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
);
692 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
693 * we have to be careful as someone else may be manipulating b_state as well.
695 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
697 unsigned long old_state
;
698 unsigned long new_state
;
700 flags
&= EXT4_MAP_FLAGS
;
702 /* Dummy buffer_head? Set non-atomically. */
704 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
708 * Someone else may be modifying b_state. Be careful! This is ugly but
709 * once we get rid of using bh as a container for mapping information
710 * to pass to / from get_block functions, this can go away.
713 old_state
= READ_ONCE(bh
->b_state
);
714 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
716 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
719 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
720 struct buffer_head
*bh
, int flags
)
722 struct ext4_map_blocks map
;
725 if (ext4_has_inline_data(inode
))
729 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
731 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
734 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
735 ext4_update_bh_state(bh
, map
.m_flags
);
736 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
742 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
743 struct buffer_head
*bh
, int create
)
745 return _ext4_get_block(inode
, iblock
, bh
,
746 create
? EXT4_GET_BLOCKS_CREATE
: 0);
750 * Get block function used when preparing for buffered write if we require
751 * creating an unwritten extent if blocks haven't been allocated. The extent
752 * will be converted to written after the IO is complete.
754 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
755 struct buffer_head
*bh_result
, int create
)
757 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
758 inode
->i_ino
, create
);
759 return _ext4_get_block(inode
, iblock
, bh_result
,
760 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
763 /* Maximum number of blocks we map for direct IO at once. */
764 #define DIO_MAX_BLOCKS 4096
766 static handle_t
*start_dio_trans(struct inode
*inode
,
767 struct buffer_head
*bh_result
)
771 /* Trim mapping request to maximum we can map at once for DIO */
772 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
773 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
774 dio_credits
= ext4_chunk_trans_blocks(inode
,
775 bh_result
->b_size
>> inode
->i_blkbits
);
776 return ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
779 /* Get block function for DIO reads and writes to inodes without extents */
780 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
781 struct buffer_head
*bh
, int create
)
786 /* We don't expect handle for direct IO */
787 WARN_ON_ONCE(ext4_journal_current_handle());
790 handle
= start_dio_trans(inode
, bh
);
792 return PTR_ERR(handle
);
794 ret
= _ext4_get_block(inode
, iblock
, bh
,
795 create
? EXT4_GET_BLOCKS_CREATE
: 0);
797 ext4_journal_stop(handle
);
802 * Get block function for AIO DIO writes when we create unwritten extent if
803 * blocks are not allocated yet. The extent will be converted to written
804 * after IO is complete.
806 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
807 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
812 /* We don't expect handle for direct IO */
813 WARN_ON_ONCE(ext4_journal_current_handle());
815 handle
= start_dio_trans(inode
, bh_result
);
817 return PTR_ERR(handle
);
818 ret
= _ext4_get_block(inode
, iblock
, bh_result
,
819 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
820 ext4_journal_stop(handle
);
823 * When doing DIO using unwritten extents, we need io_end to convert
824 * unwritten extents to written on IO completion. We allocate io_end
825 * once we spot unwritten extent and store it in b_private. Generic
826 * DIO code keeps b_private set and furthermore passes the value to
827 * our completion callback in 'private' argument.
829 if (!ret
&& buffer_unwritten(bh_result
)) {
830 if (!bh_result
->b_private
) {
831 ext4_io_end_t
*io_end
;
833 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
836 bh_result
->b_private
= io_end
;
837 ext4_set_io_unwritten_flag(inode
, io_end
);
839 set_buffer_defer_completion(bh_result
);
846 * Get block function for non-AIO DIO writes when we create unwritten extent if
847 * blocks are not allocated yet. The extent will be converted to written
848 * after IO is complete from ext4_ext_direct_IO() function.
850 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
851 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
856 /* We don't expect handle for direct IO */
857 WARN_ON_ONCE(ext4_journal_current_handle());
859 handle
= start_dio_trans(inode
, bh_result
);
861 return PTR_ERR(handle
);
862 ret
= _ext4_get_block(inode
, iblock
, bh_result
,
863 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
864 ext4_journal_stop(handle
);
867 * Mark inode as having pending DIO writes to unwritten extents.
868 * ext4_ext_direct_IO() checks this flag and converts extents to
871 if (!ret
&& buffer_unwritten(bh_result
))
872 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
877 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
878 struct buffer_head
*bh_result
, int create
)
882 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
883 inode
->i_ino
, create
);
884 /* We don't expect handle for direct IO */
885 WARN_ON_ONCE(ext4_journal_current_handle());
887 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
889 * Blocks should have been preallocated! ext4_file_write_iter() checks
892 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
899 * `handle' can be NULL if create is zero
901 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
902 ext4_lblk_t block
, int map_flags
)
904 struct ext4_map_blocks map
;
905 struct buffer_head
*bh
;
906 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
909 J_ASSERT(handle
!= NULL
|| create
== 0);
913 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
916 return create
? ERR_PTR(-ENOSPC
) : NULL
;
920 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
922 return ERR_PTR(-ENOMEM
);
923 if (map
.m_flags
& EXT4_MAP_NEW
) {
924 J_ASSERT(create
!= 0);
925 J_ASSERT(handle
!= NULL
);
928 * Now that we do not always journal data, we should
929 * keep in mind whether this should always journal the
930 * new buffer as metadata. For now, regular file
931 * writes use ext4_get_block instead, so it's not a
935 BUFFER_TRACE(bh
, "call get_create_access");
936 err
= ext4_journal_get_create_access(handle
, bh
);
941 if (!buffer_uptodate(bh
)) {
942 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
943 set_buffer_uptodate(bh
);
946 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
947 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
951 BUFFER_TRACE(bh
, "not a new buffer");
958 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
959 ext4_lblk_t block
, int map_flags
)
961 struct buffer_head
*bh
;
963 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
966 if (!bh
|| buffer_uptodate(bh
))
968 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
970 if (buffer_uptodate(bh
))
973 return ERR_PTR(-EIO
);
976 int ext4_walk_page_buffers(handle_t
*handle
,
977 struct buffer_head
*head
,
981 int (*fn
)(handle_t
*handle
,
982 struct buffer_head
*bh
))
984 struct buffer_head
*bh
;
985 unsigned block_start
, block_end
;
986 unsigned blocksize
= head
->b_size
;
988 struct buffer_head
*next
;
990 for (bh
= head
, block_start
= 0;
991 ret
== 0 && (bh
!= head
|| !block_start
);
992 block_start
= block_end
, bh
= next
) {
993 next
= bh
->b_this_page
;
994 block_end
= block_start
+ blocksize
;
995 if (block_end
<= from
|| block_start
>= to
) {
996 if (partial
&& !buffer_uptodate(bh
))
1000 err
= (*fn
)(handle
, bh
);
1008 * To preserve ordering, it is essential that the hole instantiation and
1009 * the data write be encapsulated in a single transaction. We cannot
1010 * close off a transaction and start a new one between the ext4_get_block()
1011 * and the commit_write(). So doing the jbd2_journal_start at the start of
1012 * prepare_write() is the right place.
1014 * Also, this function can nest inside ext4_writepage(). In that case, we
1015 * *know* that ext4_writepage() has generated enough buffer credits to do the
1016 * whole page. So we won't block on the journal in that case, which is good,
1017 * because the caller may be PF_MEMALLOC.
1019 * By accident, ext4 can be reentered when a transaction is open via
1020 * quota file writes. If we were to commit the transaction while thus
1021 * reentered, there can be a deadlock - we would be holding a quota
1022 * lock, and the commit would never complete if another thread had a
1023 * transaction open and was blocking on the quota lock - a ranking
1026 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1027 * will _not_ run commit under these circumstances because handle->h_ref
1028 * is elevated. We'll still have enough credits for the tiny quotafile
1031 int do_journal_get_write_access(handle_t
*handle
,
1032 struct buffer_head
*bh
)
1034 int dirty
= buffer_dirty(bh
);
1037 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1040 * __block_write_begin() could have dirtied some buffers. Clean
1041 * the dirty bit as jbd2_journal_get_write_access() could complain
1042 * otherwise about fs integrity issues. Setting of the dirty bit
1043 * by __block_write_begin() isn't a real problem here as we clear
1044 * the bit before releasing a page lock and thus writeback cannot
1045 * ever write the buffer.
1048 clear_buffer_dirty(bh
);
1049 BUFFER_TRACE(bh
, "get write access");
1050 ret
= ext4_journal_get_write_access(handle
, bh
);
1052 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1056 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1057 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1058 get_block_t
*get_block
)
1060 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1061 unsigned to
= from
+ len
;
1062 struct inode
*inode
= page
->mapping
->host
;
1063 unsigned block_start
, block_end
;
1066 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1068 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1069 bool decrypt
= false;
1071 BUG_ON(!PageLocked(page
));
1072 BUG_ON(from
> PAGE_CACHE_SIZE
);
1073 BUG_ON(to
> PAGE_CACHE_SIZE
);
1076 if (!page_has_buffers(page
))
1077 create_empty_buffers(page
, blocksize
, 0);
1078 head
= page_buffers(page
);
1079 bbits
= ilog2(blocksize
);
1080 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
1082 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1083 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1084 block_end
= block_start
+ blocksize
;
1085 if (block_end
<= from
|| block_start
>= to
) {
1086 if (PageUptodate(page
)) {
1087 if (!buffer_uptodate(bh
))
1088 set_buffer_uptodate(bh
);
1093 clear_buffer_new(bh
);
1094 if (!buffer_mapped(bh
)) {
1095 WARN_ON(bh
->b_size
!= blocksize
);
1096 err
= get_block(inode
, block
, bh
, 1);
1099 if (buffer_new(bh
)) {
1100 unmap_underlying_metadata(bh
->b_bdev
,
1102 if (PageUptodate(page
)) {
1103 clear_buffer_new(bh
);
1104 set_buffer_uptodate(bh
);
1105 mark_buffer_dirty(bh
);
1108 if (block_end
> to
|| block_start
< from
)
1109 zero_user_segments(page
, to
, block_end
,
1114 if (PageUptodate(page
)) {
1115 if (!buffer_uptodate(bh
))
1116 set_buffer_uptodate(bh
);
1119 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1120 !buffer_unwritten(bh
) &&
1121 (block_start
< from
|| block_end
> to
)) {
1122 ll_rw_block(READ
, 1, &bh
);
1124 decrypt
= ext4_encrypted_inode(inode
) &&
1125 S_ISREG(inode
->i_mode
);
1129 * If we issued read requests, let them complete.
1131 while (wait_bh
> wait
) {
1132 wait_on_buffer(*--wait_bh
);
1133 if (!buffer_uptodate(*wait_bh
))
1137 page_zero_new_buffers(page
, from
, to
);
1139 err
= ext4_decrypt(page
);
1144 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1145 loff_t pos
, unsigned len
, unsigned flags
,
1146 struct page
**pagep
, void **fsdata
)
1148 struct inode
*inode
= mapping
->host
;
1149 int ret
, needed_blocks
;
1156 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1158 * Reserve one block more for addition to orphan list in case
1159 * we allocate blocks but write fails for some reason
1161 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1162 index
= pos
>> PAGE_CACHE_SHIFT
;
1163 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1166 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1167 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1176 * grab_cache_page_write_begin() can take a long time if the
1177 * system is thrashing due to memory pressure, or if the page
1178 * is being written back. So grab it first before we start
1179 * the transaction handle. This also allows us to allocate
1180 * the page (if needed) without using GFP_NOFS.
1183 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1189 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1190 if (IS_ERR(handle
)) {
1191 page_cache_release(page
);
1192 return PTR_ERR(handle
);
1196 if (page
->mapping
!= mapping
) {
1197 /* The page got truncated from under us */
1199 page_cache_release(page
);
1200 ext4_journal_stop(handle
);
1203 /* In case writeback began while the page was unlocked */
1204 wait_for_stable_page(page
);
1206 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1207 if (ext4_should_dioread_nolock(inode
))
1208 ret
= ext4_block_write_begin(page
, pos
, len
,
1209 ext4_get_block_unwritten
);
1211 ret
= ext4_block_write_begin(page
, pos
, len
,
1214 if (ext4_should_dioread_nolock(inode
))
1215 ret
= __block_write_begin(page
, pos
, len
,
1216 ext4_get_block_unwritten
);
1218 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1220 if (!ret
&& ext4_should_journal_data(inode
)) {
1221 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1223 do_journal_get_write_access
);
1229 * __block_write_begin may have instantiated a few blocks
1230 * outside i_size. Trim these off again. Don't need
1231 * i_size_read because we hold i_mutex.
1233 * Add inode to orphan list in case we crash before
1236 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1237 ext4_orphan_add(handle
, inode
);
1239 ext4_journal_stop(handle
);
1240 if (pos
+ len
> inode
->i_size
) {
1241 ext4_truncate_failed_write(inode
);
1243 * If truncate failed early the inode might
1244 * still be on the orphan list; we need to
1245 * make sure the inode is removed from the
1246 * orphan list in that case.
1249 ext4_orphan_del(NULL
, inode
);
1252 if (ret
== -ENOSPC
&&
1253 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1255 page_cache_release(page
);
1262 /* For write_end() in data=journal mode */
1263 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1266 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1268 set_buffer_uptodate(bh
);
1269 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1270 clear_buffer_meta(bh
);
1271 clear_buffer_prio(bh
);
1276 * We need to pick up the new inode size which generic_commit_write gave us
1277 * `file' can be NULL - eg, when called from page_symlink().
1279 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1280 * buffers are managed internally.
1282 static int ext4_write_end(struct file
*file
,
1283 struct address_space
*mapping
,
1284 loff_t pos
, unsigned len
, unsigned copied
,
1285 struct page
*page
, void *fsdata
)
1287 handle_t
*handle
= ext4_journal_current_handle();
1288 struct inode
*inode
= mapping
->host
;
1289 loff_t old_size
= inode
->i_size
;
1291 int i_size_changed
= 0;
1293 trace_ext4_write_end(inode
, pos
, len
, copied
);
1294 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1295 ret
= ext4_jbd2_file_inode(handle
, inode
);
1298 page_cache_release(page
);
1303 if (ext4_has_inline_data(inode
)) {
1304 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1310 copied
= block_write_end(file
, mapping
, pos
,
1311 len
, copied
, page
, fsdata
);
1313 * it's important to update i_size while still holding page lock:
1314 * page writeout could otherwise come in and zero beyond i_size.
1316 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1318 page_cache_release(page
);
1321 pagecache_isize_extended(inode
, old_size
, pos
);
1323 * Don't mark the inode dirty under page lock. First, it unnecessarily
1324 * makes the holding time of page lock longer. Second, it forces lock
1325 * ordering of page lock and transaction start for journaling
1329 ext4_mark_inode_dirty(handle
, inode
);
1331 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1332 /* if we have allocated more blocks and copied
1333 * less. We will have blocks allocated outside
1334 * inode->i_size. So truncate them
1336 ext4_orphan_add(handle
, inode
);
1338 ret2
= ext4_journal_stop(handle
);
1342 if (pos
+ len
> inode
->i_size
) {
1343 ext4_truncate_failed_write(inode
);
1345 * If truncate failed early the inode might still be
1346 * on the orphan list; we need to make sure the inode
1347 * is removed from the orphan list in that case.
1350 ext4_orphan_del(NULL
, inode
);
1353 return ret
? ret
: copied
;
1357 * This is a private version of page_zero_new_buffers() which doesn't
1358 * set the buffer to be dirty, since in data=journalled mode we need
1359 * to call ext4_handle_dirty_metadata() instead.
1361 static void zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1363 unsigned int block_start
= 0, block_end
;
1364 struct buffer_head
*head
, *bh
;
1366 bh
= head
= page_buffers(page
);
1368 block_end
= block_start
+ bh
->b_size
;
1369 if (buffer_new(bh
)) {
1370 if (block_end
> from
&& block_start
< to
) {
1371 if (!PageUptodate(page
)) {
1372 unsigned start
, size
;
1374 start
= max(from
, block_start
);
1375 size
= min(to
, block_end
) - start
;
1377 zero_user(page
, start
, size
);
1378 set_buffer_uptodate(bh
);
1380 clear_buffer_new(bh
);
1383 block_start
= block_end
;
1384 bh
= bh
->b_this_page
;
1385 } while (bh
!= head
);
1388 static int ext4_journalled_write_end(struct file
*file
,
1389 struct address_space
*mapping
,
1390 loff_t pos
, unsigned len
, unsigned copied
,
1391 struct page
*page
, void *fsdata
)
1393 handle_t
*handle
= ext4_journal_current_handle();
1394 struct inode
*inode
= mapping
->host
;
1395 loff_t old_size
= inode
->i_size
;
1399 int size_changed
= 0;
1401 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1402 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1405 BUG_ON(!ext4_handle_valid(handle
));
1407 if (ext4_has_inline_data(inode
))
1408 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1412 if (!PageUptodate(page
))
1414 zero_new_buffers(page
, from
+copied
, to
);
1417 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1418 to
, &partial
, write_end_fn
);
1420 SetPageUptodate(page
);
1422 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1423 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1424 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1426 page_cache_release(page
);
1429 pagecache_isize_extended(inode
, old_size
, pos
);
1432 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1437 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1438 /* if we have allocated more blocks and copied
1439 * less. We will have blocks allocated outside
1440 * inode->i_size. So truncate them
1442 ext4_orphan_add(handle
, inode
);
1444 ret2
= ext4_journal_stop(handle
);
1447 if (pos
+ len
> inode
->i_size
) {
1448 ext4_truncate_failed_write(inode
);
1450 * If truncate failed early the inode might still be
1451 * on the orphan list; we need to make sure the inode
1452 * is removed from the orphan list in that case.
1455 ext4_orphan_del(NULL
, inode
);
1458 return ret
? ret
: copied
;
1462 * Reserve space for a single cluster
1464 static int ext4_da_reserve_space(struct inode
*inode
)
1466 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1467 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1471 * We will charge metadata quota at writeout time; this saves
1472 * us from metadata over-estimation, though we may go over by
1473 * a small amount in the end. Here we just reserve for data.
1475 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1479 spin_lock(&ei
->i_block_reservation_lock
);
1480 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1481 spin_unlock(&ei
->i_block_reservation_lock
);
1482 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1485 ei
->i_reserved_data_blocks
++;
1486 trace_ext4_da_reserve_space(inode
);
1487 spin_unlock(&ei
->i_block_reservation_lock
);
1489 return 0; /* success */
1492 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1494 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1495 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1498 return; /* Nothing to release, exit */
1500 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1502 trace_ext4_da_release_space(inode
, to_free
);
1503 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1505 * if there aren't enough reserved blocks, then the
1506 * counter is messed up somewhere. Since this
1507 * function is called from invalidate page, it's
1508 * harmless to return without any action.
1510 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1511 "ino %lu, to_free %d with only %d reserved "
1512 "data blocks", inode
->i_ino
, to_free
,
1513 ei
->i_reserved_data_blocks
);
1515 to_free
= ei
->i_reserved_data_blocks
;
1517 ei
->i_reserved_data_blocks
-= to_free
;
1519 /* update fs dirty data blocks counter */
1520 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1522 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1524 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1527 static void ext4_da_page_release_reservation(struct page
*page
,
1528 unsigned int offset
,
1529 unsigned int length
)
1531 int to_release
= 0, contiguous_blks
= 0;
1532 struct buffer_head
*head
, *bh
;
1533 unsigned int curr_off
= 0;
1534 struct inode
*inode
= page
->mapping
->host
;
1535 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1536 unsigned int stop
= offset
+ length
;
1540 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1542 head
= page_buffers(page
);
1545 unsigned int next_off
= curr_off
+ bh
->b_size
;
1547 if (next_off
> stop
)
1550 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1553 clear_buffer_delay(bh
);
1554 } else if (contiguous_blks
) {
1555 lblk
= page
->index
<<
1556 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1557 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1559 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1560 contiguous_blks
= 0;
1562 curr_off
= next_off
;
1563 } while ((bh
= bh
->b_this_page
) != head
);
1565 if (contiguous_blks
) {
1566 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1567 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1568 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1571 /* If we have released all the blocks belonging to a cluster, then we
1572 * need to release the reserved space for that cluster. */
1573 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1574 while (num_clusters
> 0) {
1575 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1576 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1577 if (sbi
->s_cluster_ratio
== 1 ||
1578 !ext4_find_delalloc_cluster(inode
, lblk
))
1579 ext4_da_release_space(inode
, 1);
1586 * Delayed allocation stuff
1589 struct mpage_da_data
{
1590 struct inode
*inode
;
1591 struct writeback_control
*wbc
;
1593 pgoff_t first_page
; /* The first page to write */
1594 pgoff_t next_page
; /* Current page to examine */
1595 pgoff_t last_page
; /* Last page to examine */
1597 * Extent to map - this can be after first_page because that can be
1598 * fully mapped. We somewhat abuse m_flags to store whether the extent
1599 * is delalloc or unwritten.
1601 struct ext4_map_blocks map
;
1602 struct ext4_io_submit io_submit
; /* IO submission data */
1605 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1610 struct pagevec pvec
;
1611 struct inode
*inode
= mpd
->inode
;
1612 struct address_space
*mapping
= inode
->i_mapping
;
1614 /* This is necessary when next_page == 0. */
1615 if (mpd
->first_page
>= mpd
->next_page
)
1618 index
= mpd
->first_page
;
1619 end
= mpd
->next_page
- 1;
1621 ext4_lblk_t start
, last
;
1622 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1623 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1624 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1627 pagevec_init(&pvec
, 0);
1628 while (index
<= end
) {
1629 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1632 for (i
= 0; i
< nr_pages
; i
++) {
1633 struct page
*page
= pvec
.pages
[i
];
1634 if (page
->index
> end
)
1636 BUG_ON(!PageLocked(page
));
1637 BUG_ON(PageWriteback(page
));
1639 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1640 ClearPageUptodate(page
);
1644 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1645 pagevec_release(&pvec
);
1649 static void ext4_print_free_blocks(struct inode
*inode
)
1651 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1652 struct super_block
*sb
= inode
->i_sb
;
1653 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1655 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1656 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1657 ext4_count_free_clusters(sb
)));
1658 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1659 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1660 (long long) EXT4_C2B(EXT4_SB(sb
),
1661 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1662 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1663 (long long) EXT4_C2B(EXT4_SB(sb
),
1664 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1665 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1666 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1667 ei
->i_reserved_data_blocks
);
1671 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1673 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1677 * This function is grabs code from the very beginning of
1678 * ext4_map_blocks, but assumes that the caller is from delayed write
1679 * time. This function looks up the requested blocks and sets the
1680 * buffer delay bit under the protection of i_data_sem.
1682 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1683 struct ext4_map_blocks
*map
,
1684 struct buffer_head
*bh
)
1686 struct extent_status es
;
1688 sector_t invalid_block
= ~((sector_t
) 0xffff);
1689 #ifdef ES_AGGRESSIVE_TEST
1690 struct ext4_map_blocks orig_map
;
1692 memcpy(&orig_map
, map
, sizeof(*map
));
1695 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1699 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1700 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1701 (unsigned long) map
->m_lblk
);
1703 /* Lookup extent status tree firstly */
1704 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1705 if (ext4_es_is_hole(&es
)) {
1707 down_read(&EXT4_I(inode
)->i_data_sem
);
1712 * Delayed extent could be allocated by fallocate.
1713 * So we need to check it.
1715 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1716 map_bh(bh
, inode
->i_sb
, invalid_block
);
1718 set_buffer_delay(bh
);
1722 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1723 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1724 if (retval
> map
->m_len
)
1725 retval
= map
->m_len
;
1726 map
->m_len
= retval
;
1727 if (ext4_es_is_written(&es
))
1728 map
->m_flags
|= EXT4_MAP_MAPPED
;
1729 else if (ext4_es_is_unwritten(&es
))
1730 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1734 #ifdef ES_AGGRESSIVE_TEST
1735 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1741 * Try to see if we can get the block without requesting a new
1742 * file system block.
1744 down_read(&EXT4_I(inode
)->i_data_sem
);
1745 if (ext4_has_inline_data(inode
))
1747 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1748 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1750 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1756 * XXX: __block_prepare_write() unmaps passed block,
1760 * If the block was allocated from previously allocated cluster,
1761 * then we don't need to reserve it again. However we still need
1762 * to reserve metadata for every block we're going to write.
1764 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1765 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1766 ret
= ext4_da_reserve_space(inode
);
1768 /* not enough space to reserve */
1774 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1775 ~0, EXTENT_STATUS_DELAYED
);
1781 map_bh(bh
, inode
->i_sb
, invalid_block
);
1783 set_buffer_delay(bh
);
1784 } else if (retval
> 0) {
1786 unsigned int status
;
1788 if (unlikely(retval
!= map
->m_len
)) {
1789 ext4_warning(inode
->i_sb
,
1790 "ES len assertion failed for inode "
1791 "%lu: retval %d != map->m_len %d",
1792 inode
->i_ino
, retval
, map
->m_len
);
1796 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1797 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1798 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1799 map
->m_pblk
, status
);
1805 up_read((&EXT4_I(inode
)->i_data_sem
));
1811 * This is a special get_block_t callback which is used by
1812 * ext4_da_write_begin(). It will either return mapped block or
1813 * reserve space for a single block.
1815 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1816 * We also have b_blocknr = -1 and b_bdev initialized properly
1818 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1819 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1820 * initialized properly.
1822 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1823 struct buffer_head
*bh
, int create
)
1825 struct ext4_map_blocks map
;
1828 BUG_ON(create
== 0);
1829 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1831 map
.m_lblk
= iblock
;
1835 * first, we need to know whether the block is allocated already
1836 * preallocated blocks are unmapped but should treated
1837 * the same as allocated blocks.
1839 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1843 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1844 ext4_update_bh_state(bh
, map
.m_flags
);
1846 if (buffer_unwritten(bh
)) {
1847 /* A delayed write to unwritten bh should be marked
1848 * new and mapped. Mapped ensures that we don't do
1849 * get_block multiple times when we write to the same
1850 * offset and new ensures that we do proper zero out
1851 * for partial write.
1854 set_buffer_mapped(bh
);
1859 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1865 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1871 static int __ext4_journalled_writepage(struct page
*page
,
1874 struct address_space
*mapping
= page
->mapping
;
1875 struct inode
*inode
= mapping
->host
;
1876 struct buffer_head
*page_bufs
= NULL
;
1877 handle_t
*handle
= NULL
;
1878 int ret
= 0, err
= 0;
1879 int inline_data
= ext4_has_inline_data(inode
);
1880 struct buffer_head
*inode_bh
= NULL
;
1882 ClearPageChecked(page
);
1885 BUG_ON(page
->index
!= 0);
1886 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1887 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1888 if (inode_bh
== NULL
)
1891 page_bufs
= page_buffers(page
);
1896 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1900 * We need to release the page lock before we start the
1901 * journal, so grab a reference so the page won't disappear
1902 * out from under us.
1907 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1908 ext4_writepage_trans_blocks(inode
));
1909 if (IS_ERR(handle
)) {
1910 ret
= PTR_ERR(handle
);
1912 goto out_no_pagelock
;
1914 BUG_ON(!ext4_handle_valid(handle
));
1918 if (page
->mapping
!= mapping
) {
1919 /* The page got truncated from under us */
1920 ext4_journal_stop(handle
);
1926 BUFFER_TRACE(inode_bh
, "get write access");
1927 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1929 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1932 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1933 do_journal_get_write_access
);
1935 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1940 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1941 err
= ext4_journal_stop(handle
);
1945 if (!ext4_has_inline_data(inode
))
1946 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1948 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1957 * Note that we don't need to start a transaction unless we're journaling data
1958 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1959 * need to file the inode to the transaction's list in ordered mode because if
1960 * we are writing back data added by write(), the inode is already there and if
1961 * we are writing back data modified via mmap(), no one guarantees in which
1962 * transaction the data will hit the disk. In case we are journaling data, we
1963 * cannot start transaction directly because transaction start ranks above page
1964 * lock so we have to do some magic.
1966 * This function can get called via...
1967 * - ext4_writepages after taking page lock (have journal handle)
1968 * - journal_submit_inode_data_buffers (no journal handle)
1969 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1970 * - grab_page_cache when doing write_begin (have journal handle)
1972 * We don't do any block allocation in this function. If we have page with
1973 * multiple blocks we need to write those buffer_heads that are mapped. This
1974 * is important for mmaped based write. So if we do with blocksize 1K
1975 * truncate(f, 1024);
1976 * a = mmap(f, 0, 4096);
1978 * truncate(f, 4096);
1979 * we have in the page first buffer_head mapped via page_mkwrite call back
1980 * but other buffer_heads would be unmapped but dirty (dirty done via the
1981 * do_wp_page). So writepage should write the first block. If we modify
1982 * the mmap area beyond 1024 we will again get a page_fault and the
1983 * page_mkwrite callback will do the block allocation and mark the
1984 * buffer_heads mapped.
1986 * We redirty the page if we have any buffer_heads that is either delay or
1987 * unwritten in the page.
1989 * We can get recursively called as show below.
1991 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1994 * But since we don't do any block allocation we should not deadlock.
1995 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1997 static int ext4_writepage(struct page
*page
,
1998 struct writeback_control
*wbc
)
2003 struct buffer_head
*page_bufs
= NULL
;
2004 struct inode
*inode
= page
->mapping
->host
;
2005 struct ext4_io_submit io_submit
;
2006 bool keep_towrite
= false;
2008 trace_ext4_writepage(page
);
2009 size
= i_size_read(inode
);
2010 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2011 len
= size
& ~PAGE_CACHE_MASK
;
2013 len
= PAGE_CACHE_SIZE
;
2015 page_bufs
= page_buffers(page
);
2017 * We cannot do block allocation or other extent handling in this
2018 * function. If there are buffers needing that, we have to redirty
2019 * the page. But we may reach here when we do a journal commit via
2020 * journal_submit_inode_data_buffers() and in that case we must write
2021 * allocated buffers to achieve data=ordered mode guarantees.
2023 * Also, if there is only one buffer per page (the fs block
2024 * size == the page size), if one buffer needs block
2025 * allocation or needs to modify the extent tree to clear the
2026 * unwritten flag, we know that the page can't be written at
2027 * all, so we might as well refuse the write immediately.
2028 * Unfortunately if the block size != page size, we can't as
2029 * easily detect this case using ext4_walk_page_buffers(), but
2030 * for the extremely common case, this is an optimization that
2031 * skips a useless round trip through ext4_bio_write_page().
2033 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2034 ext4_bh_delay_or_unwritten
)) {
2035 redirty_page_for_writepage(wbc
, page
);
2036 if ((current
->flags
& PF_MEMALLOC
) ||
2037 (inode
->i_sb
->s_blocksize
== PAGE_CACHE_SIZE
)) {
2039 * For memory cleaning there's no point in writing only
2040 * some buffers. So just bail out. Warn if we came here
2041 * from direct reclaim.
2043 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2048 keep_towrite
= true;
2051 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2053 * It's mmapped pagecache. Add buffers and journal it. There
2054 * doesn't seem much point in redirtying the page here.
2056 return __ext4_journalled_writepage(page
, len
);
2058 ext4_io_submit_init(&io_submit
, wbc
);
2059 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2060 if (!io_submit
.io_end
) {
2061 redirty_page_for_writepage(wbc
, page
);
2065 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2066 ext4_io_submit(&io_submit
);
2067 /* Drop io_end reference we got from init */
2068 ext4_put_io_end_defer(io_submit
.io_end
);
2072 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2075 loff_t size
= i_size_read(mpd
->inode
);
2078 BUG_ON(page
->index
!= mpd
->first_page
);
2079 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2080 len
= size
& ~PAGE_CACHE_MASK
;
2082 len
= PAGE_CACHE_SIZE
;
2083 clear_page_dirty_for_io(page
);
2084 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2086 mpd
->wbc
->nr_to_write
--;
2092 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2095 * mballoc gives us at most this number of blocks...
2096 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2097 * The rest of mballoc seems to handle chunks up to full group size.
2099 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2102 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2104 * @mpd - extent of blocks
2105 * @lblk - logical number of the block in the file
2106 * @bh - buffer head we want to add to the extent
2108 * The function is used to collect contig. blocks in the same state. If the
2109 * buffer doesn't require mapping for writeback and we haven't started the
2110 * extent of buffers to map yet, the function returns 'true' immediately - the
2111 * caller can write the buffer right away. Otherwise the function returns true
2112 * if the block has been added to the extent, false if the block couldn't be
2115 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2116 struct buffer_head
*bh
)
2118 struct ext4_map_blocks
*map
= &mpd
->map
;
2120 /* Buffer that doesn't need mapping for writeback? */
2121 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2122 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2123 /* So far no extent to map => we write the buffer right away */
2124 if (map
->m_len
== 0)
2129 /* First block in the extent? */
2130 if (map
->m_len
== 0) {
2133 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2137 /* Don't go larger than mballoc is willing to allocate */
2138 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2141 /* Can we merge the block to our big extent? */
2142 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2143 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2151 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2153 * @mpd - extent of blocks for mapping
2154 * @head - the first buffer in the page
2155 * @bh - buffer we should start processing from
2156 * @lblk - logical number of the block in the file corresponding to @bh
2158 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2159 * the page for IO if all buffers in this page were mapped and there's no
2160 * accumulated extent of buffers to map or add buffers in the page to the
2161 * extent of buffers to map. The function returns 1 if the caller can continue
2162 * by processing the next page, 0 if it should stop adding buffers to the
2163 * extent to map because we cannot extend it anymore. It can also return value
2164 * < 0 in case of error during IO submission.
2166 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2167 struct buffer_head
*head
,
2168 struct buffer_head
*bh
,
2171 struct inode
*inode
= mpd
->inode
;
2173 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2174 >> inode
->i_blkbits
;
2177 BUG_ON(buffer_locked(bh
));
2179 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2180 /* Found extent to map? */
2183 /* Everything mapped so far and we hit EOF */
2186 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2187 /* So far everything mapped? Submit the page for IO. */
2188 if (mpd
->map
.m_len
== 0) {
2189 err
= mpage_submit_page(mpd
, head
->b_page
);
2193 return lblk
< blocks
;
2197 * mpage_map_buffers - update buffers corresponding to changed extent and
2198 * submit fully mapped pages for IO
2200 * @mpd - description of extent to map, on return next extent to map
2202 * Scan buffers corresponding to changed extent (we expect corresponding pages
2203 * to be already locked) and update buffer state according to new extent state.
2204 * We map delalloc buffers to their physical location, clear unwritten bits,
2205 * and mark buffers as uninit when we perform writes to unwritten extents
2206 * and do extent conversion after IO is finished. If the last page is not fully
2207 * mapped, we update @map to the next extent in the last page that needs
2208 * mapping. Otherwise we submit the page for IO.
2210 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2212 struct pagevec pvec
;
2214 struct inode
*inode
= mpd
->inode
;
2215 struct buffer_head
*head
, *bh
;
2216 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
2222 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2223 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2224 lblk
= start
<< bpp_bits
;
2225 pblock
= mpd
->map
.m_pblk
;
2227 pagevec_init(&pvec
, 0);
2228 while (start
<= end
) {
2229 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2233 for (i
= 0; i
< nr_pages
; i
++) {
2234 struct page
*page
= pvec
.pages
[i
];
2236 if (page
->index
> end
)
2238 /* Up to 'end' pages must be contiguous */
2239 BUG_ON(page
->index
!= start
);
2240 bh
= head
= page_buffers(page
);
2242 if (lblk
< mpd
->map
.m_lblk
)
2244 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2246 * Buffer after end of mapped extent.
2247 * Find next buffer in the page to map.
2250 mpd
->map
.m_flags
= 0;
2252 * FIXME: If dioread_nolock supports
2253 * blocksize < pagesize, we need to make
2254 * sure we add size mapped so far to
2255 * io_end->size as the following call
2256 * can submit the page for IO.
2258 err
= mpage_process_page_bufs(mpd
, head
,
2260 pagevec_release(&pvec
);
2265 if (buffer_delay(bh
)) {
2266 clear_buffer_delay(bh
);
2267 bh
->b_blocknr
= pblock
++;
2269 clear_buffer_unwritten(bh
);
2270 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2273 * FIXME: This is going to break if dioread_nolock
2274 * supports blocksize < pagesize as we will try to
2275 * convert potentially unmapped parts of inode.
2277 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
2278 /* Page fully mapped - let IO run! */
2279 err
= mpage_submit_page(mpd
, page
);
2281 pagevec_release(&pvec
);
2286 pagevec_release(&pvec
);
2288 /* Extent fully mapped and matches with page boundary. We are done. */
2290 mpd
->map
.m_flags
= 0;
2294 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2296 struct inode
*inode
= mpd
->inode
;
2297 struct ext4_map_blocks
*map
= &mpd
->map
;
2298 int get_blocks_flags
;
2299 int err
, dioread_nolock
;
2301 trace_ext4_da_write_pages_extent(inode
, map
);
2303 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2304 * to convert an unwritten extent to be initialized (in the case
2305 * where we have written into one or more preallocated blocks). It is
2306 * possible that we're going to need more metadata blocks than
2307 * previously reserved. However we must not fail because we're in
2308 * writeback and there is nothing we can do about it so it might result
2309 * in data loss. So use reserved blocks to allocate metadata if
2312 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2313 * the blocks in question are delalloc blocks. This indicates
2314 * that the blocks and quotas has already been checked when
2315 * the data was copied into the page cache.
2317 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2318 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
2319 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2321 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2322 if (map
->m_flags
& (1 << BH_Delay
))
2323 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2325 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2328 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2329 if (!mpd
->io_submit
.io_end
->handle
&&
2330 ext4_handle_valid(handle
)) {
2331 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2332 handle
->h_rsv_handle
= NULL
;
2334 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2337 BUG_ON(map
->m_len
== 0);
2338 if (map
->m_flags
& EXT4_MAP_NEW
) {
2339 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2342 for (i
= 0; i
< map
->m_len
; i
++)
2343 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2349 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2350 * mpd->len and submit pages underlying it for IO
2352 * @handle - handle for journal operations
2353 * @mpd - extent to map
2354 * @give_up_on_write - we set this to true iff there is a fatal error and there
2355 * is no hope of writing the data. The caller should discard
2356 * dirty pages to avoid infinite loops.
2358 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2359 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2360 * them to initialized or split the described range from larger unwritten
2361 * extent. Note that we need not map all the described range since allocation
2362 * can return less blocks or the range is covered by more unwritten extents. We
2363 * cannot map more because we are limited by reserved transaction credits. On
2364 * the other hand we always make sure that the last touched page is fully
2365 * mapped so that it can be written out (and thus forward progress is
2366 * guaranteed). After mapping we submit all mapped pages for IO.
2368 static int mpage_map_and_submit_extent(handle_t
*handle
,
2369 struct mpage_da_data
*mpd
,
2370 bool *give_up_on_write
)
2372 struct inode
*inode
= mpd
->inode
;
2373 struct ext4_map_blocks
*map
= &mpd
->map
;
2378 mpd
->io_submit
.io_end
->offset
=
2379 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2381 err
= mpage_map_one_extent(handle
, mpd
);
2383 struct super_block
*sb
= inode
->i_sb
;
2385 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2386 goto invalidate_dirty_pages
;
2388 * Let the uper layers retry transient errors.
2389 * In the case of ENOSPC, if ext4_count_free_blocks()
2390 * is non-zero, a commit should free up blocks.
2392 if ((err
== -ENOMEM
) ||
2393 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2395 goto update_disksize
;
2398 ext4_msg(sb
, KERN_CRIT
,
2399 "Delayed block allocation failed for "
2400 "inode %lu at logical offset %llu with"
2401 " max blocks %u with error %d",
2403 (unsigned long long)map
->m_lblk
,
2404 (unsigned)map
->m_len
, -err
);
2405 ext4_msg(sb
, KERN_CRIT
,
2406 "This should not happen!! Data will "
2409 ext4_print_free_blocks(inode
);
2410 invalidate_dirty_pages
:
2411 *give_up_on_write
= true;
2416 * Update buffer state, submit mapped pages, and get us new
2419 err
= mpage_map_and_submit_buffers(mpd
);
2421 goto update_disksize
;
2422 } while (map
->m_len
);
2426 * Update on-disk size after IO is submitted. Races with
2427 * truncate are avoided by checking i_size under i_data_sem.
2429 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2430 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2434 down_write(&EXT4_I(inode
)->i_data_sem
);
2435 i_size
= i_size_read(inode
);
2436 if (disksize
> i_size
)
2438 if (disksize
> EXT4_I(inode
)->i_disksize
)
2439 EXT4_I(inode
)->i_disksize
= disksize
;
2440 err2
= ext4_mark_inode_dirty(handle
, inode
);
2441 up_write(&EXT4_I(inode
)->i_data_sem
);
2443 ext4_error(inode
->i_sb
,
2444 "Failed to mark inode %lu dirty",
2453 * Calculate the total number of credits to reserve for one writepages
2454 * iteration. This is called from ext4_writepages(). We map an extent of
2455 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2456 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2457 * bpp - 1 blocks in bpp different extents.
2459 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2461 int bpp
= ext4_journal_blocks_per_page(inode
);
2463 return ext4_meta_trans_blocks(inode
,
2464 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2468 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2469 * and underlying extent to map
2471 * @mpd - where to look for pages
2473 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2474 * IO immediately. When we find a page which isn't mapped we start accumulating
2475 * extent of buffers underlying these pages that needs mapping (formed by
2476 * either delayed or unwritten buffers). We also lock the pages containing
2477 * these buffers. The extent found is returned in @mpd structure (starting at
2478 * mpd->lblk with length mpd->len blocks).
2480 * Note that this function can attach bios to one io_end structure which are
2481 * neither logically nor physically contiguous. Although it may seem as an
2482 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2483 * case as we need to track IO to all buffers underlying a page in one io_end.
2485 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2487 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2488 struct pagevec pvec
;
2489 unsigned int nr_pages
;
2490 long left
= mpd
->wbc
->nr_to_write
;
2491 pgoff_t index
= mpd
->first_page
;
2492 pgoff_t end
= mpd
->last_page
;
2495 int blkbits
= mpd
->inode
->i_blkbits
;
2497 struct buffer_head
*head
;
2499 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2500 tag
= PAGECACHE_TAG_TOWRITE
;
2502 tag
= PAGECACHE_TAG_DIRTY
;
2504 pagevec_init(&pvec
, 0);
2506 mpd
->next_page
= index
;
2507 while (index
<= end
) {
2508 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2509 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2513 for (i
= 0; i
< nr_pages
; i
++) {
2514 struct page
*page
= pvec
.pages
[i
];
2517 * At this point, the page may be truncated or
2518 * invalidated (changing page->mapping to NULL), or
2519 * even swizzled back from swapper_space to tmpfs file
2520 * mapping. However, page->index will not change
2521 * because we have a reference on the page.
2523 if (page
->index
> end
)
2527 * Accumulated enough dirty pages? This doesn't apply
2528 * to WB_SYNC_ALL mode. For integrity sync we have to
2529 * keep going because someone may be concurrently
2530 * dirtying pages, and we might have synced a lot of
2531 * newly appeared dirty pages, but have not synced all
2532 * of the old dirty pages.
2534 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2537 /* If we can't merge this page, we are done. */
2538 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2543 * If the page is no longer dirty, or its mapping no
2544 * longer corresponds to inode we are writing (which
2545 * means it has been truncated or invalidated), or the
2546 * page is already under writeback and we are not doing
2547 * a data integrity writeback, skip the page
2549 if (!PageDirty(page
) ||
2550 (PageWriteback(page
) &&
2551 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2552 unlikely(page
->mapping
!= mapping
)) {
2557 wait_on_page_writeback(page
);
2558 BUG_ON(PageWriteback(page
));
2560 if (mpd
->map
.m_len
== 0)
2561 mpd
->first_page
= page
->index
;
2562 mpd
->next_page
= page
->index
+ 1;
2563 /* Add all dirty buffers to mpd */
2564 lblk
= ((ext4_lblk_t
)page
->index
) <<
2565 (PAGE_CACHE_SHIFT
- blkbits
);
2566 head
= page_buffers(page
);
2567 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2573 pagevec_release(&pvec
);
2578 pagevec_release(&pvec
);
2582 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2585 struct address_space
*mapping
= data
;
2586 int ret
= ext4_writepage(page
, wbc
);
2587 mapping_set_error(mapping
, ret
);
2591 static int ext4_writepages(struct address_space
*mapping
,
2592 struct writeback_control
*wbc
)
2594 pgoff_t writeback_index
= 0;
2595 long nr_to_write
= wbc
->nr_to_write
;
2596 int range_whole
= 0;
2598 handle_t
*handle
= NULL
;
2599 struct mpage_da_data mpd
;
2600 struct inode
*inode
= mapping
->host
;
2601 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2602 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2604 struct blk_plug plug
;
2605 bool give_up_on_write
= false;
2607 trace_ext4_writepages(inode
, wbc
);
2609 if (dax_mapping(mapping
))
2610 return dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2614 * No pages to write? This is mainly a kludge to avoid starting
2615 * a transaction for special inodes like journal inode on last iput()
2616 * because that could violate lock ordering on umount
2618 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2619 goto out_writepages
;
2621 if (ext4_should_journal_data(inode
)) {
2622 struct blk_plug plug
;
2624 blk_start_plug(&plug
);
2625 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2626 blk_finish_plug(&plug
);
2627 goto out_writepages
;
2631 * If the filesystem has aborted, it is read-only, so return
2632 * right away instead of dumping stack traces later on that
2633 * will obscure the real source of the problem. We test
2634 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2635 * the latter could be true if the filesystem is mounted
2636 * read-only, and in that case, ext4_writepages should
2637 * *never* be called, so if that ever happens, we would want
2640 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2642 goto out_writepages
;
2645 if (ext4_should_dioread_nolock(inode
)) {
2647 * We may need to convert up to one extent per block in
2648 * the page and we may dirty the inode.
2650 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2654 * If we have inline data and arrive here, it means that
2655 * we will soon create the block for the 1st page, so
2656 * we'd better clear the inline data here.
2658 if (ext4_has_inline_data(inode
)) {
2659 /* Just inode will be modified... */
2660 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2661 if (IS_ERR(handle
)) {
2662 ret
= PTR_ERR(handle
);
2663 goto out_writepages
;
2665 BUG_ON(ext4_test_inode_state(inode
,
2666 EXT4_STATE_MAY_INLINE_DATA
));
2667 ext4_destroy_inline_data(handle
, inode
);
2668 ext4_journal_stop(handle
);
2671 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2674 if (wbc
->range_cyclic
) {
2675 writeback_index
= mapping
->writeback_index
;
2676 if (writeback_index
)
2678 mpd
.first_page
= writeback_index
;
2681 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2682 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2687 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2689 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2690 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2692 blk_start_plug(&plug
);
2693 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2694 /* For each extent of pages we use new io_end */
2695 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2696 if (!mpd
.io_submit
.io_end
) {
2702 * We have two constraints: We find one extent to map and we
2703 * must always write out whole page (makes a difference when
2704 * blocksize < pagesize) so that we don't block on IO when we
2705 * try to write out the rest of the page. Journalled mode is
2706 * not supported by delalloc.
2708 BUG_ON(ext4_should_journal_data(inode
));
2709 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2711 /* start a new transaction */
2712 handle
= ext4_journal_start_with_reserve(inode
,
2713 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2714 if (IS_ERR(handle
)) {
2715 ret
= PTR_ERR(handle
);
2716 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2717 "%ld pages, ino %lu; err %d", __func__
,
2718 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2719 /* Release allocated io_end */
2720 ext4_put_io_end(mpd
.io_submit
.io_end
);
2724 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2725 ret
= mpage_prepare_extent_to_map(&mpd
);
2728 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2732 * We scanned the whole range (or exhausted
2733 * nr_to_write), submitted what was mapped and
2734 * didn't find anything needing mapping. We are
2740 ext4_journal_stop(handle
);
2741 /* Submit prepared bio */
2742 ext4_io_submit(&mpd
.io_submit
);
2743 /* Unlock pages we didn't use */
2744 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2745 /* Drop our io_end reference we got from init */
2746 ext4_put_io_end(mpd
.io_submit
.io_end
);
2748 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2750 * Commit the transaction which would
2751 * free blocks released in the transaction
2754 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2758 /* Fatal error - ENOMEM, EIO... */
2762 blk_finish_plug(&plug
);
2763 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2765 mpd
.last_page
= writeback_index
- 1;
2771 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2773 * Set the writeback_index so that range_cyclic
2774 * mode will write it back later
2776 mapping
->writeback_index
= mpd
.first_page
;
2779 trace_ext4_writepages_result(inode
, wbc
, ret
,
2780 nr_to_write
- wbc
->nr_to_write
);
2784 static int ext4_nonda_switch(struct super_block
*sb
)
2786 s64 free_clusters
, dirty_clusters
;
2787 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2790 * switch to non delalloc mode if we are running low
2791 * on free block. The free block accounting via percpu
2792 * counters can get slightly wrong with percpu_counter_batch getting
2793 * accumulated on each CPU without updating global counters
2794 * Delalloc need an accurate free block accounting. So switch
2795 * to non delalloc when we are near to error range.
2798 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2800 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2802 * Start pushing delalloc when 1/2 of free blocks are dirty.
2804 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2805 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2807 if (2 * free_clusters
< 3 * dirty_clusters
||
2808 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2810 * free block count is less than 150% of dirty blocks
2811 * or free blocks is less than watermark
2818 /* We always reserve for an inode update; the superblock could be there too */
2819 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2821 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2824 if (pos
+ len
<= 0x7fffffffULL
)
2827 /* We might need to update the superblock to set LARGE_FILE */
2831 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2832 loff_t pos
, unsigned len
, unsigned flags
,
2833 struct page
**pagep
, void **fsdata
)
2835 int ret
, retries
= 0;
2838 struct inode
*inode
= mapping
->host
;
2841 index
= pos
>> PAGE_CACHE_SHIFT
;
2843 if (ext4_nonda_switch(inode
->i_sb
)) {
2844 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2845 return ext4_write_begin(file
, mapping
, pos
,
2846 len
, flags
, pagep
, fsdata
);
2848 *fsdata
= (void *)0;
2849 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2851 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2852 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2862 * grab_cache_page_write_begin() can take a long time if the
2863 * system is thrashing due to memory pressure, or if the page
2864 * is being written back. So grab it first before we start
2865 * the transaction handle. This also allows us to allocate
2866 * the page (if needed) without using GFP_NOFS.
2869 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2875 * With delayed allocation, we don't log the i_disksize update
2876 * if there is delayed block allocation. But we still need
2877 * to journalling the i_disksize update if writes to the end
2878 * of file which has an already mapped buffer.
2881 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2882 ext4_da_write_credits(inode
, pos
, len
));
2883 if (IS_ERR(handle
)) {
2884 page_cache_release(page
);
2885 return PTR_ERR(handle
);
2889 if (page
->mapping
!= mapping
) {
2890 /* The page got truncated from under us */
2892 page_cache_release(page
);
2893 ext4_journal_stop(handle
);
2896 /* In case writeback began while the page was unlocked */
2897 wait_for_stable_page(page
);
2899 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2900 ret
= ext4_block_write_begin(page
, pos
, len
,
2901 ext4_da_get_block_prep
);
2903 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2907 ext4_journal_stop(handle
);
2909 * block_write_begin may have instantiated a few blocks
2910 * outside i_size. Trim these off again. Don't need
2911 * i_size_read because we hold i_mutex.
2913 if (pos
+ len
> inode
->i_size
)
2914 ext4_truncate_failed_write(inode
);
2916 if (ret
== -ENOSPC
&&
2917 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2920 page_cache_release(page
);
2929 * Check if we should update i_disksize
2930 * when write to the end of file but not require block allocation
2932 static int ext4_da_should_update_i_disksize(struct page
*page
,
2933 unsigned long offset
)
2935 struct buffer_head
*bh
;
2936 struct inode
*inode
= page
->mapping
->host
;
2940 bh
= page_buffers(page
);
2941 idx
= offset
>> inode
->i_blkbits
;
2943 for (i
= 0; i
< idx
; i
++)
2944 bh
= bh
->b_this_page
;
2946 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2951 static int ext4_da_write_end(struct file
*file
,
2952 struct address_space
*mapping
,
2953 loff_t pos
, unsigned len
, unsigned copied
,
2954 struct page
*page
, void *fsdata
)
2956 struct inode
*inode
= mapping
->host
;
2958 handle_t
*handle
= ext4_journal_current_handle();
2960 unsigned long start
, end
;
2961 int write_mode
= (int)(unsigned long)fsdata
;
2963 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2964 return ext4_write_end(file
, mapping
, pos
,
2965 len
, copied
, page
, fsdata
);
2967 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2968 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2969 end
= start
+ copied
- 1;
2972 * generic_write_end() will run mark_inode_dirty() if i_size
2973 * changes. So let's piggyback the i_disksize mark_inode_dirty
2976 new_i_size
= pos
+ copied
;
2977 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2978 if (ext4_has_inline_data(inode
) ||
2979 ext4_da_should_update_i_disksize(page
, end
)) {
2980 ext4_update_i_disksize(inode
, new_i_size
);
2981 /* We need to mark inode dirty even if
2982 * new_i_size is less that inode->i_size
2983 * bu greater than i_disksize.(hint delalloc)
2985 ext4_mark_inode_dirty(handle
, inode
);
2989 if (write_mode
!= CONVERT_INLINE_DATA
&&
2990 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2991 ext4_has_inline_data(inode
))
2992 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2995 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3001 ret2
= ext4_journal_stop(handle
);
3005 return ret
? ret
: copied
;
3008 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3009 unsigned int length
)
3012 * Drop reserved blocks
3014 BUG_ON(!PageLocked(page
));
3015 if (!page_has_buffers(page
))
3018 ext4_da_page_release_reservation(page
, offset
, length
);
3021 ext4_invalidatepage(page
, offset
, length
);
3027 * Force all delayed allocation blocks to be allocated for a given inode.
3029 int ext4_alloc_da_blocks(struct inode
*inode
)
3031 trace_ext4_alloc_da_blocks(inode
);
3033 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3037 * We do something simple for now. The filemap_flush() will
3038 * also start triggering a write of the data blocks, which is
3039 * not strictly speaking necessary (and for users of
3040 * laptop_mode, not even desirable). However, to do otherwise
3041 * would require replicating code paths in:
3043 * ext4_writepages() ->
3044 * write_cache_pages() ---> (via passed in callback function)
3045 * __mpage_da_writepage() -->
3046 * mpage_add_bh_to_extent()
3047 * mpage_da_map_blocks()
3049 * The problem is that write_cache_pages(), located in
3050 * mm/page-writeback.c, marks pages clean in preparation for
3051 * doing I/O, which is not desirable if we're not planning on
3054 * We could call write_cache_pages(), and then redirty all of
3055 * the pages by calling redirty_page_for_writepage() but that
3056 * would be ugly in the extreme. So instead we would need to
3057 * replicate parts of the code in the above functions,
3058 * simplifying them because we wouldn't actually intend to
3059 * write out the pages, but rather only collect contiguous
3060 * logical block extents, call the multi-block allocator, and
3061 * then update the buffer heads with the block allocations.
3063 * For now, though, we'll cheat by calling filemap_flush(),
3064 * which will map the blocks, and start the I/O, but not
3065 * actually wait for the I/O to complete.
3067 return filemap_flush(inode
->i_mapping
);
3071 * bmap() is special. It gets used by applications such as lilo and by
3072 * the swapper to find the on-disk block of a specific piece of data.
3074 * Naturally, this is dangerous if the block concerned is still in the
3075 * journal. If somebody makes a swapfile on an ext4 data-journaling
3076 * filesystem and enables swap, then they may get a nasty shock when the
3077 * data getting swapped to that swapfile suddenly gets overwritten by
3078 * the original zero's written out previously to the journal and
3079 * awaiting writeback in the kernel's buffer cache.
3081 * So, if we see any bmap calls here on a modified, data-journaled file,
3082 * take extra steps to flush any blocks which might be in the cache.
3084 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3086 struct inode
*inode
= mapping
->host
;
3091 * We can get here for an inline file via the FIBMAP ioctl
3093 if (ext4_has_inline_data(inode
))
3096 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3097 test_opt(inode
->i_sb
, DELALLOC
)) {
3099 * With delalloc we want to sync the file
3100 * so that we can make sure we allocate
3103 filemap_write_and_wait(mapping
);
3106 if (EXT4_JOURNAL(inode
) &&
3107 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3109 * This is a REALLY heavyweight approach, but the use of
3110 * bmap on dirty files is expected to be extremely rare:
3111 * only if we run lilo or swapon on a freshly made file
3112 * do we expect this to happen.
3114 * (bmap requires CAP_SYS_RAWIO so this does not
3115 * represent an unprivileged user DOS attack --- we'd be
3116 * in trouble if mortal users could trigger this path at
3119 * NB. EXT4_STATE_JDATA is not set on files other than
3120 * regular files. If somebody wants to bmap a directory
3121 * or symlink and gets confused because the buffer
3122 * hasn't yet been flushed to disk, they deserve
3123 * everything they get.
3126 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3127 journal
= EXT4_JOURNAL(inode
);
3128 jbd2_journal_lock_updates(journal
);
3129 err
= jbd2_journal_flush(journal
);
3130 jbd2_journal_unlock_updates(journal
);
3136 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3139 static int ext4_readpage(struct file
*file
, struct page
*page
)
3142 struct inode
*inode
= page
->mapping
->host
;
3144 trace_ext4_readpage(page
);
3146 if (ext4_has_inline_data(inode
))
3147 ret
= ext4_readpage_inline(inode
, page
);
3150 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3156 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3157 struct list_head
*pages
, unsigned nr_pages
)
3159 struct inode
*inode
= mapping
->host
;
3161 /* If the file has inline data, no need to do readpages. */
3162 if (ext4_has_inline_data(inode
))
3165 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3168 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3169 unsigned int length
)
3171 trace_ext4_invalidatepage(page
, offset
, length
);
3173 /* No journalling happens on data buffers when this function is used */
3174 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3176 block_invalidatepage(page
, offset
, length
);
3179 static int __ext4_journalled_invalidatepage(struct page
*page
,
3180 unsigned int offset
,
3181 unsigned int length
)
3183 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3185 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3188 * If it's a full truncate we just forget about the pending dirtying
3190 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
3191 ClearPageChecked(page
);
3193 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3196 /* Wrapper for aops... */
3197 static void ext4_journalled_invalidatepage(struct page
*page
,
3198 unsigned int offset
,
3199 unsigned int length
)
3201 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3204 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3206 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3208 trace_ext4_releasepage(page
);
3210 /* Page has dirty journalled data -> cannot release */
3211 if (PageChecked(page
))
3214 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3216 return try_to_free_buffers(page
);
3219 #ifdef CONFIG_FS_DAX
3220 int ext4_dax_mmap_get_block(struct inode
*inode
, sector_t iblock
,
3221 struct buffer_head
*bh_result
, int create
)
3225 struct ext4_map_blocks map
;
3226 handle_t
*handle
= NULL
;
3229 ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n",
3230 inode
->i_ino
, create
);
3231 map
.m_lblk
= iblock
;
3232 map
.m_len
= bh_result
->b_size
>> inode
->i_blkbits
;
3233 credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3235 flags
|= EXT4_GET_BLOCKS_PRE_IO
| EXT4_GET_BLOCKS_CREATE_ZERO
;
3236 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, credits
);
3237 if (IS_ERR(handle
)) {
3238 ret
= PTR_ERR(handle
);
3243 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
3245 err
= ext4_journal_stop(handle
);
3246 if (ret
>= 0 && err
< 0)
3251 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3255 * We are protected by i_mmap_sem so we know block cannot go
3256 * away from under us even though we dropped i_data_sem.
3257 * Convert extent to written and write zeros there.
3259 * Note: We may get here even when create == 0.
3261 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, credits
);
3262 if (IS_ERR(handle
)) {
3263 ret
= PTR_ERR(handle
);
3267 err
= ext4_map_blocks(handle
, inode
, &map
,
3268 EXT4_GET_BLOCKS_CONVERT
| EXT4_GET_BLOCKS_CREATE_ZERO
);
3271 err2
= ext4_journal_stop(handle
);
3272 if (err2
< 0 && ret
> 0)
3276 WARN_ON_ONCE(ret
== 0 && create
);
3278 map_bh(bh_result
, inode
->i_sb
, map
.m_pblk
);
3280 * At least for now we have to clear BH_New so that DAX code
3281 * doesn't attempt to zero blocks again in a racy way.
3283 map
.m_flags
&= ~EXT4_MAP_NEW
;
3284 ext4_update_bh_state(bh_result
, map
.m_flags
);
3285 bh_result
->b_size
= map
.m_len
<< inode
->i_blkbits
;
3292 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3293 ssize_t size
, void *private)
3295 ext4_io_end_t
*io_end
= private;
3297 /* if not async direct IO just return */
3301 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3302 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3303 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3305 io_end
->offset
= offset
;
3306 io_end
->size
= size
;
3307 ext4_put_io_end(io_end
);
3311 * For ext4 extent files, ext4 will do direct-io write to holes,
3312 * preallocated extents, and those write extend the file, no need to
3313 * fall back to buffered IO.
3315 * For holes, we fallocate those blocks, mark them as unwritten
3316 * If those blocks were preallocated, we mark sure they are split, but
3317 * still keep the range to write as unwritten.
3319 * The unwritten extents will be converted to written when DIO is completed.
3320 * For async direct IO, since the IO may still pending when return, we
3321 * set up an end_io call back function, which will do the conversion
3322 * when async direct IO completed.
3324 * If the O_DIRECT write will extend the file then add this inode to the
3325 * orphan list. So recovery will truncate it back to the original size
3326 * if the machine crashes during the write.
3329 static ssize_t
ext4_ext_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
3332 struct file
*file
= iocb
->ki_filp
;
3333 struct inode
*inode
= file
->f_mapping
->host
;
3335 size_t count
= iov_iter_count(iter
);
3337 get_block_t
*get_block_func
= NULL
;
3339 loff_t final_size
= offset
+ count
;
3341 /* Use the old path for reads and writes beyond i_size. */
3342 if (iov_iter_rw(iter
) != WRITE
|| final_size
> inode
->i_size
)
3343 return ext4_ind_direct_IO(iocb
, iter
, offset
);
3345 BUG_ON(iocb
->private == NULL
);
3348 * Make all waiters for direct IO properly wait also for extent
3349 * conversion. This also disallows race between truncate() and
3350 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3352 if (iov_iter_rw(iter
) == WRITE
)
3353 inode_dio_begin(inode
);
3355 /* If we do a overwrite dio, i_mutex locking can be released */
3356 overwrite
= *((int *)iocb
->private);
3359 inode_unlock(inode
);
3362 * We could direct write to holes and fallocate.
3364 * Allocated blocks to fill the hole are marked as unwritten to prevent
3365 * parallel buffered read to expose the stale data before DIO complete
3368 * As to previously fallocated extents, ext4 get_block will just simply
3369 * mark the buffer mapped but still keep the extents unwritten.
3371 * For non AIO case, we will convert those unwritten extents to written
3372 * after return back from blockdev_direct_IO. That way we save us from
3373 * allocating io_end structure and also the overhead of offloading
3374 * the extent convertion to a workqueue.
3376 * For async DIO, the conversion needs to be deferred when the
3377 * IO is completed. The ext4 end_io callback function will be
3378 * called to take care of the conversion work. Here for async
3379 * case, we allocate an io_end structure to hook to the iocb.
3381 iocb
->private = NULL
;
3383 get_block_func
= ext4_dio_get_block_overwrite
;
3384 else if (is_sync_kiocb(iocb
)) {
3385 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3386 dio_flags
= DIO_LOCKING
;
3388 get_block_func
= ext4_dio_get_block_unwritten_async
;
3389 dio_flags
= DIO_LOCKING
;
3391 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3392 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3395 ret
= dax_do_io(iocb
, inode
, iter
, offset
, get_block_func
,
3396 ext4_end_io_dio
, dio_flags
);
3398 ret
= __blockdev_direct_IO(iocb
, inode
,
3399 inode
->i_sb
->s_bdev
, iter
, offset
,
3401 ext4_end_io_dio
, NULL
, dio_flags
);
3403 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3404 EXT4_STATE_DIO_UNWRITTEN
)) {
3407 * for non AIO case, since the IO is already
3408 * completed, we could do the conversion right here
3410 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3414 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3417 if (iov_iter_rw(iter
) == WRITE
)
3418 inode_dio_end(inode
);
3419 /* take i_mutex locking again if we do a ovewrite dio */
3426 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
3429 struct file
*file
= iocb
->ki_filp
;
3430 struct inode
*inode
= file
->f_mapping
->host
;
3431 size_t count
= iov_iter_count(iter
);
3434 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3435 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3440 * If we are doing data journalling we don't support O_DIRECT
3442 if (ext4_should_journal_data(inode
))
3445 /* Let buffer I/O handle the inline data case. */
3446 if (ext4_has_inline_data(inode
))
3449 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3450 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3451 ret
= ext4_ext_direct_IO(iocb
, iter
, offset
);
3453 ret
= ext4_ind_direct_IO(iocb
, iter
, offset
);
3454 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3459 * Pages can be marked dirty completely asynchronously from ext4's journalling
3460 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3461 * much here because ->set_page_dirty is called under VFS locks. The page is
3462 * not necessarily locked.
3464 * We cannot just dirty the page and leave attached buffers clean, because the
3465 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3466 * or jbddirty because all the journalling code will explode.
3468 * So what we do is to mark the page "pending dirty" and next time writepage
3469 * is called, propagate that into the buffers appropriately.
3471 static int ext4_journalled_set_page_dirty(struct page
*page
)
3473 SetPageChecked(page
);
3474 return __set_page_dirty_nobuffers(page
);
3477 static const struct address_space_operations ext4_aops
= {
3478 .readpage
= ext4_readpage
,
3479 .readpages
= ext4_readpages
,
3480 .writepage
= ext4_writepage
,
3481 .writepages
= ext4_writepages
,
3482 .write_begin
= ext4_write_begin
,
3483 .write_end
= ext4_write_end
,
3485 .invalidatepage
= ext4_invalidatepage
,
3486 .releasepage
= ext4_releasepage
,
3487 .direct_IO
= ext4_direct_IO
,
3488 .migratepage
= buffer_migrate_page
,
3489 .is_partially_uptodate
= block_is_partially_uptodate
,
3490 .error_remove_page
= generic_error_remove_page
,
3493 static const struct address_space_operations ext4_journalled_aops
= {
3494 .readpage
= ext4_readpage
,
3495 .readpages
= ext4_readpages
,
3496 .writepage
= ext4_writepage
,
3497 .writepages
= ext4_writepages
,
3498 .write_begin
= ext4_write_begin
,
3499 .write_end
= ext4_journalled_write_end
,
3500 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3502 .invalidatepage
= ext4_journalled_invalidatepage
,
3503 .releasepage
= ext4_releasepage
,
3504 .direct_IO
= ext4_direct_IO
,
3505 .is_partially_uptodate
= block_is_partially_uptodate
,
3506 .error_remove_page
= generic_error_remove_page
,
3509 static const struct address_space_operations ext4_da_aops
= {
3510 .readpage
= ext4_readpage
,
3511 .readpages
= ext4_readpages
,
3512 .writepage
= ext4_writepage
,
3513 .writepages
= ext4_writepages
,
3514 .write_begin
= ext4_da_write_begin
,
3515 .write_end
= ext4_da_write_end
,
3517 .invalidatepage
= ext4_da_invalidatepage
,
3518 .releasepage
= ext4_releasepage
,
3519 .direct_IO
= ext4_direct_IO
,
3520 .migratepage
= buffer_migrate_page
,
3521 .is_partially_uptodate
= block_is_partially_uptodate
,
3522 .error_remove_page
= generic_error_remove_page
,
3525 void ext4_set_aops(struct inode
*inode
)
3527 switch (ext4_inode_journal_mode(inode
)) {
3528 case EXT4_INODE_ORDERED_DATA_MODE
:
3529 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3531 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3532 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3534 case EXT4_INODE_JOURNAL_DATA_MODE
:
3535 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3540 if (test_opt(inode
->i_sb
, DELALLOC
))
3541 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3543 inode
->i_mapping
->a_ops
= &ext4_aops
;
3546 static int __ext4_block_zero_page_range(handle_t
*handle
,
3547 struct address_space
*mapping
, loff_t from
, loff_t length
)
3549 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3550 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3551 unsigned blocksize
, pos
;
3553 struct inode
*inode
= mapping
->host
;
3554 struct buffer_head
*bh
;
3558 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3559 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3563 blocksize
= inode
->i_sb
->s_blocksize
;
3565 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3567 if (!page_has_buffers(page
))
3568 create_empty_buffers(page
, blocksize
, 0);
3570 /* Find the buffer that contains "offset" */
3571 bh
= page_buffers(page
);
3573 while (offset
>= pos
) {
3574 bh
= bh
->b_this_page
;
3578 if (buffer_freed(bh
)) {
3579 BUFFER_TRACE(bh
, "freed: skip");
3582 if (!buffer_mapped(bh
)) {
3583 BUFFER_TRACE(bh
, "unmapped");
3584 ext4_get_block(inode
, iblock
, bh
, 0);
3585 /* unmapped? It's a hole - nothing to do */
3586 if (!buffer_mapped(bh
)) {
3587 BUFFER_TRACE(bh
, "still unmapped");
3592 /* Ok, it's mapped. Make sure it's up-to-date */
3593 if (PageUptodate(page
))
3594 set_buffer_uptodate(bh
);
3596 if (!buffer_uptodate(bh
)) {
3598 ll_rw_block(READ
, 1, &bh
);
3600 /* Uhhuh. Read error. Complain and punt. */
3601 if (!buffer_uptodate(bh
))
3603 if (S_ISREG(inode
->i_mode
) &&
3604 ext4_encrypted_inode(inode
)) {
3605 /* We expect the key to be set. */
3606 BUG_ON(!ext4_has_encryption_key(inode
));
3607 BUG_ON(blocksize
!= PAGE_CACHE_SIZE
);
3608 WARN_ON_ONCE(ext4_decrypt(page
));
3611 if (ext4_should_journal_data(inode
)) {
3612 BUFFER_TRACE(bh
, "get write access");
3613 err
= ext4_journal_get_write_access(handle
, bh
);
3617 zero_user(page
, offset
, length
);
3618 BUFFER_TRACE(bh
, "zeroed end of block");
3620 if (ext4_should_journal_data(inode
)) {
3621 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3624 mark_buffer_dirty(bh
);
3625 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3626 err
= ext4_jbd2_file_inode(handle
, inode
);
3631 page_cache_release(page
);
3636 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3637 * starting from file offset 'from'. The range to be zero'd must
3638 * be contained with in one block. If the specified range exceeds
3639 * the end of the block it will be shortened to end of the block
3640 * that cooresponds to 'from'
3642 static int ext4_block_zero_page_range(handle_t
*handle
,
3643 struct address_space
*mapping
, loff_t from
, loff_t length
)
3645 struct inode
*inode
= mapping
->host
;
3646 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3647 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3648 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3651 * correct length if it does not fall between
3652 * 'from' and the end of the block
3654 if (length
> max
|| length
< 0)
3658 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3659 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3663 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3664 * up to the end of the block which corresponds to `from'.
3665 * This required during truncate. We need to physically zero the tail end
3666 * of that block so it doesn't yield old data if the file is later grown.
3668 static int ext4_block_truncate_page(handle_t
*handle
,
3669 struct address_space
*mapping
, loff_t from
)
3671 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3674 struct inode
*inode
= mapping
->host
;
3676 blocksize
= inode
->i_sb
->s_blocksize
;
3677 length
= blocksize
- (offset
& (blocksize
- 1));
3679 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3682 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3683 loff_t lstart
, loff_t length
)
3685 struct super_block
*sb
= inode
->i_sb
;
3686 struct address_space
*mapping
= inode
->i_mapping
;
3687 unsigned partial_start
, partial_end
;
3688 ext4_fsblk_t start
, end
;
3689 loff_t byte_end
= (lstart
+ length
- 1);
3692 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3693 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3695 start
= lstart
>> sb
->s_blocksize_bits
;
3696 end
= byte_end
>> sb
->s_blocksize_bits
;
3698 /* Handle partial zero within the single block */
3700 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3701 err
= ext4_block_zero_page_range(handle
, mapping
,
3705 /* Handle partial zero out on the start of the range */
3706 if (partial_start
) {
3707 err
= ext4_block_zero_page_range(handle
, mapping
,
3708 lstart
, sb
->s_blocksize
);
3712 /* Handle partial zero out on the end of the range */
3713 if (partial_end
!= sb
->s_blocksize
- 1)
3714 err
= ext4_block_zero_page_range(handle
, mapping
,
3715 byte_end
- partial_end
,
3720 int ext4_can_truncate(struct inode
*inode
)
3722 if (S_ISREG(inode
->i_mode
))
3724 if (S_ISDIR(inode
->i_mode
))
3726 if (S_ISLNK(inode
->i_mode
))
3727 return !ext4_inode_is_fast_symlink(inode
);
3732 * We have to make sure i_disksize gets properly updated before we truncate
3733 * page cache due to hole punching or zero range. Otherwise i_disksize update
3734 * can get lost as it may have been postponed to submission of writeback but
3735 * that will never happen after we truncate page cache.
3737 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3741 loff_t size
= i_size_read(inode
);
3743 WARN_ON(!inode_is_locked(inode
));
3744 if (offset
> size
|| offset
+ len
< size
)
3747 if (EXT4_I(inode
)->i_disksize
>= size
)
3750 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3752 return PTR_ERR(handle
);
3753 ext4_update_i_disksize(inode
, size
);
3754 ext4_mark_inode_dirty(handle
, inode
);
3755 ext4_journal_stop(handle
);
3761 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3762 * associated with the given offset and length
3764 * @inode: File inode
3765 * @offset: The offset where the hole will begin
3766 * @len: The length of the hole
3768 * Returns: 0 on success or negative on failure
3771 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3773 struct super_block
*sb
= inode
->i_sb
;
3774 ext4_lblk_t first_block
, stop_block
;
3775 struct address_space
*mapping
= inode
->i_mapping
;
3776 loff_t first_block_offset
, last_block_offset
;
3778 unsigned int credits
;
3781 if (!S_ISREG(inode
->i_mode
))
3784 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3787 * Write out all dirty pages to avoid race conditions
3788 * Then release them.
3790 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3791 ret
= filemap_write_and_wait_range(mapping
, offset
,
3792 offset
+ length
- 1);
3799 /* No need to punch hole beyond i_size */
3800 if (offset
>= inode
->i_size
)
3804 * If the hole extends beyond i_size, set the hole
3805 * to end after the page that contains i_size
3807 if (offset
+ length
> inode
->i_size
) {
3808 length
= inode
->i_size
+
3809 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3813 if (offset
& (sb
->s_blocksize
- 1) ||
3814 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3816 * Attach jinode to inode for jbd2 if we do any zeroing of
3819 ret
= ext4_inode_attach_jinode(inode
);
3825 /* Wait all existing dio workers, newcomers will block on i_mutex */
3826 ext4_inode_block_unlocked_dio(inode
);
3827 inode_dio_wait(inode
);
3830 * Prevent page faults from reinstantiating pages we have released from
3833 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3834 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3835 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3837 /* Now release the pages and zero block aligned part of pages*/
3838 if (last_block_offset
> first_block_offset
) {
3839 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
3842 truncate_pagecache_range(inode
, first_block_offset
,
3846 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3847 credits
= ext4_writepage_trans_blocks(inode
);
3849 credits
= ext4_blocks_for_truncate(inode
);
3850 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3851 if (IS_ERR(handle
)) {
3852 ret
= PTR_ERR(handle
);
3853 ext4_std_error(sb
, ret
);
3857 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3862 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3863 EXT4_BLOCK_SIZE_BITS(sb
);
3864 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3866 /* If there are no blocks to remove, return now */
3867 if (first_block
>= stop_block
)
3870 down_write(&EXT4_I(inode
)->i_data_sem
);
3871 ext4_discard_preallocations(inode
);
3873 ret
= ext4_es_remove_extent(inode
, first_block
,
3874 stop_block
- first_block
);
3876 up_write(&EXT4_I(inode
)->i_data_sem
);
3880 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3881 ret
= ext4_ext_remove_space(inode
, first_block
,
3884 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3887 up_write(&EXT4_I(inode
)->i_data_sem
);
3889 ext4_handle_sync(handle
);
3891 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3892 ext4_mark_inode_dirty(handle
, inode
);
3894 ext4_journal_stop(handle
);
3896 up_write(&EXT4_I(inode
)->i_mmap_sem
);
3897 ext4_inode_resume_unlocked_dio(inode
);
3899 inode_unlock(inode
);
3903 int ext4_inode_attach_jinode(struct inode
*inode
)
3905 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3906 struct jbd2_inode
*jinode
;
3908 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3911 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3912 spin_lock(&inode
->i_lock
);
3915 spin_unlock(&inode
->i_lock
);
3918 ei
->jinode
= jinode
;
3919 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3922 spin_unlock(&inode
->i_lock
);
3923 if (unlikely(jinode
!= NULL
))
3924 jbd2_free_inode(jinode
);
3931 * We block out ext4_get_block() block instantiations across the entire
3932 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3933 * simultaneously on behalf of the same inode.
3935 * As we work through the truncate and commit bits of it to the journal there
3936 * is one core, guiding principle: the file's tree must always be consistent on
3937 * disk. We must be able to restart the truncate after a crash.
3939 * The file's tree may be transiently inconsistent in memory (although it
3940 * probably isn't), but whenever we close off and commit a journal transaction,
3941 * the contents of (the filesystem + the journal) must be consistent and
3942 * restartable. It's pretty simple, really: bottom up, right to left (although
3943 * left-to-right works OK too).
3945 * Note that at recovery time, journal replay occurs *before* the restart of
3946 * truncate against the orphan inode list.
3948 * The committed inode has the new, desired i_size (which is the same as
3949 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3950 * that this inode's truncate did not complete and it will again call
3951 * ext4_truncate() to have another go. So there will be instantiated blocks
3952 * to the right of the truncation point in a crashed ext4 filesystem. But
3953 * that's fine - as long as they are linked from the inode, the post-crash
3954 * ext4_truncate() run will find them and release them.
3956 void ext4_truncate(struct inode
*inode
)
3958 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3959 unsigned int credits
;
3961 struct address_space
*mapping
= inode
->i_mapping
;
3964 * There is a possibility that we're either freeing the inode
3965 * or it's a completely new inode. In those cases we might not
3966 * have i_mutex locked because it's not necessary.
3968 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3969 WARN_ON(!inode_is_locked(inode
));
3970 trace_ext4_truncate_enter(inode
);
3972 if (!ext4_can_truncate(inode
))
3975 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3977 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3978 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3980 if (ext4_has_inline_data(inode
)) {
3983 ext4_inline_data_truncate(inode
, &has_inline
);
3988 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3989 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3990 if (ext4_inode_attach_jinode(inode
) < 0)
3994 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3995 credits
= ext4_writepage_trans_blocks(inode
);
3997 credits
= ext4_blocks_for_truncate(inode
);
3999 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4000 if (IS_ERR(handle
)) {
4001 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
4005 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4006 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4009 * We add the inode to the orphan list, so that if this
4010 * truncate spans multiple transactions, and we crash, we will
4011 * resume the truncate when the filesystem recovers. It also
4012 * marks the inode dirty, to catch the new size.
4014 * Implication: the file must always be in a sane, consistent
4015 * truncatable state while each transaction commits.
4017 if (ext4_orphan_add(handle
, inode
))
4020 down_write(&EXT4_I(inode
)->i_data_sem
);
4022 ext4_discard_preallocations(inode
);
4024 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4025 ext4_ext_truncate(handle
, inode
);
4027 ext4_ind_truncate(handle
, inode
);
4029 up_write(&ei
->i_data_sem
);
4032 ext4_handle_sync(handle
);
4036 * If this was a simple ftruncate() and the file will remain alive,
4037 * then we need to clear up the orphan record which we created above.
4038 * However, if this was a real unlink then we were called by
4039 * ext4_evict_inode(), and we allow that function to clean up the
4040 * orphan info for us.
4043 ext4_orphan_del(handle
, inode
);
4045 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4046 ext4_mark_inode_dirty(handle
, inode
);
4047 ext4_journal_stop(handle
);
4049 trace_ext4_truncate_exit(inode
);
4053 * ext4_get_inode_loc returns with an extra refcount against the inode's
4054 * underlying buffer_head on success. If 'in_mem' is true, we have all
4055 * data in memory that is needed to recreate the on-disk version of this
4058 static int __ext4_get_inode_loc(struct inode
*inode
,
4059 struct ext4_iloc
*iloc
, int in_mem
)
4061 struct ext4_group_desc
*gdp
;
4062 struct buffer_head
*bh
;
4063 struct super_block
*sb
= inode
->i_sb
;
4065 int inodes_per_block
, inode_offset
;
4068 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4069 return -EFSCORRUPTED
;
4071 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4072 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4077 * Figure out the offset within the block group inode table
4079 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4080 inode_offset
= ((inode
->i_ino
- 1) %
4081 EXT4_INODES_PER_GROUP(sb
));
4082 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4083 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4085 bh
= sb_getblk(sb
, block
);
4088 if (!buffer_uptodate(bh
)) {
4092 * If the buffer has the write error flag, we have failed
4093 * to write out another inode in the same block. In this
4094 * case, we don't have to read the block because we may
4095 * read the old inode data successfully.
4097 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4098 set_buffer_uptodate(bh
);
4100 if (buffer_uptodate(bh
)) {
4101 /* someone brought it uptodate while we waited */
4107 * If we have all information of the inode in memory and this
4108 * is the only valid inode in the block, we need not read the
4112 struct buffer_head
*bitmap_bh
;
4115 start
= inode_offset
& ~(inodes_per_block
- 1);
4117 /* Is the inode bitmap in cache? */
4118 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4119 if (unlikely(!bitmap_bh
))
4123 * If the inode bitmap isn't in cache then the
4124 * optimisation may end up performing two reads instead
4125 * of one, so skip it.
4127 if (!buffer_uptodate(bitmap_bh
)) {
4131 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4132 if (i
== inode_offset
)
4134 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4138 if (i
== start
+ inodes_per_block
) {
4139 /* all other inodes are free, so skip I/O */
4140 memset(bh
->b_data
, 0, bh
->b_size
);
4141 set_buffer_uptodate(bh
);
4149 * If we need to do any I/O, try to pre-readahead extra
4150 * blocks from the inode table.
4152 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4153 ext4_fsblk_t b
, end
, table
;
4155 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4157 table
= ext4_inode_table(sb
, gdp
);
4158 /* s_inode_readahead_blks is always a power of 2 */
4159 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4163 num
= EXT4_INODES_PER_GROUP(sb
);
4164 if (ext4_has_group_desc_csum(sb
))
4165 num
-= ext4_itable_unused_count(sb
, gdp
);
4166 table
+= num
/ inodes_per_block
;
4170 sb_breadahead(sb
, b
++);
4174 * There are other valid inodes in the buffer, this inode
4175 * has in-inode xattrs, or we don't have this inode in memory.
4176 * Read the block from disk.
4178 trace_ext4_load_inode(inode
);
4180 bh
->b_end_io
= end_buffer_read_sync
;
4181 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4183 if (!buffer_uptodate(bh
)) {
4184 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4185 "unable to read itable block");
4195 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4197 /* We have all inode data except xattrs in memory here. */
4198 return __ext4_get_inode_loc(inode
, iloc
,
4199 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4202 void ext4_set_inode_flags(struct inode
*inode
)
4204 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4205 unsigned int new_fl
= 0;
4207 if (flags
& EXT4_SYNC_FL
)
4209 if (flags
& EXT4_APPEND_FL
)
4211 if (flags
& EXT4_IMMUTABLE_FL
)
4212 new_fl
|= S_IMMUTABLE
;
4213 if (flags
& EXT4_NOATIME_FL
)
4214 new_fl
|= S_NOATIME
;
4215 if (flags
& EXT4_DIRSYNC_FL
)
4216 new_fl
|= S_DIRSYNC
;
4217 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
))
4219 inode_set_flags(inode
, new_fl
,
4220 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4223 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4224 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4226 unsigned int vfs_fl
;
4227 unsigned long old_fl
, new_fl
;
4230 vfs_fl
= ei
->vfs_inode
.i_flags
;
4231 old_fl
= ei
->i_flags
;
4232 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4233 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4235 if (vfs_fl
& S_SYNC
)
4236 new_fl
|= EXT4_SYNC_FL
;
4237 if (vfs_fl
& S_APPEND
)
4238 new_fl
|= EXT4_APPEND_FL
;
4239 if (vfs_fl
& S_IMMUTABLE
)
4240 new_fl
|= EXT4_IMMUTABLE_FL
;
4241 if (vfs_fl
& S_NOATIME
)
4242 new_fl
|= EXT4_NOATIME_FL
;
4243 if (vfs_fl
& S_DIRSYNC
)
4244 new_fl
|= EXT4_DIRSYNC_FL
;
4245 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4248 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4249 struct ext4_inode_info
*ei
)
4252 struct inode
*inode
= &(ei
->vfs_inode
);
4253 struct super_block
*sb
= inode
->i_sb
;
4255 if (ext4_has_feature_huge_file(sb
)) {
4256 /* we are using combined 48 bit field */
4257 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4258 le32_to_cpu(raw_inode
->i_blocks_lo
);
4259 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4260 /* i_blocks represent file system block size */
4261 return i_blocks
<< (inode
->i_blkbits
- 9);
4266 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4270 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4271 struct ext4_inode
*raw_inode
,
4272 struct ext4_inode_info
*ei
)
4274 __le32
*magic
= (void *)raw_inode
+
4275 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4276 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4277 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4278 ext4_find_inline_data_nolock(inode
);
4280 EXT4_I(inode
)->i_inline_off
= 0;
4283 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4285 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
, EXT4_FEATURE_RO_COMPAT_PROJECT
))
4287 *projid
= EXT4_I(inode
)->i_projid
;
4291 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4293 struct ext4_iloc iloc
;
4294 struct ext4_inode
*raw_inode
;
4295 struct ext4_inode_info
*ei
;
4296 struct inode
*inode
;
4297 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4304 inode
= iget_locked(sb
, ino
);
4306 return ERR_PTR(-ENOMEM
);
4307 if (!(inode
->i_state
& I_NEW
))
4313 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4316 raw_inode
= ext4_raw_inode(&iloc
);
4318 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4319 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4320 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4321 EXT4_INODE_SIZE(inode
->i_sb
)) {
4322 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4323 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4324 EXT4_INODE_SIZE(inode
->i_sb
));
4325 ret
= -EFSCORRUPTED
;
4329 ei
->i_extra_isize
= 0;
4331 /* Precompute checksum seed for inode metadata */
4332 if (ext4_has_metadata_csum(sb
)) {
4333 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4335 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4336 __le32 gen
= raw_inode
->i_generation
;
4337 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4339 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4343 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4344 EXT4_ERROR_INODE(inode
, "checksum invalid");
4349 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4350 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4351 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4352 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_PROJECT
) &&
4353 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4354 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4355 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4357 i_projid
= EXT4_DEF_PROJID
;
4359 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4360 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4361 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4363 i_uid_write(inode
, i_uid
);
4364 i_gid_write(inode
, i_gid
);
4365 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4366 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4368 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4369 ei
->i_inline_off
= 0;
4370 ei
->i_dir_start_lookup
= 0;
4371 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4372 /* We now have enough fields to check if the inode was active or not.
4373 * This is needed because nfsd might try to access dead inodes
4374 * the test is that same one that e2fsck uses
4375 * NeilBrown 1999oct15
4377 if (inode
->i_nlink
== 0) {
4378 if ((inode
->i_mode
== 0 ||
4379 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4380 ino
!= EXT4_BOOT_LOADER_INO
) {
4381 /* this inode is deleted */
4385 /* The only unlinked inodes we let through here have
4386 * valid i_mode and are being read by the orphan
4387 * recovery code: that's fine, we're about to complete
4388 * the process of deleting those.
4389 * OR it is the EXT4_BOOT_LOADER_INO which is
4390 * not initialized on a new filesystem. */
4392 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4393 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4394 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4395 if (ext4_has_feature_64bit(sb
))
4397 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4398 inode
->i_size
= ext4_isize(raw_inode
);
4399 ei
->i_disksize
= inode
->i_size
;
4401 ei
->i_reserved_quota
= 0;
4403 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4404 ei
->i_block_group
= iloc
.block_group
;
4405 ei
->i_last_alloc_group
= ~0;
4407 * NOTE! The in-memory inode i_data array is in little-endian order
4408 * even on big-endian machines: we do NOT byteswap the block numbers!
4410 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4411 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4412 INIT_LIST_HEAD(&ei
->i_orphan
);
4415 * Set transaction id's of transactions that have to be committed
4416 * to finish f[data]sync. We set them to currently running transaction
4417 * as we cannot be sure that the inode or some of its metadata isn't
4418 * part of the transaction - the inode could have been reclaimed and
4419 * now it is reread from disk.
4422 transaction_t
*transaction
;
4425 read_lock(&journal
->j_state_lock
);
4426 if (journal
->j_running_transaction
)
4427 transaction
= journal
->j_running_transaction
;
4429 transaction
= journal
->j_committing_transaction
;
4431 tid
= transaction
->t_tid
;
4433 tid
= journal
->j_commit_sequence
;
4434 read_unlock(&journal
->j_state_lock
);
4435 ei
->i_sync_tid
= tid
;
4436 ei
->i_datasync_tid
= tid
;
4439 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4440 if (ei
->i_extra_isize
== 0) {
4441 /* The extra space is currently unused. Use it. */
4442 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4443 EXT4_GOOD_OLD_INODE_SIZE
;
4445 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4449 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4450 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4451 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4452 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4454 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4455 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4456 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4457 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4459 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4464 if (ei
->i_file_acl
&&
4465 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4466 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4468 ret
= -EFSCORRUPTED
;
4470 } else if (!ext4_has_inline_data(inode
)) {
4471 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4472 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4473 (S_ISLNK(inode
->i_mode
) &&
4474 !ext4_inode_is_fast_symlink(inode
))))
4475 /* Validate extent which is part of inode */
4476 ret
= ext4_ext_check_inode(inode
);
4477 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4478 (S_ISLNK(inode
->i_mode
) &&
4479 !ext4_inode_is_fast_symlink(inode
))) {
4480 /* Validate block references which are part of inode */
4481 ret
= ext4_ind_check_inode(inode
);
4487 if (S_ISREG(inode
->i_mode
)) {
4488 inode
->i_op
= &ext4_file_inode_operations
;
4489 inode
->i_fop
= &ext4_file_operations
;
4490 ext4_set_aops(inode
);
4491 } else if (S_ISDIR(inode
->i_mode
)) {
4492 inode
->i_op
= &ext4_dir_inode_operations
;
4493 inode
->i_fop
= &ext4_dir_operations
;
4494 } else if (S_ISLNK(inode
->i_mode
)) {
4495 if (ext4_encrypted_inode(inode
)) {
4496 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4497 ext4_set_aops(inode
);
4498 } else if (ext4_inode_is_fast_symlink(inode
)) {
4499 inode
->i_link
= (char *)ei
->i_data
;
4500 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4501 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4502 sizeof(ei
->i_data
) - 1);
4504 inode
->i_op
= &ext4_symlink_inode_operations
;
4505 ext4_set_aops(inode
);
4507 inode_nohighmem(inode
);
4508 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4509 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4510 inode
->i_op
= &ext4_special_inode_operations
;
4511 if (raw_inode
->i_block
[0])
4512 init_special_inode(inode
, inode
->i_mode
,
4513 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4515 init_special_inode(inode
, inode
->i_mode
,
4516 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4517 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4518 make_bad_inode(inode
);
4520 ret
= -EFSCORRUPTED
;
4521 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4525 ext4_set_inode_flags(inode
);
4526 unlock_new_inode(inode
);
4532 return ERR_PTR(ret
);
4535 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4537 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4538 return ERR_PTR(-EFSCORRUPTED
);
4539 return ext4_iget(sb
, ino
);
4542 static int ext4_inode_blocks_set(handle_t
*handle
,
4543 struct ext4_inode
*raw_inode
,
4544 struct ext4_inode_info
*ei
)
4546 struct inode
*inode
= &(ei
->vfs_inode
);
4547 u64 i_blocks
= inode
->i_blocks
;
4548 struct super_block
*sb
= inode
->i_sb
;
4550 if (i_blocks
<= ~0U) {
4552 * i_blocks can be represented in a 32 bit variable
4553 * as multiple of 512 bytes
4555 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4556 raw_inode
->i_blocks_high
= 0;
4557 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4560 if (!ext4_has_feature_huge_file(sb
))
4563 if (i_blocks
<= 0xffffffffffffULL
) {
4565 * i_blocks can be represented in a 48 bit variable
4566 * as multiple of 512 bytes
4568 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4569 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4570 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4572 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4573 /* i_block is stored in file system block size */
4574 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4575 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4576 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4581 struct other_inode
{
4582 unsigned long orig_ino
;
4583 struct ext4_inode
*raw_inode
;
4586 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4589 struct other_inode
*oi
= (struct other_inode
*) data
;
4591 if ((inode
->i_ino
!= ino
) ||
4592 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4593 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4594 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4596 spin_lock(&inode
->i_lock
);
4597 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4598 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4599 (inode
->i_state
& I_DIRTY_TIME
)) {
4600 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4602 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4603 spin_unlock(&inode
->i_lock
);
4605 spin_lock(&ei
->i_raw_lock
);
4606 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4607 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4608 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4609 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4610 spin_unlock(&ei
->i_raw_lock
);
4611 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4614 spin_unlock(&inode
->i_lock
);
4619 * Opportunistically update the other time fields for other inodes in
4620 * the same inode table block.
4622 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4623 unsigned long orig_ino
, char *buf
)
4625 struct other_inode oi
;
4627 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4628 int inode_size
= EXT4_INODE_SIZE(sb
);
4630 oi
.orig_ino
= orig_ino
;
4632 * Calculate the first inode in the inode table block. Inode
4633 * numbers are one-based. That is, the first inode in a block
4634 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4636 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4637 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4638 if (ino
== orig_ino
)
4640 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4641 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4646 * Post the struct inode info into an on-disk inode location in the
4647 * buffer-cache. This gobbles the caller's reference to the
4648 * buffer_head in the inode location struct.
4650 * The caller must have write access to iloc->bh.
4652 static int ext4_do_update_inode(handle_t
*handle
,
4653 struct inode
*inode
,
4654 struct ext4_iloc
*iloc
)
4656 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4657 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4658 struct buffer_head
*bh
= iloc
->bh
;
4659 struct super_block
*sb
= inode
->i_sb
;
4660 int err
= 0, rc
, block
;
4661 int need_datasync
= 0, set_large_file
= 0;
4666 spin_lock(&ei
->i_raw_lock
);
4668 /* For fields not tracked in the in-memory inode,
4669 * initialise them to zero for new inodes. */
4670 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4671 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4673 ext4_get_inode_flags(ei
);
4674 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4675 i_uid
= i_uid_read(inode
);
4676 i_gid
= i_gid_read(inode
);
4677 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4678 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4679 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4680 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4682 * Fix up interoperability with old kernels. Otherwise, old inodes get
4683 * re-used with the upper 16 bits of the uid/gid intact
4686 raw_inode
->i_uid_high
=
4687 cpu_to_le16(high_16_bits(i_uid
));
4688 raw_inode
->i_gid_high
=
4689 cpu_to_le16(high_16_bits(i_gid
));
4691 raw_inode
->i_uid_high
= 0;
4692 raw_inode
->i_gid_high
= 0;
4695 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4696 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4697 raw_inode
->i_uid_high
= 0;
4698 raw_inode
->i_gid_high
= 0;
4700 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4702 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4703 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4704 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4705 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4707 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4709 spin_unlock(&ei
->i_raw_lock
);
4712 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4713 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4714 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4715 raw_inode
->i_file_acl_high
=
4716 cpu_to_le16(ei
->i_file_acl
>> 32);
4717 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4718 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4719 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4722 if (ei
->i_disksize
> 0x7fffffffULL
) {
4723 if (!ext4_has_feature_large_file(sb
) ||
4724 EXT4_SB(sb
)->s_es
->s_rev_level
==
4725 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4728 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4729 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4730 if (old_valid_dev(inode
->i_rdev
)) {
4731 raw_inode
->i_block
[0] =
4732 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4733 raw_inode
->i_block
[1] = 0;
4735 raw_inode
->i_block
[0] = 0;
4736 raw_inode
->i_block
[1] =
4737 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4738 raw_inode
->i_block
[2] = 0;
4740 } else if (!ext4_has_inline_data(inode
)) {
4741 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4742 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4745 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4746 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4747 if (ei
->i_extra_isize
) {
4748 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4749 raw_inode
->i_version_hi
=
4750 cpu_to_le32(inode
->i_version
>> 32);
4751 raw_inode
->i_extra_isize
=
4752 cpu_to_le16(ei
->i_extra_isize
);
4756 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
4757 EXT4_FEATURE_RO_COMPAT_PROJECT
) &&
4758 i_projid
!= EXT4_DEF_PROJID
);
4760 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4761 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4762 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4764 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4765 spin_unlock(&ei
->i_raw_lock
);
4766 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4767 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4770 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4771 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4774 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4775 if (set_large_file
) {
4776 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4777 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4780 ext4_update_dynamic_rev(sb
);
4781 ext4_set_feature_large_file(sb
);
4782 ext4_handle_sync(handle
);
4783 err
= ext4_handle_dirty_super(handle
, sb
);
4785 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4788 ext4_std_error(inode
->i_sb
, err
);
4793 * ext4_write_inode()
4795 * We are called from a few places:
4797 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4798 * Here, there will be no transaction running. We wait for any running
4799 * transaction to commit.
4801 * - Within flush work (sys_sync(), kupdate and such).
4802 * We wait on commit, if told to.
4804 * - Within iput_final() -> write_inode_now()
4805 * We wait on commit, if told to.
4807 * In all cases it is actually safe for us to return without doing anything,
4808 * because the inode has been copied into a raw inode buffer in
4809 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4812 * Note that we are absolutely dependent upon all inode dirtiers doing the
4813 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4814 * which we are interested.
4816 * It would be a bug for them to not do this. The code:
4818 * mark_inode_dirty(inode)
4820 * inode->i_size = expr;
4822 * is in error because write_inode() could occur while `stuff()' is running,
4823 * and the new i_size will be lost. Plus the inode will no longer be on the
4824 * superblock's dirty inode list.
4826 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4830 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4833 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4834 if (ext4_journal_current_handle()) {
4835 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4841 * No need to force transaction in WB_SYNC_NONE mode. Also
4842 * ext4_sync_fs() will force the commit after everything is
4845 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4848 err
= ext4_force_commit(inode
->i_sb
);
4850 struct ext4_iloc iloc
;
4852 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4856 * sync(2) will flush the whole buffer cache. No need to do
4857 * it here separately for each inode.
4859 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4860 sync_dirty_buffer(iloc
.bh
);
4861 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4862 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4863 "IO error syncing inode");
4872 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4873 * buffers that are attached to a page stradding i_size and are undergoing
4874 * commit. In that case we have to wait for commit to finish and try again.
4876 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4880 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4881 tid_t commit_tid
= 0;
4884 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4886 * All buffers in the last page remain valid? Then there's nothing to
4887 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4890 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4893 page
= find_lock_page(inode
->i_mapping
,
4894 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4897 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4898 PAGE_CACHE_SIZE
- offset
);
4900 page_cache_release(page
);
4904 read_lock(&journal
->j_state_lock
);
4905 if (journal
->j_committing_transaction
)
4906 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4907 read_unlock(&journal
->j_state_lock
);
4909 jbd2_log_wait_commit(journal
, commit_tid
);
4916 * Called from notify_change.
4918 * We want to trap VFS attempts to truncate the file as soon as
4919 * possible. In particular, we want to make sure that when the VFS
4920 * shrinks i_size, we put the inode on the orphan list and modify
4921 * i_disksize immediately, so that during the subsequent flushing of
4922 * dirty pages and freeing of disk blocks, we can guarantee that any
4923 * commit will leave the blocks being flushed in an unused state on
4924 * disk. (On recovery, the inode will get truncated and the blocks will
4925 * be freed, so we have a strong guarantee that no future commit will
4926 * leave these blocks visible to the user.)
4928 * Another thing we have to assure is that if we are in ordered mode
4929 * and inode is still attached to the committing transaction, we must
4930 * we start writeout of all the dirty pages which are being truncated.
4931 * This way we are sure that all the data written in the previous
4932 * transaction are already on disk (truncate waits for pages under
4935 * Called with inode->i_mutex down.
4937 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4939 struct inode
*inode
= d_inode(dentry
);
4942 const unsigned int ia_valid
= attr
->ia_valid
;
4944 error
= inode_change_ok(inode
, attr
);
4948 if (is_quota_modification(inode
, attr
)) {
4949 error
= dquot_initialize(inode
);
4953 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4954 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4957 /* (user+group)*(old+new) structure, inode write (sb,
4958 * inode block, ? - but truncate inode update has it) */
4959 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4960 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4961 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4962 if (IS_ERR(handle
)) {
4963 error
= PTR_ERR(handle
);
4966 error
= dquot_transfer(inode
, attr
);
4968 ext4_journal_stop(handle
);
4971 /* Update corresponding info in inode so that everything is in
4972 * one transaction */
4973 if (attr
->ia_valid
& ATTR_UID
)
4974 inode
->i_uid
= attr
->ia_uid
;
4975 if (attr
->ia_valid
& ATTR_GID
)
4976 inode
->i_gid
= attr
->ia_gid
;
4977 error
= ext4_mark_inode_dirty(handle
, inode
);
4978 ext4_journal_stop(handle
);
4981 if (attr
->ia_valid
& ATTR_SIZE
) {
4983 loff_t oldsize
= inode
->i_size
;
4984 int shrink
= (attr
->ia_size
<= inode
->i_size
);
4986 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4987 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4989 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4992 if (!S_ISREG(inode
->i_mode
))
4995 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4996 inode_inc_iversion(inode
);
4998 if (ext4_should_order_data(inode
) &&
4999 (attr
->ia_size
< inode
->i_size
)) {
5000 error
= ext4_begin_ordered_truncate(inode
,
5005 if (attr
->ia_size
!= inode
->i_size
) {
5006 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5007 if (IS_ERR(handle
)) {
5008 error
= PTR_ERR(handle
);
5011 if (ext4_handle_valid(handle
) && shrink
) {
5012 error
= ext4_orphan_add(handle
, inode
);
5016 * Update c/mtime on truncate up, ext4_truncate() will
5017 * update c/mtime in shrink case below
5020 inode
->i_mtime
= ext4_current_time(inode
);
5021 inode
->i_ctime
= inode
->i_mtime
;
5023 down_write(&EXT4_I(inode
)->i_data_sem
);
5024 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5025 rc
= ext4_mark_inode_dirty(handle
, inode
);
5029 * We have to update i_size under i_data_sem together
5030 * with i_disksize to avoid races with writeback code
5031 * running ext4_wb_update_i_disksize().
5034 i_size_write(inode
, attr
->ia_size
);
5035 up_write(&EXT4_I(inode
)->i_data_sem
);
5036 ext4_journal_stop(handle
);
5039 ext4_orphan_del(NULL
, inode
);
5044 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5047 * Blocks are going to be removed from the inode. Wait
5048 * for dio in flight. Temporarily disable
5049 * dioread_nolock to prevent livelock.
5052 if (!ext4_should_journal_data(inode
)) {
5053 ext4_inode_block_unlocked_dio(inode
);
5054 inode_dio_wait(inode
);
5055 ext4_inode_resume_unlocked_dio(inode
);
5057 ext4_wait_for_tail_page_commit(inode
);
5059 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5061 * Truncate pagecache after we've waited for commit
5062 * in data=journal mode to make pages freeable.
5064 truncate_pagecache(inode
, inode
->i_size
);
5066 ext4_truncate(inode
);
5067 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5071 setattr_copy(inode
, attr
);
5072 mark_inode_dirty(inode
);
5076 * If the call to ext4_truncate failed to get a transaction handle at
5077 * all, we need to clean up the in-core orphan list manually.
5079 if (orphan
&& inode
->i_nlink
)
5080 ext4_orphan_del(NULL
, inode
);
5082 if (!rc
&& (ia_valid
& ATTR_MODE
))
5083 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5086 ext4_std_error(inode
->i_sb
, error
);
5092 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5095 struct inode
*inode
;
5096 unsigned long long delalloc_blocks
;
5098 inode
= d_inode(dentry
);
5099 generic_fillattr(inode
, stat
);
5102 * If there is inline data in the inode, the inode will normally not
5103 * have data blocks allocated (it may have an external xattr block).
5104 * Report at least one sector for such files, so tools like tar, rsync,
5105 * others doen't incorrectly think the file is completely sparse.
5107 if (unlikely(ext4_has_inline_data(inode
)))
5108 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5111 * We can't update i_blocks if the block allocation is delayed
5112 * otherwise in the case of system crash before the real block
5113 * allocation is done, we will have i_blocks inconsistent with
5114 * on-disk file blocks.
5115 * We always keep i_blocks updated together with real
5116 * allocation. But to not confuse with user, stat
5117 * will return the blocks that include the delayed allocation
5118 * blocks for this file.
5120 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5121 EXT4_I(inode
)->i_reserved_data_blocks
);
5122 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5126 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5129 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5130 return ext4_ind_trans_blocks(inode
, lblocks
);
5131 return ext4_ext_index_trans_blocks(inode
, pextents
);
5135 * Account for index blocks, block groups bitmaps and block group
5136 * descriptor blocks if modify datablocks and index blocks
5137 * worse case, the indexs blocks spread over different block groups
5139 * If datablocks are discontiguous, they are possible to spread over
5140 * different block groups too. If they are contiguous, with flexbg,
5141 * they could still across block group boundary.
5143 * Also account for superblock, inode, quota and xattr blocks
5145 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5148 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5154 * How many index blocks need to touch to map @lblocks logical blocks
5155 * to @pextents physical extents?
5157 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5162 * Now let's see how many group bitmaps and group descriptors need
5165 groups
= idxblocks
+ pextents
;
5167 if (groups
> ngroups
)
5169 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5170 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5172 /* bitmaps and block group descriptor blocks */
5173 ret
+= groups
+ gdpblocks
;
5175 /* Blocks for super block, inode, quota and xattr blocks */
5176 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5182 * Calculate the total number of credits to reserve to fit
5183 * the modification of a single pages into a single transaction,
5184 * which may include multiple chunks of block allocations.
5186 * This could be called via ext4_write_begin()
5188 * We need to consider the worse case, when
5189 * one new block per extent.
5191 int ext4_writepage_trans_blocks(struct inode
*inode
)
5193 int bpp
= ext4_journal_blocks_per_page(inode
);
5196 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5198 /* Account for data blocks for journalled mode */
5199 if (ext4_should_journal_data(inode
))
5205 * Calculate the journal credits for a chunk of data modification.
5207 * This is called from DIO, fallocate or whoever calling
5208 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5210 * journal buffers for data blocks are not included here, as DIO
5211 * and fallocate do no need to journal data buffers.
5213 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5215 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5219 * The caller must have previously called ext4_reserve_inode_write().
5220 * Give this, we know that the caller already has write access to iloc->bh.
5222 int ext4_mark_iloc_dirty(handle_t
*handle
,
5223 struct inode
*inode
, struct ext4_iloc
*iloc
)
5227 if (IS_I_VERSION(inode
))
5228 inode_inc_iversion(inode
);
5230 /* the do_update_inode consumes one bh->b_count */
5233 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5234 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5240 * On success, We end up with an outstanding reference count against
5241 * iloc->bh. This _must_ be cleaned up later.
5245 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5246 struct ext4_iloc
*iloc
)
5250 err
= ext4_get_inode_loc(inode
, iloc
);
5252 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5253 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5259 ext4_std_error(inode
->i_sb
, err
);
5264 * Expand an inode by new_extra_isize bytes.
5265 * Returns 0 on success or negative error number on failure.
5267 static int ext4_expand_extra_isize(struct inode
*inode
,
5268 unsigned int new_extra_isize
,
5269 struct ext4_iloc iloc
,
5272 struct ext4_inode
*raw_inode
;
5273 struct ext4_xattr_ibody_header
*header
;
5275 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5278 raw_inode
= ext4_raw_inode(&iloc
);
5280 header
= IHDR(inode
, raw_inode
);
5282 /* No extended attributes present */
5283 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5284 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5285 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5287 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5291 /* try to expand with EAs present */
5292 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5297 * What we do here is to mark the in-core inode as clean with respect to inode
5298 * dirtiness (it may still be data-dirty).
5299 * This means that the in-core inode may be reaped by prune_icache
5300 * without having to perform any I/O. This is a very good thing,
5301 * because *any* task may call prune_icache - even ones which
5302 * have a transaction open against a different journal.
5304 * Is this cheating? Not really. Sure, we haven't written the
5305 * inode out, but prune_icache isn't a user-visible syncing function.
5306 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5307 * we start and wait on commits.
5309 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5311 struct ext4_iloc iloc
;
5312 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5313 static unsigned int mnt_count
;
5317 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5318 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5321 if (ext4_handle_valid(handle
) &&
5322 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5323 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5325 * We need extra buffer credits since we may write into EA block
5326 * with this same handle. If journal_extend fails, then it will
5327 * only result in a minor loss of functionality for that inode.
5328 * If this is felt to be critical, then e2fsck should be run to
5329 * force a large enough s_min_extra_isize.
5331 if ((jbd2_journal_extend(handle
,
5332 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5333 ret
= ext4_expand_extra_isize(inode
,
5334 sbi
->s_want_extra_isize
,
5337 ext4_set_inode_state(inode
,
5338 EXT4_STATE_NO_EXPAND
);
5340 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5341 ext4_warning(inode
->i_sb
,
5342 "Unable to expand inode %lu. Delete"
5343 " some EAs or run e2fsck.",
5346 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5351 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5355 * ext4_dirty_inode() is called from __mark_inode_dirty()
5357 * We're really interested in the case where a file is being extended.
5358 * i_size has been changed by generic_commit_write() and we thus need
5359 * to include the updated inode in the current transaction.
5361 * Also, dquot_alloc_block() will always dirty the inode when blocks
5362 * are allocated to the file.
5364 * If the inode is marked synchronous, we don't honour that here - doing
5365 * so would cause a commit on atime updates, which we don't bother doing.
5366 * We handle synchronous inodes at the highest possible level.
5368 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5369 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5370 * to copy into the on-disk inode structure are the timestamp files.
5372 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5376 if (flags
== I_DIRTY_TIME
)
5378 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5382 ext4_mark_inode_dirty(handle
, inode
);
5384 ext4_journal_stop(handle
);
5391 * Bind an inode's backing buffer_head into this transaction, to prevent
5392 * it from being flushed to disk early. Unlike
5393 * ext4_reserve_inode_write, this leaves behind no bh reference and
5394 * returns no iloc structure, so the caller needs to repeat the iloc
5395 * lookup to mark the inode dirty later.
5397 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5399 struct ext4_iloc iloc
;
5403 err
= ext4_get_inode_loc(inode
, &iloc
);
5405 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5406 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5408 err
= ext4_handle_dirty_metadata(handle
,
5414 ext4_std_error(inode
->i_sb
, err
);
5419 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5426 * We have to be very careful here: changing a data block's
5427 * journaling status dynamically is dangerous. If we write a
5428 * data block to the journal, change the status and then delete
5429 * that block, we risk forgetting to revoke the old log record
5430 * from the journal and so a subsequent replay can corrupt data.
5431 * So, first we make sure that the journal is empty and that
5432 * nobody is changing anything.
5435 journal
= EXT4_JOURNAL(inode
);
5438 if (is_journal_aborted(journal
))
5440 /* We have to allocate physical blocks for delalloc blocks
5441 * before flushing journal. otherwise delalloc blocks can not
5442 * be allocated any more. even more truncate on delalloc blocks
5443 * could trigger BUG by flushing delalloc blocks in journal.
5444 * There is no delalloc block in non-journal data mode.
5446 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5447 err
= ext4_alloc_da_blocks(inode
);
5452 /* Wait for all existing dio workers */
5453 ext4_inode_block_unlocked_dio(inode
);
5454 inode_dio_wait(inode
);
5456 jbd2_journal_lock_updates(journal
);
5459 * OK, there are no updates running now, and all cached data is
5460 * synced to disk. We are now in a completely consistent state
5461 * which doesn't have anything in the journal, and we know that
5462 * no filesystem updates are running, so it is safe to modify
5463 * the inode's in-core data-journaling state flag now.
5467 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5469 err
= jbd2_journal_flush(journal
);
5471 jbd2_journal_unlock_updates(journal
);
5472 ext4_inode_resume_unlocked_dio(inode
);
5475 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5477 ext4_set_aops(inode
);
5479 jbd2_journal_unlock_updates(journal
);
5480 ext4_inode_resume_unlocked_dio(inode
);
5482 /* Finally we can mark the inode as dirty. */
5484 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5486 return PTR_ERR(handle
);
5488 err
= ext4_mark_inode_dirty(handle
, inode
);
5489 ext4_handle_sync(handle
);
5490 ext4_journal_stop(handle
);
5491 ext4_std_error(inode
->i_sb
, err
);
5496 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5498 return !buffer_mapped(bh
);
5501 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5503 struct page
*page
= vmf
->page
;
5507 struct file
*file
= vma
->vm_file
;
5508 struct inode
*inode
= file_inode(file
);
5509 struct address_space
*mapping
= inode
->i_mapping
;
5511 get_block_t
*get_block
;
5514 sb_start_pagefault(inode
->i_sb
);
5515 file_update_time(vma
->vm_file
);
5517 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5518 /* Delalloc case is easy... */
5519 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5520 !ext4_should_journal_data(inode
) &&
5521 !ext4_nonda_switch(inode
->i_sb
)) {
5523 ret
= block_page_mkwrite(vma
, vmf
,
5524 ext4_da_get_block_prep
);
5525 } while (ret
== -ENOSPC
&&
5526 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5531 size
= i_size_read(inode
);
5532 /* Page got truncated from under us? */
5533 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5535 ret
= VM_FAULT_NOPAGE
;
5539 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5540 len
= size
& ~PAGE_CACHE_MASK
;
5542 len
= PAGE_CACHE_SIZE
;
5544 * Return if we have all the buffers mapped. This avoids the need to do
5545 * journal_start/journal_stop which can block and take a long time
5547 if (page_has_buffers(page
)) {
5548 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5550 ext4_bh_unmapped
)) {
5551 /* Wait so that we don't change page under IO */
5552 wait_for_stable_page(page
);
5553 ret
= VM_FAULT_LOCKED
;
5558 /* OK, we need to fill the hole... */
5559 if (ext4_should_dioread_nolock(inode
))
5560 get_block
= ext4_get_block_unwritten
;
5562 get_block
= ext4_get_block
;
5564 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5565 ext4_writepage_trans_blocks(inode
));
5566 if (IS_ERR(handle
)) {
5567 ret
= VM_FAULT_SIGBUS
;
5570 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5571 if (!ret
&& ext4_should_journal_data(inode
)) {
5572 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5573 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5575 ret
= VM_FAULT_SIGBUS
;
5576 ext4_journal_stop(handle
);
5579 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5581 ext4_journal_stop(handle
);
5582 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5585 ret
= block_page_mkwrite_return(ret
);
5587 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5588 sb_end_pagefault(inode
->i_sb
);
5592 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5594 struct inode
*inode
= file_inode(vma
->vm_file
);
5597 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5598 err
= filemap_fault(vma
, vmf
);
5599 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5605 * Find the first extent at or after @lblk in an inode that is not a hole.
5606 * Search for @map_len blocks at most. The extent is returned in @result.
5608 * The function returns 1 if we found an extent. The function returns 0 in
5609 * case there is no extent at or after @lblk and in that case also sets
5610 * @result->es_len to 0. In case of error, the error code is returned.
5612 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
5613 unsigned int map_len
, struct extent_status
*result
)
5615 struct ext4_map_blocks map
;
5616 struct extent_status es
= {};
5620 map
.m_len
= map_len
;
5623 * For non-extent based files this loop may iterate several times since
5624 * we do not determine full hole size.
5626 while (map
.m_len
> 0) {
5627 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
5630 /* There's extent covering m_lblk? Just return it. */
5634 ext4_es_store_pblock(result
, map
.m_pblk
);
5635 result
->es_lblk
= map
.m_lblk
;
5636 result
->es_len
= map
.m_len
;
5637 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
5638 status
= EXTENT_STATUS_UNWRITTEN
;
5640 status
= EXTENT_STATUS_WRITTEN
;
5641 ext4_es_store_status(result
, status
);
5644 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
5645 map
.m_lblk
+ map
.m_len
- 1,
5647 /* Is delalloc data before next block in extent tree? */
5648 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
5649 ext4_lblk_t offset
= 0;
5651 if (es
.es_lblk
< lblk
)
5652 offset
= lblk
- es
.es_lblk
;
5653 result
->es_lblk
= es
.es_lblk
+ offset
;
5654 ext4_es_store_pblock(result
,
5655 ext4_es_pblock(&es
) + offset
);
5656 result
->es_len
= es
.es_len
- offset
;
5657 ext4_es_store_status(result
, ext4_es_status(&es
));
5661 /* There's a hole at m_lblk, advance us after it */
5662 map
.m_lblk
+= map
.m_len
;
5663 map_len
-= map
.m_len
;
5664 map
.m_len
= map_len
;