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/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
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 static 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 void ext4_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length);
141 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
147 * Test whether an inode is a fast symlink.
148 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 int ext4_inode_is_fast_symlink(struct inode *inode)
152 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
153 int ea_blocks = EXT4_I(inode)->i_file_acl ?
154 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156 if (ext4_has_inline_data(inode))
159 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 return S_ISLNK(inode->i_mode) && inode->i_size &&
162 (inode->i_size < EXT4_N_BLOCKS * 4);
166 * Restart the transaction associated with *handle. This does a commit,
167 * so before we call here everything must be consistently dirtied against
170 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
176 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
177 * moment, get_block can be called only for blocks inside i_size since
178 * page cache has been already dropped and writes are blocked by
179 * i_mutex. So we can safely drop the i_data_sem here.
181 BUG_ON(EXT4_JOURNAL(inode) == NULL);
182 jbd_debug(2, "restarting handle %p\n", handle);
183 up_write(&EXT4_I(inode)->i_data_sem);
184 ret = ext4_journal_restart(handle, nblocks);
185 down_write(&EXT4_I(inode)->i_data_sem);
186 ext4_discard_preallocations(inode);
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_evict_inode(struct inode *inode)
199 * Credits for final inode cleanup and freeing:
200 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
201 * (xattr block freeing), bitmap, group descriptor (inode freeing)
203 int extra_credits = 6;
204 struct ext4_xattr_inode_array *ea_inode_array = NULL;
205 bool freeze_protected = false;
207 trace_ext4_evict_inode(inode);
209 if (inode->i_nlink) {
211 * When journalling data dirty buffers are tracked only in the
212 * journal. So although mm thinks everything is clean and
213 * ready for reaping the inode might still have some pages to
214 * write in the running transaction or waiting to be
215 * checkpointed. Thus calling jbd2_journal_invalidatepage()
216 * (via truncate_inode_pages()) to discard these buffers can
217 * cause data loss. Also even if we did not discard these
218 * buffers, we would have no way to find them after the inode
219 * is reaped and thus user could see stale data if he tries to
220 * read them before the transaction is checkpointed. So be
221 * careful and force everything to disk here... We use
222 * ei->i_datasync_tid to store the newest transaction
223 * containing inode's data.
225 * Note that directories do not have this problem because they
226 * don't use page cache.
228 if (inode->i_ino != EXT4_JOURNAL_INO &&
229 ext4_should_journal_data(inode) &&
230 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
231 inode->i_data.nrpages) {
232 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
233 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
235 jbd2_complete_transaction(journal, commit_tid);
236 filemap_write_and_wait(&inode->i_data);
238 truncate_inode_pages_final(&inode->i_data);
243 if (is_bad_inode(inode))
245 dquot_initialize(inode);
247 if (ext4_should_order_data(inode))
248 ext4_begin_ordered_truncate(inode, 0);
249 truncate_inode_pages_final(&inode->i_data);
252 * Protect us against freezing - iput() caller didn't have to have any
253 * protection against it. When we are in a running transaction though,
254 * we are already protected against freezing and we cannot grab further
255 * protection due to lock ordering constraints.
257 if (!ext4_journal_current_handle()) {
258 sb_start_intwrite(inode->i_sb);
259 freeze_protected = true;
262 if (!IS_NOQUOTA(inode))
263 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
266 * Block bitmap, group descriptor, and inode are accounted in both
267 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
269 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
270 ext4_blocks_for_truncate(inode) + extra_credits - 3);
271 if (IS_ERR(handle)) {
272 ext4_std_error(inode->i_sb, PTR_ERR(handle));
274 * If we're going to skip the normal cleanup, we still need to
275 * make sure that the in-core orphan linked list is properly
278 ext4_orphan_del(NULL, inode);
279 if (freeze_protected)
280 sb_end_intwrite(inode->i_sb);
285 ext4_handle_sync(handle);
288 * Set inode->i_size to 0 before calling ext4_truncate(). We need
289 * special handling of symlinks here because i_size is used to
290 * determine whether ext4_inode_info->i_data contains symlink data or
291 * block mappings. Setting i_size to 0 will remove its fast symlink
292 * status. Erase i_data so that it becomes a valid empty block map.
294 if (ext4_inode_is_fast_symlink(inode))
295 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
297 err = ext4_mark_inode_dirty(handle, inode);
299 ext4_warning(inode->i_sb,
300 "couldn't mark inode dirty (err %d)", err);
303 if (inode->i_blocks) {
304 err = ext4_truncate(inode);
306 ext4_error(inode->i_sb,
307 "couldn't truncate inode %lu (err %d)",
313 /* Remove xattr references. */
314 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
317 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
319 ext4_journal_stop(handle);
320 ext4_orphan_del(NULL, inode);
321 if (freeze_protected)
322 sb_end_intwrite(inode->i_sb);
323 ext4_xattr_inode_array_free(ea_inode_array);
328 * Kill off the orphan record which ext4_truncate created.
329 * AKPM: I think this can be inside the above `if'.
330 * Note that ext4_orphan_del() has to be able to cope with the
331 * deletion of a non-existent orphan - this is because we don't
332 * know if ext4_truncate() actually created an orphan record.
333 * (Well, we could do this if we need to, but heck - it works)
335 ext4_orphan_del(handle, inode);
336 EXT4_I(inode)->i_dtime = get_seconds();
339 * One subtle ordering requirement: if anything has gone wrong
340 * (transaction abort, IO errors, whatever), then we can still
341 * do these next steps (the fs will already have been marked as
342 * having errors), but we can't free the inode if the mark_dirty
345 if (ext4_mark_inode_dirty(handle, inode))
346 /* If that failed, just do the required in-core inode clear. */
347 ext4_clear_inode(inode);
349 ext4_free_inode(handle, inode);
350 ext4_journal_stop(handle);
351 if (freeze_protected)
352 sb_end_intwrite(inode->i_sb);
353 ext4_xattr_inode_array_free(ea_inode_array);
356 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
360 qsize_t *ext4_get_reserved_space(struct inode *inode)
362 return &EXT4_I(inode)->i_reserved_quota;
367 * Called with i_data_sem down, which is important since we can call
368 * ext4_discard_preallocations() from here.
370 void ext4_da_update_reserve_space(struct inode *inode,
371 int used, int quota_claim)
373 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
374 struct ext4_inode_info *ei = EXT4_I(inode);
376 spin_lock(&ei->i_block_reservation_lock);
377 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
378 if (unlikely(used > ei->i_reserved_data_blocks)) {
379 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
380 "with only %d reserved data blocks",
381 __func__, inode->i_ino, used,
382 ei->i_reserved_data_blocks);
384 used = ei->i_reserved_data_blocks;
387 /* Update per-inode reservations */
388 ei->i_reserved_data_blocks -= used;
389 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
391 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
393 /* Update quota subsystem for data blocks */
395 dquot_claim_block(inode, EXT4_C2B(sbi, used));
398 * We did fallocate with an offset that is already delayed
399 * allocated. So on delayed allocated writeback we should
400 * not re-claim the quota for fallocated blocks.
402 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
406 * If we have done all the pending block allocations and if
407 * there aren't any writers on the inode, we can discard the
408 * inode's preallocations.
410 if ((ei->i_reserved_data_blocks == 0) &&
411 (atomic_read(&inode->i_writecount) == 0))
412 ext4_discard_preallocations(inode);
415 static int __check_block_validity(struct inode *inode, const char *func,
417 struct ext4_map_blocks *map)
419 if (ext4_has_feature_journal(inode->i_sb) &&
421 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
423 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
424 ext4_error_inode(inode, func, line, map->m_pblk,
425 "lblock %lu mapped to illegal pblock %llu "
426 "(length %d)", (unsigned long) map->m_lblk,
427 map->m_pblk, map->m_len);
428 return -EFSCORRUPTED;
433 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
438 if (ext4_encrypted_inode(inode))
439 return fscrypt_zeroout_range(inode, lblk, pblk, len);
441 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
448 #define check_block_validity(inode, map) \
449 __check_block_validity((inode), __func__, __LINE__, (map))
451 #ifdef ES_AGGRESSIVE_TEST
452 static void ext4_map_blocks_es_recheck(handle_t *handle,
454 struct ext4_map_blocks *es_map,
455 struct ext4_map_blocks *map,
462 * There is a race window that the result is not the same.
463 * e.g. xfstests #223 when dioread_nolock enables. The reason
464 * is that we lookup a block mapping in extent status tree with
465 * out taking i_data_sem. So at the time the unwritten extent
466 * could be converted.
468 down_read(&EXT4_I(inode)->i_data_sem);
469 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
470 retval = ext4_ext_map_blocks(handle, inode, map, flags &
471 EXT4_GET_BLOCKS_KEEP_SIZE);
473 retval = ext4_ind_map_blocks(handle, inode, map, flags &
474 EXT4_GET_BLOCKS_KEEP_SIZE);
476 up_read((&EXT4_I(inode)->i_data_sem));
479 * We don't check m_len because extent will be collpased in status
480 * tree. So the m_len might not equal.
482 if (es_map->m_lblk != map->m_lblk ||
483 es_map->m_flags != map->m_flags ||
484 es_map->m_pblk != map->m_pblk) {
485 printk("ES cache assertion failed for inode: %lu "
486 "es_cached ex [%d/%d/%llu/%x] != "
487 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
488 inode->i_ino, es_map->m_lblk, es_map->m_len,
489 es_map->m_pblk, es_map->m_flags, map->m_lblk,
490 map->m_len, map->m_pblk, map->m_flags,
494 #endif /* ES_AGGRESSIVE_TEST */
497 * The ext4_map_blocks() function tries to look up the requested blocks,
498 * and returns if the blocks are already mapped.
500 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
501 * and store the allocated blocks in the result buffer head and mark it
504 * If file type is extents based, it will call ext4_ext_map_blocks(),
505 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
508 * On success, it returns the number of blocks being mapped or allocated. if
509 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
510 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
512 * It returns 0 if plain look up failed (blocks have not been allocated), in
513 * that case, @map is returned as unmapped but we still do fill map->m_len to
514 * indicate the length of a hole starting at map->m_lblk.
516 * It returns the error in case of allocation failure.
518 int ext4_map_blocks(handle_t *handle, struct inode *inode,
519 struct ext4_map_blocks *map, int flags)
521 struct extent_status es;
524 #ifdef ES_AGGRESSIVE_TEST
525 struct ext4_map_blocks orig_map;
527 memcpy(&orig_map, map, sizeof(*map));
531 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
532 "logical block %lu\n", inode->i_ino, flags, map->m_len,
533 (unsigned long) map->m_lblk);
536 * ext4_map_blocks returns an int, and m_len is an unsigned int
538 if (unlikely(map->m_len > INT_MAX))
539 map->m_len = INT_MAX;
541 /* We can handle the block number less than EXT_MAX_BLOCKS */
542 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
543 return -EFSCORRUPTED;
545 /* Lookup extent status tree firstly */
546 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
547 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
548 map->m_pblk = ext4_es_pblock(&es) +
549 map->m_lblk - es.es_lblk;
550 map->m_flags |= ext4_es_is_written(&es) ?
551 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
552 retval = es.es_len - (map->m_lblk - es.es_lblk);
553 if (retval > map->m_len)
556 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
558 retval = es.es_len - (map->m_lblk - es.es_lblk);
559 if (retval > map->m_len)
566 #ifdef ES_AGGRESSIVE_TEST
567 ext4_map_blocks_es_recheck(handle, inode, map,
574 * Try to see if we can get the block without requesting a new
577 down_read(&EXT4_I(inode)->i_data_sem);
578 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
579 retval = ext4_ext_map_blocks(handle, inode, map, flags &
580 EXT4_GET_BLOCKS_KEEP_SIZE);
582 retval = ext4_ind_map_blocks(handle, inode, map, flags &
583 EXT4_GET_BLOCKS_KEEP_SIZE);
588 if (unlikely(retval != map->m_len)) {
589 ext4_warning(inode->i_sb,
590 "ES len assertion failed for inode "
591 "%lu: retval %d != map->m_len %d",
592 inode->i_ino, retval, map->m_len);
596 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
597 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
598 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
599 !(status & EXTENT_STATUS_WRITTEN) &&
600 ext4_find_delalloc_range(inode, map->m_lblk,
601 map->m_lblk + map->m_len - 1))
602 status |= EXTENT_STATUS_DELAYED;
603 ret = ext4_es_insert_extent(inode, map->m_lblk,
604 map->m_len, map->m_pblk, status);
608 up_read((&EXT4_I(inode)->i_data_sem));
611 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
612 ret = check_block_validity(inode, map);
617 /* If it is only a block(s) look up */
618 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
622 * Returns if the blocks have already allocated
624 * Note that if blocks have been preallocated
625 * ext4_ext_get_block() returns the create = 0
626 * with buffer head unmapped.
628 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
630 * If we need to convert extent to unwritten
631 * we continue and do the actual work in
632 * ext4_ext_map_blocks()
634 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
638 * Here we clear m_flags because after allocating an new extent,
639 * it will be set again.
641 map->m_flags &= ~EXT4_MAP_FLAGS;
644 * New blocks allocate and/or writing to unwritten extent
645 * will possibly result in updating i_data, so we take
646 * the write lock of i_data_sem, and call get_block()
647 * with create == 1 flag.
649 down_write(&EXT4_I(inode)->i_data_sem);
652 * We need to check for EXT4 here because migrate
653 * could have changed the inode type in between
655 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
656 retval = ext4_ext_map_blocks(handle, inode, map, flags);
658 retval = ext4_ind_map_blocks(handle, inode, map, flags);
660 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
662 * We allocated new blocks which will result in
663 * i_data's format changing. Force the migrate
664 * to fail by clearing migrate flags
666 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
670 * Update reserved blocks/metadata blocks after successful
671 * block allocation which had been deferred till now. We don't
672 * support fallocate for non extent files. So we can update
673 * reserve space here.
676 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
677 ext4_da_update_reserve_space(inode, retval, 1);
683 if (unlikely(retval != map->m_len)) {
684 ext4_warning(inode->i_sb,
685 "ES len assertion failed for inode "
686 "%lu: retval %d != map->m_len %d",
687 inode->i_ino, retval, map->m_len);
692 * We have to zeroout blocks before inserting them into extent
693 * status tree. Otherwise someone could look them up there and
694 * use them before they are really zeroed. We also have to
695 * unmap metadata before zeroing as otherwise writeback can
696 * overwrite zeros with stale data from block device.
698 if (flags & EXT4_GET_BLOCKS_ZERO &&
699 map->m_flags & EXT4_MAP_MAPPED &&
700 map->m_flags & EXT4_MAP_NEW) {
701 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
703 ret = ext4_issue_zeroout(inode, map->m_lblk,
704 map->m_pblk, map->m_len);
712 * If the extent has been zeroed out, we don't need to update
713 * extent status tree.
715 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
716 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
717 if (ext4_es_is_written(&es))
720 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
721 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
722 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
723 !(status & EXTENT_STATUS_WRITTEN) &&
724 ext4_find_delalloc_range(inode, map->m_lblk,
725 map->m_lblk + map->m_len - 1))
726 status |= EXTENT_STATUS_DELAYED;
727 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
728 map->m_pblk, status);
736 up_write((&EXT4_I(inode)->i_data_sem));
737 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
738 ret = check_block_validity(inode, map);
743 * Inodes with freshly allocated blocks where contents will be
744 * visible after transaction commit must be on transaction's
747 if (map->m_flags & EXT4_MAP_NEW &&
748 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
749 !(flags & EXT4_GET_BLOCKS_ZERO) &&
750 !ext4_is_quota_file(inode) &&
751 ext4_should_order_data(inode)) {
753 (loff_t)map->m_lblk << inode->i_blkbits;
754 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
756 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
757 ret = ext4_jbd2_inode_add_wait(handle, inode,
760 ret = ext4_jbd2_inode_add_write(handle, inode,
770 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
771 * we have to be careful as someone else may be manipulating b_state as well.
773 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
775 unsigned long old_state;
776 unsigned long new_state;
778 flags &= EXT4_MAP_FLAGS;
780 /* Dummy buffer_head? Set non-atomically. */
782 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
786 * Someone else may be modifying b_state. Be careful! This is ugly but
787 * once we get rid of using bh as a container for mapping information
788 * to pass to / from get_block functions, this can go away.
791 old_state = READ_ONCE(bh->b_state);
792 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
794 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
797 static int _ext4_get_block(struct inode *inode, sector_t iblock,
798 struct buffer_head *bh, int flags)
800 struct ext4_map_blocks map;
803 if (ext4_has_inline_data(inode))
807 map.m_len = bh->b_size >> inode->i_blkbits;
809 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
812 map_bh(bh, inode->i_sb, map.m_pblk);
813 ext4_update_bh_state(bh, map.m_flags);
814 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
816 } else if (ret == 0) {
817 /* hole case, need to fill in bh->b_size */
818 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
823 int ext4_get_block(struct inode *inode, sector_t iblock,
824 struct buffer_head *bh, int create)
826 return _ext4_get_block(inode, iblock, bh,
827 create ? EXT4_GET_BLOCKS_CREATE : 0);
831 * Get block function used when preparing for buffered write if we require
832 * creating an unwritten extent if blocks haven't been allocated. The extent
833 * will be converted to written after the IO is complete.
835 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
836 struct buffer_head *bh_result, int create)
838 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
839 inode->i_ino, create);
840 return _ext4_get_block(inode, iblock, bh_result,
841 EXT4_GET_BLOCKS_IO_CREATE_EXT);
844 /* Maximum number of blocks we map for direct IO at once. */
845 #define DIO_MAX_BLOCKS 4096
848 * Get blocks function for the cases that need to start a transaction -
849 * generally difference cases of direct IO and DAX IO. It also handles retries
852 static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
853 struct buffer_head *bh_result, int flags)
860 /* Trim mapping request to maximum we can map at once for DIO */
861 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
862 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
863 dio_credits = ext4_chunk_trans_blocks(inode,
864 bh_result->b_size >> inode->i_blkbits);
866 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
868 return PTR_ERR(handle);
870 ret = _ext4_get_block(inode, iblock, bh_result, flags);
871 ext4_journal_stop(handle);
873 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
878 /* Get block function for DIO reads and writes to inodes without extents */
879 int ext4_dio_get_block(struct inode *inode, sector_t iblock,
880 struct buffer_head *bh, int create)
882 /* We don't expect handle for direct IO */
883 WARN_ON_ONCE(ext4_journal_current_handle());
886 return _ext4_get_block(inode, iblock, bh, 0);
887 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
891 * Get block function for AIO DIO writes when we create unwritten extent if
892 * blocks are not allocated yet. The extent will be converted to written
893 * after IO is complete.
895 static int ext4_dio_get_block_unwritten_async(struct inode *inode,
896 sector_t iblock, struct buffer_head *bh_result, int create)
900 /* We don't expect handle for direct IO */
901 WARN_ON_ONCE(ext4_journal_current_handle());
903 ret = ext4_get_block_trans(inode, iblock, bh_result,
904 EXT4_GET_BLOCKS_IO_CREATE_EXT);
907 * When doing DIO using unwritten extents, we need io_end to convert
908 * unwritten extents to written on IO completion. We allocate io_end
909 * once we spot unwritten extent and store it in b_private. Generic
910 * DIO code keeps b_private set and furthermore passes the value to
911 * our completion callback in 'private' argument.
913 if (!ret && buffer_unwritten(bh_result)) {
914 if (!bh_result->b_private) {
915 ext4_io_end_t *io_end;
917 io_end = ext4_init_io_end(inode, GFP_KERNEL);
920 bh_result->b_private = io_end;
921 ext4_set_io_unwritten_flag(inode, io_end);
923 set_buffer_defer_completion(bh_result);
930 * Get block function for non-AIO DIO writes when we create unwritten extent if
931 * blocks are not allocated yet. The extent will be converted to written
932 * after IO is complete by ext4_direct_IO_write().
934 static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
935 sector_t iblock, struct buffer_head *bh_result, int create)
939 /* We don't expect handle for direct IO */
940 WARN_ON_ONCE(ext4_journal_current_handle());
942 ret = ext4_get_block_trans(inode, iblock, bh_result,
943 EXT4_GET_BLOCKS_IO_CREATE_EXT);
946 * Mark inode as having pending DIO writes to unwritten extents.
947 * ext4_direct_IO_write() checks this flag and converts extents to
950 if (!ret && buffer_unwritten(bh_result))
951 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
956 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
957 struct buffer_head *bh_result, int create)
961 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
962 inode->i_ino, create);
963 /* We don't expect handle for direct IO */
964 WARN_ON_ONCE(ext4_journal_current_handle());
966 ret = _ext4_get_block(inode, iblock, bh_result, 0);
968 * Blocks should have been preallocated! ext4_file_write_iter() checks
971 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
978 * `handle' can be NULL if create is zero
980 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
981 ext4_lblk_t block, int map_flags)
983 struct ext4_map_blocks map;
984 struct buffer_head *bh;
985 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
988 J_ASSERT(handle != NULL || create == 0);
992 err = ext4_map_blocks(handle, inode, &map, map_flags);
995 return create ? ERR_PTR(-ENOSPC) : NULL;
999 bh = sb_getblk(inode->i_sb, map.m_pblk);
1001 return ERR_PTR(-ENOMEM);
1002 if (map.m_flags & EXT4_MAP_NEW) {
1003 J_ASSERT(create != 0);
1004 J_ASSERT(handle != NULL);
1007 * Now that we do not always journal data, we should
1008 * keep in mind whether this should always journal the
1009 * new buffer as metadata. For now, regular file
1010 * writes use ext4_get_block instead, so it's not a
1014 BUFFER_TRACE(bh, "call get_create_access");
1015 err = ext4_journal_get_create_access(handle, bh);
1016 if (unlikely(err)) {
1020 if (!buffer_uptodate(bh)) {
1021 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1022 set_buffer_uptodate(bh);
1025 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1026 err = ext4_handle_dirty_metadata(handle, inode, bh);
1030 BUFFER_TRACE(bh, "not a new buffer");
1034 return ERR_PTR(err);
1037 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1038 ext4_lblk_t block, int map_flags)
1040 struct buffer_head *bh;
1042 bh = ext4_getblk(handle, inode, block, map_flags);
1045 if (!bh || buffer_uptodate(bh))
1047 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1049 if (buffer_uptodate(bh))
1052 return ERR_PTR(-EIO);
1055 /* Read a contiguous batch of blocks. */
1056 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1057 bool wait, struct buffer_head **bhs)
1061 for (i = 0; i < bh_count; i++) {
1062 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1063 if (IS_ERR(bhs[i])) {
1064 err = PTR_ERR(bhs[i]);
1070 for (i = 0; i < bh_count; i++)
1071 /* Note that NULL bhs[i] is valid because of holes. */
1072 if (bhs[i] && !buffer_uptodate(bhs[i]))
1073 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1079 for (i = 0; i < bh_count; i++)
1081 wait_on_buffer(bhs[i]);
1083 for (i = 0; i < bh_count; i++) {
1084 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1092 for (i = 0; i < bh_count; i++) {
1099 int ext4_walk_page_buffers(handle_t *handle,
1100 struct buffer_head *head,
1104 int (*fn)(handle_t *handle,
1105 struct buffer_head *bh))
1107 struct buffer_head *bh;
1108 unsigned block_start, block_end;
1109 unsigned blocksize = head->b_size;
1111 struct buffer_head *next;
1113 for (bh = head, block_start = 0;
1114 ret == 0 && (bh != head || !block_start);
1115 block_start = block_end, bh = next) {
1116 next = bh->b_this_page;
1117 block_end = block_start + blocksize;
1118 if (block_end <= from || block_start >= to) {
1119 if (partial && !buffer_uptodate(bh))
1123 err = (*fn)(handle, bh);
1131 * To preserve ordering, it is essential that the hole instantiation and
1132 * the data write be encapsulated in a single transaction. We cannot
1133 * close off a transaction and start a new one between the ext4_get_block()
1134 * and the commit_write(). So doing the jbd2_journal_start at the start of
1135 * prepare_write() is the right place.
1137 * Also, this function can nest inside ext4_writepage(). In that case, we
1138 * *know* that ext4_writepage() has generated enough buffer credits to do the
1139 * whole page. So we won't block on the journal in that case, which is good,
1140 * because the caller may be PF_MEMALLOC.
1142 * By accident, ext4 can be reentered when a transaction is open via
1143 * quota file writes. If we were to commit the transaction while thus
1144 * reentered, there can be a deadlock - we would be holding a quota
1145 * lock, and the commit would never complete if another thread had a
1146 * transaction open and was blocking on the quota lock - a ranking
1149 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1150 * will _not_ run commit under these circumstances because handle->h_ref
1151 * is elevated. We'll still have enough credits for the tiny quotafile
1154 int do_journal_get_write_access(handle_t *handle,
1155 struct buffer_head *bh)
1157 int dirty = buffer_dirty(bh);
1160 if (!buffer_mapped(bh) || buffer_freed(bh))
1163 * __block_write_begin() could have dirtied some buffers. Clean
1164 * the dirty bit as jbd2_journal_get_write_access() could complain
1165 * otherwise about fs integrity issues. Setting of the dirty bit
1166 * by __block_write_begin() isn't a real problem here as we clear
1167 * the bit before releasing a page lock and thus writeback cannot
1168 * ever write the buffer.
1171 clear_buffer_dirty(bh);
1172 BUFFER_TRACE(bh, "get write access");
1173 ret = ext4_journal_get_write_access(handle, bh);
1175 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1179 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1180 static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1181 get_block_t *get_block)
1183 unsigned from = pos & (PAGE_SIZE - 1);
1184 unsigned to = from + len;
1185 struct inode *inode = page->mapping->host;
1186 unsigned block_start, block_end;
1189 unsigned blocksize = inode->i_sb->s_blocksize;
1191 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1192 bool decrypt = false;
1194 BUG_ON(!PageLocked(page));
1195 BUG_ON(from > PAGE_SIZE);
1196 BUG_ON(to > PAGE_SIZE);
1199 if (!page_has_buffers(page))
1200 create_empty_buffers(page, blocksize, 0);
1201 head = page_buffers(page);
1202 bbits = ilog2(blocksize);
1203 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1205 for (bh = head, block_start = 0; bh != head || !block_start;
1206 block++, block_start = block_end, bh = bh->b_this_page) {
1207 block_end = block_start + blocksize;
1208 if (block_end <= from || block_start >= to) {
1209 if (PageUptodate(page)) {
1210 if (!buffer_uptodate(bh))
1211 set_buffer_uptodate(bh);
1216 clear_buffer_new(bh);
1217 if (!buffer_mapped(bh)) {
1218 WARN_ON(bh->b_size != blocksize);
1219 err = get_block(inode, block, bh, 1);
1222 if (buffer_new(bh)) {
1223 clean_bdev_bh_alias(bh);
1224 if (PageUptodate(page)) {
1225 clear_buffer_new(bh);
1226 set_buffer_uptodate(bh);
1227 mark_buffer_dirty(bh);
1230 if (block_end > to || block_start < from)
1231 zero_user_segments(page, to, block_end,
1236 if (PageUptodate(page)) {
1237 if (!buffer_uptodate(bh))
1238 set_buffer_uptodate(bh);
1241 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1242 !buffer_unwritten(bh) &&
1243 (block_start < from || block_end > to)) {
1244 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1246 decrypt = ext4_encrypted_inode(inode) &&
1247 S_ISREG(inode->i_mode);
1251 * If we issued read requests, let them complete.
1253 while (wait_bh > wait) {
1254 wait_on_buffer(*--wait_bh);
1255 if (!buffer_uptodate(*wait_bh))
1259 page_zero_new_buffers(page, from, to);
1261 err = fscrypt_decrypt_page(page->mapping->host, page,
1262 PAGE_SIZE, 0, page->index);
1267 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1268 loff_t pos, unsigned len, unsigned flags,
1269 struct page **pagep, void **fsdata)
1271 struct inode *inode = mapping->host;
1272 int ret, needed_blocks;
1279 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1282 trace_ext4_write_begin(inode, pos, len, flags);
1284 * Reserve one block more for addition to orphan list in case
1285 * we allocate blocks but write fails for some reason
1287 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1288 index = pos >> PAGE_SHIFT;
1289 from = pos & (PAGE_SIZE - 1);
1292 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1293 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1302 * grab_cache_page_write_begin() can take a long time if the
1303 * system is thrashing due to memory pressure, or if the page
1304 * is being written back. So grab it first before we start
1305 * the transaction handle. This also allows us to allocate
1306 * the page (if needed) without using GFP_NOFS.
1309 page = grab_cache_page_write_begin(mapping, index, flags);
1313 * The same as page allocation, we prealloc buffer heads before
1314 * starting the handle.
1316 if (!page_has_buffers(page))
1317 create_empty_buffers(page, inode->i_sb->s_blocksize, 0);
1322 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1323 if (IS_ERR(handle)) {
1325 return PTR_ERR(handle);
1329 if (page->mapping != mapping) {
1330 /* The page got truncated from under us */
1333 ext4_journal_stop(handle);
1336 /* In case writeback began while the page was unlocked */
1337 wait_for_stable_page(page);
1339 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1340 if (ext4_should_dioread_nolock(inode))
1341 ret = ext4_block_write_begin(page, pos, len,
1342 ext4_get_block_unwritten);
1344 ret = ext4_block_write_begin(page, pos, len,
1347 if (ext4_should_dioread_nolock(inode))
1348 ret = __block_write_begin(page, pos, len,
1349 ext4_get_block_unwritten);
1351 ret = __block_write_begin(page, pos, len, ext4_get_block);
1353 if (!ret && ext4_should_journal_data(inode)) {
1354 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1356 do_journal_get_write_access);
1362 * __block_write_begin may have instantiated a few blocks
1363 * outside i_size. Trim these off again. Don't need
1364 * i_size_read because we hold i_mutex.
1366 * Add inode to orphan list in case we crash before
1369 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1370 ext4_orphan_add(handle, inode);
1372 ext4_journal_stop(handle);
1373 if (pos + len > inode->i_size) {
1374 ext4_truncate_failed_write(inode);
1376 * If truncate failed early the inode might
1377 * still be on the orphan list; we need to
1378 * make sure the inode is removed from the
1379 * orphan list in that case.
1382 ext4_orphan_del(NULL, inode);
1385 if (ret == -ENOSPC &&
1386 ext4_should_retry_alloc(inode->i_sb, &retries))
1395 /* For write_end() in data=journal mode */
1396 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1399 if (!buffer_mapped(bh) || buffer_freed(bh))
1401 set_buffer_uptodate(bh);
1402 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1403 clear_buffer_meta(bh);
1404 clear_buffer_prio(bh);
1409 * We need to pick up the new inode size which generic_commit_write gave us
1410 * `file' can be NULL - eg, when called from page_symlink().
1412 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1413 * buffers are managed internally.
1415 static int ext4_write_end(struct file *file,
1416 struct address_space *mapping,
1417 loff_t pos, unsigned len, unsigned copied,
1418 struct page *page, void *fsdata)
1420 handle_t *handle = ext4_journal_current_handle();
1421 struct inode *inode = mapping->host;
1422 loff_t old_size = inode->i_size;
1424 int i_size_changed = 0;
1425 int inline_data = ext4_has_inline_data(inode);
1427 trace_ext4_write_end(inode, pos, len, copied);
1429 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1430 ret = ext4_write_inline_data_end(inode, pos, len,
1439 copied = block_write_end(file, mapping, pos,
1440 len, copied, page, fsdata);
1442 * it's important to update i_size while still holding page lock:
1443 * page writeout could otherwise come in and zero beyond i_size.
1445 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1450 pagecache_isize_extended(inode, old_size, pos);
1452 * Don't mark the inode dirty under page lock. First, it unnecessarily
1453 * makes the holding time of page lock longer. Second, it forces lock
1454 * ordering of page lock and transaction start for journaling
1457 if (i_size_changed || inline_data)
1458 ext4_mark_inode_dirty(handle, inode);
1460 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1461 /* if we have allocated more blocks and copied
1462 * less. We will have blocks allocated outside
1463 * inode->i_size. So truncate them
1465 ext4_orphan_add(handle, inode);
1467 ret2 = ext4_journal_stop(handle);
1471 if (pos + len > inode->i_size) {
1472 ext4_truncate_failed_write(inode);
1474 * If truncate failed early the inode might still be
1475 * on the orphan list; we need to make sure the inode
1476 * is removed from the orphan list in that case.
1479 ext4_orphan_del(NULL, inode);
1482 return ret ? ret : copied;
1486 * This is a private version of page_zero_new_buffers() which doesn't
1487 * set the buffer to be dirty, since in data=journalled mode we need
1488 * to call ext4_handle_dirty_metadata() instead.
1490 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1492 unsigned from, unsigned to)
1494 unsigned int block_start = 0, block_end;
1495 struct buffer_head *head, *bh;
1497 bh = head = page_buffers(page);
1499 block_end = block_start + bh->b_size;
1500 if (buffer_new(bh)) {
1501 if (block_end > from && block_start < to) {
1502 if (!PageUptodate(page)) {
1503 unsigned start, size;
1505 start = max(from, block_start);
1506 size = min(to, block_end) - start;
1508 zero_user(page, start, size);
1509 write_end_fn(handle, bh);
1511 clear_buffer_new(bh);
1514 block_start = block_end;
1515 bh = bh->b_this_page;
1516 } while (bh != head);
1519 static int ext4_journalled_write_end(struct file *file,
1520 struct address_space *mapping,
1521 loff_t pos, unsigned len, unsigned copied,
1522 struct page *page, void *fsdata)
1524 handle_t *handle = ext4_journal_current_handle();
1525 struct inode *inode = mapping->host;
1526 loff_t old_size = inode->i_size;
1530 int size_changed = 0;
1531 int inline_data = ext4_has_inline_data(inode);
1533 trace_ext4_journalled_write_end(inode, pos, len, copied);
1534 from = pos & (PAGE_SIZE - 1);
1537 BUG_ON(!ext4_handle_valid(handle));
1540 ret = ext4_write_inline_data_end(inode, pos, len,
1548 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1550 ext4_journalled_zero_new_buffers(handle, page, from, to);
1552 if (unlikely(copied < len))
1553 ext4_journalled_zero_new_buffers(handle, page,
1555 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1556 from + copied, &partial,
1559 SetPageUptodate(page);
1561 size_changed = ext4_update_inode_size(inode, pos + copied);
1562 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1563 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1568 pagecache_isize_extended(inode, old_size, pos);
1570 if (size_changed || inline_data) {
1571 ret2 = ext4_mark_inode_dirty(handle, inode);
1576 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1577 /* if we have allocated more blocks and copied
1578 * less. We will have blocks allocated outside
1579 * inode->i_size. So truncate them
1581 ext4_orphan_add(handle, inode);
1584 ret2 = ext4_journal_stop(handle);
1587 if (pos + len > inode->i_size) {
1588 ext4_truncate_failed_write(inode);
1590 * If truncate failed early the inode might still be
1591 * on the orphan list; we need to make sure the inode
1592 * is removed from the orphan list in that case.
1595 ext4_orphan_del(NULL, inode);
1598 return ret ? ret : copied;
1602 * Reserve space for a single cluster
1604 static int ext4_da_reserve_space(struct inode *inode)
1606 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1607 struct ext4_inode_info *ei = EXT4_I(inode);
1611 * We will charge metadata quota at writeout time; this saves
1612 * us from metadata over-estimation, though we may go over by
1613 * a small amount in the end. Here we just reserve for data.
1615 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1619 spin_lock(&ei->i_block_reservation_lock);
1620 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1621 spin_unlock(&ei->i_block_reservation_lock);
1622 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1625 ei->i_reserved_data_blocks++;
1626 trace_ext4_da_reserve_space(inode);
1627 spin_unlock(&ei->i_block_reservation_lock);
1629 return 0; /* success */
1632 static void ext4_da_release_space(struct inode *inode, int to_free)
1634 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1635 struct ext4_inode_info *ei = EXT4_I(inode);
1638 return; /* Nothing to release, exit */
1640 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1642 trace_ext4_da_release_space(inode, to_free);
1643 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1645 * if there aren't enough reserved blocks, then the
1646 * counter is messed up somewhere. Since this
1647 * function is called from invalidate page, it's
1648 * harmless to return without any action.
1650 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1651 "ino %lu, to_free %d with only %d reserved "
1652 "data blocks", inode->i_ino, to_free,
1653 ei->i_reserved_data_blocks);
1655 to_free = ei->i_reserved_data_blocks;
1657 ei->i_reserved_data_blocks -= to_free;
1659 /* update fs dirty data blocks counter */
1660 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1662 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1664 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1667 static void ext4_da_page_release_reservation(struct page *page,
1668 unsigned int offset,
1669 unsigned int length)
1671 int to_release = 0, contiguous_blks = 0;
1672 struct buffer_head *head, *bh;
1673 unsigned int curr_off = 0;
1674 struct inode *inode = page->mapping->host;
1675 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1676 unsigned int stop = offset + length;
1680 BUG_ON(stop > PAGE_SIZE || stop < length);
1682 head = page_buffers(page);
1685 unsigned int next_off = curr_off + bh->b_size;
1687 if (next_off > stop)
1690 if ((offset <= curr_off) && (buffer_delay(bh))) {
1693 clear_buffer_delay(bh);
1694 } else if (contiguous_blks) {
1695 lblk = page->index <<
1696 (PAGE_SHIFT - inode->i_blkbits);
1697 lblk += (curr_off >> inode->i_blkbits) -
1699 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1700 contiguous_blks = 0;
1702 curr_off = next_off;
1703 } while ((bh = bh->b_this_page) != head);
1705 if (contiguous_blks) {
1706 lblk = page->index << (PAGE_SHIFT - inode->i_blkbits);
1707 lblk += (curr_off >> inode->i_blkbits) - contiguous_blks;
1708 ext4_es_remove_extent(inode, lblk, contiguous_blks);
1711 /* If we have released all the blocks belonging to a cluster, then we
1712 * need to release the reserved space for that cluster. */
1713 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1714 while (num_clusters > 0) {
1715 lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) +
1716 ((num_clusters - 1) << sbi->s_cluster_bits);
1717 if (sbi->s_cluster_ratio == 1 ||
1718 !ext4_find_delalloc_cluster(inode, lblk))
1719 ext4_da_release_space(inode, 1);
1726 * Delayed allocation stuff
1729 struct mpage_da_data {
1730 struct inode *inode;
1731 struct writeback_control *wbc;
1733 pgoff_t first_page; /* The first page to write */
1734 pgoff_t next_page; /* Current page to examine */
1735 pgoff_t last_page; /* Last page to examine */
1737 * Extent to map - this can be after first_page because that can be
1738 * fully mapped. We somewhat abuse m_flags to store whether the extent
1739 * is delalloc or unwritten.
1741 struct ext4_map_blocks map;
1742 struct ext4_io_submit io_submit; /* IO submission data */
1743 unsigned int do_map:1;
1746 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1751 struct pagevec pvec;
1752 struct inode *inode = mpd->inode;
1753 struct address_space *mapping = inode->i_mapping;
1755 /* This is necessary when next_page == 0. */
1756 if (mpd->first_page >= mpd->next_page)
1759 index = mpd->first_page;
1760 end = mpd->next_page - 1;
1762 ext4_lblk_t start, last;
1763 start = index << (PAGE_SHIFT - inode->i_blkbits);
1764 last = end << (PAGE_SHIFT - inode->i_blkbits);
1767 * avoid racing with extent status tree scans made by
1768 * ext4_insert_delayed_block()
1770 down_write(&EXT4_I(inode)->i_data_sem);
1771 ext4_es_remove_extent(inode, start, last - start + 1);
1772 up_write(&EXT4_I(inode)->i_data_sem);
1775 pagevec_init(&pvec, 0);
1776 while (index <= end) {
1777 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1780 for (i = 0; i < nr_pages; i++) {
1781 struct page *page = pvec.pages[i];
1783 BUG_ON(!PageLocked(page));
1784 BUG_ON(PageWriteback(page));
1786 if (page_mapped(page))
1787 clear_page_dirty_for_io(page);
1788 block_invalidatepage(page, 0, PAGE_SIZE);
1789 ClearPageUptodate(page);
1793 pagevec_release(&pvec);
1797 static void ext4_print_free_blocks(struct inode *inode)
1799 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1800 struct super_block *sb = inode->i_sb;
1801 struct ext4_inode_info *ei = EXT4_I(inode);
1803 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1804 EXT4_C2B(EXT4_SB(inode->i_sb),
1805 ext4_count_free_clusters(sb)));
1806 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1807 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1808 (long long) EXT4_C2B(EXT4_SB(sb),
1809 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1810 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1811 (long long) EXT4_C2B(EXT4_SB(sb),
1812 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1813 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1814 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1815 ei->i_reserved_data_blocks);
1819 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1821 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1825 * This function is grabs code from the very beginning of
1826 * ext4_map_blocks, but assumes that the caller is from delayed write
1827 * time. This function looks up the requested blocks and sets the
1828 * buffer delay bit under the protection of i_data_sem.
1830 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1831 struct ext4_map_blocks *map,
1832 struct buffer_head *bh)
1834 struct extent_status es;
1836 sector_t invalid_block = ~((sector_t) 0xffff);
1837 #ifdef ES_AGGRESSIVE_TEST
1838 struct ext4_map_blocks orig_map;
1840 memcpy(&orig_map, map, sizeof(*map));
1843 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1847 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1848 "logical block %lu\n", inode->i_ino, map->m_len,
1849 (unsigned long) map->m_lblk);
1851 /* Lookup extent status tree firstly */
1852 if (ext4_es_lookup_extent(inode, iblock, &es)) {
1853 if (ext4_es_is_hole(&es)) {
1855 down_read(&EXT4_I(inode)->i_data_sem);
1860 * Delayed extent could be allocated by fallocate.
1861 * So we need to check it.
1863 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1864 map_bh(bh, inode->i_sb, invalid_block);
1866 set_buffer_delay(bh);
1870 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1871 retval = es.es_len - (iblock - es.es_lblk);
1872 if (retval > map->m_len)
1873 retval = map->m_len;
1874 map->m_len = retval;
1875 if (ext4_es_is_written(&es))
1876 map->m_flags |= EXT4_MAP_MAPPED;
1877 else if (ext4_es_is_unwritten(&es))
1878 map->m_flags |= EXT4_MAP_UNWRITTEN;
1882 #ifdef ES_AGGRESSIVE_TEST
1883 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1889 * Try to see if we can get the block without requesting a new
1890 * file system block.
1892 down_read(&EXT4_I(inode)->i_data_sem);
1893 if (ext4_has_inline_data(inode))
1895 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1896 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1898 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1904 * XXX: __block_prepare_write() unmaps passed block,
1908 * If the block was allocated from previously allocated cluster,
1909 * then we don't need to reserve it again. However we still need
1910 * to reserve metadata for every block we're going to write.
1912 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
1913 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
1914 ret = ext4_da_reserve_space(inode);
1916 /* not enough space to reserve */
1922 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1923 ~0, EXTENT_STATUS_DELAYED);
1929 map_bh(bh, inode->i_sb, invalid_block);
1931 set_buffer_delay(bh);
1932 } else if (retval > 0) {
1934 unsigned int status;
1936 if (unlikely(retval != map->m_len)) {
1937 ext4_warning(inode->i_sb,
1938 "ES len assertion failed for inode "
1939 "%lu: retval %d != map->m_len %d",
1940 inode->i_ino, retval, map->m_len);
1944 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1945 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1946 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1947 map->m_pblk, status);
1953 up_read((&EXT4_I(inode)->i_data_sem));
1959 * This is a special get_block_t callback which is used by
1960 * ext4_da_write_begin(). It will either return mapped block or
1961 * reserve space for a single block.
1963 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1964 * We also have b_blocknr = -1 and b_bdev initialized properly
1966 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1967 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1968 * initialized properly.
1970 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1971 struct buffer_head *bh, int create)
1973 struct ext4_map_blocks map;
1976 BUG_ON(create == 0);
1977 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1979 map.m_lblk = iblock;
1983 * first, we need to know whether the block is allocated already
1984 * preallocated blocks are unmapped but should treated
1985 * the same as allocated blocks.
1987 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1991 map_bh(bh, inode->i_sb, map.m_pblk);
1992 ext4_update_bh_state(bh, map.m_flags);
1994 if (buffer_unwritten(bh)) {
1995 /* A delayed write to unwritten bh should be marked
1996 * new and mapped. Mapped ensures that we don't do
1997 * get_block multiple times when we write to the same
1998 * offset and new ensures that we do proper zero out
1999 * for partial write.
2002 set_buffer_mapped(bh);
2007 static int bget_one(handle_t *handle, struct buffer_head *bh)
2013 static int bput_one(handle_t *handle, struct buffer_head *bh)
2019 static int __ext4_journalled_writepage(struct page *page,
2022 struct address_space *mapping = page->mapping;
2023 struct inode *inode = mapping->host;
2024 struct buffer_head *page_bufs = NULL;
2025 handle_t *handle = NULL;
2026 int ret = 0, err = 0;
2027 int inline_data = ext4_has_inline_data(inode);
2028 struct buffer_head *inode_bh = NULL;
2030 ClearPageChecked(page);
2033 BUG_ON(page->index != 0);
2034 BUG_ON(len > ext4_get_max_inline_size(inode));
2035 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2036 if (inode_bh == NULL)
2039 page_bufs = page_buffers(page);
2044 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2048 * We need to release the page lock before we start the
2049 * journal, so grab a reference so the page won't disappear
2050 * out from under us.
2055 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2056 ext4_writepage_trans_blocks(inode));
2057 if (IS_ERR(handle)) {
2058 ret = PTR_ERR(handle);
2060 goto out_no_pagelock;
2062 BUG_ON(!ext4_handle_valid(handle));
2066 if (page->mapping != mapping) {
2067 /* The page got truncated from under us */
2068 ext4_journal_stop(handle);
2074 ret = ext4_mark_inode_dirty(handle, inode);
2076 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2077 do_journal_get_write_access);
2079 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2084 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2085 err = ext4_journal_stop(handle);
2089 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2093 if (!inline_data && page_bufs)
2094 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2101 * Note that we don't need to start a transaction unless we're journaling data
2102 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2103 * need to file the inode to the transaction's list in ordered mode because if
2104 * we are writing back data added by write(), the inode is already there and if
2105 * we are writing back data modified via mmap(), no one guarantees in which
2106 * transaction the data will hit the disk. In case we are journaling data, we
2107 * cannot start transaction directly because transaction start ranks above page
2108 * lock so we have to do some magic.
2110 * This function can get called via...
2111 * - ext4_writepages after taking page lock (have journal handle)
2112 * - journal_submit_inode_data_buffers (no journal handle)
2113 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2114 * - grab_page_cache when doing write_begin (have journal handle)
2116 * We don't do any block allocation in this function. If we have page with
2117 * multiple blocks we need to write those buffer_heads that are mapped. This
2118 * is important for mmaped based write. So if we do with blocksize 1K
2119 * truncate(f, 1024);
2120 * a = mmap(f, 0, 4096);
2122 * truncate(f, 4096);
2123 * we have in the page first buffer_head mapped via page_mkwrite call back
2124 * but other buffer_heads would be unmapped but dirty (dirty done via the
2125 * do_wp_page). So writepage should write the first block. If we modify
2126 * the mmap area beyond 1024 we will again get a page_fault and the
2127 * page_mkwrite callback will do the block allocation and mark the
2128 * buffer_heads mapped.
2130 * We redirty the page if we have any buffer_heads that is either delay or
2131 * unwritten in the page.
2133 * We can get recursively called as show below.
2135 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2138 * But since we don't do any block allocation we should not deadlock.
2139 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2141 static int ext4_writepage(struct page *page,
2142 struct writeback_control *wbc)
2147 struct buffer_head *page_bufs = NULL;
2148 struct inode *inode = page->mapping->host;
2149 struct ext4_io_submit io_submit;
2150 bool keep_towrite = false;
2152 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2153 inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2158 trace_ext4_writepage(page);
2159 size = i_size_read(inode);
2160 if (page->index == size >> PAGE_SHIFT)
2161 len = size & ~PAGE_MASK;
2165 /* Should never happen but for bugs in other kernel subsystems */
2166 if (!page_has_buffers(page)) {
2167 ext4_warning_inode(inode,
2168 "page %lu does not have buffers attached", page->index);
2169 ClearPageDirty(page);
2174 page_bufs = page_buffers(page);
2176 * We cannot do block allocation or other extent handling in this
2177 * function. If there are buffers needing that, we have to redirty
2178 * the page. But we may reach here when we do a journal commit via
2179 * journal_submit_inode_data_buffers() and in that case we must write
2180 * allocated buffers to achieve data=ordered mode guarantees.
2182 * Also, if there is only one buffer per page (the fs block
2183 * size == the page size), if one buffer needs block
2184 * allocation or needs to modify the extent tree to clear the
2185 * unwritten flag, we know that the page can't be written at
2186 * all, so we might as well refuse the write immediately.
2187 * Unfortunately if the block size != page size, we can't as
2188 * easily detect this case using ext4_walk_page_buffers(), but
2189 * for the extremely common case, this is an optimization that
2190 * skips a useless round trip through ext4_bio_write_page().
2192 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2193 ext4_bh_delay_or_unwritten)) {
2194 redirty_page_for_writepage(wbc, page);
2195 if ((current->flags & PF_MEMALLOC) ||
2196 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2198 * For memory cleaning there's no point in writing only
2199 * some buffers. So just bail out. Warn if we came here
2200 * from direct reclaim.
2202 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2207 keep_towrite = true;
2210 if (PageChecked(page) && ext4_should_journal_data(inode))
2212 * It's mmapped pagecache. Add buffers and journal it. There
2213 * doesn't seem much point in redirtying the page here.
2215 return __ext4_journalled_writepage(page, len);
2217 ext4_io_submit_init(&io_submit, wbc);
2218 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2219 if (!io_submit.io_end) {
2220 redirty_page_for_writepage(wbc, page);
2224 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2225 ext4_io_submit(&io_submit);
2226 /* Drop io_end reference we got from init */
2227 ext4_put_io_end_defer(io_submit.io_end);
2231 static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2237 BUG_ON(page->index != mpd->first_page);
2238 clear_page_dirty_for_io(page);
2240 * We have to be very careful here! Nothing protects writeback path
2241 * against i_size changes and the page can be writeably mapped into
2242 * page tables. So an application can be growing i_size and writing
2243 * data through mmap while writeback runs. clear_page_dirty_for_io()
2244 * write-protects our page in page tables and the page cannot get
2245 * written to again until we release page lock. So only after
2246 * clear_page_dirty_for_io() we are safe to sample i_size for
2247 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2248 * on the barrier provided by TestClearPageDirty in
2249 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2250 * after page tables are updated.
2252 size = i_size_read(mpd->inode);
2253 if (page->index == size >> PAGE_SHIFT)
2254 len = size & ~PAGE_MASK;
2257 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2259 mpd->wbc->nr_to_write--;
2265 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2268 * mballoc gives us at most this number of blocks...
2269 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2270 * The rest of mballoc seems to handle chunks up to full group size.
2272 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2275 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2277 * @mpd - extent of blocks
2278 * @lblk - logical number of the block in the file
2279 * @bh - buffer head we want to add to the extent
2281 * The function is used to collect contig. blocks in the same state. If the
2282 * buffer doesn't require mapping for writeback and we haven't started the
2283 * extent of buffers to map yet, the function returns 'true' immediately - the
2284 * caller can write the buffer right away. Otherwise the function returns true
2285 * if the block has been added to the extent, false if the block couldn't be
2288 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2289 struct buffer_head *bh)
2291 struct ext4_map_blocks *map = &mpd->map;
2293 /* Buffer that doesn't need mapping for writeback? */
2294 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2295 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2296 /* So far no extent to map => we write the buffer right away */
2297 if (map->m_len == 0)
2302 /* First block in the extent? */
2303 if (map->m_len == 0) {
2304 /* We cannot map unless handle is started... */
2309 map->m_flags = bh->b_state & BH_FLAGS;
2313 /* Don't go larger than mballoc is willing to allocate */
2314 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2317 /* Can we merge the block to our big extent? */
2318 if (lblk == map->m_lblk + map->m_len &&
2319 (bh->b_state & BH_FLAGS) == map->m_flags) {
2327 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2329 * @mpd - extent of blocks for mapping
2330 * @head - the first buffer in the page
2331 * @bh - buffer we should start processing from
2332 * @lblk - logical number of the block in the file corresponding to @bh
2334 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2335 * the page for IO if all buffers in this page were mapped and there's no
2336 * accumulated extent of buffers to map or add buffers in the page to the
2337 * extent of buffers to map. The function returns 1 if the caller can continue
2338 * by processing the next page, 0 if it should stop adding buffers to the
2339 * extent to map because we cannot extend it anymore. It can also return value
2340 * < 0 in case of error during IO submission.
2342 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2343 struct buffer_head *head,
2344 struct buffer_head *bh,
2347 struct inode *inode = mpd->inode;
2349 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2350 >> inode->i_blkbits;
2353 BUG_ON(buffer_locked(bh));
2355 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2356 /* Found extent to map? */
2359 /* Buffer needs mapping and handle is not started? */
2362 /* Everything mapped so far and we hit EOF */
2365 } while (lblk++, (bh = bh->b_this_page) != head);
2366 /* So far everything mapped? Submit the page for IO. */
2367 if (mpd->map.m_len == 0) {
2368 err = mpage_submit_page(mpd, head->b_page);
2372 return lblk < blocks;
2376 * mpage_map_buffers - update buffers corresponding to changed extent and
2377 * submit fully mapped pages for IO
2379 * @mpd - description of extent to map, on return next extent to map
2381 * Scan buffers corresponding to changed extent (we expect corresponding pages
2382 * to be already locked) and update buffer state according to new extent state.
2383 * We map delalloc buffers to their physical location, clear unwritten bits,
2384 * and mark buffers as uninit when we perform writes to unwritten extents
2385 * and do extent conversion after IO is finished. If the last page is not fully
2386 * mapped, we update @map to the next extent in the last page that needs
2387 * mapping. Otherwise we submit the page for IO.
2389 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2391 struct pagevec pvec;
2393 struct inode *inode = mpd->inode;
2394 struct buffer_head *head, *bh;
2395 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2401 start = mpd->map.m_lblk >> bpp_bits;
2402 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2403 lblk = start << bpp_bits;
2404 pblock = mpd->map.m_pblk;
2406 pagevec_init(&pvec, 0);
2407 while (start <= end) {
2408 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2412 for (i = 0; i < nr_pages; i++) {
2413 struct page *page = pvec.pages[i];
2415 bh = head = page_buffers(page);
2417 if (lblk < mpd->map.m_lblk)
2419 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2421 * Buffer after end of mapped extent.
2422 * Find next buffer in the page to map.
2425 mpd->map.m_flags = 0;
2427 * FIXME: If dioread_nolock supports
2428 * blocksize < pagesize, we need to make
2429 * sure we add size mapped so far to
2430 * io_end->size as the following call
2431 * can submit the page for IO.
2433 err = mpage_process_page_bufs(mpd, head,
2435 pagevec_release(&pvec);
2440 if (buffer_delay(bh)) {
2441 clear_buffer_delay(bh);
2442 bh->b_blocknr = pblock++;
2444 clear_buffer_unwritten(bh);
2445 } while (lblk++, (bh = bh->b_this_page) != head);
2448 * FIXME: This is going to break if dioread_nolock
2449 * supports blocksize < pagesize as we will try to
2450 * convert potentially unmapped parts of inode.
2452 mpd->io_submit.io_end->size += PAGE_SIZE;
2453 /* Page fully mapped - let IO run! */
2454 err = mpage_submit_page(mpd, page);
2456 pagevec_release(&pvec);
2460 pagevec_release(&pvec);
2462 /* Extent fully mapped and matches with page boundary. We are done. */
2464 mpd->map.m_flags = 0;
2468 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2470 struct inode *inode = mpd->inode;
2471 struct ext4_map_blocks *map = &mpd->map;
2472 int get_blocks_flags;
2473 int err, dioread_nolock;
2475 trace_ext4_da_write_pages_extent(inode, map);
2477 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2478 * to convert an unwritten extent to be initialized (in the case
2479 * where we have written into one or more preallocated blocks). It is
2480 * possible that we're going to need more metadata blocks than
2481 * previously reserved. However we must not fail because we're in
2482 * writeback and there is nothing we can do about it so it might result
2483 * in data loss. So use reserved blocks to allocate metadata if
2486 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2487 * the blocks in question are delalloc blocks. This indicates
2488 * that the blocks and quotas has already been checked when
2489 * the data was copied into the page cache.
2491 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2492 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2493 EXT4_GET_BLOCKS_IO_SUBMIT;
2494 dioread_nolock = ext4_should_dioread_nolock(inode);
2496 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2497 if (map->m_flags & (1 << BH_Delay))
2498 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2500 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2503 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2504 if (!mpd->io_submit.io_end->handle &&
2505 ext4_handle_valid(handle)) {
2506 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2507 handle->h_rsv_handle = NULL;
2509 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2512 BUG_ON(map->m_len == 0);
2513 if (map->m_flags & EXT4_MAP_NEW) {
2514 clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk,
2521 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2522 * mpd->len and submit pages underlying it for IO
2524 * @handle - handle for journal operations
2525 * @mpd - extent to map
2526 * @give_up_on_write - we set this to true iff there is a fatal error and there
2527 * is no hope of writing the data. The caller should discard
2528 * dirty pages to avoid infinite loops.
2530 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2531 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2532 * them to initialized or split the described range from larger unwritten
2533 * extent. Note that we need not map all the described range since allocation
2534 * can return less blocks or the range is covered by more unwritten extents. We
2535 * cannot map more because we are limited by reserved transaction credits. On
2536 * the other hand we always make sure that the last touched page is fully
2537 * mapped so that it can be written out (and thus forward progress is
2538 * guaranteed). After mapping we submit all mapped pages for IO.
2540 static int mpage_map_and_submit_extent(handle_t *handle,
2541 struct mpage_da_data *mpd,
2542 bool *give_up_on_write)
2544 struct inode *inode = mpd->inode;
2545 struct ext4_map_blocks *map = &mpd->map;
2550 mpd->io_submit.io_end->offset =
2551 ((loff_t)map->m_lblk) << inode->i_blkbits;
2553 err = mpage_map_one_extent(handle, mpd);
2555 struct super_block *sb = inode->i_sb;
2557 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2558 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2559 goto invalidate_dirty_pages;
2561 * Let the uper layers retry transient errors.
2562 * In the case of ENOSPC, if ext4_count_free_blocks()
2563 * is non-zero, a commit should free up blocks.
2565 if ((err == -ENOMEM) ||
2566 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2568 goto update_disksize;
2571 ext4_msg(sb, KERN_CRIT,
2572 "Delayed block allocation failed for "
2573 "inode %lu at logical offset %llu with"
2574 " max blocks %u with error %d",
2576 (unsigned long long)map->m_lblk,
2577 (unsigned)map->m_len, -err);
2578 ext4_msg(sb, KERN_CRIT,
2579 "This should not happen!! Data will "
2582 ext4_print_free_blocks(inode);
2583 invalidate_dirty_pages:
2584 *give_up_on_write = true;
2589 * Update buffer state, submit mapped pages, and get us new
2592 err = mpage_map_and_submit_buffers(mpd);
2594 goto update_disksize;
2595 } while (map->m_len);
2599 * Update on-disk size after IO is submitted. Races with
2600 * truncate are avoided by checking i_size under i_data_sem.
2602 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2603 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2607 down_write(&EXT4_I(inode)->i_data_sem);
2608 i_size = i_size_read(inode);
2609 if (disksize > i_size)
2611 if (disksize > EXT4_I(inode)->i_disksize)
2612 EXT4_I(inode)->i_disksize = disksize;
2613 up_write(&EXT4_I(inode)->i_data_sem);
2614 err2 = ext4_mark_inode_dirty(handle, inode);
2616 ext4_error(inode->i_sb,
2617 "Failed to mark inode %lu dirty",
2626 * Calculate the total number of credits to reserve for one writepages
2627 * iteration. This is called from ext4_writepages(). We map an extent of
2628 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2629 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2630 * bpp - 1 blocks in bpp different extents.
2632 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2634 int bpp = ext4_journal_blocks_per_page(inode);
2636 return ext4_meta_trans_blocks(inode,
2637 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2641 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2642 * and underlying extent to map
2644 * @mpd - where to look for pages
2646 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2647 * IO immediately. When we find a page which isn't mapped we start accumulating
2648 * extent of buffers underlying these pages that needs mapping (formed by
2649 * either delayed or unwritten buffers). We also lock the pages containing
2650 * these buffers. The extent found is returned in @mpd structure (starting at
2651 * mpd->lblk with length mpd->len blocks).
2653 * Note that this function can attach bios to one io_end structure which are
2654 * neither logically nor physically contiguous. Although it may seem as an
2655 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2656 * case as we need to track IO to all buffers underlying a page in one io_end.
2658 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2660 struct address_space *mapping = mpd->inode->i_mapping;
2661 struct pagevec pvec;
2662 unsigned int nr_pages;
2663 long left = mpd->wbc->nr_to_write;
2664 pgoff_t index = mpd->first_page;
2665 pgoff_t end = mpd->last_page;
2668 int blkbits = mpd->inode->i_blkbits;
2670 struct buffer_head *head;
2672 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2673 tag = PAGECACHE_TAG_TOWRITE;
2675 tag = PAGECACHE_TAG_DIRTY;
2677 pagevec_init(&pvec, 0);
2679 mpd->next_page = index;
2680 while (index <= end) {
2681 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2682 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2686 for (i = 0; i < nr_pages; i++) {
2687 struct page *page = pvec.pages[i];
2690 * At this point, the page may be truncated or
2691 * invalidated (changing page->mapping to NULL), or
2692 * even swizzled back from swapper_space to tmpfs file
2693 * mapping. However, page->index will not change
2694 * because we have a reference on the page.
2696 if (page->index > end)
2700 * Accumulated enough dirty pages? This doesn't apply
2701 * to WB_SYNC_ALL mode. For integrity sync we have to
2702 * keep going because someone may be concurrently
2703 * dirtying pages, and we might have synced a lot of
2704 * newly appeared dirty pages, but have not synced all
2705 * of the old dirty pages.
2707 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2710 /* If we can't merge this page, we are done. */
2711 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2716 * If the page is no longer dirty, or its mapping no
2717 * longer corresponds to inode we are writing (which
2718 * means it has been truncated or invalidated), or the
2719 * page is already under writeback and we are not doing
2720 * a data integrity writeback, skip the page
2722 if (!PageDirty(page) ||
2723 (PageWriteback(page) &&
2724 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2725 unlikely(page->mapping != mapping)) {
2730 wait_on_page_writeback(page);
2731 BUG_ON(PageWriteback(page));
2734 * Should never happen but for buggy code in
2735 * other subsystems that call
2736 * set_page_dirty() without properly warning
2737 * the file system first. See [1] for more
2740 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2742 if (!page_has_buffers(page)) {
2743 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index);
2744 ClearPageDirty(page);
2749 if (mpd->map.m_len == 0)
2750 mpd->first_page = page->index;
2751 mpd->next_page = page->index + 1;
2752 /* Add all dirty buffers to mpd */
2753 lblk = ((ext4_lblk_t)page->index) <<
2754 (PAGE_SHIFT - blkbits);
2755 head = page_buffers(page);
2756 err = mpage_process_page_bufs(mpd, head, head, lblk);
2762 pagevec_release(&pvec);
2767 pagevec_release(&pvec);
2771 static int __writepage(struct page *page, struct writeback_control *wbc,
2774 struct address_space *mapping = data;
2775 int ret = ext4_writepage(page, wbc);
2776 mapping_set_error(mapping, ret);
2780 static int ext4_writepages(struct address_space *mapping,
2781 struct writeback_control *wbc)
2783 pgoff_t writeback_index = 0;
2784 long nr_to_write = wbc->nr_to_write;
2785 int range_whole = 0;
2787 handle_t *handle = NULL;
2788 struct mpage_da_data mpd;
2789 struct inode *inode = mapping->host;
2790 int needed_blocks, rsv_blocks = 0, ret = 0;
2791 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2793 struct blk_plug plug;
2794 bool give_up_on_write = false;
2796 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2799 percpu_down_read(&sbi->s_writepages_rwsem);
2800 trace_ext4_writepages(inode, wbc);
2802 if (dax_mapping(mapping)) {
2803 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev,
2805 goto out_writepages;
2809 * No pages to write? This is mainly a kludge to avoid starting
2810 * a transaction for special inodes like journal inode on last iput()
2811 * because that could violate lock ordering on umount
2813 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2814 goto out_writepages;
2816 if (ext4_should_journal_data(inode)) {
2817 struct blk_plug plug;
2819 blk_start_plug(&plug);
2820 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2821 blk_finish_plug(&plug);
2822 goto out_writepages;
2826 * If the filesystem has aborted, it is read-only, so return
2827 * right away instead of dumping stack traces later on that
2828 * will obscure the real source of the problem. We test
2829 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2830 * the latter could be true if the filesystem is mounted
2831 * read-only, and in that case, ext4_writepages should
2832 * *never* be called, so if that ever happens, we would want
2835 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2836 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2838 goto out_writepages;
2841 if (ext4_should_dioread_nolock(inode)) {
2843 * We may need to convert up to one extent per block in
2844 * the page and we may dirty the inode.
2846 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2847 PAGE_SIZE >> inode->i_blkbits);
2851 * If we have inline data and arrive here, it means that
2852 * we will soon create the block for the 1st page, so
2853 * we'd better clear the inline data here.
2855 if (ext4_has_inline_data(inode)) {
2856 /* Just inode will be modified... */
2857 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2858 if (IS_ERR(handle)) {
2859 ret = PTR_ERR(handle);
2860 goto out_writepages;
2862 BUG_ON(ext4_test_inode_state(inode,
2863 EXT4_STATE_MAY_INLINE_DATA));
2864 ext4_destroy_inline_data(handle, inode);
2865 ext4_journal_stop(handle);
2868 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2871 if (wbc->range_cyclic) {
2872 writeback_index = mapping->writeback_index;
2873 if (writeback_index)
2875 mpd.first_page = writeback_index;
2878 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2879 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2884 ext4_io_submit_init(&mpd.io_submit, wbc);
2886 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2887 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2889 blk_start_plug(&plug);
2892 * First writeback pages that don't need mapping - we can avoid
2893 * starting a transaction unnecessarily and also avoid being blocked
2894 * in the block layer on device congestion while having transaction
2898 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2899 if (!mpd.io_submit.io_end) {
2903 ret = mpage_prepare_extent_to_map(&mpd);
2904 /* Submit prepared bio */
2905 ext4_io_submit(&mpd.io_submit);
2906 ext4_put_io_end_defer(mpd.io_submit.io_end);
2907 mpd.io_submit.io_end = NULL;
2908 /* Unlock pages we didn't use */
2909 mpage_release_unused_pages(&mpd, false);
2913 while (!done && mpd.first_page <= mpd.last_page) {
2914 /* For each extent of pages we use new io_end */
2915 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2916 if (!mpd.io_submit.io_end) {
2922 * We have two constraints: We find one extent to map and we
2923 * must always write out whole page (makes a difference when
2924 * blocksize < pagesize) so that we don't block on IO when we
2925 * try to write out the rest of the page. Journalled mode is
2926 * not supported by delalloc.
2928 BUG_ON(ext4_should_journal_data(inode));
2929 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2931 /* start a new transaction */
2932 handle = ext4_journal_start_with_reserve(inode,
2933 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2934 if (IS_ERR(handle)) {
2935 ret = PTR_ERR(handle);
2936 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2937 "%ld pages, ino %lu; err %d", __func__,
2938 wbc->nr_to_write, inode->i_ino, ret);
2939 /* Release allocated io_end */
2940 ext4_put_io_end(mpd.io_submit.io_end);
2941 mpd.io_submit.io_end = NULL;
2946 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2947 ret = mpage_prepare_extent_to_map(&mpd);
2950 ret = mpage_map_and_submit_extent(handle, &mpd,
2954 * We scanned the whole range (or exhausted
2955 * nr_to_write), submitted what was mapped and
2956 * didn't find anything needing mapping. We are
2963 * Caution: If the handle is synchronous,
2964 * ext4_journal_stop() can wait for transaction commit
2965 * to finish which may depend on writeback of pages to
2966 * complete or on page lock to be released. In that
2967 * case, we have to wait until after after we have
2968 * submitted all the IO, released page locks we hold,
2969 * and dropped io_end reference (for extent conversion
2970 * to be able to complete) before stopping the handle.
2972 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2973 ext4_journal_stop(handle);
2977 /* Submit prepared bio */
2978 ext4_io_submit(&mpd.io_submit);
2979 /* Unlock pages we didn't use */
2980 mpage_release_unused_pages(&mpd, give_up_on_write);
2982 * Drop our io_end reference we got from init. We have
2983 * to be careful and use deferred io_end finishing if
2984 * we are still holding the transaction as we can
2985 * release the last reference to io_end which may end
2986 * up doing unwritten extent conversion.
2989 ext4_put_io_end_defer(mpd.io_submit.io_end);
2990 ext4_journal_stop(handle);
2992 ext4_put_io_end(mpd.io_submit.io_end);
2993 mpd.io_submit.io_end = NULL;
2995 if (ret == -ENOSPC && sbi->s_journal) {
2997 * Commit the transaction which would
2998 * free blocks released in the transaction
3001 jbd2_journal_force_commit_nested(sbi->s_journal);
3005 /* Fatal error - ENOMEM, EIO... */
3010 blk_finish_plug(&plug);
3011 if (!ret && !cycled && wbc->nr_to_write > 0) {
3013 mpd.last_page = writeback_index - 1;
3019 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3021 * Set the writeback_index so that range_cyclic
3022 * mode will write it back later
3024 mapping->writeback_index = mpd.first_page;
3027 trace_ext4_writepages_result(inode, wbc, ret,
3028 nr_to_write - wbc->nr_to_write);
3029 percpu_up_read(&sbi->s_writepages_rwsem);
3033 static int ext4_nonda_switch(struct super_block *sb)
3035 s64 free_clusters, dirty_clusters;
3036 struct ext4_sb_info *sbi = EXT4_SB(sb);
3039 * switch to non delalloc mode if we are running low
3040 * on free block. The free block accounting via percpu
3041 * counters can get slightly wrong with percpu_counter_batch getting
3042 * accumulated on each CPU without updating global counters
3043 * Delalloc need an accurate free block accounting. So switch
3044 * to non delalloc when we are near to error range.
3047 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
3049 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
3051 * Start pushing delalloc when 1/2 of free blocks are dirty.
3053 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3054 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
3056 if (2 * free_clusters < 3 * dirty_clusters ||
3057 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3059 * free block count is less than 150% of dirty blocks
3060 * or free blocks is less than watermark
3067 /* We always reserve for an inode update; the superblock could be there too */
3068 static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3070 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3073 if (pos + len <= 0x7fffffffULL)
3076 /* We might need to update the superblock to set LARGE_FILE */
3080 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3081 loff_t pos, unsigned len, unsigned flags,
3082 struct page **pagep, void **fsdata)
3084 int ret, retries = 0;
3087 struct inode *inode = mapping->host;
3090 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3093 index = pos >> PAGE_SHIFT;
3095 if (ext4_nonda_switch(inode->i_sb) ||
3096 S_ISLNK(inode->i_mode)) {
3097 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3098 return ext4_write_begin(file, mapping, pos,
3099 len, flags, pagep, fsdata);
3101 *fsdata = (void *)0;
3102 trace_ext4_da_write_begin(inode, pos, len, flags);
3104 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3105 ret = ext4_da_write_inline_data_begin(mapping, inode,
3115 * grab_cache_page_write_begin() can take a long time if the
3116 * system is thrashing due to memory pressure, or if the page
3117 * is being written back. So grab it first before we start
3118 * the transaction handle. This also allows us to allocate
3119 * the page (if needed) without using GFP_NOFS.
3122 page = grab_cache_page_write_begin(mapping, index, flags);
3128 * With delayed allocation, we don't log the i_disksize update
3129 * if there is delayed block allocation. But we still need
3130 * to journalling the i_disksize update if writes to the end
3131 * of file which has an already mapped buffer.
3134 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3135 ext4_da_write_credits(inode, pos, len));
3136 if (IS_ERR(handle)) {
3138 return PTR_ERR(handle);
3142 if (page->mapping != mapping) {
3143 /* The page got truncated from under us */
3146 ext4_journal_stop(handle);
3149 /* In case writeback began while the page was unlocked */
3150 wait_for_stable_page(page);
3152 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3153 ret = ext4_block_write_begin(page, pos, len,
3154 ext4_da_get_block_prep);
3156 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3160 ext4_journal_stop(handle);
3162 * block_write_begin may have instantiated a few blocks
3163 * outside i_size. Trim these off again. Don't need
3164 * i_size_read because we hold i_mutex.
3166 if (pos + len > inode->i_size)
3167 ext4_truncate_failed_write(inode);
3169 if (ret == -ENOSPC &&
3170 ext4_should_retry_alloc(inode->i_sb, &retries))
3182 * Check if we should update i_disksize
3183 * when write to the end of file but not require block allocation
3185 static int ext4_da_should_update_i_disksize(struct page *page,
3186 unsigned long offset)
3188 struct buffer_head *bh;
3189 struct inode *inode = page->mapping->host;
3193 bh = page_buffers(page);
3194 idx = offset >> inode->i_blkbits;
3196 for (i = 0; i < idx; i++)
3197 bh = bh->b_this_page;
3199 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3204 static int ext4_da_write_end(struct file *file,
3205 struct address_space *mapping,
3206 loff_t pos, unsigned len, unsigned copied,
3207 struct page *page, void *fsdata)
3209 struct inode *inode = mapping->host;
3211 handle_t *handle = ext4_journal_current_handle();
3213 unsigned long start, end;
3214 int write_mode = (int)(unsigned long)fsdata;
3216 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3217 return ext4_write_end(file, mapping, pos,
3218 len, copied, page, fsdata);
3220 trace_ext4_da_write_end(inode, pos, len, copied);
3221 start = pos & (PAGE_SIZE - 1);
3222 end = start + copied - 1;
3225 * generic_write_end() will run mark_inode_dirty() if i_size
3226 * changes. So let's piggyback the i_disksize mark_inode_dirty
3229 new_i_size = pos + copied;
3230 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3231 if (ext4_has_inline_data(inode) ||
3232 ext4_da_should_update_i_disksize(page, end)) {
3233 ext4_update_i_disksize(inode, new_i_size);
3234 /* We need to mark inode dirty even if
3235 * new_i_size is less that inode->i_size
3236 * bu greater than i_disksize.(hint delalloc)
3238 ext4_mark_inode_dirty(handle, inode);
3242 if (write_mode != CONVERT_INLINE_DATA &&
3243 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3244 ext4_has_inline_data(inode))
3245 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3248 ret2 = generic_write_end(file, mapping, pos, len, copied,
3254 ret2 = ext4_journal_stop(handle);
3258 return ret ? ret : copied;
3261 static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
3262 unsigned int length)
3265 * Drop reserved blocks
3267 BUG_ON(!PageLocked(page));
3268 if (!page_has_buffers(page))
3271 ext4_da_page_release_reservation(page, offset, length);
3274 ext4_invalidatepage(page, offset, length);
3280 * Force all delayed allocation blocks to be allocated for a given inode.
3282 int ext4_alloc_da_blocks(struct inode *inode)
3284 trace_ext4_alloc_da_blocks(inode);
3286 if (!EXT4_I(inode)->i_reserved_data_blocks)
3290 * We do something simple for now. The filemap_flush() will
3291 * also start triggering a write of the data blocks, which is
3292 * not strictly speaking necessary (and for users of
3293 * laptop_mode, not even desirable). However, to do otherwise
3294 * would require replicating code paths in:
3296 * ext4_writepages() ->
3297 * write_cache_pages() ---> (via passed in callback function)
3298 * __mpage_da_writepage() -->
3299 * mpage_add_bh_to_extent()
3300 * mpage_da_map_blocks()
3302 * The problem is that write_cache_pages(), located in
3303 * mm/page-writeback.c, marks pages clean in preparation for
3304 * doing I/O, which is not desirable if we're not planning on
3307 * We could call write_cache_pages(), and then redirty all of
3308 * the pages by calling redirty_page_for_writepage() but that
3309 * would be ugly in the extreme. So instead we would need to
3310 * replicate parts of the code in the above functions,
3311 * simplifying them because we wouldn't actually intend to
3312 * write out the pages, but rather only collect contiguous
3313 * logical block extents, call the multi-block allocator, and
3314 * then update the buffer heads with the block allocations.
3316 * For now, though, we'll cheat by calling filemap_flush(),
3317 * which will map the blocks, and start the I/O, but not
3318 * actually wait for the I/O to complete.
3320 return filemap_flush(inode->i_mapping);
3324 * bmap() is special. It gets used by applications such as lilo and by
3325 * the swapper to find the on-disk block of a specific piece of data.
3327 * Naturally, this is dangerous if the block concerned is still in the
3328 * journal. If somebody makes a swapfile on an ext4 data-journaling
3329 * filesystem and enables swap, then they may get a nasty shock when the
3330 * data getting swapped to that swapfile suddenly gets overwritten by
3331 * the original zero's written out previously to the journal and
3332 * awaiting writeback in the kernel's buffer cache.
3334 * So, if we see any bmap calls here on a modified, data-journaled file,
3335 * take extra steps to flush any blocks which might be in the cache.
3337 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3339 struct inode *inode = mapping->host;
3344 * We can get here for an inline file via the FIBMAP ioctl
3346 if (ext4_has_inline_data(inode))
3349 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3350 test_opt(inode->i_sb, DELALLOC)) {
3352 * With delalloc we want to sync the file
3353 * so that we can make sure we allocate
3356 filemap_write_and_wait(mapping);
3359 if (EXT4_JOURNAL(inode) &&
3360 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3362 * This is a REALLY heavyweight approach, but the use of
3363 * bmap on dirty files is expected to be extremely rare:
3364 * only if we run lilo or swapon on a freshly made file
3365 * do we expect this to happen.
3367 * (bmap requires CAP_SYS_RAWIO so this does not
3368 * represent an unprivileged user DOS attack --- we'd be
3369 * in trouble if mortal users could trigger this path at
3372 * NB. EXT4_STATE_JDATA is not set on files other than
3373 * regular files. If somebody wants to bmap a directory
3374 * or symlink and gets confused because the buffer
3375 * hasn't yet been flushed to disk, they deserve
3376 * everything they get.
3379 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3380 journal = EXT4_JOURNAL(inode);
3381 jbd2_journal_lock_updates(journal);
3382 err = jbd2_journal_flush(journal);
3383 jbd2_journal_unlock_updates(journal);
3389 return generic_block_bmap(mapping, block, ext4_get_block);
3392 static int ext4_readpage(struct file *file, struct page *page)
3395 struct inode *inode = page->mapping->host;
3397 trace_ext4_readpage(page);
3399 if (ext4_has_inline_data(inode))
3400 ret = ext4_readpage_inline(inode, page);
3403 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
3409 ext4_readpages(struct file *file, struct address_space *mapping,
3410 struct list_head *pages, unsigned nr_pages)
3412 struct inode *inode = mapping->host;
3414 /* If the file has inline data, no need to do readpages. */
3415 if (ext4_has_inline_data(inode))
3418 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
3421 static void ext4_invalidatepage(struct page *page, unsigned int offset,
3422 unsigned int length)
3424 trace_ext4_invalidatepage(page, offset, length);
3426 /* No journalling happens on data buffers when this function is used */
3427 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3429 block_invalidatepage(page, offset, length);
3432 static int __ext4_journalled_invalidatepage(struct page *page,
3433 unsigned int offset,
3434 unsigned int length)
3436 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3438 trace_ext4_journalled_invalidatepage(page, offset, length);
3441 * If it's a full truncate we just forget about the pending dirtying
3443 if (offset == 0 && length == PAGE_SIZE)
3444 ClearPageChecked(page);
3446 return jbd2_journal_invalidatepage(journal, page, offset, length);
3449 /* Wrapper for aops... */
3450 static void ext4_journalled_invalidatepage(struct page *page,
3451 unsigned int offset,
3452 unsigned int length)
3454 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3457 static int ext4_releasepage(struct page *page, gfp_t wait)
3459 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3461 trace_ext4_releasepage(page);
3463 /* Page has dirty journalled data -> cannot release */
3464 if (PageChecked(page))
3467 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3469 return try_to_free_buffers(page);
3472 #ifdef CONFIG_FS_DAX
3473 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3474 unsigned flags, struct iomap *iomap)
3476 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3477 unsigned int blkbits = inode->i_blkbits;
3478 unsigned long first_block, last_block;
3479 struct ext4_map_blocks map;
3482 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3484 first_block = offset >> blkbits;
3485 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3486 EXT4_MAX_LOGICAL_BLOCK);
3488 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3491 map.m_lblk = first_block;
3492 map.m_len = last_block - first_block + 1;
3494 if (!(flags & IOMAP_WRITE)) {
3495 ret = ext4_map_blocks(NULL, inode, &map, 0);
3501 /* Trim mapping request to maximum we can map at once for DIO */
3502 if (map.m_len > DIO_MAX_BLOCKS)
3503 map.m_len = DIO_MAX_BLOCKS;
3504 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3507 * Either we allocate blocks and then we don't get unwritten
3508 * extent so we have reserved enough credits, or the blocks
3509 * are already allocated and unwritten and in that case
3510 * extent conversion fits in the credits as well.
3512 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3515 return PTR_ERR(handle);
3517 ret = ext4_map_blocks(handle, inode, &map,
3518 EXT4_GET_BLOCKS_CREATE_ZERO);
3520 ext4_journal_stop(handle);
3521 if (ret == -ENOSPC &&
3522 ext4_should_retry_alloc(inode->i_sb, &retries))
3528 * If we added blocks beyond i_size, we need to make sure they
3529 * will get truncated if we crash before updating i_size in
3530 * ext4_iomap_end(). For faults we don't need to do that (and
3531 * even cannot because for orphan list operations inode_lock is
3532 * required) - if we happen to instantiate block beyond i_size,
3533 * it is because we race with truncate which has already added
3534 * the inode to the orphan list.
3536 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3537 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3540 err = ext4_orphan_add(handle, inode);
3542 ext4_journal_stop(handle);
3546 ext4_journal_stop(handle);
3550 iomap->bdev = inode->i_sb->s_bdev;
3551 iomap->dax_dev = sbi->s_daxdev;
3552 iomap->offset = first_block << blkbits;
3555 iomap->type = IOMAP_HOLE;
3556 iomap->blkno = IOMAP_NULL_BLOCK;
3557 iomap->length = (u64)map.m_len << blkbits;
3559 if (map.m_flags & EXT4_MAP_MAPPED) {
3560 iomap->type = IOMAP_MAPPED;
3561 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3562 iomap->type = IOMAP_UNWRITTEN;
3567 iomap->blkno = (sector_t)map.m_pblk << (blkbits - 9);
3568 iomap->length = (u64)map.m_len << blkbits;
3571 if (map.m_flags & EXT4_MAP_NEW)
3572 iomap->flags |= IOMAP_F_NEW;
3576 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3577 ssize_t written, unsigned flags, struct iomap *iomap)
3581 int blkbits = inode->i_blkbits;
3582 bool truncate = false;
3584 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3587 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3588 if (IS_ERR(handle)) {
3589 ret = PTR_ERR(handle);
3592 if (ext4_update_inode_size(inode, offset + written))
3593 ext4_mark_inode_dirty(handle, inode);
3595 * We may need to truncate allocated but not written blocks beyond EOF.
3597 if (iomap->offset + iomap->length >
3598 ALIGN(inode->i_size, 1 << blkbits)) {
3599 ext4_lblk_t written_blk, end_blk;
3601 written_blk = (offset + written) >> blkbits;
3602 end_blk = (offset + length) >> blkbits;
3603 if (written_blk < end_blk && ext4_can_truncate(inode))
3607 * Remove inode from orphan list if we were extending a inode and
3608 * everything went fine.
3610 if (!truncate && inode->i_nlink &&
3611 !list_empty(&EXT4_I(inode)->i_orphan))
3612 ext4_orphan_del(handle, inode);
3613 ext4_journal_stop(handle);
3615 ext4_truncate_failed_write(inode);
3618 * If truncate failed early the inode might still be on the
3619 * orphan list; we need to make sure the inode is removed from
3620 * the orphan list in that case.
3623 ext4_orphan_del(NULL, inode);
3628 const struct iomap_ops ext4_iomap_ops = {
3629 .iomap_begin = ext4_iomap_begin,
3630 .iomap_end = ext4_iomap_end,
3635 static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3636 ssize_t size, void *private)
3638 ext4_io_end_t *io_end = private;
3640 /* if not async direct IO just return */
3644 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3645 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3646 io_end, io_end->inode->i_ino, iocb, offset, size);
3649 * Error during AIO DIO. We cannot convert unwritten extents as the
3650 * data was not written. Just clear the unwritten flag and drop io_end.
3653 ext4_clear_io_unwritten_flag(io_end);
3656 io_end->offset = offset;
3657 io_end->size = size;
3658 ext4_put_io_end(io_end);
3664 * Handling of direct IO writes.
3666 * For ext4 extent files, ext4 will do direct-io write even to holes,
3667 * preallocated extents, and those write extend the file, no need to
3668 * fall back to buffered IO.
3670 * For holes, we fallocate those blocks, mark them as unwritten
3671 * If those blocks were preallocated, we mark sure they are split, but
3672 * still keep the range to write as unwritten.
3674 * The unwritten extents will be converted to written when DIO is completed.
3675 * For async direct IO, since the IO may still pending when return, we
3676 * set up an end_io call back function, which will do the conversion
3677 * when async direct IO completed.
3679 * If the O_DIRECT write will extend the file then add this inode to the
3680 * orphan list. So recovery will truncate it back to the original size
3681 * if the machine crashes during the write.
3684 static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3686 struct file *file = iocb->ki_filp;
3687 struct inode *inode = file->f_mapping->host;
3689 loff_t offset = iocb->ki_pos;
3690 size_t count = iov_iter_count(iter);
3692 get_block_t *get_block_func = NULL;
3694 loff_t final_size = offset + count;
3698 if (final_size > inode->i_size) {
3699 /* Credits for sb + inode write */
3700 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3701 if (IS_ERR(handle)) {
3702 ret = PTR_ERR(handle);
3705 ret = ext4_orphan_add(handle, inode);
3707 ext4_journal_stop(handle);
3711 ext4_update_i_disksize(inode, inode->i_size);
3712 ext4_journal_stop(handle);
3715 BUG_ON(iocb->private == NULL);
3718 * Make all waiters for direct IO properly wait also for extent
3719 * conversion. This also disallows race between truncate() and
3720 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3722 inode_dio_begin(inode);
3724 /* If we do a overwrite dio, i_mutex locking can be released */
3725 overwrite = *((int *)iocb->private);
3728 inode_unlock(inode);
3731 * For extent mapped files we could direct write to holes and fallocate.
3733 * Allocated blocks to fill the hole are marked as unwritten to prevent
3734 * parallel buffered read to expose the stale data before DIO complete
3737 * As to previously fallocated extents, ext4 get_block will just simply
3738 * mark the buffer mapped but still keep the extents unwritten.
3740 * For non AIO case, we will convert those unwritten extents to written
3741 * after return back from blockdev_direct_IO. That way we save us from
3742 * allocating io_end structure and also the overhead of offloading
3743 * the extent convertion to a workqueue.
3745 * For async DIO, the conversion needs to be deferred when the
3746 * IO is completed. The ext4 end_io callback function will be
3747 * called to take care of the conversion work. Here for async
3748 * case, we allocate an io_end structure to hook to the iocb.
3750 iocb->private = NULL;
3752 get_block_func = ext4_dio_get_block_overwrite;
3753 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3754 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3755 get_block_func = ext4_dio_get_block;
3756 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3757 } else if (is_sync_kiocb(iocb)) {
3758 get_block_func = ext4_dio_get_block_unwritten_sync;
3759 dio_flags = DIO_LOCKING;
3761 get_block_func = ext4_dio_get_block_unwritten_async;
3762 dio_flags = DIO_LOCKING;
3764 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3765 get_block_func, ext4_end_io_dio, NULL,
3768 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3769 EXT4_STATE_DIO_UNWRITTEN)) {
3772 * for non AIO case, since the IO is already
3773 * completed, we could do the conversion right here
3775 err = ext4_convert_unwritten_extents(NULL, inode,
3779 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3782 inode_dio_end(inode);
3783 /* take i_mutex locking again if we do a ovewrite dio */
3787 if (ret < 0 && final_size > inode->i_size)
3788 ext4_truncate_failed_write(inode);
3790 /* Handle extending of i_size after direct IO write */
3794 /* Credits for sb + inode write */
3795 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3796 if (IS_ERR(handle)) {
3798 * We wrote the data but cannot extend
3799 * i_size. Bail out. In async io case, we do
3800 * not return error here because we have
3801 * already submmitted the corresponding
3802 * bio. Returning error here makes the caller
3803 * think that this IO is done and failed
3804 * resulting in race with bio's completion
3808 ret = PTR_ERR(handle);
3810 ext4_orphan_del(NULL, inode);
3815 ext4_orphan_del(handle, inode);
3817 loff_t end = offset + ret;
3818 if (end > inode->i_size) {
3819 ext4_update_i_disksize(inode, end);
3820 i_size_write(inode, end);
3822 * We're going to return a positive `ret'
3823 * here due to non-zero-length I/O, so there's
3824 * no way of reporting error returns from
3825 * ext4_mark_inode_dirty() to userspace. So
3828 ext4_mark_inode_dirty(handle, inode);
3831 err = ext4_journal_stop(handle);
3839 static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3841 struct address_space *mapping = iocb->ki_filp->f_mapping;
3842 struct inode *inode = mapping->host;
3843 size_t count = iov_iter_count(iter);
3845 loff_t offset = iocb->ki_pos;
3846 loff_t size = i_size_read(inode);
3852 * Shared inode_lock is enough for us - it protects against concurrent
3853 * writes & truncates and since we take care of writing back page cache,
3854 * we are protected against page writeback as well.
3856 if (iocb->ki_flags & IOCB_NOWAIT) {
3857 if (!inode_trylock_shared(inode))
3860 inode_lock_shared(inode);
3863 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3864 iocb->ki_pos + count - 1);
3867 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3868 iter, ext4_dio_get_block, NULL, NULL, 0);
3870 inode_unlock_shared(inode);
3874 static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3876 struct file *file = iocb->ki_filp;
3877 struct inode *inode = file->f_mapping->host;
3878 size_t count = iov_iter_count(iter);
3879 loff_t offset = iocb->ki_pos;
3882 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3883 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3888 * If we are doing data journalling we don't support O_DIRECT
3890 if (ext4_should_journal_data(inode))
3893 /* Let buffer I/O handle the inline data case. */
3894 if (ext4_has_inline_data(inode))
3897 /* DAX uses iomap path now */
3898 if (WARN_ON_ONCE(IS_DAX(inode)))
3901 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3902 if (iov_iter_rw(iter) == READ)
3903 ret = ext4_direct_IO_read(iocb, iter);
3905 ret = ext4_direct_IO_write(iocb, iter);
3906 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3911 * Pages can be marked dirty completely asynchronously from ext4's journalling
3912 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3913 * much here because ->set_page_dirty is called under VFS locks. The page is
3914 * not necessarily locked.
3916 * We cannot just dirty the page and leave attached buffers clean, because the
3917 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3918 * or jbddirty because all the journalling code will explode.
3920 * So what we do is to mark the page "pending dirty" and next time writepage
3921 * is called, propagate that into the buffers appropriately.
3923 static int ext4_journalled_set_page_dirty(struct page *page)
3925 SetPageChecked(page);
3926 return __set_page_dirty_nobuffers(page);
3929 static int ext4_set_page_dirty(struct page *page)
3931 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3932 WARN_ON_ONCE(!page_has_buffers(page));
3933 return __set_page_dirty_buffers(page);
3936 static const struct address_space_operations ext4_aops = {
3937 .readpage = ext4_readpage,
3938 .readpages = ext4_readpages,
3939 .writepage = ext4_writepage,
3940 .writepages = ext4_writepages,
3941 .write_begin = ext4_write_begin,
3942 .write_end = ext4_write_end,
3943 .set_page_dirty = ext4_set_page_dirty,
3945 .invalidatepage = ext4_invalidatepage,
3946 .releasepage = ext4_releasepage,
3947 .direct_IO = ext4_direct_IO,
3948 .migratepage = buffer_migrate_page,
3949 .is_partially_uptodate = block_is_partially_uptodate,
3950 .error_remove_page = generic_error_remove_page,
3953 static const struct address_space_operations ext4_journalled_aops = {
3954 .readpage = ext4_readpage,
3955 .readpages = ext4_readpages,
3956 .writepage = ext4_writepage,
3957 .writepages = ext4_writepages,
3958 .write_begin = ext4_write_begin,
3959 .write_end = ext4_journalled_write_end,
3960 .set_page_dirty = ext4_journalled_set_page_dirty,
3962 .invalidatepage = ext4_journalled_invalidatepage,
3963 .releasepage = ext4_releasepage,
3964 .direct_IO = ext4_direct_IO,
3965 .is_partially_uptodate = block_is_partially_uptodate,
3966 .error_remove_page = generic_error_remove_page,
3969 static const struct address_space_operations ext4_da_aops = {
3970 .readpage = ext4_readpage,
3971 .readpages = ext4_readpages,
3972 .writepage = ext4_writepage,
3973 .writepages = ext4_writepages,
3974 .write_begin = ext4_da_write_begin,
3975 .write_end = ext4_da_write_end,
3976 .set_page_dirty = ext4_set_page_dirty,
3978 .invalidatepage = ext4_da_invalidatepage,
3979 .releasepage = ext4_releasepage,
3980 .direct_IO = ext4_direct_IO,
3981 .migratepage = buffer_migrate_page,
3982 .is_partially_uptodate = block_is_partially_uptodate,
3983 .error_remove_page = generic_error_remove_page,
3986 void ext4_set_aops(struct inode *inode)
3988 switch (ext4_inode_journal_mode(inode)) {
3989 case EXT4_INODE_ORDERED_DATA_MODE:
3990 case EXT4_INODE_WRITEBACK_DATA_MODE:
3992 case EXT4_INODE_JOURNAL_DATA_MODE:
3993 inode->i_mapping->a_ops = &ext4_journalled_aops;
3998 if (test_opt(inode->i_sb, DELALLOC))
3999 inode->i_mapping->a_ops = &ext4_da_aops;
4001 inode->i_mapping->a_ops = &ext4_aops;
4004 static int __ext4_block_zero_page_range(handle_t *handle,
4005 struct address_space *mapping, loff_t from, loff_t length)
4007 ext4_fsblk_t index = from >> PAGE_SHIFT;
4008 unsigned offset = from & (PAGE_SIZE-1);
4009 unsigned blocksize, pos;
4011 struct inode *inode = mapping->host;
4012 struct buffer_head *bh;
4016 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
4017 mapping_gfp_constraint(mapping, ~__GFP_FS));
4021 blocksize = inode->i_sb->s_blocksize;
4023 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4025 if (!page_has_buffers(page))
4026 create_empty_buffers(page, blocksize, 0);
4028 /* Find the buffer that contains "offset" */
4029 bh = page_buffers(page);
4031 while (offset >= pos) {
4032 bh = bh->b_this_page;
4036 if (buffer_freed(bh)) {
4037 BUFFER_TRACE(bh, "freed: skip");
4040 if (!buffer_mapped(bh)) {
4041 BUFFER_TRACE(bh, "unmapped");
4042 ext4_get_block(inode, iblock, bh, 0);
4043 /* unmapped? It's a hole - nothing to do */
4044 if (!buffer_mapped(bh)) {
4045 BUFFER_TRACE(bh, "still unmapped");
4050 /* Ok, it's mapped. Make sure it's up-to-date */
4051 if (PageUptodate(page))
4052 set_buffer_uptodate(bh);
4054 if (!buffer_uptodate(bh)) {
4056 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4058 /* Uhhuh. Read error. Complain and punt. */
4059 if (!buffer_uptodate(bh))
4061 if (S_ISREG(inode->i_mode) &&
4062 ext4_encrypted_inode(inode)) {
4063 /* We expect the key to be set. */
4064 BUG_ON(!fscrypt_has_encryption_key(inode));
4065 BUG_ON(blocksize != PAGE_SIZE);
4066 WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host,
4067 page, PAGE_SIZE, 0, page->index));
4070 if (ext4_should_journal_data(inode)) {
4071 BUFFER_TRACE(bh, "get write access");
4072 err = ext4_journal_get_write_access(handle, bh);
4076 zero_user(page, offset, length);
4077 BUFFER_TRACE(bh, "zeroed end of block");
4079 if (ext4_should_journal_data(inode)) {
4080 err = ext4_handle_dirty_metadata(handle, inode, bh);
4083 mark_buffer_dirty(bh);
4084 if (ext4_should_order_data(inode))
4085 err = ext4_jbd2_inode_add_write(handle, inode, from,
4096 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4097 * starting from file offset 'from'. The range to be zero'd must
4098 * be contained with in one block. If the specified range exceeds
4099 * the end of the block it will be shortened to end of the block
4100 * that cooresponds to 'from'
4102 static int ext4_block_zero_page_range(handle_t *handle,
4103 struct address_space *mapping, loff_t from, loff_t length)
4105 struct inode *inode = mapping->host;
4106 unsigned offset = from & (PAGE_SIZE-1);
4107 unsigned blocksize = inode->i_sb->s_blocksize;
4108 unsigned max = blocksize - (offset & (blocksize - 1));
4111 * correct length if it does not fall between
4112 * 'from' and the end of the block
4114 if (length > max || length < 0)
4117 if (IS_DAX(inode)) {
4118 return iomap_zero_range(inode, from, length, NULL,
4121 return __ext4_block_zero_page_range(handle, mapping, from, length);
4125 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4126 * up to the end of the block which corresponds to `from'.
4127 * This required during truncate. We need to physically zero the tail end
4128 * of that block so it doesn't yield old data if the file is later grown.
4130 static int ext4_block_truncate_page(handle_t *handle,
4131 struct address_space *mapping, loff_t from)
4133 unsigned offset = from & (PAGE_SIZE-1);
4136 struct inode *inode = mapping->host;
4138 /* If we are processing an encrypted inode during orphan list handling */
4139 if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode))
4142 blocksize = inode->i_sb->s_blocksize;
4143 length = blocksize - (offset & (blocksize - 1));
4145 return ext4_block_zero_page_range(handle, mapping, from, length);
4148 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4149 loff_t lstart, loff_t length)
4151 struct super_block *sb = inode->i_sb;
4152 struct address_space *mapping = inode->i_mapping;
4153 unsigned partial_start, partial_end;
4154 ext4_fsblk_t start, end;
4155 loff_t byte_end = (lstart + length - 1);
4158 partial_start = lstart & (sb->s_blocksize - 1);
4159 partial_end = byte_end & (sb->s_blocksize - 1);
4161 start = lstart >> sb->s_blocksize_bits;
4162 end = byte_end >> sb->s_blocksize_bits;
4164 /* Handle partial zero within the single block */
4166 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4167 err = ext4_block_zero_page_range(handle, mapping,
4171 /* Handle partial zero out on the start of the range */
4172 if (partial_start) {
4173 err = ext4_block_zero_page_range(handle, mapping,
4174 lstart, sb->s_blocksize);
4178 /* Handle partial zero out on the end of the range */
4179 if (partial_end != sb->s_blocksize - 1)
4180 err = ext4_block_zero_page_range(handle, mapping,
4181 byte_end - partial_end,
4186 int ext4_can_truncate(struct inode *inode)
4188 if (S_ISREG(inode->i_mode))
4190 if (S_ISDIR(inode->i_mode))
4192 if (S_ISLNK(inode->i_mode))
4193 return !ext4_inode_is_fast_symlink(inode);
4198 * We have to make sure i_disksize gets properly updated before we truncate
4199 * page cache due to hole punching or zero range. Otherwise i_disksize update
4200 * can get lost as it may have been postponed to submission of writeback but
4201 * that will never happen after we truncate page cache.
4203 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4207 loff_t size = i_size_read(inode);
4209 WARN_ON(!inode_is_locked(inode));
4210 if (offset > size || offset + len < size)
4213 if (EXT4_I(inode)->i_disksize >= size)
4216 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4218 return PTR_ERR(handle);
4219 ext4_update_i_disksize(inode, size);
4220 ext4_mark_inode_dirty(handle, inode);
4221 ext4_journal_stop(handle);
4227 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4228 * associated with the given offset and length
4230 * @inode: File inode
4231 * @offset: The offset where the hole will begin
4232 * @len: The length of the hole
4234 * Returns: 0 on success or negative on failure
4237 int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4239 struct super_block *sb = inode->i_sb;
4240 ext4_lblk_t first_block, stop_block;
4241 struct address_space *mapping = inode->i_mapping;
4242 loff_t first_block_offset, last_block_offset, max_length;
4243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4245 unsigned int credits;
4248 if (!S_ISREG(inode->i_mode))
4251 trace_ext4_punch_hole(inode, offset, length, 0);
4253 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4254 if (ext4_has_inline_data(inode)) {
4255 down_write(&EXT4_I(inode)->i_mmap_sem);
4256 ret = ext4_convert_inline_data(inode);
4257 up_write(&EXT4_I(inode)->i_mmap_sem);
4263 * Write out all dirty pages to avoid race conditions
4264 * Then release them.
4266 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4267 ret = filemap_write_and_wait_range(mapping, offset,
4268 offset + length - 1);
4275 /* No need to punch hole beyond i_size */
4276 if (offset >= inode->i_size)
4280 * If the hole extends beyond i_size, set the hole
4281 * to end after the page that contains i_size
4283 if (offset + length > inode->i_size) {
4284 length = inode->i_size +
4285 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4290 * For punch hole the length + offset needs to be within one block
4291 * before last range. Adjust the length if it goes beyond that limit.
4293 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
4294 if (offset + length > max_length)
4295 length = max_length - offset;
4297 if (offset & (sb->s_blocksize - 1) ||
4298 (offset + length) & (sb->s_blocksize - 1)) {
4300 * Attach jinode to inode for jbd2 if we do any zeroing of
4303 ret = ext4_inode_attach_jinode(inode);
4309 /* Wait all existing dio workers, newcomers will block on i_mutex */
4310 ext4_inode_block_unlocked_dio(inode);
4311 inode_dio_wait(inode);
4314 * Prevent page faults from reinstantiating pages we have released from
4317 down_write(&EXT4_I(inode)->i_mmap_sem);
4318 first_block_offset = round_up(offset, sb->s_blocksize);
4319 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4321 /* Now release the pages and zero block aligned part of pages*/
4322 if (last_block_offset > first_block_offset) {
4323 ret = ext4_update_disksize_before_punch(inode, offset, length);
4326 truncate_pagecache_range(inode, first_block_offset,
4330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4331 credits = ext4_writepage_trans_blocks(inode);
4333 credits = ext4_blocks_for_truncate(inode);
4334 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4335 if (IS_ERR(handle)) {
4336 ret = PTR_ERR(handle);
4337 ext4_std_error(sb, ret);
4341 ret = ext4_zero_partial_blocks(handle, inode, offset,
4346 first_block = (offset + sb->s_blocksize - 1) >>
4347 EXT4_BLOCK_SIZE_BITS(sb);
4348 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4350 /* If there are blocks to remove, do it */
4351 if (stop_block > first_block) {
4353 down_write(&EXT4_I(inode)->i_data_sem);
4354 ext4_discard_preallocations(inode);
4356 ret = ext4_es_remove_extent(inode, first_block,
4357 stop_block - first_block);
4359 up_write(&EXT4_I(inode)->i_data_sem);
4363 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4364 ret = ext4_ext_remove_space(inode, first_block,
4367 ret = ext4_ind_remove_space(handle, inode, first_block,
4370 up_write(&EXT4_I(inode)->i_data_sem);
4373 ext4_handle_sync(handle);
4375 inode->i_mtime = inode->i_ctime = current_time(inode);
4376 ext4_mark_inode_dirty(handle, inode);
4378 ext4_update_inode_fsync_trans(handle, inode, 1);
4380 ext4_journal_stop(handle);
4382 up_write(&EXT4_I(inode)->i_mmap_sem);
4383 ext4_inode_resume_unlocked_dio(inode);
4385 inode_unlock(inode);
4389 int ext4_inode_attach_jinode(struct inode *inode)
4391 struct ext4_inode_info *ei = EXT4_I(inode);
4392 struct jbd2_inode *jinode;
4394 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4397 jinode = jbd2_alloc_inode(GFP_KERNEL);
4398 spin_lock(&inode->i_lock);
4401 spin_unlock(&inode->i_lock);
4404 ei->jinode = jinode;
4405 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4408 spin_unlock(&inode->i_lock);
4409 if (unlikely(jinode != NULL))
4410 jbd2_free_inode(jinode);
4417 * We block out ext4_get_block() block instantiations across the entire
4418 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4419 * simultaneously on behalf of the same inode.
4421 * As we work through the truncate and commit bits of it to the journal there
4422 * is one core, guiding principle: the file's tree must always be consistent on
4423 * disk. We must be able to restart the truncate after a crash.
4425 * The file's tree may be transiently inconsistent in memory (although it
4426 * probably isn't), but whenever we close off and commit a journal transaction,
4427 * the contents of (the filesystem + the journal) must be consistent and
4428 * restartable. It's pretty simple, really: bottom up, right to left (although
4429 * left-to-right works OK too).
4431 * Note that at recovery time, journal replay occurs *before* the restart of
4432 * truncate against the orphan inode list.
4434 * The committed inode has the new, desired i_size (which is the same as
4435 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4436 * that this inode's truncate did not complete and it will again call
4437 * ext4_truncate() to have another go. So there will be instantiated blocks
4438 * to the right of the truncation point in a crashed ext4 filesystem. But
4439 * that's fine - as long as they are linked from the inode, the post-crash
4440 * ext4_truncate() run will find them and release them.
4442 int ext4_truncate(struct inode *inode)
4444 struct ext4_inode_info *ei = EXT4_I(inode);
4445 unsigned int credits;
4448 struct address_space *mapping = inode->i_mapping;
4451 * There is a possibility that we're either freeing the inode
4452 * or it's a completely new inode. In those cases we might not
4453 * have i_mutex locked because it's not necessary.
4455 if (!(inode->i_state & (I_NEW|I_FREEING)))
4456 WARN_ON(!inode_is_locked(inode));
4457 trace_ext4_truncate_enter(inode);
4459 if (!ext4_can_truncate(inode))
4462 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4464 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4465 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4467 if (ext4_has_inline_data(inode)) {
4470 err = ext4_inline_data_truncate(inode, &has_inline);
4477 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4478 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4479 if (ext4_inode_attach_jinode(inode) < 0)
4483 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4484 credits = ext4_writepage_trans_blocks(inode);
4486 credits = ext4_blocks_for_truncate(inode);
4488 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4490 return PTR_ERR(handle);
4492 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4493 ext4_block_truncate_page(handle, mapping, inode->i_size);
4496 * We add the inode to the orphan list, so that if this
4497 * truncate spans multiple transactions, and we crash, we will
4498 * resume the truncate when the filesystem recovers. It also
4499 * marks the inode dirty, to catch the new size.
4501 * Implication: the file must always be in a sane, consistent
4502 * truncatable state while each transaction commits.
4504 err = ext4_orphan_add(handle, inode);
4508 down_write(&EXT4_I(inode)->i_data_sem);
4510 ext4_discard_preallocations(inode);
4512 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4513 err = ext4_ext_truncate(handle, inode);
4515 ext4_ind_truncate(handle, inode);
4517 up_write(&ei->i_data_sem);
4522 ext4_handle_sync(handle);
4526 * If this was a simple ftruncate() and the file will remain alive,
4527 * then we need to clear up the orphan record which we created above.
4528 * However, if this was a real unlink then we were called by
4529 * ext4_evict_inode(), and we allow that function to clean up the
4530 * orphan info for us.
4533 ext4_orphan_del(handle, inode);
4535 inode->i_mtime = inode->i_ctime = current_time(inode);
4536 ext4_mark_inode_dirty(handle, inode);
4537 ext4_journal_stop(handle);
4539 trace_ext4_truncate_exit(inode);
4544 * ext4_get_inode_loc returns with an extra refcount against the inode's
4545 * underlying buffer_head on success. If 'in_mem' is true, we have all
4546 * data in memory that is needed to recreate the on-disk version of this
4549 static int __ext4_get_inode_loc(struct inode *inode,
4550 struct ext4_iloc *iloc, int in_mem)
4552 struct ext4_group_desc *gdp;
4553 struct buffer_head *bh;
4554 struct super_block *sb = inode->i_sb;
4556 int inodes_per_block, inode_offset;
4559 if (inode->i_ino < EXT4_ROOT_INO ||
4560 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4561 return -EFSCORRUPTED;
4563 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4564 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4569 * Figure out the offset within the block group inode table
4571 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4572 inode_offset = ((inode->i_ino - 1) %
4573 EXT4_INODES_PER_GROUP(sb));
4574 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4576 block = ext4_inode_table(sb, gdp);
4577 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4578 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4579 ext4_error(sb, "Invalid inode table block %llu in "
4580 "block_group %u", block, iloc->block_group);
4581 return -EFSCORRUPTED;
4583 block += (inode_offset / inodes_per_block);
4585 bh = sb_getblk(sb, block);
4588 if (!buffer_uptodate(bh)) {
4592 * If the buffer has the write error flag, we have failed
4593 * to write out another inode in the same block. In this
4594 * case, we don't have to read the block because we may
4595 * read the old inode data successfully.
4597 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4598 set_buffer_uptodate(bh);
4600 if (buffer_uptodate(bh)) {
4601 /* someone brought it uptodate while we waited */
4607 * If we have all information of the inode in memory and this
4608 * is the only valid inode in the block, we need not read the
4612 struct buffer_head *bitmap_bh;
4615 start = inode_offset & ~(inodes_per_block - 1);
4617 /* Is the inode bitmap in cache? */
4618 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4619 if (unlikely(!bitmap_bh))
4623 * If the inode bitmap isn't in cache then the
4624 * optimisation may end up performing two reads instead
4625 * of one, so skip it.
4627 if (!buffer_uptodate(bitmap_bh)) {
4631 for (i = start; i < start + inodes_per_block; i++) {
4632 if (i == inode_offset)
4634 if (ext4_test_bit(i, bitmap_bh->b_data))
4638 if (i == start + inodes_per_block) {
4639 /* all other inodes are free, so skip I/O */
4640 memset(bh->b_data, 0, bh->b_size);
4641 set_buffer_uptodate(bh);
4649 * If we need to do any I/O, try to pre-readahead extra
4650 * blocks from the inode table.
4652 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4653 ext4_fsblk_t b, end, table;
4655 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4657 table = ext4_inode_table(sb, gdp);
4658 /* s_inode_readahead_blks is always a power of 2 */
4659 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4663 num = EXT4_INODES_PER_GROUP(sb);
4664 if (ext4_has_group_desc_csum(sb))
4665 num -= ext4_itable_unused_count(sb, gdp);
4666 table += num / inodes_per_block;
4670 sb_breadahead_unmovable(sb, b++);
4674 * There are other valid inodes in the buffer, this inode
4675 * has in-inode xattrs, or we don't have this inode in memory.
4676 * Read the block from disk.
4678 trace_ext4_load_inode(inode);
4680 bh->b_end_io = end_buffer_read_sync;
4681 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4683 if (!buffer_uptodate(bh)) {
4684 EXT4_ERROR_INODE_BLOCK(inode, block,
4685 "unable to read itable block");
4695 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4697 /* We have all inode data except xattrs in memory here. */
4698 return __ext4_get_inode_loc(inode, iloc,
4699 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4702 void ext4_set_inode_flags(struct inode *inode)
4704 unsigned int flags = EXT4_I(inode)->i_flags;
4705 unsigned int new_fl = 0;
4707 if (flags & EXT4_SYNC_FL)
4709 if (flags & EXT4_APPEND_FL)
4711 if (flags & EXT4_IMMUTABLE_FL)
4712 new_fl |= S_IMMUTABLE;
4713 if (flags & EXT4_NOATIME_FL)
4714 new_fl |= S_NOATIME;
4715 if (flags & EXT4_DIRSYNC_FL)
4716 new_fl |= S_DIRSYNC;
4717 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) &&
4718 !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) &&
4719 !ext4_encrypted_inode(inode))
4721 inode_set_flags(inode, new_fl,
4722 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
4725 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4726 struct ext4_inode_info *ei)
4729 struct inode *inode = &(ei->vfs_inode);
4730 struct super_block *sb = inode->i_sb;
4732 if (ext4_has_feature_huge_file(sb)) {
4733 /* we are using combined 48 bit field */
4734 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4735 le32_to_cpu(raw_inode->i_blocks_lo);
4736 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4737 /* i_blocks represent file system block size */
4738 return i_blocks << (inode->i_blkbits - 9);
4743 return le32_to_cpu(raw_inode->i_blocks_lo);
4747 static inline int ext4_iget_extra_inode(struct inode *inode,
4748 struct ext4_inode *raw_inode,
4749 struct ext4_inode_info *ei)
4751 __le32 *magic = (void *)raw_inode +
4752 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4754 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4755 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4758 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4759 err = ext4_find_inline_data_nolock(inode);
4760 if (!err && ext4_has_inline_data(inode))
4761 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4764 EXT4_I(inode)->i_inline_off = 0;
4768 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4770 if (!ext4_has_feature_project(inode->i_sb))
4772 *projid = EXT4_I(inode)->i_projid;
4776 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4777 ext4_iget_flags flags, const char *function,
4780 struct ext4_iloc iloc;
4781 struct ext4_inode *raw_inode;
4782 struct ext4_inode_info *ei;
4783 struct inode *inode;
4784 journal_t *journal = EXT4_SB(sb)->s_journal;
4792 if ((!(flags & EXT4_IGET_SPECIAL) &&
4793 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4794 (ino < EXT4_ROOT_INO) ||
4795 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4796 if (flags & EXT4_IGET_HANDLE)
4797 return ERR_PTR(-ESTALE);
4798 __ext4_error(sb, function, line,
4799 "inode #%lu: comm %s: iget: illegal inode #",
4800 ino, current->comm);
4801 return ERR_PTR(-EFSCORRUPTED);
4804 inode = iget_locked(sb, ino);
4806 return ERR_PTR(-ENOMEM);
4807 if (!(inode->i_state & I_NEW))
4813 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4816 raw_inode = ext4_raw_inode(&iloc);
4818 if ((flags & EXT4_IGET_HANDLE) &&
4819 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4824 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4825 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4826 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4827 EXT4_INODE_SIZE(inode->i_sb) ||
4828 (ei->i_extra_isize & 3)) {
4829 ext4_error_inode(inode, function, line, 0,
4830 "iget: bad extra_isize %u "
4833 EXT4_INODE_SIZE(inode->i_sb));
4834 ret = -EFSCORRUPTED;
4838 ei->i_extra_isize = 0;
4840 /* Precompute checksum seed for inode metadata */
4841 if (ext4_has_metadata_csum(sb)) {
4842 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4844 __le32 inum = cpu_to_le32(inode->i_ino);
4845 __le32 gen = raw_inode->i_generation;
4846 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4848 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4852 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4853 ext4_error_inode(inode, function, line, 0,
4854 "iget: checksum invalid");
4859 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4860 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4861 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4862 if (ext4_has_feature_project(sb) &&
4863 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4864 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4865 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4867 i_projid = EXT4_DEF_PROJID;
4869 if (!(test_opt(inode->i_sb, NO_UID32))) {
4870 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4871 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4873 i_uid_write(inode, i_uid);
4874 i_gid_write(inode, i_gid);
4875 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4876 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4878 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4879 ei->i_inline_off = 0;
4880 ei->i_dir_start_lookup = 0;
4881 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4882 /* We now have enough fields to check if the inode was active or not.
4883 * This is needed because nfsd might try to access dead inodes
4884 * the test is that same one that e2fsck uses
4885 * NeilBrown 1999oct15
4887 if (inode->i_nlink == 0) {
4888 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4889 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4890 ino != EXT4_BOOT_LOADER_INO) {
4891 /* this inode is deleted or unallocated */
4892 if (flags & EXT4_IGET_SPECIAL) {
4893 ext4_error_inode(inode, function, line, 0,
4894 "iget: special inode unallocated");
4895 ret = -EFSCORRUPTED;
4900 /* The only unlinked inodes we let through here have
4901 * valid i_mode and are being read by the orphan
4902 * recovery code: that's fine, we're about to complete
4903 * the process of deleting those.
4904 * OR it is the EXT4_BOOT_LOADER_INO which is
4905 * not initialized on a new filesystem. */
4907 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4908 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4909 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4910 if (ext4_has_feature_64bit(sb))
4912 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4913 inode->i_size = ext4_isize(sb, raw_inode);
4914 if ((size = i_size_read(inode)) < 0) {
4915 ext4_error_inode(inode, function, line, 0,
4916 "iget: bad i_size value: %lld", size);
4917 ret = -EFSCORRUPTED;
4921 * If dir_index is not enabled but there's dir with INDEX flag set,
4922 * we'd normally treat htree data as empty space. But with metadata
4923 * checksumming that corrupts checksums so forbid that.
4925 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4926 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4927 EXT4_ERROR_INODE(inode,
4928 "iget: Dir with htree data on filesystem without dir_index feature.");
4929 ret = -EFSCORRUPTED;
4932 ei->i_disksize = inode->i_size;
4934 ei->i_reserved_quota = 0;
4936 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4937 ei->i_block_group = iloc.block_group;
4938 ei->i_last_alloc_group = ~0;
4940 * NOTE! The in-memory inode i_data array is in little-endian order
4941 * even on big-endian machines: we do NOT byteswap the block numbers!
4943 for (block = 0; block < EXT4_N_BLOCKS; block++)
4944 ei->i_data[block] = raw_inode->i_block[block];
4945 INIT_LIST_HEAD(&ei->i_orphan);
4948 * Set transaction id's of transactions that have to be committed
4949 * to finish f[data]sync. We set them to currently running transaction
4950 * as we cannot be sure that the inode or some of its metadata isn't
4951 * part of the transaction - the inode could have been reclaimed and
4952 * now it is reread from disk.
4955 transaction_t *transaction;
4958 read_lock(&journal->j_state_lock);
4959 if (journal->j_running_transaction)
4960 transaction = journal->j_running_transaction;
4962 transaction = journal->j_committing_transaction;
4964 tid = transaction->t_tid;
4966 tid = journal->j_commit_sequence;
4967 read_unlock(&journal->j_state_lock);
4968 ei->i_sync_tid = tid;
4969 ei->i_datasync_tid = tid;
4972 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4973 if (ei->i_extra_isize == 0) {
4974 /* The extra space is currently unused. Use it. */
4975 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4976 ei->i_extra_isize = sizeof(struct ext4_inode) -
4977 EXT4_GOOD_OLD_INODE_SIZE;
4979 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4985 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4986 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4987 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4988 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4990 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4991 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4992 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4993 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4995 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5000 if (ei->i_file_acl &&
5001 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
5002 ext4_error_inode(inode, function, line, 0,
5003 "iget: bad extended attribute block %llu",
5005 ret = -EFSCORRUPTED;
5007 } else if (!ext4_has_inline_data(inode)) {
5008 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5009 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5010 (S_ISLNK(inode->i_mode) &&
5011 !ext4_inode_is_fast_symlink(inode))))
5012 /* Validate extent which is part of inode */
5013 ret = ext4_ext_check_inode(inode);
5014 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5015 (S_ISLNK(inode->i_mode) &&
5016 !ext4_inode_is_fast_symlink(inode))) {
5017 /* Validate block references which are part of inode */
5018 ret = ext4_ind_check_inode(inode);
5024 if (S_ISREG(inode->i_mode)) {
5025 inode->i_op = &ext4_file_inode_operations;
5026 inode->i_fop = &ext4_file_operations;
5027 ext4_set_aops(inode);
5028 } else if (S_ISDIR(inode->i_mode)) {
5029 inode->i_op = &ext4_dir_inode_operations;
5030 inode->i_fop = &ext4_dir_operations;
5031 } else if (S_ISLNK(inode->i_mode)) {
5032 if (ext4_encrypted_inode(inode)) {
5033 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5034 ext4_set_aops(inode);
5035 } else if (ext4_inode_is_fast_symlink(inode)) {
5036 inode->i_link = (char *)ei->i_data;
5037 inode->i_op = &ext4_fast_symlink_inode_operations;
5038 nd_terminate_link(ei->i_data, inode->i_size,
5039 sizeof(ei->i_data) - 1);
5041 inode->i_op = &ext4_symlink_inode_operations;
5042 ext4_set_aops(inode);
5044 inode_nohighmem(inode);
5045 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5046 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5047 inode->i_op = &ext4_special_inode_operations;
5048 if (raw_inode->i_block[0])
5049 init_special_inode(inode, inode->i_mode,
5050 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5052 init_special_inode(inode, inode->i_mode,
5053 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5054 } else if (ino == EXT4_BOOT_LOADER_INO) {
5055 make_bad_inode(inode);
5057 ret = -EFSCORRUPTED;
5058 ext4_error_inode(inode, function, line, 0,
5059 "iget: bogus i_mode (%o)", inode->i_mode);
5063 ext4_set_inode_flags(inode);
5065 unlock_new_inode(inode);
5071 return ERR_PTR(ret);
5074 static int ext4_inode_blocks_set(handle_t *handle,
5075 struct ext4_inode *raw_inode,
5076 struct ext4_inode_info *ei)
5078 struct inode *inode = &(ei->vfs_inode);
5079 u64 i_blocks = READ_ONCE(inode->i_blocks);
5080 struct super_block *sb = inode->i_sb;
5082 if (i_blocks <= ~0U) {
5084 * i_blocks can be represented in a 32 bit variable
5085 * as multiple of 512 bytes
5087 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5088 raw_inode->i_blocks_high = 0;
5089 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5092 if (!ext4_has_feature_huge_file(sb))
5095 if (i_blocks <= 0xffffffffffffULL) {
5097 * i_blocks can be represented in a 48 bit variable
5098 * as multiple of 512 bytes
5100 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5101 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5102 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5104 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5105 /* i_block is stored in file system block size */
5106 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5107 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5108 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5113 struct other_inode {
5114 unsigned long orig_ino;
5115 struct ext4_inode *raw_inode;
5118 static int other_inode_match(struct inode * inode, unsigned long ino,
5121 struct other_inode *oi = (struct other_inode *) data;
5123 if ((inode->i_ino != ino) ||
5124 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5126 ((inode->i_state & I_DIRTY_TIME) == 0))
5128 spin_lock(&inode->i_lock);
5129 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5130 I_DIRTY_INODE)) == 0) &&
5131 (inode->i_state & I_DIRTY_TIME)) {
5132 struct ext4_inode_info *ei = EXT4_I(inode);
5134 inode->i_state &= ~I_DIRTY_TIME;
5135 spin_unlock(&inode->i_lock);
5137 spin_lock(&ei->i_raw_lock);
5138 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5139 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5140 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5141 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5142 spin_unlock(&ei->i_raw_lock);
5143 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5146 spin_unlock(&inode->i_lock);
5151 * Opportunistically update the other time fields for other inodes in
5152 * the same inode table block.
5154 static void ext4_update_other_inodes_time(struct super_block *sb,
5155 unsigned long orig_ino, char *buf)
5157 struct other_inode oi;
5159 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5160 int inode_size = EXT4_INODE_SIZE(sb);
5162 oi.orig_ino = orig_ino;
5164 * Calculate the first inode in the inode table block. Inode
5165 * numbers are one-based. That is, the first inode in a block
5166 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5168 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5169 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5170 if (ino == orig_ino)
5172 oi.raw_inode = (struct ext4_inode *) buf;
5173 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5178 * Post the struct inode info into an on-disk inode location in the
5179 * buffer-cache. This gobbles the caller's reference to the
5180 * buffer_head in the inode location struct.
5182 * The caller must have write access to iloc->bh.
5184 static int ext4_do_update_inode(handle_t *handle,
5185 struct inode *inode,
5186 struct ext4_iloc *iloc)
5188 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5189 struct ext4_inode_info *ei = EXT4_I(inode);
5190 struct buffer_head *bh = iloc->bh;
5191 struct super_block *sb = inode->i_sb;
5193 int need_datasync = 0, set_large_file = 0;
5198 spin_lock(&ei->i_raw_lock);
5200 /* For fields not tracked in the in-memory inode,
5201 * initialise them to zero for new inodes. */
5202 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5203 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5205 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5207 spin_unlock(&ei->i_raw_lock);
5211 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5212 i_uid = i_uid_read(inode);
5213 i_gid = i_gid_read(inode);
5214 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5215 if (!(test_opt(inode->i_sb, NO_UID32))) {
5216 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5217 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5219 * Fix up interoperability with old kernels. Otherwise, old inodes get
5220 * re-used with the upper 16 bits of the uid/gid intact
5222 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5223 raw_inode->i_uid_high = 0;
5224 raw_inode->i_gid_high = 0;
5226 raw_inode->i_uid_high =
5227 cpu_to_le16(high_16_bits(i_uid));
5228 raw_inode->i_gid_high =
5229 cpu_to_le16(high_16_bits(i_gid));
5232 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5233 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5234 raw_inode->i_uid_high = 0;
5235 raw_inode->i_gid_high = 0;
5237 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5239 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5240 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5241 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5242 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5244 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5245 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5246 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5247 raw_inode->i_file_acl_high =
5248 cpu_to_le16(ei->i_file_acl >> 32);
5249 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5250 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5251 ext4_isize_set(raw_inode, ei->i_disksize);
5254 if (ei->i_disksize > 0x7fffffffULL) {
5255 if (!ext4_has_feature_large_file(sb) ||
5256 EXT4_SB(sb)->s_es->s_rev_level ==
5257 cpu_to_le32(EXT4_GOOD_OLD_REV))
5260 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5261 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5262 if (old_valid_dev(inode->i_rdev)) {
5263 raw_inode->i_block[0] =
5264 cpu_to_le32(old_encode_dev(inode->i_rdev));
5265 raw_inode->i_block[1] = 0;
5267 raw_inode->i_block[0] = 0;
5268 raw_inode->i_block[1] =
5269 cpu_to_le32(new_encode_dev(inode->i_rdev));
5270 raw_inode->i_block[2] = 0;
5272 } else if (!ext4_has_inline_data(inode)) {
5273 for (block = 0; block < EXT4_N_BLOCKS; block++)
5274 raw_inode->i_block[block] = ei->i_data[block];
5277 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5278 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5279 if (ei->i_extra_isize) {
5280 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5281 raw_inode->i_version_hi =
5282 cpu_to_le32(inode->i_version >> 32);
5283 raw_inode->i_extra_isize =
5284 cpu_to_le16(ei->i_extra_isize);
5288 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5289 i_projid != EXT4_DEF_PROJID);
5291 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5292 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5293 raw_inode->i_projid = cpu_to_le32(i_projid);
5295 ext4_inode_csum_set(inode, raw_inode, ei);
5296 spin_unlock(&ei->i_raw_lock);
5297 if (inode->i_sb->s_flags & MS_LAZYTIME)
5298 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5301 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5302 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5305 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5306 if (set_large_file) {
5307 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5308 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5311 ext4_update_dynamic_rev(sb);
5312 ext4_set_feature_large_file(sb);
5313 ext4_handle_sync(handle);
5314 err = ext4_handle_dirty_super(handle, sb);
5316 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5319 ext4_std_error(inode->i_sb, err);
5324 * ext4_write_inode()
5326 * We are called from a few places:
5328 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5329 * Here, there will be no transaction running. We wait for any running
5330 * transaction to commit.
5332 * - Within flush work (sys_sync(), kupdate and such).
5333 * We wait on commit, if told to.
5335 * - Within iput_final() -> write_inode_now()
5336 * We wait on commit, if told to.
5338 * In all cases it is actually safe for us to return without doing anything,
5339 * because the inode has been copied into a raw inode buffer in
5340 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5343 * Note that we are absolutely dependent upon all inode dirtiers doing the
5344 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5345 * which we are interested.
5347 * It would be a bug for them to not do this. The code:
5349 * mark_inode_dirty(inode)
5351 * inode->i_size = expr;
5353 * is in error because write_inode() could occur while `stuff()' is running,
5354 * and the new i_size will be lost. Plus the inode will no longer be on the
5355 * superblock's dirty inode list.
5357 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5361 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5362 sb_rdonly(inode->i_sb))
5365 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5368 if (EXT4_SB(inode->i_sb)->s_journal) {
5369 if (ext4_journal_current_handle()) {
5370 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5376 * No need to force transaction in WB_SYNC_NONE mode. Also
5377 * ext4_sync_fs() will force the commit after everything is
5380 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5383 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5384 EXT4_I(inode)->i_sync_tid);
5386 struct ext4_iloc iloc;
5388 err = __ext4_get_inode_loc(inode, &iloc, 0);
5392 * sync(2) will flush the whole buffer cache. No need to do
5393 * it here separately for each inode.
5395 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5396 sync_dirty_buffer(iloc.bh);
5397 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5398 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5399 "IO error syncing inode");
5408 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5409 * buffers that are attached to a page stradding i_size and are undergoing
5410 * commit. In that case we have to wait for commit to finish and try again.
5412 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5416 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5417 tid_t commit_tid = 0;
5420 offset = inode->i_size & (PAGE_SIZE - 1);
5422 * If the page is fully truncated, we don't need to wait for any commit
5423 * (and we even should not as __ext4_journalled_invalidatepage() may
5424 * strip all buffers from the page but keep the page dirty which can then
5425 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5426 * buffers). Also we don't need to wait for any commit if all buffers in
5427 * the page remain valid. This is most beneficial for the common case of
5428 * blocksize == PAGESIZE.
5430 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5433 page = find_lock_page(inode->i_mapping,
5434 inode->i_size >> PAGE_SHIFT);
5437 ret = __ext4_journalled_invalidatepage(page, offset,
5438 PAGE_SIZE - offset);
5444 read_lock(&journal->j_state_lock);
5445 if (journal->j_committing_transaction)
5446 commit_tid = journal->j_committing_transaction->t_tid;
5447 read_unlock(&journal->j_state_lock);
5449 jbd2_log_wait_commit(journal, commit_tid);
5456 * Called from notify_change.
5458 * We want to trap VFS attempts to truncate the file as soon as
5459 * possible. In particular, we want to make sure that when the VFS
5460 * shrinks i_size, we put the inode on the orphan list and modify
5461 * i_disksize immediately, so that during the subsequent flushing of
5462 * dirty pages and freeing of disk blocks, we can guarantee that any
5463 * commit will leave the blocks being flushed in an unused state on
5464 * disk. (On recovery, the inode will get truncated and the blocks will
5465 * be freed, so we have a strong guarantee that no future commit will
5466 * leave these blocks visible to the user.)
5468 * Another thing we have to assure is that if we are in ordered mode
5469 * and inode is still attached to the committing transaction, we must
5470 * we start writeout of all the dirty pages which are being truncated.
5471 * This way we are sure that all the data written in the previous
5472 * transaction are already on disk (truncate waits for pages under
5475 * Called with inode->i_mutex down.
5477 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5479 struct inode *inode = d_inode(dentry);
5482 const unsigned int ia_valid = attr->ia_valid;
5484 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5487 if (unlikely(IS_IMMUTABLE(inode)))
5490 if (unlikely(IS_APPEND(inode) &&
5491 (ia_valid & (ATTR_MODE | ATTR_UID |
5492 ATTR_GID | ATTR_TIMES_SET))))
5495 error = setattr_prepare(dentry, attr);
5499 if (is_quota_modification(inode, attr)) {
5500 error = dquot_initialize(inode);
5504 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5505 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5508 /* (user+group)*(old+new) structure, inode write (sb,
5509 * inode block, ? - but truncate inode update has it) */
5510 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5511 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5512 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5513 if (IS_ERR(handle)) {
5514 error = PTR_ERR(handle);
5518 /* dquot_transfer() calls back ext4_get_inode_usage() which
5519 * counts xattr inode references.
5521 down_read(&EXT4_I(inode)->xattr_sem);
5522 error = dquot_transfer(inode, attr);
5523 up_read(&EXT4_I(inode)->xattr_sem);
5526 ext4_journal_stop(handle);
5529 /* Update corresponding info in inode so that everything is in
5530 * one transaction */
5531 if (attr->ia_valid & ATTR_UID)
5532 inode->i_uid = attr->ia_uid;
5533 if (attr->ia_valid & ATTR_GID)
5534 inode->i_gid = attr->ia_gid;
5535 error = ext4_mark_inode_dirty(handle, inode);
5536 ext4_journal_stop(handle);
5539 if (attr->ia_valid & ATTR_SIZE) {
5541 loff_t oldsize = inode->i_size;
5542 int shrink = (attr->ia_size <= inode->i_size);
5544 if (ext4_encrypted_inode(inode)) {
5545 error = fscrypt_get_encryption_info(inode);
5548 if (!fscrypt_has_encryption_key(inode))
5552 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5553 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5555 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5558 if (!S_ISREG(inode->i_mode))
5561 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5562 inode_inc_iversion(inode);
5564 if (ext4_should_order_data(inode) &&
5565 (attr->ia_size < inode->i_size)) {
5566 error = ext4_begin_ordered_truncate(inode,
5571 if (attr->ia_size != inode->i_size) {
5572 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5573 if (IS_ERR(handle)) {
5574 error = PTR_ERR(handle);
5577 if (ext4_handle_valid(handle) && shrink) {
5578 error = ext4_orphan_add(handle, inode);
5582 * Update c/mtime on truncate up, ext4_truncate() will
5583 * update c/mtime in shrink case below
5586 inode->i_mtime = current_time(inode);
5587 inode->i_ctime = inode->i_mtime;
5589 down_write(&EXT4_I(inode)->i_data_sem);
5590 EXT4_I(inode)->i_disksize = attr->ia_size;
5591 rc = ext4_mark_inode_dirty(handle, inode);
5595 * We have to update i_size under i_data_sem together
5596 * with i_disksize to avoid races with writeback code
5597 * running ext4_wb_update_i_disksize().
5600 i_size_write(inode, attr->ia_size);
5601 up_write(&EXT4_I(inode)->i_data_sem);
5602 ext4_journal_stop(handle);
5604 if (orphan && inode->i_nlink)
5605 ext4_orphan_del(NULL, inode);
5610 pagecache_isize_extended(inode, oldsize, inode->i_size);
5613 * Blocks are going to be removed from the inode. Wait
5614 * for dio in flight. Temporarily disable
5615 * dioread_nolock to prevent livelock.
5618 if (!ext4_should_journal_data(inode)) {
5619 ext4_inode_block_unlocked_dio(inode);
5620 inode_dio_wait(inode);
5621 ext4_inode_resume_unlocked_dio(inode);
5623 ext4_wait_for_tail_page_commit(inode);
5625 down_write(&EXT4_I(inode)->i_mmap_sem);
5627 * Truncate pagecache after we've waited for commit
5628 * in data=journal mode to make pages freeable.
5630 truncate_pagecache(inode, inode->i_size);
5632 rc = ext4_truncate(inode);
5636 up_write(&EXT4_I(inode)->i_mmap_sem);
5640 setattr_copy(inode, attr);
5641 mark_inode_dirty(inode);
5645 * If the call to ext4_truncate failed to get a transaction handle at
5646 * all, we need to clean up the in-core orphan list manually.
5648 if (orphan && inode->i_nlink)
5649 ext4_orphan_del(NULL, inode);
5651 if (!error && (ia_valid & ATTR_MODE))
5652 rc = posix_acl_chmod(inode, inode->i_mode);
5655 ext4_std_error(inode->i_sb, error);
5661 int ext4_getattr(const struct path *path, struct kstat *stat,
5662 u32 request_mask, unsigned int query_flags)
5664 struct inode *inode = d_inode(path->dentry);
5665 struct ext4_inode *raw_inode;
5666 struct ext4_inode_info *ei = EXT4_I(inode);
5669 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5670 stat->result_mask |= STATX_BTIME;
5671 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5672 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5675 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5676 if (flags & EXT4_APPEND_FL)
5677 stat->attributes |= STATX_ATTR_APPEND;
5678 if (flags & EXT4_COMPR_FL)
5679 stat->attributes |= STATX_ATTR_COMPRESSED;
5680 if (flags & EXT4_ENCRYPT_FL)
5681 stat->attributes |= STATX_ATTR_ENCRYPTED;
5682 if (flags & EXT4_IMMUTABLE_FL)
5683 stat->attributes |= STATX_ATTR_IMMUTABLE;
5684 if (flags & EXT4_NODUMP_FL)
5685 stat->attributes |= STATX_ATTR_NODUMP;
5687 stat->attributes_mask |= (STATX_ATTR_APPEND |
5688 STATX_ATTR_COMPRESSED |
5689 STATX_ATTR_ENCRYPTED |
5690 STATX_ATTR_IMMUTABLE |
5693 generic_fillattr(inode, stat);
5697 int ext4_file_getattr(const struct path *path, struct kstat *stat,
5698 u32 request_mask, unsigned int query_flags)
5700 struct inode *inode = d_inode(path->dentry);
5701 u64 delalloc_blocks;
5703 ext4_getattr(path, stat, request_mask, query_flags);
5706 * If there is inline data in the inode, the inode will normally not
5707 * have data blocks allocated (it may have an external xattr block).
5708 * Report at least one sector for such files, so tools like tar, rsync,
5709 * others don't incorrectly think the file is completely sparse.
5711 if (unlikely(ext4_has_inline_data(inode)))
5712 stat->blocks += (stat->size + 511) >> 9;
5715 * We can't update i_blocks if the block allocation is delayed
5716 * otherwise in the case of system crash before the real block
5717 * allocation is done, we will have i_blocks inconsistent with
5718 * on-disk file blocks.
5719 * We always keep i_blocks updated together with real
5720 * allocation. But to not confuse with user, stat
5721 * will return the blocks that include the delayed allocation
5722 * blocks for this file.
5724 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5725 EXT4_I(inode)->i_reserved_data_blocks);
5726 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5730 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5733 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5734 return ext4_ind_trans_blocks(inode, lblocks);
5735 return ext4_ext_index_trans_blocks(inode, pextents);
5739 * Account for index blocks, block groups bitmaps and block group
5740 * descriptor blocks if modify datablocks and index blocks
5741 * worse case, the indexs blocks spread over different block groups
5743 * If datablocks are discontiguous, they are possible to spread over
5744 * different block groups too. If they are contiguous, with flexbg,
5745 * they could still across block group boundary.
5747 * Also account for superblock, inode, quota and xattr blocks
5749 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5752 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5758 * How many index blocks need to touch to map @lblocks logical blocks
5759 * to @pextents physical extents?
5761 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5766 * Now let's see how many group bitmaps and group descriptors need
5769 groups = idxblocks + pextents;
5771 if (groups > ngroups)
5773 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5774 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5776 /* bitmaps and block group descriptor blocks */
5777 ret += groups + gdpblocks;
5779 /* Blocks for super block, inode, quota and xattr blocks */
5780 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5786 * Calculate the total number of credits to reserve to fit
5787 * the modification of a single pages into a single transaction,
5788 * which may include multiple chunks of block allocations.
5790 * This could be called via ext4_write_begin()
5792 * We need to consider the worse case, when
5793 * one new block per extent.
5795 int ext4_writepage_trans_blocks(struct inode *inode)
5797 int bpp = ext4_journal_blocks_per_page(inode);
5800 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5802 /* Account for data blocks for journalled mode */
5803 if (ext4_should_journal_data(inode))
5809 * Calculate the journal credits for a chunk of data modification.
5811 * This is called from DIO, fallocate or whoever calling
5812 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5814 * journal buffers for data blocks are not included here, as DIO
5815 * and fallocate do no need to journal data buffers.
5817 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5819 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5823 * The caller must have previously called ext4_reserve_inode_write().
5824 * Give this, we know that the caller already has write access to iloc->bh.
5826 int ext4_mark_iloc_dirty(handle_t *handle,
5827 struct inode *inode, struct ext4_iloc *iloc)
5831 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5835 if (IS_I_VERSION(inode))
5836 inode_inc_iversion(inode);
5838 /* the do_update_inode consumes one bh->b_count */
5841 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5842 err = ext4_do_update_inode(handle, inode, iloc);
5848 * On success, We end up with an outstanding reference count against
5849 * iloc->bh. This _must_ be cleaned up later.
5853 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5854 struct ext4_iloc *iloc)
5858 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5861 err = ext4_get_inode_loc(inode, iloc);
5863 BUFFER_TRACE(iloc->bh, "get_write_access");
5864 err = ext4_journal_get_write_access(handle, iloc->bh);
5870 ext4_std_error(inode->i_sb, err);
5874 static int __ext4_expand_extra_isize(struct inode *inode,
5875 unsigned int new_extra_isize,
5876 struct ext4_iloc *iloc,
5877 handle_t *handle, int *no_expand)
5879 struct ext4_inode *raw_inode;
5880 struct ext4_xattr_ibody_header *header;
5881 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5882 struct ext4_inode_info *ei = EXT4_I(inode);
5885 /* this was checked at iget time, but double check for good measure */
5886 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5887 (ei->i_extra_isize & 3)) {
5888 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5890 EXT4_INODE_SIZE(inode->i_sb));
5891 return -EFSCORRUPTED;
5893 if ((new_extra_isize < ei->i_extra_isize) ||
5894 (new_extra_isize < 4) ||
5895 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5896 return -EINVAL; /* Should never happen */
5898 raw_inode = ext4_raw_inode(iloc);
5900 header = IHDR(inode, raw_inode);
5902 /* No extended attributes present */
5903 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5904 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5905 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5906 EXT4_I(inode)->i_extra_isize, 0,
5907 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5908 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5913 * We may need to allocate external xattr block so we need quotas
5914 * initialized. Here we can be called with various locks held so we
5915 * cannot affort to initialize quotas ourselves. So just bail.
5917 if (dquot_initialize_needed(inode))
5920 /* try to expand with EAs present */
5921 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5925 * Inode size expansion failed; don't try again
5934 * Expand an inode by new_extra_isize bytes.
5935 * Returns 0 on success or negative error number on failure.
5937 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5938 unsigned int new_extra_isize,
5939 struct ext4_iloc iloc,
5945 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5949 * In nojournal mode, we can immediately attempt to expand
5950 * the inode. When journaled, we first need to obtain extra
5951 * buffer credits since we may write into the EA block
5952 * with this same handle. If journal_extend fails, then it will
5953 * only result in a minor loss of functionality for that inode.
5954 * If this is felt to be critical, then e2fsck should be run to
5955 * force a large enough s_min_extra_isize.
5957 if (ext4_handle_valid(handle) &&
5958 jbd2_journal_extend(handle,
5959 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5962 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5965 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5966 handle, &no_expand);
5967 ext4_write_unlock_xattr(inode, &no_expand);
5972 int ext4_expand_extra_isize(struct inode *inode,
5973 unsigned int new_extra_isize,
5974 struct ext4_iloc *iloc)
5980 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5985 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5986 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5987 if (IS_ERR(handle)) {
5988 error = PTR_ERR(handle);
5993 ext4_write_lock_xattr(inode, &no_expand);
5995 BUFFER_TRACE(iloc->bh, "get_write_access");
5996 error = ext4_journal_get_write_access(handle, iloc->bh);
6002 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6003 handle, &no_expand);
6005 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6010 ext4_write_unlock_xattr(inode, &no_expand);
6011 ext4_journal_stop(handle);
6016 * What we do here is to mark the in-core inode as clean with respect to inode
6017 * dirtiness (it may still be data-dirty).
6018 * This means that the in-core inode may be reaped by prune_icache
6019 * without having to perform any I/O. This is a very good thing,
6020 * because *any* task may call prune_icache - even ones which
6021 * have a transaction open against a different journal.
6023 * Is this cheating? Not really. Sure, we haven't written the
6024 * inode out, but prune_icache isn't a user-visible syncing function.
6025 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6026 * we start and wait on commits.
6028 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6030 struct ext4_iloc iloc;
6031 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6035 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6036 err = ext4_reserve_inode_write(handle, inode, &iloc);
6040 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6041 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6044 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6048 * ext4_dirty_inode() is called from __mark_inode_dirty()
6050 * We're really interested in the case where a file is being extended.
6051 * i_size has been changed by generic_commit_write() and we thus need
6052 * to include the updated inode in the current transaction.
6054 * Also, dquot_alloc_block() will always dirty the inode when blocks
6055 * are allocated to the file.
6057 * If the inode is marked synchronous, we don't honour that here - doing
6058 * so would cause a commit on atime updates, which we don't bother doing.
6059 * We handle synchronous inodes at the highest possible level.
6061 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6062 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6063 * to copy into the on-disk inode structure are the timestamp files.
6065 void ext4_dirty_inode(struct inode *inode, int flags)
6069 if (flags == I_DIRTY_TIME)
6071 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6075 ext4_mark_inode_dirty(handle, inode);
6077 ext4_journal_stop(handle);
6084 * Bind an inode's backing buffer_head into this transaction, to prevent
6085 * it from being flushed to disk early. Unlike
6086 * ext4_reserve_inode_write, this leaves behind no bh reference and
6087 * returns no iloc structure, so the caller needs to repeat the iloc
6088 * lookup to mark the inode dirty later.
6090 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
6092 struct ext4_iloc iloc;
6096 err = ext4_get_inode_loc(inode, &iloc);
6098 BUFFER_TRACE(iloc.bh, "get_write_access");
6099 err = jbd2_journal_get_write_access(handle, iloc.bh);
6101 err = ext4_handle_dirty_metadata(handle,
6107 ext4_std_error(inode->i_sb, err);
6112 int ext4_change_inode_journal_flag(struct inode *inode, int val)
6117 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6120 * We have to be very careful here: changing a data block's
6121 * journaling status dynamically is dangerous. If we write a
6122 * data block to the journal, change the status and then delete
6123 * that block, we risk forgetting to revoke the old log record
6124 * from the journal and so a subsequent replay can corrupt data.
6125 * So, first we make sure that the journal is empty and that
6126 * nobody is changing anything.
6129 journal = EXT4_JOURNAL(inode);
6132 if (is_journal_aborted(journal))
6135 /* Wait for all existing dio workers */
6136 ext4_inode_block_unlocked_dio(inode);
6137 inode_dio_wait(inode);
6140 * Before flushing the journal and switching inode's aops, we have
6141 * to flush all dirty data the inode has. There can be outstanding
6142 * delayed allocations, there can be unwritten extents created by
6143 * fallocate or buffered writes in dioread_nolock mode covered by
6144 * dirty data which can be converted only after flushing the dirty
6145 * data (and journalled aops don't know how to handle these cases).
6148 down_write(&EXT4_I(inode)->i_mmap_sem);
6149 err = filemap_write_and_wait(inode->i_mapping);
6151 up_write(&EXT4_I(inode)->i_mmap_sem);
6152 ext4_inode_resume_unlocked_dio(inode);
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);
6173 jbd2_journal_unlock_updates(journal);
6174 percpu_up_write(&sbi->s_writepages_rwsem);
6175 ext4_inode_resume_unlocked_dio(inode);
6178 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6180 ext4_set_aops(inode);
6182 jbd2_journal_unlock_updates(journal);
6183 percpu_up_write(&sbi->s_writepages_rwsem);
6186 up_write(&EXT4_I(inode)->i_mmap_sem);
6187 ext4_inode_resume_unlocked_dio(inode);
6189 /* Finally we can mark the inode as dirty. */
6191 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6193 return PTR_ERR(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 buffer_head *bh)
6205 return !buffer_mapped(bh);
6208 int ext4_page_mkwrite(struct vm_fault *vmf)
6210 struct vm_area_struct *vma = vmf->vma;
6211 struct page *page = vmf->page;
6215 struct file *file = vma->vm_file;
6216 struct inode *inode = file_inode(file);
6217 struct address_space *mapping = inode->i_mapping;
6219 get_block_t *get_block;
6222 if (unlikely(IS_IMMUTABLE(inode)))
6223 return VM_FAULT_SIGBUS;
6225 sb_start_pagefault(inode->i_sb);
6226 file_update_time(vma->vm_file);
6228 down_read(&EXT4_I(inode)->i_mmap_sem);
6230 ret = ext4_convert_inline_data(inode);
6234 /* Delalloc case is easy... */
6235 if (test_opt(inode->i_sb, DELALLOC) &&
6236 !ext4_should_journal_data(inode) &&
6237 !ext4_nonda_switch(inode->i_sb)) {
6239 ret = block_page_mkwrite(vma, vmf,
6240 ext4_da_get_block_prep);
6241 } while (ret == -ENOSPC &&
6242 ext4_should_retry_alloc(inode->i_sb, &retries));
6247 size = i_size_read(inode);
6248 /* Page got truncated from under us? */
6249 if (page->mapping != mapping || page_offset(page) > size) {
6251 ret = VM_FAULT_NOPAGE;
6255 if (page->index == size >> PAGE_SHIFT)
6256 len = size & ~PAGE_MASK;
6260 * Return if we have all the buffers mapped. This avoids the need to do
6261 * journal_start/journal_stop which can block and take a long time
6263 if (page_has_buffers(page)) {
6264 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6266 ext4_bh_unmapped)) {
6267 /* Wait so that we don't change page under IO */
6268 wait_for_stable_page(page);
6269 ret = VM_FAULT_LOCKED;
6274 /* OK, we need to fill the hole... */
6275 if (ext4_should_dioread_nolock(inode))
6276 get_block = ext4_get_block_unwritten;
6278 get_block = ext4_get_block;
6280 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6281 ext4_writepage_trans_blocks(inode));
6282 if (IS_ERR(handle)) {
6283 ret = VM_FAULT_SIGBUS;
6286 ret = block_page_mkwrite(vma, vmf, get_block);
6287 if (!ret && ext4_should_journal_data(inode)) {
6288 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6289 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6291 ret = VM_FAULT_SIGBUS;
6292 ext4_journal_stop(handle);
6295 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6297 ext4_journal_stop(handle);
6298 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6301 ret = block_page_mkwrite_return(ret);
6303 up_read(&EXT4_I(inode)->i_mmap_sem);
6304 sb_end_pagefault(inode->i_sb);
6308 int ext4_filemap_fault(struct vm_fault *vmf)
6310 struct inode *inode = file_inode(vmf->vma->vm_file);
6313 down_read(&EXT4_I(inode)->i_mmap_sem);
6314 err = filemap_fault(vmf);
6315 up_read(&EXT4_I(inode)->i_mmap_sem);
6321 * Find the first extent at or after @lblk in an inode that is not a hole.
6322 * Search for @map_len blocks at most. The extent is returned in @result.
6324 * The function returns 1 if we found an extent. The function returns 0 in
6325 * case there is no extent at or after @lblk and in that case also sets
6326 * @result->es_len to 0. In case of error, the error code is returned.
6328 int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
6329 unsigned int map_len, struct extent_status *result)
6331 struct ext4_map_blocks map;
6332 struct extent_status es = {};
6336 map.m_len = map_len;
6339 * For non-extent based files this loop may iterate several times since
6340 * we do not determine full hole size.
6342 while (map.m_len > 0) {
6343 ret = ext4_map_blocks(NULL, inode, &map, 0);
6346 /* There's extent covering m_lblk? Just return it. */
6350 ext4_es_store_pblock(result, map.m_pblk);
6351 result->es_lblk = map.m_lblk;
6352 result->es_len = map.m_len;
6353 if (map.m_flags & EXT4_MAP_UNWRITTEN)
6354 status = EXTENT_STATUS_UNWRITTEN;
6356 status = EXTENT_STATUS_WRITTEN;
6357 ext4_es_store_status(result, status);
6360 ext4_es_find_delayed_extent_range(inode, map.m_lblk,
6361 map.m_lblk + map.m_len - 1,
6363 /* Is delalloc data before next block in extent tree? */
6364 if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
6365 ext4_lblk_t offset = 0;
6367 if (es.es_lblk < lblk)
6368 offset = lblk - es.es_lblk;
6369 result->es_lblk = es.es_lblk + offset;
6370 ext4_es_store_pblock(result,
6371 ext4_es_pblock(&es) + offset);
6372 result->es_len = es.es_len - offset;
6373 ext4_es_store_status(result, ext4_es_status(&es));
6377 /* There's a hole at m_lblk, advance us after it */
6378 map.m_lblk += map.m_len;
6379 map_len -= map.m_len;
6380 map.m_len = map_len;