight to the transaction handle:
6238 * there's no delalloc; page truncated will be checked later; the
6239 * early return w/ all buffers mapped (calculates size/len) can't
6240 * be used; and there's no dioread_nolock, so only ext4_get_block.
6242 if (ext4_should_journal_data(inode))
6245 /* Delalloc case is easy... */
6246 if (test_opt(inode->i_sb, DELALLOC) &&
6247 !ext4_nonda_switch(inode->i_sb)) {
6249 err = block_page_mkwrite(vma, vmf,
6250 ext4_da_get_block_prep);
6251 } while (err == -ENOSPC &&
6252 ext4_should_retry_alloc(inode->i_sb, &retries));
6257 size = i_size_read(inode);
6258 /* Page got truncated from under us? */
6259 if (page->mapping != mapping || page_offset(page) > size) {
6261 ret = VM_FAULT_NOPAGE;
6265 if (page->index == size >> PAGE_SHIFT)
6266 len = size & ~PAGE_MASK;
6270 * Return if we have all the buffers mapped. This avoids the need to do
6271 * journal_start/journal_stop which can block and take a long time
6273 * This cannot be done for data journalling, as we have to add the
6274 * inode to the transaction's list to writeprotect pages on commit.
6276 if (page_has_buffers(page)) {
6277 if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page),
6279 ext4_bh_unmapped)) {
6280 /* Wait so that we don't change page under IO */
6281 wait_for_stable_page(page);
6282 ret = VM_FAULT_LOCKED;
6287 /* OK, we need to fill the hole... */
6288 if (ext4_should_dioread_nolock(inode))
6289 get_block = ext4_get_block_unwritten;
6291 get_block = ext4_get_block;
6293 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6294 ext4_writepage_trans_blocks(inode));
6295 if (IS_ERR(handle)) {
6296 ret = VM_FAULT_SIGBUS;
6300 * Data journalling can't use block_page_mkwrite() because it
6301 * will set_buffer_dirty() before do_journal_get_write_access()
6302 * thus might hit warning messages for dirty metadata buffers.
6304 if (!ext4_should_journal_data(inode)) {
6305 err = block_page_mkwrite(vma, vmf, get_block);
6308 size = i_size_read(inode);
6309 /* Page got truncated from under us? */
6310 if (page->mapping != mapping || page_offset(page) > size) {
6311 ret = VM_FAULT_NOPAGE;
6315 if (page->index == size >> PAGE_SHIFT)
6316 len = size & ~PAGE_MASK;
6320 err = __block_write_begin(page, 0, len, ext4_get_block);
6322 ret = VM_FAULT_SIGBUS;
6323 if (ext4_walk_page_buffers(handle, inode,
6324 page_buffers(page), 0, len, NULL,
6325 do_journal_get_write_access))
6327 if (ext4_walk_page_buffers(handle, inode,
6328 page_buffers(page), 0, len, NULL,
6331 if (ext4_jbd2_inode_add_write(handle, inode,
6332 page_offset(page), len))
6334 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6339 ext4_journal_stop(handle);
6340 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6343 ret = block_page_mkwrite_return(err);
6345 filemap_invalidate_unlock_shared(mapping);
6346 sb_end_pagefault(inode->i_sb);
6350 ext4_journal_stop(handle);