2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
51 struct ext4_inode_info
*ei
)
53 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
59 raw
->i_checksum_lo
= 0;
60 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
61 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
62 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
63 raw
->i_checksum_hi
= 0;
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
67 EXT4_INODE_SIZE(inode
->i_sb
));
69 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
70 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
71 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
72 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
77 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
78 struct ext4_inode_info
*ei
)
80 __u32 provided
, calculated
;
82 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
83 cpu_to_le32(EXT4_OS_LINUX
) ||
84 !ext4_has_metadata_csum(inode
->i_sb
))
87 provided
= le16_to_cpu(raw
->i_checksum_lo
);
88 calculated
= ext4_inode_csum(inode
, raw
, ei
);
89 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
90 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
91 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
95 return provided
== calculated
;
98 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
99 struct ext4_inode_info
*ei
)
103 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
104 cpu_to_le32(EXT4_OS_LINUX
) ||
105 !ext4_has_metadata_csum(inode
->i_sb
))
108 csum
= ext4_inode_csum(inode
, raw
, ei
);
109 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
110 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
111 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
112 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
118 trace_ext4_begin_ordered_truncate(inode
, new_size
);
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
125 if (!EXT4_I(inode
)->jinode
)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
128 EXT4_I(inode
)->jinode
,
132 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
133 unsigned int length
);
134 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
135 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
136 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode
*inode
)
144 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
145 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
147 if (ext4_has_inline_data(inode
))
150 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
170 jbd_debug(2, "restarting handle %p\n", handle
);
171 up_write(&EXT4_I(inode
)->i_data_sem
);
172 ret
= ext4_journal_restart(handle
, nblocks
);
173 down_write(&EXT4_I(inode
)->i_data_sem
);
174 ext4_discard_preallocations(inode
);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode
*inode
)
187 trace_ext4_evict_inode(inode
);
189 if (inode
->i_nlink
) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode
) &&
209 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
210 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
211 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
212 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
214 jbd2_complete_transaction(journal
, commit_tid
);
215 filemap_write_and_wait(&inode
->i_data
);
217 truncate_inode_pages_final(&inode
->i_data
);
222 if (is_bad_inode(inode
))
224 dquot_initialize(inode
);
226 if (ext4_should_order_data(inode
))
227 ext4_begin_ordered_truncate(inode
, 0);
228 truncate_inode_pages_final(&inode
->i_data
);
231 * Protect us against freezing - iput() caller didn't have to have any
232 * protection against it
234 sb_start_intwrite(inode
->i_sb
);
235 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
236 ext4_blocks_for_truncate(inode
)+3);
237 if (IS_ERR(handle
)) {
238 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
240 * If we're going to skip the normal cleanup, we still need to
241 * make sure that the in-core orphan linked list is properly
244 ext4_orphan_del(NULL
, inode
);
245 sb_end_intwrite(inode
->i_sb
);
250 ext4_handle_sync(handle
);
252 err
= ext4_mark_inode_dirty(handle
, inode
);
254 ext4_warning(inode
->i_sb
,
255 "couldn't mark inode dirty (err %d)", err
);
259 ext4_truncate(inode
);
262 * ext4_ext_truncate() doesn't reserve any slop when it
263 * restarts journal transactions; therefore there may not be
264 * enough credits left in the handle to remove the inode from
265 * the orphan list and set the dtime field.
267 if (!ext4_handle_has_enough_credits(handle
, 3)) {
268 err
= ext4_journal_extend(handle
, 3);
270 err
= ext4_journal_restart(handle
, 3);
272 ext4_warning(inode
->i_sb
,
273 "couldn't extend journal (err %d)", err
);
275 ext4_journal_stop(handle
);
276 ext4_orphan_del(NULL
, inode
);
277 sb_end_intwrite(inode
->i_sb
);
283 * Kill off the orphan record which ext4_truncate created.
284 * AKPM: I think this can be inside the above `if'.
285 * Note that ext4_orphan_del() has to be able to cope with the
286 * deletion of a non-existent orphan - this is because we don't
287 * know if ext4_truncate() actually created an orphan record.
288 * (Well, we could do this if we need to, but heck - it works)
290 ext4_orphan_del(handle
, inode
);
291 EXT4_I(inode
)->i_dtime
= get_seconds();
294 * One subtle ordering requirement: if anything has gone wrong
295 * (transaction abort, IO errors, whatever), then we can still
296 * do these next steps (the fs will already have been marked as
297 * having errors), but we can't free the inode if the mark_dirty
300 if (ext4_mark_inode_dirty(handle
, inode
))
301 /* If that failed, just do the required in-core inode clear. */
302 ext4_clear_inode(inode
);
304 ext4_free_inode(handle
, inode
);
305 ext4_journal_stop(handle
);
306 sb_end_intwrite(inode
->i_sb
);
309 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
313 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
315 return &EXT4_I(inode
)->i_reserved_quota
;
320 * Called with i_data_sem down, which is important since we can call
321 * ext4_discard_preallocations() from here.
323 void ext4_da_update_reserve_space(struct inode
*inode
,
324 int used
, int quota_claim
)
326 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
327 struct ext4_inode_info
*ei
= EXT4_I(inode
);
329 spin_lock(&ei
->i_block_reservation_lock
);
330 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
331 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
332 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
333 "with only %d reserved data blocks",
334 __func__
, inode
->i_ino
, used
,
335 ei
->i_reserved_data_blocks
);
337 used
= ei
->i_reserved_data_blocks
;
340 /* Update per-inode reservations */
341 ei
->i_reserved_data_blocks
-= used
;
342 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
344 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
346 /* Update quota subsystem for data blocks */
348 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
351 * We did fallocate with an offset that is already delayed
352 * allocated. So on delayed allocated writeback we should
353 * not re-claim the quota for fallocated blocks.
355 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
359 * If we have done all the pending block allocations and if
360 * there aren't any writers on the inode, we can discard the
361 * inode's preallocations.
363 if ((ei
->i_reserved_data_blocks
== 0) &&
364 (atomic_read(&inode
->i_writecount
) == 0))
365 ext4_discard_preallocations(inode
);
368 static int __check_block_validity(struct inode
*inode
, const char *func
,
370 struct ext4_map_blocks
*map
)
372 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
374 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
375 "lblock %lu mapped to illegal pblock "
376 "(length %d)", (unsigned long) map
->m_lblk
,
378 return -EFSCORRUPTED
;
383 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
388 if (ext4_encrypted_inode(inode
))
389 return ext4_encrypted_zeroout(inode
, lblk
, pblk
, len
);
391 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
398 #define check_block_validity(inode, map) \
399 __check_block_validity((inode), __func__, __LINE__, (map))
401 #ifdef ES_AGGRESSIVE_TEST
402 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
404 struct ext4_map_blocks
*es_map
,
405 struct ext4_map_blocks
*map
,
412 * There is a race window that the result is not the same.
413 * e.g. xfstests #223 when dioread_nolock enables. The reason
414 * is that we lookup a block mapping in extent status tree with
415 * out taking i_data_sem. So at the time the unwritten extent
416 * could be converted.
418 down_read(&EXT4_I(inode
)->i_data_sem
);
419 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
420 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
421 EXT4_GET_BLOCKS_KEEP_SIZE
);
423 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
424 EXT4_GET_BLOCKS_KEEP_SIZE
);
426 up_read((&EXT4_I(inode
)->i_data_sem
));
429 * We don't check m_len because extent will be collpased in status
430 * tree. So the m_len might not equal.
432 if (es_map
->m_lblk
!= map
->m_lblk
||
433 es_map
->m_flags
!= map
->m_flags
||
434 es_map
->m_pblk
!= map
->m_pblk
) {
435 printk("ES cache assertion failed for inode: %lu "
436 "es_cached ex [%d/%d/%llu/%x] != "
437 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
438 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
439 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
440 map
->m_len
, map
->m_pblk
, map
->m_flags
,
444 #endif /* ES_AGGRESSIVE_TEST */
447 * The ext4_map_blocks() function tries to look up the requested blocks,
448 * and returns if the blocks are already mapped.
450 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
451 * and store the allocated blocks in the result buffer head and mark it
454 * If file type is extents based, it will call ext4_ext_map_blocks(),
455 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
458 * On success, it returns the number of blocks being mapped or allocated. if
459 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
460 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
462 * It returns 0 if plain look up failed (blocks have not been allocated), in
463 * that case, @map is returned as unmapped but we still do fill map->m_len to
464 * indicate the length of a hole starting at map->m_lblk.
466 * It returns the error in case of allocation failure.
468 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
469 struct ext4_map_blocks
*map
, int flags
)
471 struct extent_status es
;
474 #ifdef ES_AGGRESSIVE_TEST
475 struct ext4_map_blocks orig_map
;
477 memcpy(&orig_map
, map
, sizeof(*map
));
481 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
482 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
483 (unsigned long) map
->m_lblk
);
486 * ext4_map_blocks returns an int, and m_len is an unsigned int
488 if (unlikely(map
->m_len
> INT_MAX
))
489 map
->m_len
= INT_MAX
;
491 /* We can handle the block number less than EXT_MAX_BLOCKS */
492 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
493 return -EFSCORRUPTED
;
495 /* Lookup extent status tree firstly */
496 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
497 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
498 map
->m_pblk
= ext4_es_pblock(&es
) +
499 map
->m_lblk
- es
.es_lblk
;
500 map
->m_flags
|= ext4_es_is_written(&es
) ?
501 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
502 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
503 if (retval
> map
->m_len
)
506 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
508 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
509 if (retval
> map
->m_len
)
516 #ifdef ES_AGGRESSIVE_TEST
517 ext4_map_blocks_es_recheck(handle
, inode
, map
,
524 * Try to see if we can get the block without requesting a new
527 down_read(&EXT4_I(inode
)->i_data_sem
);
528 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
529 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
530 EXT4_GET_BLOCKS_KEEP_SIZE
);
532 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
533 EXT4_GET_BLOCKS_KEEP_SIZE
);
538 if (unlikely(retval
!= map
->m_len
)) {
539 ext4_warning(inode
->i_sb
,
540 "ES len assertion failed for inode "
541 "%lu: retval %d != map->m_len %d",
542 inode
->i_ino
, retval
, map
->m_len
);
546 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
547 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
548 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
549 !(status
& EXTENT_STATUS_WRITTEN
) &&
550 ext4_find_delalloc_range(inode
, map
->m_lblk
,
551 map
->m_lblk
+ map
->m_len
- 1))
552 status
|= EXTENT_STATUS_DELAYED
;
553 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
554 map
->m_len
, map
->m_pblk
, status
);
558 up_read((&EXT4_I(inode
)->i_data_sem
));
561 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
562 ret
= check_block_validity(inode
, map
);
567 /* If it is only a block(s) look up */
568 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
572 * Returns if the blocks have already allocated
574 * Note that if blocks have been preallocated
575 * ext4_ext_get_block() returns the create = 0
576 * with buffer head unmapped.
578 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
580 * If we need to convert extent to unwritten
581 * we continue and do the actual work in
582 * ext4_ext_map_blocks()
584 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
588 * Here we clear m_flags because after allocating an new extent,
589 * it will be set again.
591 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
594 * New blocks allocate and/or writing to unwritten extent
595 * will possibly result in updating i_data, so we take
596 * the write lock of i_data_sem, and call get_block()
597 * with create == 1 flag.
599 down_write(&EXT4_I(inode
)->i_data_sem
);
602 * We need to check for EXT4 here because migrate
603 * could have changed the inode type in between
605 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
606 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
608 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
610 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
612 * We allocated new blocks which will result in
613 * i_data's format changing. Force the migrate
614 * to fail by clearing migrate flags
616 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
620 * Update reserved blocks/metadata blocks after successful
621 * block allocation which had been deferred till now. We don't
622 * support fallocate for non extent files. So we can update
623 * reserve space here.
626 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
627 ext4_da_update_reserve_space(inode
, retval
, 1);
633 if (unlikely(retval
!= map
->m_len
)) {
634 ext4_warning(inode
->i_sb
,
635 "ES len assertion failed for inode "
636 "%lu: retval %d != map->m_len %d",
637 inode
->i_ino
, retval
, map
->m_len
);
642 * We have to zeroout blocks before inserting them into extent
643 * status tree. Otherwise someone could look them up there and
644 * use them before they are really zeroed.
646 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
647 map
->m_flags
& EXT4_MAP_MAPPED
&&
648 map
->m_flags
& EXT4_MAP_NEW
) {
649 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
650 map
->m_pblk
, map
->m_len
);
658 * If the extent has been zeroed out, we don't need to update
659 * extent status tree.
661 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
662 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
663 if (ext4_es_is_written(&es
))
666 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
667 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
668 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
669 !(status
& EXTENT_STATUS_WRITTEN
) &&
670 ext4_find_delalloc_range(inode
, map
->m_lblk
,
671 map
->m_lblk
+ map
->m_len
- 1))
672 status
|= EXTENT_STATUS_DELAYED
;
673 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
674 map
->m_pblk
, status
);
682 up_write((&EXT4_I(inode
)->i_data_sem
));
683 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
684 ret
= check_block_validity(inode
, map
);
689 * Inodes with freshly allocated blocks where contents will be
690 * visible after transaction commit must be on transaction's
693 if (map
->m_flags
& EXT4_MAP_NEW
&&
694 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
695 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
696 !IS_NOQUOTA(inode
) &&
697 ext4_should_order_data(inode
)) {
698 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
699 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
701 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
710 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
711 * we have to be careful as someone else may be manipulating b_state as well.
713 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
715 unsigned long old_state
;
716 unsigned long new_state
;
718 flags
&= EXT4_MAP_FLAGS
;
720 /* Dummy buffer_head? Set non-atomically. */
722 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
726 * Someone else may be modifying b_state. Be careful! This is ugly but
727 * once we get rid of using bh as a container for mapping information
728 * to pass to / from get_block functions, this can go away.
731 old_state
= READ_ONCE(bh
->b_state
);
732 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
734 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
737 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
738 struct buffer_head
*bh
, int flags
)
740 struct ext4_map_blocks map
;
743 if (ext4_has_inline_data(inode
))
747 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
749 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
752 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
753 ext4_update_bh_state(bh
, map
.m_flags
);
754 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
760 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
761 struct buffer_head
*bh
, int create
)
763 return _ext4_get_block(inode
, iblock
, bh
,
764 create
? EXT4_GET_BLOCKS_CREATE
: 0);
768 * Get block function used when preparing for buffered write if we require
769 * creating an unwritten extent if blocks haven't been allocated. The extent
770 * will be converted to written after the IO is complete.
772 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
773 struct buffer_head
*bh_result
, int create
)
775 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
776 inode
->i_ino
, create
);
777 return _ext4_get_block(inode
, iblock
, bh_result
,
778 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
781 /* Maximum number of blocks we map for direct IO at once. */
782 #define DIO_MAX_BLOCKS 4096
785 * Get blocks function for the cases that need to start a transaction -
786 * generally difference cases of direct IO and DAX IO. It also handles retries
789 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
790 struct buffer_head
*bh_result
, int flags
)
797 /* Trim mapping request to maximum we can map at once for DIO */
798 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
799 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
800 dio_credits
= ext4_chunk_trans_blocks(inode
,
801 bh_result
->b_size
>> inode
->i_blkbits
);
803 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
805 return PTR_ERR(handle
);
807 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
808 ext4_journal_stop(handle
);
810 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
815 /* Get block function for DIO reads and writes to inodes without extents */
816 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
817 struct buffer_head
*bh
, int create
)
819 /* We don't expect handle for direct IO */
820 WARN_ON_ONCE(ext4_journal_current_handle());
823 return _ext4_get_block(inode
, iblock
, bh
, 0);
824 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
828 * Get block function for AIO DIO writes when we create unwritten extent if
829 * blocks are not allocated yet. The extent will be converted to written
830 * after IO is complete.
832 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
833 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
837 /* We don't expect handle for direct IO */
838 WARN_ON_ONCE(ext4_journal_current_handle());
840 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
841 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
844 * When doing DIO using unwritten extents, we need io_end to convert
845 * unwritten extents to written on IO completion. We allocate io_end
846 * once we spot unwritten extent and store it in b_private. Generic
847 * DIO code keeps b_private set and furthermore passes the value to
848 * our completion callback in 'private' argument.
850 if (!ret
&& buffer_unwritten(bh_result
)) {
851 if (!bh_result
->b_private
) {
852 ext4_io_end_t
*io_end
;
854 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
857 bh_result
->b_private
= io_end
;
858 ext4_set_io_unwritten_flag(inode
, io_end
);
860 set_buffer_defer_completion(bh_result
);
867 * Get block function for non-AIO DIO writes when we create unwritten extent if
868 * blocks are not allocated yet. The extent will be converted to written
869 * after IO is complete from ext4_ext_direct_IO() function.
871 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
872 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
876 /* We don't expect handle for direct IO */
877 WARN_ON_ONCE(ext4_journal_current_handle());
879 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
880 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
883 * Mark inode as having pending DIO writes to unwritten extents.
884 * ext4_ext_direct_IO() checks this flag and converts extents to
887 if (!ret
&& buffer_unwritten(bh_result
))
888 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
893 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
894 struct buffer_head
*bh_result
, int create
)
898 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
899 inode
->i_ino
, create
);
900 /* We don't expect handle for direct IO */
901 WARN_ON_ONCE(ext4_journal_current_handle());
903 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
905 * Blocks should have been preallocated! ext4_file_write_iter() checks
908 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
915 * `handle' can be NULL if create is zero
917 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
918 ext4_lblk_t block
, int map_flags
)
920 struct ext4_map_blocks map
;
921 struct buffer_head
*bh
;
922 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
925 J_ASSERT(handle
!= NULL
|| create
== 0);
929 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
932 return create
? ERR_PTR(-ENOSPC
) : NULL
;
936 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
938 return ERR_PTR(-ENOMEM
);
939 if (map
.m_flags
& EXT4_MAP_NEW
) {
940 J_ASSERT(create
!= 0);
941 J_ASSERT(handle
!= NULL
);
944 * Now that we do not always journal data, we should
945 * keep in mind whether this should always journal the
946 * new buffer as metadata. For now, regular file
947 * writes use ext4_get_block instead, so it's not a
951 BUFFER_TRACE(bh
, "call get_create_access");
952 err
= ext4_journal_get_create_access(handle
, bh
);
957 if (!buffer_uptodate(bh
)) {
958 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
959 set_buffer_uptodate(bh
);
962 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
963 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
967 BUFFER_TRACE(bh
, "not a new buffer");
974 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
975 ext4_lblk_t block
, int map_flags
)
977 struct buffer_head
*bh
;
979 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
982 if (!bh
|| buffer_uptodate(bh
))
984 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
986 if (buffer_uptodate(bh
))
989 return ERR_PTR(-EIO
);
992 int ext4_walk_page_buffers(handle_t
*handle
,
993 struct buffer_head
*head
,
997 int (*fn
)(handle_t
*handle
,
998 struct buffer_head
*bh
))
1000 struct buffer_head
*bh
;
1001 unsigned block_start
, block_end
;
1002 unsigned blocksize
= head
->b_size
;
1004 struct buffer_head
*next
;
1006 for (bh
= head
, block_start
= 0;
1007 ret
== 0 && (bh
!= head
|| !block_start
);
1008 block_start
= block_end
, bh
= next
) {
1009 next
= bh
->b_this_page
;
1010 block_end
= block_start
+ blocksize
;
1011 if (block_end
<= from
|| block_start
>= to
) {
1012 if (partial
&& !buffer_uptodate(bh
))
1016 err
= (*fn
)(handle
, bh
);
1024 * To preserve ordering, it is essential that the hole instantiation and
1025 * the data write be encapsulated in a single transaction. We cannot
1026 * close off a transaction and start a new one between the ext4_get_block()
1027 * and the commit_write(). So doing the jbd2_journal_start at the start of
1028 * prepare_write() is the right place.
1030 * Also, this function can nest inside ext4_writepage(). In that case, we
1031 * *know* that ext4_writepage() has generated enough buffer credits to do the
1032 * whole page. So we won't block on the journal in that case, which is good,
1033 * because the caller may be PF_MEMALLOC.
1035 * By accident, ext4 can be reentered when a transaction is open via
1036 * quota file writes. If we were to commit the transaction while thus
1037 * reentered, there can be a deadlock - we would be holding a quota
1038 * lock, and the commit would never complete if another thread had a
1039 * transaction open and was blocking on the quota lock - a ranking
1042 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1043 * will _not_ run commit under these circumstances because handle->h_ref
1044 * is elevated. We'll still have enough credits for the tiny quotafile
1047 int do_journal_get_write_access(handle_t
*handle
,
1048 struct buffer_head
*bh
)
1050 int dirty
= buffer_dirty(bh
);
1053 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1056 * __block_write_begin() could have dirtied some buffers. Clean
1057 * the dirty bit as jbd2_journal_get_write_access() could complain
1058 * otherwise about fs integrity issues. Setting of the dirty bit
1059 * by __block_write_begin() isn't a real problem here as we clear
1060 * the bit before releasing a page lock and thus writeback cannot
1061 * ever write the buffer.
1064 clear_buffer_dirty(bh
);
1065 BUFFER_TRACE(bh
, "get write access");
1066 ret
= ext4_journal_get_write_access(handle
, bh
);
1068 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1072 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1073 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1074 get_block_t
*get_block
)
1076 unsigned from
= pos
& (PAGE_SIZE
- 1);
1077 unsigned to
= from
+ len
;
1078 struct inode
*inode
= page
->mapping
->host
;
1079 unsigned block_start
, block_end
;
1082 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1084 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1085 bool decrypt
= false;
1087 BUG_ON(!PageLocked(page
));
1088 BUG_ON(from
> PAGE_SIZE
);
1089 BUG_ON(to
> PAGE_SIZE
);
1092 if (!page_has_buffers(page
))
1093 create_empty_buffers(page
, blocksize
, 0);
1094 head
= page_buffers(page
);
1095 bbits
= ilog2(blocksize
);
1096 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1098 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1099 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1100 block_end
= block_start
+ blocksize
;
1101 if (block_end
<= from
|| block_start
>= to
) {
1102 if (PageUptodate(page
)) {
1103 if (!buffer_uptodate(bh
))
1104 set_buffer_uptodate(bh
);
1109 clear_buffer_new(bh
);
1110 if (!buffer_mapped(bh
)) {
1111 WARN_ON(bh
->b_size
!= blocksize
);
1112 err
= get_block(inode
, block
, bh
, 1);
1115 if (buffer_new(bh
)) {
1116 unmap_underlying_metadata(bh
->b_bdev
,
1118 if (PageUptodate(page
)) {
1119 clear_buffer_new(bh
);
1120 set_buffer_uptodate(bh
);
1121 mark_buffer_dirty(bh
);
1124 if (block_end
> to
|| block_start
< from
)
1125 zero_user_segments(page
, to
, block_end
,
1130 if (PageUptodate(page
)) {
1131 if (!buffer_uptodate(bh
))
1132 set_buffer_uptodate(bh
);
1135 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1136 !buffer_unwritten(bh
) &&
1137 (block_start
< from
|| block_end
> to
)) {
1138 ll_rw_block(READ
, 1, &bh
);
1140 decrypt
= ext4_encrypted_inode(inode
) &&
1141 S_ISREG(inode
->i_mode
);
1145 * If we issued read requests, let them complete.
1147 while (wait_bh
> wait
) {
1148 wait_on_buffer(*--wait_bh
);
1149 if (!buffer_uptodate(*wait_bh
))
1153 page_zero_new_buffers(page
, from
, to
);
1155 err
= ext4_decrypt(page
);
1160 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1161 loff_t pos
, unsigned len
, unsigned flags
,
1162 struct page
**pagep
, void **fsdata
)
1164 struct inode
*inode
= mapping
->host
;
1165 int ret
, needed_blocks
;
1172 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1174 * Reserve one block more for addition to orphan list in case
1175 * we allocate blocks but write fails for some reason
1177 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1178 index
= pos
>> PAGE_SHIFT
;
1179 from
= pos
& (PAGE_SIZE
- 1);
1182 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1183 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1192 * grab_cache_page_write_begin() can take a long time if the
1193 * system is thrashing due to memory pressure, or if the page
1194 * is being written back. So grab it first before we start
1195 * the transaction handle. This also allows us to allocate
1196 * the page (if needed) without using GFP_NOFS.
1199 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1205 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1206 if (IS_ERR(handle
)) {
1208 return PTR_ERR(handle
);
1212 if (page
->mapping
!= mapping
) {
1213 /* The page got truncated from under us */
1216 ext4_journal_stop(handle
);
1219 /* In case writeback began while the page was unlocked */
1220 wait_for_stable_page(page
);
1222 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1223 if (ext4_should_dioread_nolock(inode
))
1224 ret
= ext4_block_write_begin(page
, pos
, len
,
1225 ext4_get_block_unwritten
);
1227 ret
= ext4_block_write_begin(page
, pos
, len
,
1230 if (ext4_should_dioread_nolock(inode
))
1231 ret
= __block_write_begin(page
, pos
, len
,
1232 ext4_get_block_unwritten
);
1234 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1236 if (!ret
&& ext4_should_journal_data(inode
)) {
1237 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1239 do_journal_get_write_access
);
1245 * __block_write_begin may have instantiated a few blocks
1246 * outside i_size. Trim these off again. Don't need
1247 * i_size_read because we hold i_mutex.
1249 * Add inode to orphan list in case we crash before
1252 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1253 ext4_orphan_add(handle
, inode
);
1255 ext4_journal_stop(handle
);
1256 if (pos
+ len
> inode
->i_size
) {
1257 ext4_truncate_failed_write(inode
);
1259 * If truncate failed early the inode might
1260 * still be on the orphan list; we need to
1261 * make sure the inode is removed from the
1262 * orphan list in that case.
1265 ext4_orphan_del(NULL
, inode
);
1268 if (ret
== -ENOSPC
&&
1269 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1278 /* For write_end() in data=journal mode */
1279 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1282 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1284 set_buffer_uptodate(bh
);
1285 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1286 clear_buffer_meta(bh
);
1287 clear_buffer_prio(bh
);
1292 * We need to pick up the new inode size which generic_commit_write gave us
1293 * `file' can be NULL - eg, when called from page_symlink().
1295 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1296 * buffers are managed internally.
1298 static int ext4_write_end(struct file
*file
,
1299 struct address_space
*mapping
,
1300 loff_t pos
, unsigned len
, unsigned copied
,
1301 struct page
*page
, void *fsdata
)
1303 handle_t
*handle
= ext4_journal_current_handle();
1304 struct inode
*inode
= mapping
->host
;
1305 loff_t old_size
= inode
->i_size
;
1307 int i_size_changed
= 0;
1309 trace_ext4_write_end(inode
, pos
, len
, copied
);
1310 if (ext4_has_inline_data(inode
)) {
1311 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1317 copied
= block_write_end(file
, mapping
, pos
,
1318 len
, copied
, page
, fsdata
);
1320 * it's important to update i_size while still holding page lock:
1321 * page writeout could otherwise come in and zero beyond i_size.
1323 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1328 pagecache_isize_extended(inode
, old_size
, pos
);
1330 * Don't mark the inode dirty under page lock. First, it unnecessarily
1331 * makes the holding time of page lock longer. Second, it forces lock
1332 * ordering of page lock and transaction start for journaling
1336 ext4_mark_inode_dirty(handle
, inode
);
1338 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1339 /* if we have allocated more blocks and copied
1340 * less. We will have blocks allocated outside
1341 * inode->i_size. So truncate them
1343 ext4_orphan_add(handle
, inode
);
1345 ret2
= ext4_journal_stop(handle
);
1349 if (pos
+ len
> inode
->i_size
) {
1350 ext4_truncate_failed_write(inode
);
1352 * If truncate failed early the inode might still be
1353 * on the orphan list; we need to make sure the inode
1354 * is removed from the orphan list in that case.
1357 ext4_orphan_del(NULL
, inode
);
1360 return ret
? ret
: copied
;
1364 * This is a private version of page_zero_new_buffers() which doesn't
1365 * set the buffer to be dirty, since in data=journalled mode we need
1366 * to call ext4_handle_dirty_metadata() instead.
1368 static void zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1370 unsigned int block_start
= 0, block_end
;
1371 struct buffer_head
*head
, *bh
;
1373 bh
= head
= page_buffers(page
);
1375 block_end
= block_start
+ bh
->b_size
;
1376 if (buffer_new(bh
)) {
1377 if (block_end
> from
&& block_start
< to
) {
1378 if (!PageUptodate(page
)) {
1379 unsigned start
, size
;
1381 start
= max(from
, block_start
);
1382 size
= min(to
, block_end
) - start
;
1384 zero_user(page
, start
, size
);
1385 set_buffer_uptodate(bh
);
1387 clear_buffer_new(bh
);
1390 block_start
= block_end
;
1391 bh
= bh
->b_this_page
;
1392 } while (bh
!= head
);
1395 static int ext4_journalled_write_end(struct file
*file
,
1396 struct address_space
*mapping
,
1397 loff_t pos
, unsigned len
, unsigned copied
,
1398 struct page
*page
, void *fsdata
)
1400 handle_t
*handle
= ext4_journal_current_handle();
1401 struct inode
*inode
= mapping
->host
;
1402 loff_t old_size
= inode
->i_size
;
1406 int size_changed
= 0;
1408 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1409 from
= pos
& (PAGE_SIZE
- 1);
1412 BUG_ON(!ext4_handle_valid(handle
));
1414 if (ext4_has_inline_data(inode
))
1415 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1419 if (!PageUptodate(page
))
1421 zero_new_buffers(page
, from
+copied
, to
);
1424 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1425 to
, &partial
, write_end_fn
);
1427 SetPageUptodate(page
);
1429 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1430 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1431 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1436 pagecache_isize_extended(inode
, old_size
, pos
);
1439 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1444 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1445 /* if we have allocated more blocks and copied
1446 * less. We will have blocks allocated outside
1447 * inode->i_size. So truncate them
1449 ext4_orphan_add(handle
, inode
);
1451 ret2
= ext4_journal_stop(handle
);
1454 if (pos
+ len
> inode
->i_size
) {
1455 ext4_truncate_failed_write(inode
);
1457 * If truncate failed early the inode might still be
1458 * on the orphan list; we need to make sure the inode
1459 * is removed from the orphan list in that case.
1462 ext4_orphan_del(NULL
, inode
);
1465 return ret
? ret
: copied
;
1469 * Reserve space for a single cluster
1471 static int ext4_da_reserve_space(struct inode
*inode
)
1473 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1474 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1478 * We will charge metadata quota at writeout time; this saves
1479 * us from metadata over-estimation, though we may go over by
1480 * a small amount in the end. Here we just reserve for data.
1482 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1486 spin_lock(&ei
->i_block_reservation_lock
);
1487 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1488 spin_unlock(&ei
->i_block_reservation_lock
);
1489 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1492 ei
->i_reserved_data_blocks
++;
1493 trace_ext4_da_reserve_space(inode
);
1494 spin_unlock(&ei
->i_block_reservation_lock
);
1496 return 0; /* success */
1499 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1501 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1502 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1505 return; /* Nothing to release, exit */
1507 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1509 trace_ext4_da_release_space(inode
, to_free
);
1510 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1512 * if there aren't enough reserved blocks, then the
1513 * counter is messed up somewhere. Since this
1514 * function is called from invalidate page, it's
1515 * harmless to return without any action.
1517 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1518 "ino %lu, to_free %d with only %d reserved "
1519 "data blocks", inode
->i_ino
, to_free
,
1520 ei
->i_reserved_data_blocks
);
1522 to_free
= ei
->i_reserved_data_blocks
;
1524 ei
->i_reserved_data_blocks
-= to_free
;
1526 /* update fs dirty data blocks counter */
1527 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1529 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1531 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1534 static void ext4_da_page_release_reservation(struct page
*page
,
1535 unsigned int offset
,
1536 unsigned int length
)
1538 int to_release
= 0, contiguous_blks
= 0;
1539 struct buffer_head
*head
, *bh
;
1540 unsigned int curr_off
= 0;
1541 struct inode
*inode
= page
->mapping
->host
;
1542 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1543 unsigned int stop
= offset
+ length
;
1547 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1549 head
= page_buffers(page
);
1552 unsigned int next_off
= curr_off
+ bh
->b_size
;
1554 if (next_off
> stop
)
1557 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1560 clear_buffer_delay(bh
);
1561 } else if (contiguous_blks
) {
1562 lblk
= page
->index
<<
1563 (PAGE_SHIFT
- inode
->i_blkbits
);
1564 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1566 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1567 contiguous_blks
= 0;
1569 curr_off
= next_off
;
1570 } while ((bh
= bh
->b_this_page
) != head
);
1572 if (contiguous_blks
) {
1573 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1574 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1575 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1578 /* If we have released all the blocks belonging to a cluster, then we
1579 * need to release the reserved space for that cluster. */
1580 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1581 while (num_clusters
> 0) {
1582 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1583 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1584 if (sbi
->s_cluster_ratio
== 1 ||
1585 !ext4_find_delalloc_cluster(inode
, lblk
))
1586 ext4_da_release_space(inode
, 1);
1593 * Delayed allocation stuff
1596 struct mpage_da_data
{
1597 struct inode
*inode
;
1598 struct writeback_control
*wbc
;
1600 pgoff_t first_page
; /* The first page to write */
1601 pgoff_t next_page
; /* Current page to examine */
1602 pgoff_t last_page
; /* Last page to examine */
1604 * Extent to map - this can be after first_page because that can be
1605 * fully mapped. We somewhat abuse m_flags to store whether the extent
1606 * is delalloc or unwritten.
1608 struct ext4_map_blocks map
;
1609 struct ext4_io_submit io_submit
; /* IO submission data */
1612 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1617 struct pagevec pvec
;
1618 struct inode
*inode
= mpd
->inode
;
1619 struct address_space
*mapping
= inode
->i_mapping
;
1621 /* This is necessary when next_page == 0. */
1622 if (mpd
->first_page
>= mpd
->next_page
)
1625 index
= mpd
->first_page
;
1626 end
= mpd
->next_page
- 1;
1628 ext4_lblk_t start
, last
;
1629 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1630 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1631 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1634 pagevec_init(&pvec
, 0);
1635 while (index
<= end
) {
1636 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1639 for (i
= 0; i
< nr_pages
; i
++) {
1640 struct page
*page
= pvec
.pages
[i
];
1641 if (page
->index
> end
)
1643 BUG_ON(!PageLocked(page
));
1644 BUG_ON(PageWriteback(page
));
1646 block_invalidatepage(page
, 0, PAGE_SIZE
);
1647 ClearPageUptodate(page
);
1651 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1652 pagevec_release(&pvec
);
1656 static void ext4_print_free_blocks(struct inode
*inode
)
1658 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1659 struct super_block
*sb
= inode
->i_sb
;
1660 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1662 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1663 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1664 ext4_count_free_clusters(sb
)));
1665 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1666 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1667 (long long) EXT4_C2B(EXT4_SB(sb
),
1668 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1669 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1670 (long long) EXT4_C2B(EXT4_SB(sb
),
1671 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1672 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1673 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1674 ei
->i_reserved_data_blocks
);
1678 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1680 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1684 * This function is grabs code from the very beginning of
1685 * ext4_map_blocks, but assumes that the caller is from delayed write
1686 * time. This function looks up the requested blocks and sets the
1687 * buffer delay bit under the protection of i_data_sem.
1689 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1690 struct ext4_map_blocks
*map
,
1691 struct buffer_head
*bh
)
1693 struct extent_status es
;
1695 sector_t invalid_block
= ~((sector_t
) 0xffff);
1696 #ifdef ES_AGGRESSIVE_TEST
1697 struct ext4_map_blocks orig_map
;
1699 memcpy(&orig_map
, map
, sizeof(*map
));
1702 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1706 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1707 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1708 (unsigned long) map
->m_lblk
);
1710 /* Lookup extent status tree firstly */
1711 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1712 if (ext4_es_is_hole(&es
)) {
1714 down_read(&EXT4_I(inode
)->i_data_sem
);
1719 * Delayed extent could be allocated by fallocate.
1720 * So we need to check it.
1722 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1723 map_bh(bh
, inode
->i_sb
, invalid_block
);
1725 set_buffer_delay(bh
);
1729 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1730 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1731 if (retval
> map
->m_len
)
1732 retval
= map
->m_len
;
1733 map
->m_len
= retval
;
1734 if (ext4_es_is_written(&es
))
1735 map
->m_flags
|= EXT4_MAP_MAPPED
;
1736 else if (ext4_es_is_unwritten(&es
))
1737 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1741 #ifdef ES_AGGRESSIVE_TEST
1742 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1748 * Try to see if we can get the block without requesting a new
1749 * file system block.
1751 down_read(&EXT4_I(inode
)->i_data_sem
);
1752 if (ext4_has_inline_data(inode
))
1754 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1755 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1757 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1763 * XXX: __block_prepare_write() unmaps passed block,
1767 * If the block was allocated from previously allocated cluster,
1768 * then we don't need to reserve it again. However we still need
1769 * to reserve metadata for every block we're going to write.
1771 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1772 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1773 ret
= ext4_da_reserve_space(inode
);
1775 /* not enough space to reserve */
1781 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1782 ~0, EXTENT_STATUS_DELAYED
);
1788 map_bh(bh
, inode
->i_sb
, invalid_block
);
1790 set_buffer_delay(bh
);
1791 } else if (retval
> 0) {
1793 unsigned int status
;
1795 if (unlikely(retval
!= map
->m_len
)) {
1796 ext4_warning(inode
->i_sb
,
1797 "ES len assertion failed for inode "
1798 "%lu: retval %d != map->m_len %d",
1799 inode
->i_ino
, retval
, map
->m_len
);
1803 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1804 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1805 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1806 map
->m_pblk
, status
);
1812 up_read((&EXT4_I(inode
)->i_data_sem
));
1818 * This is a special get_block_t callback which is used by
1819 * ext4_da_write_begin(). It will either return mapped block or
1820 * reserve space for a single block.
1822 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1823 * We also have b_blocknr = -1 and b_bdev initialized properly
1825 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1826 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1827 * initialized properly.
1829 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1830 struct buffer_head
*bh
, int create
)
1832 struct ext4_map_blocks map
;
1835 BUG_ON(create
== 0);
1836 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1838 map
.m_lblk
= iblock
;
1842 * first, we need to know whether the block is allocated already
1843 * preallocated blocks are unmapped but should treated
1844 * the same as allocated blocks.
1846 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1850 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1851 ext4_update_bh_state(bh
, map
.m_flags
);
1853 if (buffer_unwritten(bh
)) {
1854 /* A delayed write to unwritten bh should be marked
1855 * new and mapped. Mapped ensures that we don't do
1856 * get_block multiple times when we write to the same
1857 * offset and new ensures that we do proper zero out
1858 * for partial write.
1861 set_buffer_mapped(bh
);
1866 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1872 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1878 static int __ext4_journalled_writepage(struct page
*page
,
1881 struct address_space
*mapping
= page
->mapping
;
1882 struct inode
*inode
= mapping
->host
;
1883 struct buffer_head
*page_bufs
= NULL
;
1884 handle_t
*handle
= NULL
;
1885 int ret
= 0, err
= 0;
1886 int inline_data
= ext4_has_inline_data(inode
);
1887 struct buffer_head
*inode_bh
= NULL
;
1889 ClearPageChecked(page
);
1892 BUG_ON(page
->index
!= 0);
1893 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1894 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1895 if (inode_bh
== NULL
)
1898 page_bufs
= page_buffers(page
);
1903 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1907 * We need to release the page lock before we start the
1908 * journal, so grab a reference so the page won't disappear
1909 * out from under us.
1914 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1915 ext4_writepage_trans_blocks(inode
));
1916 if (IS_ERR(handle
)) {
1917 ret
= PTR_ERR(handle
);
1919 goto out_no_pagelock
;
1921 BUG_ON(!ext4_handle_valid(handle
));
1925 if (page
->mapping
!= mapping
) {
1926 /* The page got truncated from under us */
1927 ext4_journal_stop(handle
);
1933 BUFFER_TRACE(inode_bh
, "get write access");
1934 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1936 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1939 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1940 do_journal_get_write_access
);
1942 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1947 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1948 err
= ext4_journal_stop(handle
);
1952 if (!ext4_has_inline_data(inode
))
1953 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1955 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1964 * Note that we don't need to start a transaction unless we're journaling data
1965 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1966 * need to file the inode to the transaction's list in ordered mode because if
1967 * we are writing back data added by write(), the inode is already there and if
1968 * we are writing back data modified via mmap(), no one guarantees in which
1969 * transaction the data will hit the disk. In case we are journaling data, we
1970 * cannot start transaction directly because transaction start ranks above page
1971 * lock so we have to do some magic.
1973 * This function can get called via...
1974 * - ext4_writepages after taking page lock (have journal handle)
1975 * - journal_submit_inode_data_buffers (no journal handle)
1976 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1977 * - grab_page_cache when doing write_begin (have journal handle)
1979 * We don't do any block allocation in this function. If we have page with
1980 * multiple blocks we need to write those buffer_heads that are mapped. This
1981 * is important for mmaped based write. So if we do with blocksize 1K
1982 * truncate(f, 1024);
1983 * a = mmap(f, 0, 4096);
1985 * truncate(f, 4096);
1986 * we have in the page first buffer_head mapped via page_mkwrite call back
1987 * but other buffer_heads would be unmapped but dirty (dirty done via the
1988 * do_wp_page). So writepage should write the first block. If we modify
1989 * the mmap area beyond 1024 we will again get a page_fault and the
1990 * page_mkwrite callback will do the block allocation and mark the
1991 * buffer_heads mapped.
1993 * We redirty the page if we have any buffer_heads that is either delay or
1994 * unwritten in the page.
1996 * We can get recursively called as show below.
1998 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2001 * But since we don't do any block allocation we should not deadlock.
2002 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2004 static int ext4_writepage(struct page
*page
,
2005 struct writeback_control
*wbc
)
2010 struct buffer_head
*page_bufs
= NULL
;
2011 struct inode
*inode
= page
->mapping
->host
;
2012 struct ext4_io_submit io_submit
;
2013 bool keep_towrite
= false;
2015 trace_ext4_writepage(page
);
2016 size
= i_size_read(inode
);
2017 if (page
->index
== size
>> PAGE_SHIFT
)
2018 len
= size
& ~PAGE_MASK
;
2022 page_bufs
= page_buffers(page
);
2024 * We cannot do block allocation or other extent handling in this
2025 * function. If there are buffers needing that, we have to redirty
2026 * the page. But we may reach here when we do a journal commit via
2027 * journal_submit_inode_data_buffers() and in that case we must write
2028 * allocated buffers to achieve data=ordered mode guarantees.
2030 * Also, if there is only one buffer per page (the fs block
2031 * size == the page size), if one buffer needs block
2032 * allocation or needs to modify the extent tree to clear the
2033 * unwritten flag, we know that the page can't be written at
2034 * all, so we might as well refuse the write immediately.
2035 * Unfortunately if the block size != page size, we can't as
2036 * easily detect this case using ext4_walk_page_buffers(), but
2037 * for the extremely common case, this is an optimization that
2038 * skips a useless round trip through ext4_bio_write_page().
2040 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2041 ext4_bh_delay_or_unwritten
)) {
2042 redirty_page_for_writepage(wbc
, page
);
2043 if ((current
->flags
& PF_MEMALLOC
) ||
2044 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2046 * For memory cleaning there's no point in writing only
2047 * some buffers. So just bail out. Warn if we came here
2048 * from direct reclaim.
2050 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2055 keep_towrite
= true;
2058 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2060 * It's mmapped pagecache. Add buffers and journal it. There
2061 * doesn't seem much point in redirtying the page here.
2063 return __ext4_journalled_writepage(page
, len
);
2065 ext4_io_submit_init(&io_submit
, wbc
);
2066 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2067 if (!io_submit
.io_end
) {
2068 redirty_page_for_writepage(wbc
, page
);
2072 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2073 ext4_io_submit(&io_submit
);
2074 /* Drop io_end reference we got from init */
2075 ext4_put_io_end_defer(io_submit
.io_end
);
2079 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2082 loff_t size
= i_size_read(mpd
->inode
);
2085 BUG_ON(page
->index
!= mpd
->first_page
);
2086 if (page
->index
== size
>> PAGE_SHIFT
)
2087 len
= size
& ~PAGE_MASK
;
2090 clear_page_dirty_for_io(page
);
2091 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2093 mpd
->wbc
->nr_to_write
--;
2099 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2102 * mballoc gives us at most this number of blocks...
2103 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2104 * The rest of mballoc seems to handle chunks up to full group size.
2106 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2109 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2111 * @mpd - extent of blocks
2112 * @lblk - logical number of the block in the file
2113 * @bh - buffer head we want to add to the extent
2115 * The function is used to collect contig. blocks in the same state. If the
2116 * buffer doesn't require mapping for writeback and we haven't started the
2117 * extent of buffers to map yet, the function returns 'true' immediately - the
2118 * caller can write the buffer right away. Otherwise the function returns true
2119 * if the block has been added to the extent, false if the block couldn't be
2122 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2123 struct buffer_head
*bh
)
2125 struct ext4_map_blocks
*map
= &mpd
->map
;
2127 /* Buffer that doesn't need mapping for writeback? */
2128 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2129 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2130 /* So far no extent to map => we write the buffer right away */
2131 if (map
->m_len
== 0)
2136 /* First block in the extent? */
2137 if (map
->m_len
== 0) {
2140 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2144 /* Don't go larger than mballoc is willing to allocate */
2145 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2148 /* Can we merge the block to our big extent? */
2149 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2150 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2158 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2160 * @mpd - extent of blocks for mapping
2161 * @head - the first buffer in the page
2162 * @bh - buffer we should start processing from
2163 * @lblk - logical number of the block in the file corresponding to @bh
2165 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2166 * the page for IO if all buffers in this page were mapped and there's no
2167 * accumulated extent of buffers to map or add buffers in the page to the
2168 * extent of buffers to map. The function returns 1 if the caller can continue
2169 * by processing the next page, 0 if it should stop adding buffers to the
2170 * extent to map because we cannot extend it anymore. It can also return value
2171 * < 0 in case of error during IO submission.
2173 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2174 struct buffer_head
*head
,
2175 struct buffer_head
*bh
,
2178 struct inode
*inode
= mpd
->inode
;
2180 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2181 >> inode
->i_blkbits
;
2184 BUG_ON(buffer_locked(bh
));
2186 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2187 /* Found extent to map? */
2190 /* Everything mapped so far and we hit EOF */
2193 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2194 /* So far everything mapped? Submit the page for IO. */
2195 if (mpd
->map
.m_len
== 0) {
2196 err
= mpage_submit_page(mpd
, head
->b_page
);
2200 return lblk
< blocks
;
2204 * mpage_map_buffers - update buffers corresponding to changed extent and
2205 * submit fully mapped pages for IO
2207 * @mpd - description of extent to map, on return next extent to map
2209 * Scan buffers corresponding to changed extent (we expect corresponding pages
2210 * to be already locked) and update buffer state according to new extent state.
2211 * We map delalloc buffers to their physical location, clear unwritten bits,
2212 * and mark buffers as uninit when we perform writes to unwritten extents
2213 * and do extent conversion after IO is finished. If the last page is not fully
2214 * mapped, we update @map to the next extent in the last page that needs
2215 * mapping. Otherwise we submit the page for IO.
2217 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2219 struct pagevec pvec
;
2221 struct inode
*inode
= mpd
->inode
;
2222 struct buffer_head
*head
, *bh
;
2223 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2229 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2230 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2231 lblk
= start
<< bpp_bits
;
2232 pblock
= mpd
->map
.m_pblk
;
2234 pagevec_init(&pvec
, 0);
2235 while (start
<= end
) {
2236 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2240 for (i
= 0; i
< nr_pages
; i
++) {
2241 struct page
*page
= pvec
.pages
[i
];
2243 if (page
->index
> end
)
2245 /* Up to 'end' pages must be contiguous */
2246 BUG_ON(page
->index
!= start
);
2247 bh
= head
= page_buffers(page
);
2249 if (lblk
< mpd
->map
.m_lblk
)
2251 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2253 * Buffer after end of mapped extent.
2254 * Find next buffer in the page to map.
2257 mpd
->map
.m_flags
= 0;
2259 * FIXME: If dioread_nolock supports
2260 * blocksize < pagesize, we need to make
2261 * sure we add size mapped so far to
2262 * io_end->size as the following call
2263 * can submit the page for IO.
2265 err
= mpage_process_page_bufs(mpd
, head
,
2267 pagevec_release(&pvec
);
2272 if (buffer_delay(bh
)) {
2273 clear_buffer_delay(bh
);
2274 bh
->b_blocknr
= pblock
++;
2276 clear_buffer_unwritten(bh
);
2277 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2280 * FIXME: This is going to break if dioread_nolock
2281 * supports blocksize < pagesize as we will try to
2282 * convert potentially unmapped parts of inode.
2284 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2285 /* Page fully mapped - let IO run! */
2286 err
= mpage_submit_page(mpd
, page
);
2288 pagevec_release(&pvec
);
2293 pagevec_release(&pvec
);
2295 /* Extent fully mapped and matches with page boundary. We are done. */
2297 mpd
->map
.m_flags
= 0;
2301 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2303 struct inode
*inode
= mpd
->inode
;
2304 struct ext4_map_blocks
*map
= &mpd
->map
;
2305 int get_blocks_flags
;
2306 int err
, dioread_nolock
;
2308 trace_ext4_da_write_pages_extent(inode
, map
);
2310 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2311 * to convert an unwritten extent to be initialized (in the case
2312 * where we have written into one or more preallocated blocks). It is
2313 * possible that we're going to need more metadata blocks than
2314 * previously reserved. However we must not fail because we're in
2315 * writeback and there is nothing we can do about it so it might result
2316 * in data loss. So use reserved blocks to allocate metadata if
2319 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2320 * the blocks in question are delalloc blocks. This indicates
2321 * that the blocks and quotas has already been checked when
2322 * the data was copied into the page cache.
2324 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2325 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2326 EXT4_GET_BLOCKS_IO_SUBMIT
;
2327 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2329 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2330 if (map
->m_flags
& (1 << BH_Delay
))
2331 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2333 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2336 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2337 if (!mpd
->io_submit
.io_end
->handle
&&
2338 ext4_handle_valid(handle
)) {
2339 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2340 handle
->h_rsv_handle
= NULL
;
2342 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2345 BUG_ON(map
->m_len
== 0);
2346 if (map
->m_flags
& EXT4_MAP_NEW
) {
2347 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2350 for (i
= 0; i
< map
->m_len
; i
++)
2351 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2357 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2358 * mpd->len and submit pages underlying it for IO
2360 * @handle - handle for journal operations
2361 * @mpd - extent to map
2362 * @give_up_on_write - we set this to true iff there is a fatal error and there
2363 * is no hope of writing the data. The caller should discard
2364 * dirty pages to avoid infinite loops.
2366 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2367 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2368 * them to initialized or split the described range from larger unwritten
2369 * extent. Note that we need not map all the described range since allocation
2370 * can return less blocks or the range is covered by more unwritten extents. We
2371 * cannot map more because we are limited by reserved transaction credits. On
2372 * the other hand we always make sure that the last touched page is fully
2373 * mapped so that it can be written out (and thus forward progress is
2374 * guaranteed). After mapping we submit all mapped pages for IO.
2376 static int mpage_map_and_submit_extent(handle_t
*handle
,
2377 struct mpage_da_data
*mpd
,
2378 bool *give_up_on_write
)
2380 struct inode
*inode
= mpd
->inode
;
2381 struct ext4_map_blocks
*map
= &mpd
->map
;
2386 mpd
->io_submit
.io_end
->offset
=
2387 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2389 err
= mpage_map_one_extent(handle
, mpd
);
2391 struct super_block
*sb
= inode
->i_sb
;
2393 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2394 goto invalidate_dirty_pages
;
2396 * Let the uper layers retry transient errors.
2397 * In the case of ENOSPC, if ext4_count_free_blocks()
2398 * is non-zero, a commit should free up blocks.
2400 if ((err
== -ENOMEM
) ||
2401 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2403 goto update_disksize
;
2406 ext4_msg(sb
, KERN_CRIT
,
2407 "Delayed block allocation failed for "
2408 "inode %lu at logical offset %llu with"
2409 " max blocks %u with error %d",
2411 (unsigned long long)map
->m_lblk
,
2412 (unsigned)map
->m_len
, -err
);
2413 ext4_msg(sb
, KERN_CRIT
,
2414 "This should not happen!! Data will "
2417 ext4_print_free_blocks(inode
);
2418 invalidate_dirty_pages
:
2419 *give_up_on_write
= true;
2424 * Update buffer state, submit mapped pages, and get us new
2427 err
= mpage_map_and_submit_buffers(mpd
);
2429 goto update_disksize
;
2430 } while (map
->m_len
);
2434 * Update on-disk size after IO is submitted. Races with
2435 * truncate are avoided by checking i_size under i_data_sem.
2437 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2438 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2442 down_write(&EXT4_I(inode
)->i_data_sem
);
2443 i_size
= i_size_read(inode
);
2444 if (disksize
> i_size
)
2446 if (disksize
> EXT4_I(inode
)->i_disksize
)
2447 EXT4_I(inode
)->i_disksize
= disksize
;
2448 err2
= ext4_mark_inode_dirty(handle
, inode
);
2449 up_write(&EXT4_I(inode
)->i_data_sem
);
2451 ext4_error(inode
->i_sb
,
2452 "Failed to mark inode %lu dirty",
2461 * Calculate the total number of credits to reserve for one writepages
2462 * iteration. This is called from ext4_writepages(). We map an extent of
2463 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2464 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2465 * bpp - 1 blocks in bpp different extents.
2467 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2469 int bpp
= ext4_journal_blocks_per_page(inode
);
2471 return ext4_meta_trans_blocks(inode
,
2472 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2476 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2477 * and underlying extent to map
2479 * @mpd - where to look for pages
2481 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2482 * IO immediately. When we find a page which isn't mapped we start accumulating
2483 * extent of buffers underlying these pages that needs mapping (formed by
2484 * either delayed or unwritten buffers). We also lock the pages containing
2485 * these buffers. The extent found is returned in @mpd structure (starting at
2486 * mpd->lblk with length mpd->len blocks).
2488 * Note that this function can attach bios to one io_end structure which are
2489 * neither logically nor physically contiguous. Although it may seem as an
2490 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2491 * case as we need to track IO to all buffers underlying a page in one io_end.
2493 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2495 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2496 struct pagevec pvec
;
2497 unsigned int nr_pages
;
2498 long left
= mpd
->wbc
->nr_to_write
;
2499 pgoff_t index
= mpd
->first_page
;
2500 pgoff_t end
= mpd
->last_page
;
2503 int blkbits
= mpd
->inode
->i_blkbits
;
2505 struct buffer_head
*head
;
2507 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2508 tag
= PAGECACHE_TAG_TOWRITE
;
2510 tag
= PAGECACHE_TAG_DIRTY
;
2512 pagevec_init(&pvec
, 0);
2514 mpd
->next_page
= index
;
2515 while (index
<= end
) {
2516 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2517 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2521 for (i
= 0; i
< nr_pages
; i
++) {
2522 struct page
*page
= pvec
.pages
[i
];
2525 * At this point, the page may be truncated or
2526 * invalidated (changing page->mapping to NULL), or
2527 * even swizzled back from swapper_space to tmpfs file
2528 * mapping. However, page->index will not change
2529 * because we have a reference on the page.
2531 if (page
->index
> end
)
2535 * Accumulated enough dirty pages? This doesn't apply
2536 * to WB_SYNC_ALL mode. For integrity sync we have to
2537 * keep going because someone may be concurrently
2538 * dirtying pages, and we might have synced a lot of
2539 * newly appeared dirty pages, but have not synced all
2540 * of the old dirty pages.
2542 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2545 /* If we can't merge this page, we are done. */
2546 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2551 * If the page is no longer dirty, or its mapping no
2552 * longer corresponds to inode we are writing (which
2553 * means it has been truncated or invalidated), or the
2554 * page is already under writeback and we are not doing
2555 * a data integrity writeback, skip the page
2557 if (!PageDirty(page
) ||
2558 (PageWriteback(page
) &&
2559 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2560 unlikely(page
->mapping
!= mapping
)) {
2565 wait_on_page_writeback(page
);
2566 BUG_ON(PageWriteback(page
));
2568 if (mpd
->map
.m_len
== 0)
2569 mpd
->first_page
= page
->index
;
2570 mpd
->next_page
= page
->index
+ 1;
2571 /* Add all dirty buffers to mpd */
2572 lblk
= ((ext4_lblk_t
)page
->index
) <<
2573 (PAGE_SHIFT
- blkbits
);
2574 head
= page_buffers(page
);
2575 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2581 pagevec_release(&pvec
);
2586 pagevec_release(&pvec
);
2590 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2593 struct address_space
*mapping
= data
;
2594 int ret
= ext4_writepage(page
, wbc
);
2595 mapping_set_error(mapping
, ret
);
2599 static int ext4_writepages(struct address_space
*mapping
,
2600 struct writeback_control
*wbc
)
2602 pgoff_t writeback_index
= 0;
2603 long nr_to_write
= wbc
->nr_to_write
;
2604 int range_whole
= 0;
2606 handle_t
*handle
= NULL
;
2607 struct mpage_da_data mpd
;
2608 struct inode
*inode
= mapping
->host
;
2609 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2610 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2612 struct blk_plug plug
;
2613 bool give_up_on_write
= false;
2615 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2616 trace_ext4_writepages(inode
, wbc
);
2618 if (dax_mapping(mapping
)) {
2619 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2621 goto out_writepages
;
2625 * No pages to write? This is mainly a kludge to avoid starting
2626 * a transaction for special inodes like journal inode on last iput()
2627 * because that could violate lock ordering on umount
2629 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2630 goto out_writepages
;
2632 if (ext4_should_journal_data(inode
)) {
2633 struct blk_plug plug
;
2635 blk_start_plug(&plug
);
2636 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2637 blk_finish_plug(&plug
);
2638 goto out_writepages
;
2642 * If the filesystem has aborted, it is read-only, so return
2643 * right away instead of dumping stack traces later on that
2644 * will obscure the real source of the problem. We test
2645 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2646 * the latter could be true if the filesystem is mounted
2647 * read-only, and in that case, ext4_writepages should
2648 * *never* be called, so if that ever happens, we would want
2651 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2653 goto out_writepages
;
2656 if (ext4_should_dioread_nolock(inode
)) {
2658 * We may need to convert up to one extent per block in
2659 * the page and we may dirty the inode.
2661 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2665 * If we have inline data and arrive here, it means that
2666 * we will soon create the block for the 1st page, so
2667 * we'd better clear the inline data here.
2669 if (ext4_has_inline_data(inode
)) {
2670 /* Just inode will be modified... */
2671 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2672 if (IS_ERR(handle
)) {
2673 ret
= PTR_ERR(handle
);
2674 goto out_writepages
;
2676 BUG_ON(ext4_test_inode_state(inode
,
2677 EXT4_STATE_MAY_INLINE_DATA
));
2678 ext4_destroy_inline_data(handle
, inode
);
2679 ext4_journal_stop(handle
);
2682 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2685 if (wbc
->range_cyclic
) {
2686 writeback_index
= mapping
->writeback_index
;
2687 if (writeback_index
)
2689 mpd
.first_page
= writeback_index
;
2692 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2693 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2698 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2700 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2701 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2703 blk_start_plug(&plug
);
2704 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2705 /* For each extent of pages we use new io_end */
2706 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2707 if (!mpd
.io_submit
.io_end
) {
2713 * We have two constraints: We find one extent to map and we
2714 * must always write out whole page (makes a difference when
2715 * blocksize < pagesize) so that we don't block on IO when we
2716 * try to write out the rest of the page. Journalled mode is
2717 * not supported by delalloc.
2719 BUG_ON(ext4_should_journal_data(inode
));
2720 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2722 /* start a new transaction */
2723 handle
= ext4_journal_start_with_reserve(inode
,
2724 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2725 if (IS_ERR(handle
)) {
2726 ret
= PTR_ERR(handle
);
2727 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2728 "%ld pages, ino %lu; err %d", __func__
,
2729 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2730 /* Release allocated io_end */
2731 ext4_put_io_end(mpd
.io_submit
.io_end
);
2735 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2736 ret
= mpage_prepare_extent_to_map(&mpd
);
2739 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2743 * We scanned the whole range (or exhausted
2744 * nr_to_write), submitted what was mapped and
2745 * didn't find anything needing mapping. We are
2751 ext4_journal_stop(handle
);
2752 /* Submit prepared bio */
2753 ext4_io_submit(&mpd
.io_submit
);
2754 /* Unlock pages we didn't use */
2755 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2756 /* Drop our io_end reference we got from init */
2757 ext4_put_io_end(mpd
.io_submit
.io_end
);
2759 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2761 * Commit the transaction which would
2762 * free blocks released in the transaction
2765 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2769 /* Fatal error - ENOMEM, EIO... */
2773 blk_finish_plug(&plug
);
2774 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2776 mpd
.last_page
= writeback_index
- 1;
2782 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2784 * Set the writeback_index so that range_cyclic
2785 * mode will write it back later
2787 mapping
->writeback_index
= mpd
.first_page
;
2790 trace_ext4_writepages_result(inode
, wbc
, ret
,
2791 nr_to_write
- wbc
->nr_to_write
);
2792 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2796 static int ext4_nonda_switch(struct super_block
*sb
)
2798 s64 free_clusters
, dirty_clusters
;
2799 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2802 * switch to non delalloc mode if we are running low
2803 * on free block. The free block accounting via percpu
2804 * counters can get slightly wrong with percpu_counter_batch getting
2805 * accumulated on each CPU without updating global counters
2806 * Delalloc need an accurate free block accounting. So switch
2807 * to non delalloc when we are near to error range.
2810 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2812 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2814 * Start pushing delalloc when 1/2 of free blocks are dirty.
2816 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2817 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2819 if (2 * free_clusters
< 3 * dirty_clusters
||
2820 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2822 * free block count is less than 150% of dirty blocks
2823 * or free blocks is less than watermark
2830 /* We always reserve for an inode update; the superblock could be there too */
2831 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2833 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2836 if (pos
+ len
<= 0x7fffffffULL
)
2839 /* We might need to update the superblock to set LARGE_FILE */
2843 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2844 loff_t pos
, unsigned len
, unsigned flags
,
2845 struct page
**pagep
, void **fsdata
)
2847 int ret
, retries
= 0;
2850 struct inode
*inode
= mapping
->host
;
2853 index
= pos
>> PAGE_SHIFT
;
2855 if (ext4_nonda_switch(inode
->i_sb
)) {
2856 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2857 return ext4_write_begin(file
, mapping
, pos
,
2858 len
, flags
, pagep
, fsdata
);
2860 *fsdata
= (void *)0;
2861 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2863 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2864 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2874 * grab_cache_page_write_begin() can take a long time if the
2875 * system is thrashing due to memory pressure, or if the page
2876 * is being written back. So grab it first before we start
2877 * the transaction handle. This also allows us to allocate
2878 * the page (if needed) without using GFP_NOFS.
2881 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2887 * With delayed allocation, we don't log the i_disksize update
2888 * if there is delayed block allocation. But we still need
2889 * to journalling the i_disksize update if writes to the end
2890 * of file which has an already mapped buffer.
2893 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2894 ext4_da_write_credits(inode
, pos
, len
));
2895 if (IS_ERR(handle
)) {
2897 return PTR_ERR(handle
);
2901 if (page
->mapping
!= mapping
) {
2902 /* The page got truncated from under us */
2905 ext4_journal_stop(handle
);
2908 /* In case writeback began while the page was unlocked */
2909 wait_for_stable_page(page
);
2911 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2912 ret
= ext4_block_write_begin(page
, pos
, len
,
2913 ext4_da_get_block_prep
);
2915 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2919 ext4_journal_stop(handle
);
2921 * block_write_begin may have instantiated a few blocks
2922 * outside i_size. Trim these off again. Don't need
2923 * i_size_read because we hold i_mutex.
2925 if (pos
+ len
> inode
->i_size
)
2926 ext4_truncate_failed_write(inode
);
2928 if (ret
== -ENOSPC
&&
2929 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2941 * Check if we should update i_disksize
2942 * when write to the end of file but not require block allocation
2944 static int ext4_da_should_update_i_disksize(struct page
*page
,
2945 unsigned long offset
)
2947 struct buffer_head
*bh
;
2948 struct inode
*inode
= page
->mapping
->host
;
2952 bh
= page_buffers(page
);
2953 idx
= offset
>> inode
->i_blkbits
;
2955 for (i
= 0; i
< idx
; i
++)
2956 bh
= bh
->b_this_page
;
2958 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2963 static int ext4_da_write_end(struct file
*file
,
2964 struct address_space
*mapping
,
2965 loff_t pos
, unsigned len
, unsigned copied
,
2966 struct page
*page
, void *fsdata
)
2968 struct inode
*inode
= mapping
->host
;
2970 handle_t
*handle
= ext4_journal_current_handle();
2972 unsigned long start
, end
;
2973 int write_mode
= (int)(unsigned long)fsdata
;
2975 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2976 return ext4_write_end(file
, mapping
, pos
,
2977 len
, copied
, page
, fsdata
);
2979 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2980 start
= pos
& (PAGE_SIZE
- 1);
2981 end
= start
+ copied
- 1;
2984 * generic_write_end() will run mark_inode_dirty() if i_size
2985 * changes. So let's piggyback the i_disksize mark_inode_dirty
2988 new_i_size
= pos
+ copied
;
2989 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2990 if (ext4_has_inline_data(inode
) ||
2991 ext4_da_should_update_i_disksize(page
, end
)) {
2992 ext4_update_i_disksize(inode
, new_i_size
);
2993 /* We need to mark inode dirty even if
2994 * new_i_size is less that inode->i_size
2995 * bu greater than i_disksize.(hint delalloc)
2997 ext4_mark_inode_dirty(handle
, inode
);
3001 if (write_mode
!= CONVERT_INLINE_DATA
&&
3002 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3003 ext4_has_inline_data(inode
))
3004 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3007 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3013 ret2
= ext4_journal_stop(handle
);
3017 return ret
? ret
: copied
;
3020 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3021 unsigned int length
)
3024 * Drop reserved blocks
3026 BUG_ON(!PageLocked(page
));
3027 if (!page_has_buffers(page
))
3030 ext4_da_page_release_reservation(page
, offset
, length
);
3033 ext4_invalidatepage(page
, offset
, length
);
3039 * Force all delayed allocation blocks to be allocated for a given inode.
3041 int ext4_alloc_da_blocks(struct inode
*inode
)
3043 trace_ext4_alloc_da_blocks(inode
);
3045 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3049 * We do something simple for now. The filemap_flush() will
3050 * also start triggering a write of the data blocks, which is
3051 * not strictly speaking necessary (and for users of
3052 * laptop_mode, not even desirable). However, to do otherwise
3053 * would require replicating code paths in:
3055 * ext4_writepages() ->
3056 * write_cache_pages() ---> (via passed in callback function)
3057 * __mpage_da_writepage() -->
3058 * mpage_add_bh_to_extent()
3059 * mpage_da_map_blocks()
3061 * The problem is that write_cache_pages(), located in
3062 * mm/page-writeback.c, marks pages clean in preparation for
3063 * doing I/O, which is not desirable if we're not planning on
3066 * We could call write_cache_pages(), and then redirty all of
3067 * the pages by calling redirty_page_for_writepage() but that
3068 * would be ugly in the extreme. So instead we would need to
3069 * replicate parts of the code in the above functions,
3070 * simplifying them because we wouldn't actually intend to
3071 * write out the pages, but rather only collect contiguous
3072 * logical block extents, call the multi-block allocator, and
3073 * then update the buffer heads with the block allocations.
3075 * For now, though, we'll cheat by calling filemap_flush(),
3076 * which will map the blocks, and start the I/O, but not
3077 * actually wait for the I/O to complete.
3079 return filemap_flush(inode
->i_mapping
);
3083 * bmap() is special. It gets used by applications such as lilo and by
3084 * the swapper to find the on-disk block of a specific piece of data.
3086 * Naturally, this is dangerous if the block concerned is still in the
3087 * journal. If somebody makes a swapfile on an ext4 data-journaling
3088 * filesystem and enables swap, then they may get a nasty shock when the
3089 * data getting swapped to that swapfile suddenly gets overwritten by
3090 * the original zero's written out previously to the journal and
3091 * awaiting writeback in the kernel's buffer cache.
3093 * So, if we see any bmap calls here on a modified, data-journaled file,
3094 * take extra steps to flush any blocks which might be in the cache.
3096 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3098 struct inode
*inode
= mapping
->host
;
3103 * We can get here for an inline file via the FIBMAP ioctl
3105 if (ext4_has_inline_data(inode
))
3108 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3109 test_opt(inode
->i_sb
, DELALLOC
)) {
3111 * With delalloc we want to sync the file
3112 * so that we can make sure we allocate
3115 filemap_write_and_wait(mapping
);
3118 if (EXT4_JOURNAL(inode
) &&
3119 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3121 * This is a REALLY heavyweight approach, but the use of
3122 * bmap on dirty files is expected to be extremely rare:
3123 * only if we run lilo or swapon on a freshly made file
3124 * do we expect this to happen.
3126 * (bmap requires CAP_SYS_RAWIO so this does not
3127 * represent an unprivileged user DOS attack --- we'd be
3128 * in trouble if mortal users could trigger this path at
3131 * NB. EXT4_STATE_JDATA is not set on files other than
3132 * regular files. If somebody wants to bmap a directory
3133 * or symlink and gets confused because the buffer
3134 * hasn't yet been flushed to disk, they deserve
3135 * everything they get.
3138 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3139 journal
= EXT4_JOURNAL(inode
);
3140 jbd2_journal_lock_updates(journal
);
3141 err
= jbd2_journal_flush(journal
);
3142 jbd2_journal_unlock_updates(journal
);
3148 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3151 static int ext4_readpage(struct file
*file
, struct page
*page
)
3154 struct inode
*inode
= page
->mapping
->host
;
3156 trace_ext4_readpage(page
);
3158 if (ext4_has_inline_data(inode
))
3159 ret
= ext4_readpage_inline(inode
, page
);
3162 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3168 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3169 struct list_head
*pages
, unsigned nr_pages
)
3171 struct inode
*inode
= mapping
->host
;
3173 /* If the file has inline data, no need to do readpages. */
3174 if (ext4_has_inline_data(inode
))
3177 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3180 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3181 unsigned int length
)
3183 trace_ext4_invalidatepage(page
, offset
, length
);
3185 /* No journalling happens on data buffers when this function is used */
3186 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3188 block_invalidatepage(page
, offset
, length
);
3191 static int __ext4_journalled_invalidatepage(struct page
*page
,
3192 unsigned int offset
,
3193 unsigned int length
)
3195 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3197 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3200 * If it's a full truncate we just forget about the pending dirtying
3202 if (offset
== 0 && length
== PAGE_SIZE
)
3203 ClearPageChecked(page
);
3205 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3208 /* Wrapper for aops... */
3209 static void ext4_journalled_invalidatepage(struct page
*page
,
3210 unsigned int offset
,
3211 unsigned int length
)
3213 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3216 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3218 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3220 trace_ext4_releasepage(page
);
3222 /* Page has dirty journalled data -> cannot release */
3223 if (PageChecked(page
))
3226 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3228 return try_to_free_buffers(page
);
3231 #ifdef CONFIG_FS_DAX
3233 * Get block function for DAX IO and mmap faults. It takes care of converting
3234 * unwritten extents to written ones and initializes new / converted blocks
3237 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3238 struct buffer_head
*bh_result
, int create
)
3242 ext4_debug("inode %lu, create flag %d\n", inode
->i_ino
, create
);
3244 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
3246 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3247 EXT4_GET_BLOCKS_PRE_IO
|
3248 EXT4_GET_BLOCKS_CREATE_ZERO
);
3252 if (buffer_unwritten(bh_result
)) {
3254 * We are protected by i_mmap_sem or i_mutex so we know block
3255 * cannot go away from under us even though we dropped
3256 * i_data_sem. Convert extent to written and write zeros there.
3258 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3259 EXT4_GET_BLOCKS_CONVERT
|
3260 EXT4_GET_BLOCKS_CREATE_ZERO
);
3265 * At least for now we have to clear BH_New so that DAX code
3266 * doesn't attempt to zero blocks again in a racy way.
3268 clear_buffer_new(bh_result
);
3272 /* Just define empty function, it will never get called. */
3273 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3274 struct buffer_head
*bh_result
, int create
)
3281 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3282 ssize_t size
, void *private)
3284 ext4_io_end_t
*io_end
= private;
3286 /* if not async direct IO just return */
3290 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3291 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3292 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3295 * Error during AIO DIO. We cannot convert unwritten extents as the
3296 * data was not written. Just clear the unwritten flag and drop io_end.
3299 ext4_clear_io_unwritten_flag(io_end
);
3302 io_end
->offset
= offset
;
3303 io_end
->size
= size
;
3304 ext4_put_io_end(io_end
);
3310 * Handling of direct IO writes.
3312 * For ext4 extent files, ext4 will do direct-io write even to holes,
3313 * preallocated extents, and those write extend the file, no need to
3314 * fall back to buffered IO.
3316 * For holes, we fallocate those blocks, mark them as unwritten
3317 * If those blocks were preallocated, we mark sure they are split, but
3318 * still keep the range to write as unwritten.
3320 * The unwritten extents will be converted to written when DIO is completed.
3321 * For async direct IO, since the IO may still pending when return, we
3322 * set up an end_io call back function, which will do the conversion
3323 * when async direct IO completed.
3325 * If the O_DIRECT write will extend the file then add this inode to the
3326 * orphan list. So recovery will truncate it back to the original size
3327 * if the machine crashes during the write.
3330 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3332 struct file
*file
= iocb
->ki_filp
;
3333 struct inode
*inode
= file
->f_mapping
->host
;
3334 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3336 loff_t offset
= iocb
->ki_pos
;
3337 size_t count
= iov_iter_count(iter
);
3339 get_block_t
*get_block_func
= NULL
;
3341 loff_t final_size
= offset
+ count
;
3345 if (final_size
> inode
->i_size
) {
3346 /* Credits for sb + inode write */
3347 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3348 if (IS_ERR(handle
)) {
3349 ret
= PTR_ERR(handle
);
3352 ret
= ext4_orphan_add(handle
, inode
);
3354 ext4_journal_stop(handle
);
3358 ei
->i_disksize
= inode
->i_size
;
3359 ext4_journal_stop(handle
);
3362 BUG_ON(iocb
->private == NULL
);
3365 * Make all waiters for direct IO properly wait also for extent
3366 * conversion. This also disallows race between truncate() and
3367 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3369 inode_dio_begin(inode
);
3371 /* If we do a overwrite dio, i_mutex locking can be released */
3372 overwrite
= *((int *)iocb
->private);
3375 inode_unlock(inode
);
3378 * For extent mapped files we could direct write to holes and fallocate.
3380 * Allocated blocks to fill the hole are marked as unwritten to prevent
3381 * parallel buffered read to expose the stale data before DIO complete
3384 * As to previously fallocated extents, ext4 get_block will just simply
3385 * mark the buffer mapped but still keep the extents unwritten.
3387 * For non AIO case, we will convert those unwritten extents to written
3388 * after return back from blockdev_direct_IO. That way we save us from
3389 * allocating io_end structure and also the overhead of offloading
3390 * the extent convertion to a workqueue.
3392 * For async DIO, the conversion needs to be deferred when the
3393 * IO is completed. The ext4 end_io callback function will be
3394 * called to take care of the conversion work. Here for async
3395 * case, we allocate an io_end structure to hook to the iocb.
3397 iocb
->private = NULL
;
3399 get_block_func
= ext4_dio_get_block_overwrite
;
3400 else if (IS_DAX(inode
)) {
3402 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3403 * writes need zeroing either because they can race with page
3404 * faults or because they use partial blocks.
3406 if (round_down(offset
, 1<<inode
->i_blkbits
) >= inode
->i_size
&&
3407 ext4_aligned_io(inode
, offset
, count
))
3408 get_block_func
= ext4_dio_get_block
;
3410 get_block_func
= ext4_dax_get_block
;
3411 dio_flags
= DIO_LOCKING
;
3412 } else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3413 round_down(offset
, 1 << inode
->i_blkbits
) >= inode
->i_size
) {
3414 get_block_func
= ext4_dio_get_block
;
3415 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3416 } else if (is_sync_kiocb(iocb
)) {
3417 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3418 dio_flags
= DIO_LOCKING
;
3420 get_block_func
= ext4_dio_get_block_unwritten_async
;
3421 dio_flags
= DIO_LOCKING
;
3423 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3424 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3426 if (IS_DAX(inode
)) {
3427 ret
= dax_do_io(iocb
, inode
, iter
, get_block_func
,
3428 ext4_end_io_dio
, dio_flags
);
3430 ret
= __blockdev_direct_IO(iocb
, inode
,
3431 inode
->i_sb
->s_bdev
, iter
,
3433 ext4_end_io_dio
, NULL
, dio_flags
);
3435 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3436 EXT4_STATE_DIO_UNWRITTEN
)) {
3439 * for non AIO case, since the IO is already
3440 * completed, we could do the conversion right here
3442 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3446 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3449 inode_dio_end(inode
);
3450 /* take i_mutex locking again if we do a ovewrite dio */
3454 if (ret
< 0 && final_size
> inode
->i_size
)
3455 ext4_truncate_failed_write(inode
);
3457 /* Handle extending of i_size after direct IO write */
3461 /* Credits for sb + inode write */
3462 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3463 if (IS_ERR(handle
)) {
3464 /* This is really bad luck. We've written the data
3465 * but cannot extend i_size. Bail out and pretend
3466 * the write failed... */
3467 ret
= PTR_ERR(handle
);
3469 ext4_orphan_del(NULL
, inode
);
3474 ext4_orphan_del(handle
, inode
);
3476 loff_t end
= offset
+ ret
;
3477 if (end
> inode
->i_size
) {
3478 ei
->i_disksize
= end
;
3479 i_size_write(inode
, end
);
3481 * We're going to return a positive `ret'
3482 * here due to non-zero-length I/O, so there's
3483 * no way of reporting error returns from
3484 * ext4_mark_inode_dirty() to userspace. So
3487 ext4_mark_inode_dirty(handle
, inode
);
3490 err
= ext4_journal_stop(handle
);
3498 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3501 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
3504 if (ext4_should_dioread_nolock(inode
)) {
3506 * Nolock dioread optimization may be dynamically disabled
3507 * via ext4_inode_block_unlocked_dio(). Check inode's state
3508 * while holding extra i_dio_count ref.
3510 inode_dio_begin(inode
);
3512 if (unlikely(ext4_test_inode_state(inode
,
3513 EXT4_STATE_DIOREAD_LOCK
)))
3514 inode_dio_end(inode
);
3518 if (IS_DAX(inode
)) {
3519 ret
= dax_do_io(iocb
, inode
, iter
, ext4_dio_get_block
,
3520 NULL
, unlocked
? 0 : DIO_LOCKING
);
3522 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3523 iter
, ext4_dio_get_block
,
3525 unlocked
? 0 : DIO_LOCKING
);
3528 inode_dio_end(inode
);
3532 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3534 struct file
*file
= iocb
->ki_filp
;
3535 struct inode
*inode
= file
->f_mapping
->host
;
3536 size_t count
= iov_iter_count(iter
);
3537 loff_t offset
= iocb
->ki_pos
;
3540 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3541 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3546 * If we are doing data journalling we don't support O_DIRECT
3548 if (ext4_should_journal_data(inode
))
3551 /* Let buffer I/O handle the inline data case. */
3552 if (ext4_has_inline_data(inode
))
3555 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3556 if (iov_iter_rw(iter
) == READ
)
3557 ret
= ext4_direct_IO_read(iocb
, iter
);
3559 ret
= ext4_direct_IO_write(iocb
, iter
);
3560 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3565 * Pages can be marked dirty completely asynchronously from ext4's journalling
3566 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3567 * much here because ->set_page_dirty is called under VFS locks. The page is
3568 * not necessarily locked.
3570 * We cannot just dirty the page and leave attached buffers clean, because the
3571 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3572 * or jbddirty because all the journalling code will explode.
3574 * So what we do is to mark the page "pending dirty" and next time writepage
3575 * is called, propagate that into the buffers appropriately.
3577 static int ext4_journalled_set_page_dirty(struct page
*page
)
3579 SetPageChecked(page
);
3580 return __set_page_dirty_nobuffers(page
);
3583 static const struct address_space_operations ext4_aops
= {
3584 .readpage
= ext4_readpage
,
3585 .readpages
= ext4_readpages
,
3586 .writepage
= ext4_writepage
,
3587 .writepages
= ext4_writepages
,
3588 .write_begin
= ext4_write_begin
,
3589 .write_end
= ext4_write_end
,
3591 .invalidatepage
= ext4_invalidatepage
,
3592 .releasepage
= ext4_releasepage
,
3593 .direct_IO
= ext4_direct_IO
,
3594 .migratepage
= buffer_migrate_page
,
3595 .is_partially_uptodate
= block_is_partially_uptodate
,
3596 .error_remove_page
= generic_error_remove_page
,
3599 static const struct address_space_operations ext4_journalled_aops
= {
3600 .readpage
= ext4_readpage
,
3601 .readpages
= ext4_readpages
,
3602 .writepage
= ext4_writepage
,
3603 .writepages
= ext4_writepages
,
3604 .write_begin
= ext4_write_begin
,
3605 .write_end
= ext4_journalled_write_end
,
3606 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3608 .invalidatepage
= ext4_journalled_invalidatepage
,
3609 .releasepage
= ext4_releasepage
,
3610 .direct_IO
= ext4_direct_IO
,
3611 .is_partially_uptodate
= block_is_partially_uptodate
,
3612 .error_remove_page
= generic_error_remove_page
,
3615 static const struct address_space_operations ext4_da_aops
= {
3616 .readpage
= ext4_readpage
,
3617 .readpages
= ext4_readpages
,
3618 .writepage
= ext4_writepage
,
3619 .writepages
= ext4_writepages
,
3620 .write_begin
= ext4_da_write_begin
,
3621 .write_end
= ext4_da_write_end
,
3623 .invalidatepage
= ext4_da_invalidatepage
,
3624 .releasepage
= ext4_releasepage
,
3625 .direct_IO
= ext4_direct_IO
,
3626 .migratepage
= buffer_migrate_page
,
3627 .is_partially_uptodate
= block_is_partially_uptodate
,
3628 .error_remove_page
= generic_error_remove_page
,
3631 void ext4_set_aops(struct inode
*inode
)
3633 switch (ext4_inode_journal_mode(inode
)) {
3634 case EXT4_INODE_ORDERED_DATA_MODE
:
3635 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3637 case EXT4_INODE_JOURNAL_DATA_MODE
:
3638 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3643 if (test_opt(inode
->i_sb
, DELALLOC
))
3644 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3646 inode
->i_mapping
->a_ops
= &ext4_aops
;
3649 static int __ext4_block_zero_page_range(handle_t
*handle
,
3650 struct address_space
*mapping
, loff_t from
, loff_t length
)
3652 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3653 unsigned offset
= from
& (PAGE_SIZE
-1);
3654 unsigned blocksize
, pos
;
3656 struct inode
*inode
= mapping
->host
;
3657 struct buffer_head
*bh
;
3661 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3662 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3666 blocksize
= inode
->i_sb
->s_blocksize
;
3668 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3670 if (!page_has_buffers(page
))
3671 create_empty_buffers(page
, blocksize
, 0);
3673 /* Find the buffer that contains "offset" */
3674 bh
= page_buffers(page
);
3676 while (offset
>= pos
) {
3677 bh
= bh
->b_this_page
;
3681 if (buffer_freed(bh
)) {
3682 BUFFER_TRACE(bh
, "freed: skip");
3685 if (!buffer_mapped(bh
)) {
3686 BUFFER_TRACE(bh
, "unmapped");
3687 ext4_get_block(inode
, iblock
, bh
, 0);
3688 /* unmapped? It's a hole - nothing to do */
3689 if (!buffer_mapped(bh
)) {
3690 BUFFER_TRACE(bh
, "still unmapped");
3695 /* Ok, it's mapped. Make sure it's up-to-date */
3696 if (PageUptodate(page
))
3697 set_buffer_uptodate(bh
);
3699 if (!buffer_uptodate(bh
)) {
3701 ll_rw_block(READ
, 1, &bh
);
3703 /* Uhhuh. Read error. Complain and punt. */
3704 if (!buffer_uptodate(bh
))
3706 if (S_ISREG(inode
->i_mode
) &&
3707 ext4_encrypted_inode(inode
)) {
3708 /* We expect the key to be set. */
3709 BUG_ON(!ext4_has_encryption_key(inode
));
3710 BUG_ON(blocksize
!= PAGE_SIZE
);
3711 WARN_ON_ONCE(ext4_decrypt(page
));
3714 if (ext4_should_journal_data(inode
)) {
3715 BUFFER_TRACE(bh
, "get write access");
3716 err
= ext4_journal_get_write_access(handle
, bh
);
3720 zero_user(page
, offset
, length
);
3721 BUFFER_TRACE(bh
, "zeroed end of block");
3723 if (ext4_should_journal_data(inode
)) {
3724 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3727 mark_buffer_dirty(bh
);
3728 if (ext4_should_order_data(inode
))
3729 err
= ext4_jbd2_inode_add_write(handle
, inode
);
3739 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3740 * starting from file offset 'from'. The range to be zero'd must
3741 * be contained with in one block. If the specified range exceeds
3742 * the end of the block it will be shortened to end of the block
3743 * that cooresponds to 'from'
3745 static int ext4_block_zero_page_range(handle_t
*handle
,
3746 struct address_space
*mapping
, loff_t from
, loff_t length
)
3748 struct inode
*inode
= mapping
->host
;
3749 unsigned offset
= from
& (PAGE_SIZE
-1);
3750 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3751 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3754 * correct length if it does not fall between
3755 * 'from' and the end of the block
3757 if (length
> max
|| length
< 0)
3761 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3762 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3766 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3767 * up to the end of the block which corresponds to `from'.
3768 * This required during truncate. We need to physically zero the tail end
3769 * of that block so it doesn't yield old data if the file is later grown.
3771 static int ext4_block_truncate_page(handle_t
*handle
,
3772 struct address_space
*mapping
, loff_t from
)
3774 unsigned offset
= from
& (PAGE_SIZE
-1);
3777 struct inode
*inode
= mapping
->host
;
3779 blocksize
= inode
->i_sb
->s_blocksize
;
3780 length
= blocksize
- (offset
& (blocksize
- 1));
3782 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3785 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3786 loff_t lstart
, loff_t length
)
3788 struct super_block
*sb
= inode
->i_sb
;
3789 struct address_space
*mapping
= inode
->i_mapping
;
3790 unsigned partial_start
, partial_end
;
3791 ext4_fsblk_t start
, end
;
3792 loff_t byte_end
= (lstart
+ length
- 1);
3795 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3796 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3798 start
= lstart
>> sb
->s_blocksize_bits
;
3799 end
= byte_end
>> sb
->s_blocksize_bits
;
3801 /* Handle partial zero within the single block */
3803 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3804 err
= ext4_block_zero_page_range(handle
, mapping
,
3808 /* Handle partial zero out on the start of the range */
3809 if (partial_start
) {
3810 err
= ext4_block_zero_page_range(handle
, mapping
,
3811 lstart
, sb
->s_blocksize
);
3815 /* Handle partial zero out on the end of the range */
3816 if (partial_end
!= sb
->s_blocksize
- 1)
3817 err
= ext4_block_zero_page_range(handle
, mapping
,
3818 byte_end
- partial_end
,
3823 int ext4_can_truncate(struct inode
*inode
)
3825 if (S_ISREG(inode
->i_mode
))
3827 if (S_ISDIR(inode
->i_mode
))
3829 if (S_ISLNK(inode
->i_mode
))
3830 return !ext4_inode_is_fast_symlink(inode
);
3835 * We have to make sure i_disksize gets properly updated before we truncate
3836 * page cache due to hole punching or zero range. Otherwise i_disksize update
3837 * can get lost as it may have been postponed to submission of writeback but
3838 * that will never happen after we truncate page cache.
3840 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3844 loff_t size
= i_size_read(inode
);
3846 WARN_ON(!inode_is_locked(inode
));
3847 if (offset
> size
|| offset
+ len
< size
)
3850 if (EXT4_I(inode
)->i_disksize
>= size
)
3853 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3855 return PTR_ERR(handle
);
3856 ext4_update_i_disksize(inode
, size
);
3857 ext4_mark_inode_dirty(handle
, inode
);
3858 ext4_journal_stop(handle
);
3864 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3865 * associated with the given offset and length
3867 * @inode: File inode
3868 * @offset: The offset where the hole will begin
3869 * @len: The length of the hole
3871 * Returns: 0 on success or negative on failure
3874 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3876 struct super_block
*sb
= inode
->i_sb
;
3877 ext4_lblk_t first_block
, stop_block
;
3878 struct address_space
*mapping
= inode
->i_mapping
;
3879 loff_t first_block_offset
, last_block_offset
;
3881 unsigned int credits
;
3884 if (!S_ISREG(inode
->i_mode
))
3887 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3890 * Write out all dirty pages to avoid race conditions
3891 * Then release them.
3893 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3894 ret
= filemap_write_and_wait_range(mapping
, offset
,
3895 offset
+ length
- 1);
3902 /* No need to punch hole beyond i_size */
3903 if (offset
>= inode
->i_size
)
3907 * If the hole extends beyond i_size, set the hole
3908 * to end after the page that contains i_size
3910 if (offset
+ length
> inode
->i_size
) {
3911 length
= inode
->i_size
+
3912 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
3916 if (offset
& (sb
->s_blocksize
- 1) ||
3917 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3919 * Attach jinode to inode for jbd2 if we do any zeroing of
3922 ret
= ext4_inode_attach_jinode(inode
);
3928 /* Wait all existing dio workers, newcomers will block on i_mutex */
3929 ext4_inode_block_unlocked_dio(inode
);
3930 inode_dio_wait(inode
);
3933 * Prevent page faults from reinstantiating pages we have released from
3936 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3937 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3938 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3940 /* Now release the pages and zero block aligned part of pages*/
3941 if (last_block_offset
> first_block_offset
) {
3942 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
3945 truncate_pagecache_range(inode
, first_block_offset
,
3949 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3950 credits
= ext4_writepage_trans_blocks(inode
);
3952 credits
= ext4_blocks_for_truncate(inode
);
3953 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3954 if (IS_ERR(handle
)) {
3955 ret
= PTR_ERR(handle
);
3956 ext4_std_error(sb
, ret
);
3960 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3965 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3966 EXT4_BLOCK_SIZE_BITS(sb
);
3967 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3969 /* If there are no blocks to remove, return now */
3970 if (first_block
>= stop_block
)
3973 down_write(&EXT4_I(inode
)->i_data_sem
);
3974 ext4_discard_preallocations(inode
);
3976 ret
= ext4_es_remove_extent(inode
, first_block
,
3977 stop_block
- first_block
);
3979 up_write(&EXT4_I(inode
)->i_data_sem
);
3983 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3984 ret
= ext4_ext_remove_space(inode
, first_block
,
3987 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3990 up_write(&EXT4_I(inode
)->i_data_sem
);
3992 ext4_handle_sync(handle
);
3994 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3995 ext4_mark_inode_dirty(handle
, inode
);
3997 ext4_journal_stop(handle
);
3999 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4000 ext4_inode_resume_unlocked_dio(inode
);
4002 inode_unlock(inode
);
4006 int ext4_inode_attach_jinode(struct inode
*inode
)
4008 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4009 struct jbd2_inode
*jinode
;
4011 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4014 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4015 spin_lock(&inode
->i_lock
);
4018 spin_unlock(&inode
->i_lock
);
4021 ei
->jinode
= jinode
;
4022 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4025 spin_unlock(&inode
->i_lock
);
4026 if (unlikely(jinode
!= NULL
))
4027 jbd2_free_inode(jinode
);
4034 * We block out ext4_get_block() block instantiations across the entire
4035 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4036 * simultaneously on behalf of the same inode.
4038 * As we work through the truncate and commit bits of it to the journal there
4039 * is one core, guiding principle: the file's tree must always be consistent on
4040 * disk. We must be able to restart the truncate after a crash.
4042 * The file's tree may be transiently inconsistent in memory (although it
4043 * probably isn't), but whenever we close off and commit a journal transaction,
4044 * the contents of (the filesystem + the journal) must be consistent and
4045 * restartable. It's pretty simple, really: bottom up, right to left (although
4046 * left-to-right works OK too).
4048 * Note that at recovery time, journal replay occurs *before* the restart of
4049 * truncate against the orphan inode list.
4051 * The committed inode has the new, desired i_size (which is the same as
4052 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4053 * that this inode's truncate did not complete and it will again call
4054 * ext4_truncate() to have another go. So there will be instantiated blocks
4055 * to the right of the truncation point in a crashed ext4 filesystem. But
4056 * that's fine - as long as they are linked from the inode, the post-crash
4057 * ext4_truncate() run will find them and release them.
4059 void ext4_truncate(struct inode
*inode
)
4061 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4062 unsigned int credits
;
4064 struct address_space
*mapping
= inode
->i_mapping
;
4067 * There is a possibility that we're either freeing the inode
4068 * or it's a completely new inode. In those cases we might not
4069 * have i_mutex locked because it's not necessary.
4071 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4072 WARN_ON(!inode_is_locked(inode
));
4073 trace_ext4_truncate_enter(inode
);
4075 if (!ext4_can_truncate(inode
))
4078 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4080 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4081 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4083 if (ext4_has_inline_data(inode
)) {
4086 ext4_inline_data_truncate(inode
, &has_inline
);
4091 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4092 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4093 if (ext4_inode_attach_jinode(inode
) < 0)
4097 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4098 credits
= ext4_writepage_trans_blocks(inode
);
4100 credits
= ext4_blocks_for_truncate(inode
);
4102 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4103 if (IS_ERR(handle
)) {
4104 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
4108 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4109 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4112 * We add the inode to the orphan list, so that if this
4113 * truncate spans multiple transactions, and we crash, we will
4114 * resume the truncate when the filesystem recovers. It also
4115 * marks the inode dirty, to catch the new size.
4117 * Implication: the file must always be in a sane, consistent
4118 * truncatable state while each transaction commits.
4120 if (ext4_orphan_add(handle
, inode
))
4123 down_write(&EXT4_I(inode
)->i_data_sem
);
4125 ext4_discard_preallocations(inode
);
4127 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4128 ext4_ext_truncate(handle
, inode
);
4130 ext4_ind_truncate(handle
, inode
);
4132 up_write(&ei
->i_data_sem
);
4135 ext4_handle_sync(handle
);
4139 * If this was a simple ftruncate() and the file will remain alive,
4140 * then we need to clear up the orphan record which we created above.
4141 * However, if this was a real unlink then we were called by
4142 * ext4_evict_inode(), and we allow that function to clean up the
4143 * orphan info for us.
4146 ext4_orphan_del(handle
, inode
);
4148 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4149 ext4_mark_inode_dirty(handle
, inode
);
4150 ext4_journal_stop(handle
);
4152 trace_ext4_truncate_exit(inode
);
4156 * ext4_get_inode_loc returns with an extra refcount against the inode's
4157 * underlying buffer_head on success. If 'in_mem' is true, we have all
4158 * data in memory that is needed to recreate the on-disk version of this
4161 static int __ext4_get_inode_loc(struct inode
*inode
,
4162 struct ext4_iloc
*iloc
, int in_mem
)
4164 struct ext4_group_desc
*gdp
;
4165 struct buffer_head
*bh
;
4166 struct super_block
*sb
= inode
->i_sb
;
4168 int inodes_per_block
, inode_offset
;
4171 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4172 return -EFSCORRUPTED
;
4174 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4175 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4180 * Figure out the offset within the block group inode table
4182 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4183 inode_offset
= ((inode
->i_ino
- 1) %
4184 EXT4_INODES_PER_GROUP(sb
));
4185 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4186 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4188 bh
= sb_getblk(sb
, block
);
4191 if (!buffer_uptodate(bh
)) {
4195 * If the buffer has the write error flag, we have failed
4196 * to write out another inode in the same block. In this
4197 * case, we don't have to read the block because we may
4198 * read the old inode data successfully.
4200 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4201 set_buffer_uptodate(bh
);
4203 if (buffer_uptodate(bh
)) {
4204 /* someone brought it uptodate while we waited */
4210 * If we have all information of the inode in memory and this
4211 * is the only valid inode in the block, we need not read the
4215 struct buffer_head
*bitmap_bh
;
4218 start
= inode_offset
& ~(inodes_per_block
- 1);
4220 /* Is the inode bitmap in cache? */
4221 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4222 if (unlikely(!bitmap_bh
))
4226 * If the inode bitmap isn't in cache then the
4227 * optimisation may end up performing two reads instead
4228 * of one, so skip it.
4230 if (!buffer_uptodate(bitmap_bh
)) {
4234 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4235 if (i
== inode_offset
)
4237 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4241 if (i
== start
+ inodes_per_block
) {
4242 /* all other inodes are free, so skip I/O */
4243 memset(bh
->b_data
, 0, bh
->b_size
);
4244 set_buffer_uptodate(bh
);
4252 * If we need to do any I/O, try to pre-readahead extra
4253 * blocks from the inode table.
4255 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4256 ext4_fsblk_t b
, end
, table
;
4258 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4260 table
= ext4_inode_table(sb
, gdp
);
4261 /* s_inode_readahead_blks is always a power of 2 */
4262 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4266 num
= EXT4_INODES_PER_GROUP(sb
);
4267 if (ext4_has_group_desc_csum(sb
))
4268 num
-= ext4_itable_unused_count(sb
, gdp
);
4269 table
+= num
/ inodes_per_block
;
4273 sb_breadahead(sb
, b
++);
4277 * There are other valid inodes in the buffer, this inode
4278 * has in-inode xattrs, or we don't have this inode in memory.
4279 * Read the block from disk.
4281 trace_ext4_load_inode(inode
);
4283 bh
->b_end_io
= end_buffer_read_sync
;
4284 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4286 if (!buffer_uptodate(bh
)) {
4287 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4288 "unable to read itable block");
4298 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4300 /* We have all inode data except xattrs in memory here. */
4301 return __ext4_get_inode_loc(inode
, iloc
,
4302 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4305 void ext4_set_inode_flags(struct inode
*inode
)
4307 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4308 unsigned int new_fl
= 0;
4310 if (flags
& EXT4_SYNC_FL
)
4312 if (flags
& EXT4_APPEND_FL
)
4314 if (flags
& EXT4_IMMUTABLE_FL
)
4315 new_fl
|= S_IMMUTABLE
;
4316 if (flags
& EXT4_NOATIME_FL
)
4317 new_fl
|= S_NOATIME
;
4318 if (flags
& EXT4_DIRSYNC_FL
)
4319 new_fl
|= S_DIRSYNC
;
4320 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
))
4322 inode_set_flags(inode
, new_fl
,
4323 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4326 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4327 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4329 unsigned int vfs_fl
;
4330 unsigned long old_fl
, new_fl
;
4333 vfs_fl
= ei
->vfs_inode
.i_flags
;
4334 old_fl
= ei
->i_flags
;
4335 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4336 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4338 if (vfs_fl
& S_SYNC
)
4339 new_fl
|= EXT4_SYNC_FL
;
4340 if (vfs_fl
& S_APPEND
)
4341 new_fl
|= EXT4_APPEND_FL
;
4342 if (vfs_fl
& S_IMMUTABLE
)
4343 new_fl
|= EXT4_IMMUTABLE_FL
;
4344 if (vfs_fl
& S_NOATIME
)
4345 new_fl
|= EXT4_NOATIME_FL
;
4346 if (vfs_fl
& S_DIRSYNC
)
4347 new_fl
|= EXT4_DIRSYNC_FL
;
4348 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4351 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4352 struct ext4_inode_info
*ei
)
4355 struct inode
*inode
= &(ei
->vfs_inode
);
4356 struct super_block
*sb
= inode
->i_sb
;
4358 if (ext4_has_feature_huge_file(sb
)) {
4359 /* we are using combined 48 bit field */
4360 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4361 le32_to_cpu(raw_inode
->i_blocks_lo
);
4362 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4363 /* i_blocks represent file system block size */
4364 return i_blocks
<< (inode
->i_blkbits
- 9);
4369 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4373 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4374 struct ext4_inode
*raw_inode
,
4375 struct ext4_inode_info
*ei
)
4377 __le32
*magic
= (void *)raw_inode
+
4378 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4379 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4380 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4381 ext4_find_inline_data_nolock(inode
);
4383 EXT4_I(inode
)->i_inline_off
= 0;
4386 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4388 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
, EXT4_FEATURE_RO_COMPAT_PROJECT
))
4390 *projid
= EXT4_I(inode
)->i_projid
;
4394 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4396 struct ext4_iloc iloc
;
4397 struct ext4_inode
*raw_inode
;
4398 struct ext4_inode_info
*ei
;
4399 struct inode
*inode
;
4400 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4407 inode
= iget_locked(sb
, ino
);
4409 return ERR_PTR(-ENOMEM
);
4410 if (!(inode
->i_state
& I_NEW
))
4416 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4419 raw_inode
= ext4_raw_inode(&iloc
);
4421 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4422 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4423 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4424 EXT4_INODE_SIZE(inode
->i_sb
)) {
4425 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4426 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4427 EXT4_INODE_SIZE(inode
->i_sb
));
4428 ret
= -EFSCORRUPTED
;
4432 ei
->i_extra_isize
= 0;
4434 /* Precompute checksum seed for inode metadata */
4435 if (ext4_has_metadata_csum(sb
)) {
4436 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4438 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4439 __le32 gen
= raw_inode
->i_generation
;
4440 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4442 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4446 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4447 EXT4_ERROR_INODE(inode
, "checksum invalid");
4452 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4453 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4454 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4455 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_PROJECT
) &&
4456 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4457 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4458 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4460 i_projid
= EXT4_DEF_PROJID
;
4462 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4463 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4464 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4466 i_uid_write(inode
, i_uid
);
4467 i_gid_write(inode
, i_gid
);
4468 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4469 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4471 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4472 ei
->i_inline_off
= 0;
4473 ei
->i_dir_start_lookup
= 0;
4474 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4475 /* We now have enough fields to check if the inode was active or not.
4476 * This is needed because nfsd might try to access dead inodes
4477 * the test is that same one that e2fsck uses
4478 * NeilBrown 1999oct15
4480 if (inode
->i_nlink
== 0) {
4481 if ((inode
->i_mode
== 0 ||
4482 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4483 ino
!= EXT4_BOOT_LOADER_INO
) {
4484 /* this inode is deleted */
4488 /* The only unlinked inodes we let through here have
4489 * valid i_mode and are being read by the orphan
4490 * recovery code: that's fine, we're about to complete
4491 * the process of deleting those.
4492 * OR it is the EXT4_BOOT_LOADER_INO which is
4493 * not initialized on a new filesystem. */
4495 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4496 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4497 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4498 if (ext4_has_feature_64bit(sb
))
4500 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4501 inode
->i_size
= ext4_isize(raw_inode
);
4502 ei
->i_disksize
= inode
->i_size
;
4504 ei
->i_reserved_quota
= 0;
4506 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4507 ei
->i_block_group
= iloc
.block_group
;
4508 ei
->i_last_alloc_group
= ~0;
4510 * NOTE! The in-memory inode i_data array is in little-endian order
4511 * even on big-endian machines: we do NOT byteswap the block numbers!
4513 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4514 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4515 INIT_LIST_HEAD(&ei
->i_orphan
);
4518 * Set transaction id's of transactions that have to be committed
4519 * to finish f[data]sync. We set them to currently running transaction
4520 * as we cannot be sure that the inode or some of its metadata isn't
4521 * part of the transaction - the inode could have been reclaimed and
4522 * now it is reread from disk.
4525 transaction_t
*transaction
;
4528 read_lock(&journal
->j_state_lock
);
4529 if (journal
->j_running_transaction
)
4530 transaction
= journal
->j_running_transaction
;
4532 transaction
= journal
->j_committing_transaction
;
4534 tid
= transaction
->t_tid
;
4536 tid
= journal
->j_commit_sequence
;
4537 read_unlock(&journal
->j_state_lock
);
4538 ei
->i_sync_tid
= tid
;
4539 ei
->i_datasync_tid
= tid
;
4542 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4543 if (ei
->i_extra_isize
== 0) {
4544 /* The extra space is currently unused. Use it. */
4545 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4546 EXT4_GOOD_OLD_INODE_SIZE
;
4548 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4552 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4553 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4554 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4555 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4557 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4558 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4559 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4560 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4562 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4567 if (ei
->i_file_acl
&&
4568 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4569 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4571 ret
= -EFSCORRUPTED
;
4573 } else if (!ext4_has_inline_data(inode
)) {
4574 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4575 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4576 (S_ISLNK(inode
->i_mode
) &&
4577 !ext4_inode_is_fast_symlink(inode
))))
4578 /* Validate extent which is part of inode */
4579 ret
= ext4_ext_check_inode(inode
);
4580 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4581 (S_ISLNK(inode
->i_mode
) &&
4582 !ext4_inode_is_fast_symlink(inode
))) {
4583 /* Validate block references which are part of inode */
4584 ret
= ext4_ind_check_inode(inode
);
4590 if (S_ISREG(inode
->i_mode
)) {
4591 inode
->i_op
= &ext4_file_inode_operations
;
4592 inode
->i_fop
= &ext4_file_operations
;
4593 ext4_set_aops(inode
);
4594 } else if (S_ISDIR(inode
->i_mode
)) {
4595 inode
->i_op
= &ext4_dir_inode_operations
;
4596 inode
->i_fop
= &ext4_dir_operations
;
4597 } else if (S_ISLNK(inode
->i_mode
)) {
4598 if (ext4_encrypted_inode(inode
)) {
4599 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4600 ext4_set_aops(inode
);
4601 } else if (ext4_inode_is_fast_symlink(inode
)) {
4602 inode
->i_link
= (char *)ei
->i_data
;
4603 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4604 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4605 sizeof(ei
->i_data
) - 1);
4607 inode
->i_op
= &ext4_symlink_inode_operations
;
4608 ext4_set_aops(inode
);
4610 inode_nohighmem(inode
);
4611 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4612 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4613 inode
->i_op
= &ext4_special_inode_operations
;
4614 if (raw_inode
->i_block
[0])
4615 init_special_inode(inode
, inode
->i_mode
,
4616 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4618 init_special_inode(inode
, inode
->i_mode
,
4619 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4620 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4621 make_bad_inode(inode
);
4623 ret
= -EFSCORRUPTED
;
4624 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4628 ext4_set_inode_flags(inode
);
4629 unlock_new_inode(inode
);
4635 return ERR_PTR(ret
);
4638 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4640 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4641 return ERR_PTR(-EFSCORRUPTED
);
4642 return ext4_iget(sb
, ino
);
4645 static int ext4_inode_blocks_set(handle_t
*handle
,
4646 struct ext4_inode
*raw_inode
,
4647 struct ext4_inode_info
*ei
)
4649 struct inode
*inode
= &(ei
->vfs_inode
);
4650 u64 i_blocks
= inode
->i_blocks
;
4651 struct super_block
*sb
= inode
->i_sb
;
4653 if (i_blocks
<= ~0U) {
4655 * i_blocks can be represented in a 32 bit variable
4656 * as multiple of 512 bytes
4658 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4659 raw_inode
->i_blocks_high
= 0;
4660 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4663 if (!ext4_has_feature_huge_file(sb
))
4666 if (i_blocks
<= 0xffffffffffffULL
) {
4668 * i_blocks can be represented in a 48 bit variable
4669 * as multiple of 512 bytes
4671 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4672 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4673 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4675 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4676 /* i_block is stored in file system block size */
4677 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4678 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4679 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4684 struct other_inode
{
4685 unsigned long orig_ino
;
4686 struct ext4_inode
*raw_inode
;
4689 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4692 struct other_inode
*oi
= (struct other_inode
*) data
;
4694 if ((inode
->i_ino
!= ino
) ||
4695 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4696 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4697 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4699 spin_lock(&inode
->i_lock
);
4700 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4701 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4702 (inode
->i_state
& I_DIRTY_TIME
)) {
4703 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4705 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4706 spin_unlock(&inode
->i_lock
);
4708 spin_lock(&ei
->i_raw_lock
);
4709 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4710 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4711 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4712 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4713 spin_unlock(&ei
->i_raw_lock
);
4714 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4717 spin_unlock(&inode
->i_lock
);
4722 * Opportunistically update the other time fields for other inodes in
4723 * the same inode table block.
4725 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4726 unsigned long orig_ino
, char *buf
)
4728 struct other_inode oi
;
4730 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4731 int inode_size
= EXT4_INODE_SIZE(sb
);
4733 oi
.orig_ino
= orig_ino
;
4735 * Calculate the first inode in the inode table block. Inode
4736 * numbers are one-based. That is, the first inode in a block
4737 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4739 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4740 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4741 if (ino
== orig_ino
)
4743 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4744 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4749 * Post the struct inode info into an on-disk inode location in the
4750 * buffer-cache. This gobbles the caller's reference to the
4751 * buffer_head in the inode location struct.
4753 * The caller must have write access to iloc->bh.
4755 static int ext4_do_update_inode(handle_t
*handle
,
4756 struct inode
*inode
,
4757 struct ext4_iloc
*iloc
)
4759 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4760 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4761 struct buffer_head
*bh
= iloc
->bh
;
4762 struct super_block
*sb
= inode
->i_sb
;
4763 int err
= 0, rc
, block
;
4764 int need_datasync
= 0, set_large_file
= 0;
4769 spin_lock(&ei
->i_raw_lock
);
4771 /* For fields not tracked in the in-memory inode,
4772 * initialise them to zero for new inodes. */
4773 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4774 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4776 ext4_get_inode_flags(ei
);
4777 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4778 i_uid
= i_uid_read(inode
);
4779 i_gid
= i_gid_read(inode
);
4780 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4781 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4782 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4783 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4785 * Fix up interoperability with old kernels. Otherwise, old inodes get
4786 * re-used with the upper 16 bits of the uid/gid intact
4789 raw_inode
->i_uid_high
=
4790 cpu_to_le16(high_16_bits(i_uid
));
4791 raw_inode
->i_gid_high
=
4792 cpu_to_le16(high_16_bits(i_gid
));
4794 raw_inode
->i_uid_high
= 0;
4795 raw_inode
->i_gid_high
= 0;
4798 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4799 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4800 raw_inode
->i_uid_high
= 0;
4801 raw_inode
->i_gid_high
= 0;
4803 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4805 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4806 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4807 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4808 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4810 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4812 spin_unlock(&ei
->i_raw_lock
);
4815 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4816 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4817 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4818 raw_inode
->i_file_acl_high
=
4819 cpu_to_le16(ei
->i_file_acl
>> 32);
4820 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4821 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4822 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4825 if (ei
->i_disksize
> 0x7fffffffULL
) {
4826 if (!ext4_has_feature_large_file(sb
) ||
4827 EXT4_SB(sb
)->s_es
->s_rev_level
==
4828 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4831 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4832 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4833 if (old_valid_dev(inode
->i_rdev
)) {
4834 raw_inode
->i_block
[0] =
4835 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4836 raw_inode
->i_block
[1] = 0;
4838 raw_inode
->i_block
[0] = 0;
4839 raw_inode
->i_block
[1] =
4840 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4841 raw_inode
->i_block
[2] = 0;
4843 } else if (!ext4_has_inline_data(inode
)) {
4844 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4845 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4848 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4849 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4850 if (ei
->i_extra_isize
) {
4851 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4852 raw_inode
->i_version_hi
=
4853 cpu_to_le32(inode
->i_version
>> 32);
4854 raw_inode
->i_extra_isize
=
4855 cpu_to_le16(ei
->i_extra_isize
);
4859 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
4860 EXT4_FEATURE_RO_COMPAT_PROJECT
) &&
4861 i_projid
!= EXT4_DEF_PROJID
);
4863 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4864 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4865 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4867 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4868 spin_unlock(&ei
->i_raw_lock
);
4869 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4870 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4873 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4874 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4877 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4878 if (set_large_file
) {
4879 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4880 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4883 ext4_update_dynamic_rev(sb
);
4884 ext4_set_feature_large_file(sb
);
4885 ext4_handle_sync(handle
);
4886 err
= ext4_handle_dirty_super(handle
, sb
);
4888 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4891 ext4_std_error(inode
->i_sb
, err
);
4896 * ext4_write_inode()
4898 * We are called from a few places:
4900 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4901 * Here, there will be no transaction running. We wait for any running
4902 * transaction to commit.
4904 * - Within flush work (sys_sync(), kupdate and such).
4905 * We wait on commit, if told to.
4907 * - Within iput_final() -> write_inode_now()
4908 * We wait on commit, if told to.
4910 * In all cases it is actually safe for us to return without doing anything,
4911 * because the inode has been copied into a raw inode buffer in
4912 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4915 * Note that we are absolutely dependent upon all inode dirtiers doing the
4916 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4917 * which we are interested.
4919 * It would be a bug for them to not do this. The code:
4921 * mark_inode_dirty(inode)
4923 * inode->i_size = expr;
4925 * is in error because write_inode() could occur while `stuff()' is running,
4926 * and the new i_size will be lost. Plus the inode will no longer be on the
4927 * superblock's dirty inode list.
4929 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4933 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4936 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4937 if (ext4_journal_current_handle()) {
4938 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4944 * No need to force transaction in WB_SYNC_NONE mode. Also
4945 * ext4_sync_fs() will force the commit after everything is
4948 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4951 err
= ext4_force_commit(inode
->i_sb
);
4953 struct ext4_iloc iloc
;
4955 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4959 * sync(2) will flush the whole buffer cache. No need to do
4960 * it here separately for each inode.
4962 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4963 sync_dirty_buffer(iloc
.bh
);
4964 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4965 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4966 "IO error syncing inode");
4975 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4976 * buffers that are attached to a page stradding i_size and are undergoing
4977 * commit. In that case we have to wait for commit to finish and try again.
4979 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4983 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4984 tid_t commit_tid
= 0;
4987 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
4989 * All buffers in the last page remain valid? Then there's nothing to
4990 * do. We do the check mainly to optimize the common PAGE_SIZE ==
4993 if (offset
> PAGE_SIZE
- (1 << inode
->i_blkbits
))
4996 page
= find_lock_page(inode
->i_mapping
,
4997 inode
->i_size
>> PAGE_SHIFT
);
5000 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5001 PAGE_SIZE
- offset
);
5007 read_lock(&journal
->j_state_lock
);
5008 if (journal
->j_committing_transaction
)
5009 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5010 read_unlock(&journal
->j_state_lock
);
5012 jbd2_log_wait_commit(journal
, commit_tid
);
5019 * Called from notify_change.
5021 * We want to trap VFS attempts to truncate the file as soon as
5022 * possible. In particular, we want to make sure that when the VFS
5023 * shrinks i_size, we put the inode on the orphan list and modify
5024 * i_disksize immediately, so that during the subsequent flushing of
5025 * dirty pages and freeing of disk blocks, we can guarantee that any
5026 * commit will leave the blocks being flushed in an unused state on
5027 * disk. (On recovery, the inode will get truncated and the blocks will
5028 * be freed, so we have a strong guarantee that no future commit will
5029 * leave these blocks visible to the user.)
5031 * Another thing we have to assure is that if we are in ordered mode
5032 * and inode is still attached to the committing transaction, we must
5033 * we start writeout of all the dirty pages which are being truncated.
5034 * This way we are sure that all the data written in the previous
5035 * transaction are already on disk (truncate waits for pages under
5038 * Called with inode->i_mutex down.
5040 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5042 struct inode
*inode
= d_inode(dentry
);
5045 const unsigned int ia_valid
= attr
->ia_valid
;
5047 error
= inode_change_ok(inode
, attr
);
5051 if (is_quota_modification(inode
, attr
)) {
5052 error
= dquot_initialize(inode
);
5056 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5057 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5060 /* (user+group)*(old+new) structure, inode write (sb,
5061 * inode block, ? - but truncate inode update has it) */
5062 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5063 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5064 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5065 if (IS_ERR(handle
)) {
5066 error
= PTR_ERR(handle
);
5069 error
= dquot_transfer(inode
, attr
);
5071 ext4_journal_stop(handle
);
5074 /* Update corresponding info in inode so that everything is in
5075 * one transaction */
5076 if (attr
->ia_valid
& ATTR_UID
)
5077 inode
->i_uid
= attr
->ia_uid
;
5078 if (attr
->ia_valid
& ATTR_GID
)
5079 inode
->i_gid
= attr
->ia_gid
;
5080 error
= ext4_mark_inode_dirty(handle
, inode
);
5081 ext4_journal_stop(handle
);
5084 if (attr
->ia_valid
& ATTR_SIZE
) {
5086 loff_t oldsize
= inode
->i_size
;
5087 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5089 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5090 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5092 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5095 if (!S_ISREG(inode
->i_mode
))
5098 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5099 inode_inc_iversion(inode
);
5101 if (ext4_should_order_data(inode
) &&
5102 (attr
->ia_size
< inode
->i_size
)) {
5103 error
= ext4_begin_ordered_truncate(inode
,
5108 if (attr
->ia_size
!= inode
->i_size
) {
5109 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5110 if (IS_ERR(handle
)) {
5111 error
= PTR_ERR(handle
);
5114 if (ext4_handle_valid(handle
) && shrink
) {
5115 error
= ext4_orphan_add(handle
, inode
);
5119 * Update c/mtime on truncate up, ext4_truncate() will
5120 * update c/mtime in shrink case below
5123 inode
->i_mtime
= ext4_current_time(inode
);
5124 inode
->i_ctime
= inode
->i_mtime
;
5126 down_write(&EXT4_I(inode
)->i_data_sem
);
5127 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5128 rc
= ext4_mark_inode_dirty(handle
, inode
);
5132 * We have to update i_size under i_data_sem together
5133 * with i_disksize to avoid races with writeback code
5134 * running ext4_wb_update_i_disksize().
5137 i_size_write(inode
, attr
->ia_size
);
5138 up_write(&EXT4_I(inode
)->i_data_sem
);
5139 ext4_journal_stop(handle
);
5142 ext4_orphan_del(NULL
, inode
);
5147 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5150 * Blocks are going to be removed from the inode. Wait
5151 * for dio in flight. Temporarily disable
5152 * dioread_nolock to prevent livelock.
5155 if (!ext4_should_journal_data(inode
)) {
5156 ext4_inode_block_unlocked_dio(inode
);
5157 inode_dio_wait(inode
);
5158 ext4_inode_resume_unlocked_dio(inode
);
5160 ext4_wait_for_tail_page_commit(inode
);
5162 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5164 * Truncate pagecache after we've waited for commit
5165 * in data=journal mode to make pages freeable.
5167 truncate_pagecache(inode
, inode
->i_size
);
5169 ext4_truncate(inode
);
5170 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5174 setattr_copy(inode
, attr
);
5175 mark_inode_dirty(inode
);
5179 * If the call to ext4_truncate failed to get a transaction handle at
5180 * all, we need to clean up the in-core orphan list manually.
5182 if (orphan
&& inode
->i_nlink
)
5183 ext4_orphan_del(NULL
, inode
);
5185 if (!rc
&& (ia_valid
& ATTR_MODE
))
5186 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5189 ext4_std_error(inode
->i_sb
, error
);
5195 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5198 struct inode
*inode
;
5199 unsigned long long delalloc_blocks
;
5201 inode
= d_inode(dentry
);
5202 generic_fillattr(inode
, stat
);
5205 * If there is inline data in the inode, the inode will normally not
5206 * have data blocks allocated (it may have an external xattr block).
5207 * Report at least one sector for such files, so tools like tar, rsync,
5208 * others doen't incorrectly think the file is completely sparse.
5210 if (unlikely(ext4_has_inline_data(inode
)))
5211 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5214 * We can't update i_blocks if the block allocation is delayed
5215 * otherwise in the case of system crash before the real block
5216 * allocation is done, we will have i_blocks inconsistent with
5217 * on-disk file blocks.
5218 * We always keep i_blocks updated together with real
5219 * allocation. But to not confuse with user, stat
5220 * will return the blocks that include the delayed allocation
5221 * blocks for this file.
5223 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5224 EXT4_I(inode
)->i_reserved_data_blocks
);
5225 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5229 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5232 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5233 return ext4_ind_trans_blocks(inode
, lblocks
);
5234 return ext4_ext_index_trans_blocks(inode
, pextents
);
5238 * Account for index blocks, block groups bitmaps and block group
5239 * descriptor blocks if modify datablocks and index blocks
5240 * worse case, the indexs blocks spread over different block groups
5242 * If datablocks are discontiguous, they are possible to spread over
5243 * different block groups too. If they are contiguous, with flexbg,
5244 * they could still across block group boundary.
5246 * Also account for superblock, inode, quota and xattr blocks
5248 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5251 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5257 * How many index blocks need to touch to map @lblocks logical blocks
5258 * to @pextents physical extents?
5260 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5265 * Now let's see how many group bitmaps and group descriptors need
5268 groups
= idxblocks
+ pextents
;
5270 if (groups
> ngroups
)
5272 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5273 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5275 /* bitmaps and block group descriptor blocks */
5276 ret
+= groups
+ gdpblocks
;
5278 /* Blocks for super block, inode, quota and xattr blocks */
5279 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5285 * Calculate the total number of credits to reserve to fit
5286 * the modification of a single pages into a single transaction,
5287 * which may include multiple chunks of block allocations.
5289 * This could be called via ext4_write_begin()
5291 * We need to consider the worse case, when
5292 * one new block per extent.
5294 int ext4_writepage_trans_blocks(struct inode
*inode
)
5296 int bpp
= ext4_journal_blocks_per_page(inode
);
5299 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5301 /* Account for data blocks for journalled mode */
5302 if (ext4_should_journal_data(inode
))
5308 * Calculate the journal credits for a chunk of data modification.
5310 * This is called from DIO, fallocate or whoever calling
5311 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5313 * journal buffers for data blocks are not included here, as DIO
5314 * and fallocate do no need to journal data buffers.
5316 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5318 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5322 * The caller must have previously called ext4_reserve_inode_write().
5323 * Give this, we know that the caller already has write access to iloc->bh.
5325 int ext4_mark_iloc_dirty(handle_t
*handle
,
5326 struct inode
*inode
, struct ext4_iloc
*iloc
)
5330 if (IS_I_VERSION(inode
))
5331 inode_inc_iversion(inode
);
5333 /* the do_update_inode consumes one bh->b_count */
5336 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5337 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5343 * On success, We end up with an outstanding reference count against
5344 * iloc->bh. This _must_ be cleaned up later.
5348 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5349 struct ext4_iloc
*iloc
)
5353 err
= ext4_get_inode_loc(inode
, iloc
);
5355 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5356 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5362 ext4_std_error(inode
->i_sb
, err
);
5367 * Expand an inode by new_extra_isize bytes.
5368 * Returns 0 on success or negative error number on failure.
5370 static int ext4_expand_extra_isize(struct inode
*inode
,
5371 unsigned int new_extra_isize
,
5372 struct ext4_iloc iloc
,
5375 struct ext4_inode
*raw_inode
;
5376 struct ext4_xattr_ibody_header
*header
;
5378 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5381 raw_inode
= ext4_raw_inode(&iloc
);
5383 header
= IHDR(inode
, raw_inode
);
5385 /* No extended attributes present */
5386 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5387 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5388 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5390 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5394 /* try to expand with EAs present */
5395 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5400 * What we do here is to mark the in-core inode as clean with respect to inode
5401 * dirtiness (it may still be data-dirty).
5402 * This means that the in-core inode may be reaped by prune_icache
5403 * without having to perform any I/O. This is a very good thing,
5404 * because *any* task may call prune_icache - even ones which
5405 * have a transaction open against a different journal.
5407 * Is this cheating? Not really. Sure, we haven't written the
5408 * inode out, but prune_icache isn't a user-visible syncing function.
5409 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5410 * we start and wait on commits.
5412 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5414 struct ext4_iloc iloc
;
5415 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5416 static unsigned int mnt_count
;
5420 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5421 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5424 if (ext4_handle_valid(handle
) &&
5425 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5426 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5428 * We need extra buffer credits since we may write into EA block
5429 * with this same handle. If journal_extend fails, then it will
5430 * only result in a minor loss of functionality for that inode.
5431 * If this is felt to be critical, then e2fsck should be run to
5432 * force a large enough s_min_extra_isize.
5434 if ((jbd2_journal_extend(handle
,
5435 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5436 ret
= ext4_expand_extra_isize(inode
,
5437 sbi
->s_want_extra_isize
,
5440 ext4_set_inode_state(inode
,
5441 EXT4_STATE_NO_EXPAND
);
5443 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5444 ext4_warning(inode
->i_sb
,
5445 "Unable to expand inode %lu. Delete"
5446 " some EAs or run e2fsck.",
5449 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5454 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5458 * ext4_dirty_inode() is called from __mark_inode_dirty()
5460 * We're really interested in the case where a file is being extended.
5461 * i_size has been changed by generic_commit_write() and we thus need
5462 * to include the updated inode in the current transaction.
5464 * Also, dquot_alloc_block() will always dirty the inode when blocks
5465 * are allocated to the file.
5467 * If the inode is marked synchronous, we don't honour that here - doing
5468 * so would cause a commit on atime updates, which we don't bother doing.
5469 * We handle synchronous inodes at the highest possible level.
5471 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5472 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5473 * to copy into the on-disk inode structure are the timestamp files.
5475 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5479 if (flags
== I_DIRTY_TIME
)
5481 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5485 ext4_mark_inode_dirty(handle
, inode
);
5487 ext4_journal_stop(handle
);
5494 * Bind an inode's backing buffer_head into this transaction, to prevent
5495 * it from being flushed to disk early. Unlike
5496 * ext4_reserve_inode_write, this leaves behind no bh reference and
5497 * returns no iloc structure, so the caller needs to repeat the iloc
5498 * lookup to mark the inode dirty later.
5500 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5502 struct ext4_iloc iloc
;
5506 err
= ext4_get_inode_loc(inode
, &iloc
);
5508 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5509 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5511 err
= ext4_handle_dirty_metadata(handle
,
5517 ext4_std_error(inode
->i_sb
, err
);
5522 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5527 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5530 * We have to be very careful here: changing a data block's
5531 * journaling status dynamically is dangerous. If we write a
5532 * data block to the journal, change the status and then delete
5533 * that block, we risk forgetting to revoke the old log record
5534 * from the journal and so a subsequent replay can corrupt data.
5535 * So, first we make sure that the journal is empty and that
5536 * nobody is changing anything.
5539 journal
= EXT4_JOURNAL(inode
);
5542 if (is_journal_aborted(journal
))
5545 /* Wait for all existing dio workers */
5546 ext4_inode_block_unlocked_dio(inode
);
5547 inode_dio_wait(inode
);
5550 * Before flushing the journal and switching inode's aops, we have
5551 * to flush all dirty data the inode has. There can be outstanding
5552 * delayed allocations, there can be unwritten extents created by
5553 * fallocate or buffered writes in dioread_nolock mode covered by
5554 * dirty data which can be converted only after flushing the dirty
5555 * data (and journalled aops don't know how to handle these cases).
5558 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5559 err
= filemap_write_and_wait(inode
->i_mapping
);
5561 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5562 ext4_inode_resume_unlocked_dio(inode
);
5567 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5568 jbd2_journal_lock_updates(journal
);
5571 * OK, there are no updates running now, and all cached data is
5572 * synced to disk. We are now in a completely consistent state
5573 * which doesn't have anything in the journal, and we know that
5574 * no filesystem updates are running, so it is safe to modify
5575 * the inode's in-core data-journaling state flag now.
5579 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5581 err
= jbd2_journal_flush(journal
);
5583 jbd2_journal_unlock_updates(journal
);
5584 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5585 ext4_inode_resume_unlocked_dio(inode
);
5588 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5590 ext4_set_aops(inode
);
5592 jbd2_journal_unlock_updates(journal
);
5593 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5596 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5597 ext4_inode_resume_unlocked_dio(inode
);
5599 /* Finally we can mark the inode as dirty. */
5601 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5603 return PTR_ERR(handle
);
5605 err
= ext4_mark_inode_dirty(handle
, inode
);
5606 ext4_handle_sync(handle
);
5607 ext4_journal_stop(handle
);
5608 ext4_std_error(inode
->i_sb
, err
);
5613 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5615 return !buffer_mapped(bh
);
5618 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5620 struct page
*page
= vmf
->page
;
5624 struct file
*file
= vma
->vm_file
;
5625 struct inode
*inode
= file_inode(file
);
5626 struct address_space
*mapping
= inode
->i_mapping
;
5628 get_block_t
*get_block
;
5631 sb_start_pagefault(inode
->i_sb
);
5632 file_update_time(vma
->vm_file
);
5634 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5635 /* Delalloc case is easy... */
5636 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5637 !ext4_should_journal_data(inode
) &&
5638 !ext4_nonda_switch(inode
->i_sb
)) {
5640 ret
= block_page_mkwrite(vma
, vmf
,
5641 ext4_da_get_block_prep
);
5642 } while (ret
== -ENOSPC
&&
5643 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5648 size
= i_size_read(inode
);
5649 /* Page got truncated from under us? */
5650 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5652 ret
= VM_FAULT_NOPAGE
;
5656 if (page
->index
== size
>> PAGE_SHIFT
)
5657 len
= size
& ~PAGE_MASK
;
5661 * Return if we have all the buffers mapped. This avoids the need to do
5662 * journal_start/journal_stop which can block and take a long time
5664 if (page_has_buffers(page
)) {
5665 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5667 ext4_bh_unmapped
)) {
5668 /* Wait so that we don't change page under IO */
5669 wait_for_stable_page(page
);
5670 ret
= VM_FAULT_LOCKED
;
5675 /* OK, we need to fill the hole... */
5676 if (ext4_should_dioread_nolock(inode
))
5677 get_block
= ext4_get_block_unwritten
;
5679 get_block
= ext4_get_block
;
5681 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5682 ext4_writepage_trans_blocks(inode
));
5683 if (IS_ERR(handle
)) {
5684 ret
= VM_FAULT_SIGBUS
;
5687 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5688 if (!ret
&& ext4_should_journal_data(inode
)) {
5689 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5690 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
5692 ret
= VM_FAULT_SIGBUS
;
5693 ext4_journal_stop(handle
);
5696 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5698 ext4_journal_stop(handle
);
5699 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5702 ret
= block_page_mkwrite_return(ret
);
5704 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5705 sb_end_pagefault(inode
->i_sb
);
5709 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5711 struct inode
*inode
= file_inode(vma
->vm_file
);
5714 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5715 err
= filemap_fault(vma
, vmf
);
5716 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5722 * Find the first extent at or after @lblk in an inode that is not a hole.
5723 * Search for @map_len blocks at most. The extent is returned in @result.
5725 * The function returns 1 if we found an extent. The function returns 0 in
5726 * case there is no extent at or after @lblk and in that case also sets
5727 * @result->es_len to 0. In case of error, the error code is returned.
5729 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
5730 unsigned int map_len
, struct extent_status
*result
)
5732 struct ext4_map_blocks map
;
5733 struct extent_status es
= {};
5737 map
.m_len
= map_len
;
5740 * For non-extent based files this loop may iterate several times since
5741 * we do not determine full hole size.
5743 while (map
.m_len
> 0) {
5744 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
5747 /* There's extent covering m_lblk? Just return it. */
5751 ext4_es_store_pblock(result
, map
.m_pblk
);
5752 result
->es_lblk
= map
.m_lblk
;
5753 result
->es_len
= map
.m_len
;
5754 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
5755 status
= EXTENT_STATUS_UNWRITTEN
;
5757 status
= EXTENT_STATUS_WRITTEN
;
5758 ext4_es_store_status(result
, status
);
5761 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
5762 map
.m_lblk
+ map
.m_len
- 1,
5764 /* Is delalloc data before next block in extent tree? */
5765 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
5766 ext4_lblk_t offset
= 0;
5768 if (es
.es_lblk
< lblk
)
5769 offset
= lblk
- es
.es_lblk
;
5770 result
->es_lblk
= es
.es_lblk
+ offset
;
5771 ext4_es_store_pblock(result
,
5772 ext4_es_pblock(&es
) + offset
);
5773 result
->es_len
= es
.es_len
- offset
;
5774 ext4_es_store_status(result
, ext4_es_status(&es
));
5778 /* There's a hole at m_lblk, advance us after it */
5779 map
.m_lblk
+= map
.m_len
;
5780 map_len
-= map
.m_len
;
5781 map
.m_len
= map_len
;