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
);
56 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
57 unsigned int csum_size
= sizeof(dummy_csum
);
59 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
60 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
62 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
63 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
65 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
66 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
67 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
68 EXT4_GOOD_OLD_INODE_SIZE
,
69 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
70 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
71 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
74 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
75 EXT4_INODE_SIZE(inode
->i_sb
) -
83 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
84 struct ext4_inode_info
*ei
)
86 __u32 provided
, calculated
;
88 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
89 cpu_to_le32(EXT4_OS_LINUX
) ||
90 !ext4_has_metadata_csum(inode
->i_sb
))
93 provided
= le16_to_cpu(raw
->i_checksum_lo
);
94 calculated
= ext4_inode_csum(inode
, raw
, ei
);
95 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
96 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
97 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
101 return provided
== calculated
;
104 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
105 struct ext4_inode_info
*ei
)
109 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
110 cpu_to_le32(EXT4_OS_LINUX
) ||
111 !ext4_has_metadata_csum(inode
->i_sb
))
114 csum
= ext4_inode_csum(inode
, raw
, ei
);
115 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
116 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
117 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
118 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
124 trace_ext4_begin_ordered_truncate(inode
, new_size
);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode
)->jinode
)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
134 EXT4_I(inode
)->jinode
,
138 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
139 unsigned int length
);
140 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
141 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
142 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
146 * Test whether an inode is a fast symlink.
148 int ext4_inode_is_fast_symlink(struct inode
*inode
)
150 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
151 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
153 if (ext4_has_inline_data(inode
))
156 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
160 * Restart the transaction associated with *handle. This does a commit,
161 * so before we call here everything must be consistently dirtied against
164 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
170 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
171 * moment, get_block can be called only for blocks inside i_size since
172 * page cache has been already dropped and writes are blocked by
173 * i_mutex. So we can safely drop the i_data_sem here.
175 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
176 jbd_debug(2, "restarting handle %p\n", handle
);
177 up_write(&EXT4_I(inode
)->i_data_sem
);
178 ret
= ext4_journal_restart(handle
, nblocks
);
179 down_write(&EXT4_I(inode
)->i_data_sem
);
180 ext4_discard_preallocations(inode
);
186 * Called at the last iput() if i_nlink is zero.
188 void ext4_evict_inode(struct inode
*inode
)
193 trace_ext4_evict_inode(inode
);
195 if (inode
->i_nlink
) {
197 * When journalling data dirty buffers are tracked only in the
198 * journal. So although mm thinks everything is clean and
199 * ready for reaping the inode might still have some pages to
200 * write in the running transaction or waiting to be
201 * checkpointed. Thus calling jbd2_journal_invalidatepage()
202 * (via truncate_inode_pages()) to discard these buffers can
203 * cause data loss. Also even if we did not discard these
204 * buffers, we would have no way to find them after the inode
205 * is reaped and thus user could see stale data if he tries to
206 * read them before the transaction is checkpointed. So be
207 * careful and force everything to disk here... We use
208 * ei->i_datasync_tid to store the newest transaction
209 * containing inode's data.
211 * Note that directories do not have this problem because they
212 * don't use page cache.
214 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
215 ext4_should_journal_data(inode
) &&
216 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
217 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
218 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
220 jbd2_complete_transaction(journal
, commit_tid
);
221 filemap_write_and_wait(&inode
->i_data
);
223 truncate_inode_pages_final(&inode
->i_data
);
228 if (is_bad_inode(inode
))
230 dquot_initialize(inode
);
232 if (ext4_should_order_data(inode
))
233 ext4_begin_ordered_truncate(inode
, 0);
234 truncate_inode_pages_final(&inode
->i_data
);
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode
->i_sb
);
241 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
242 ext4_blocks_for_truncate(inode
)+3);
243 if (IS_ERR(handle
)) {
244 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
246 * If we're going to skip the normal cleanup, we still need to
247 * make sure that the in-core orphan linked list is properly
250 ext4_orphan_del(NULL
, inode
);
251 sb_end_intwrite(inode
->i_sb
);
256 ext4_handle_sync(handle
);
258 err
= ext4_mark_inode_dirty(handle
, inode
);
260 ext4_warning(inode
->i_sb
,
261 "couldn't mark inode dirty (err %d)", err
);
265 ext4_truncate(inode
);
268 * ext4_ext_truncate() doesn't reserve any slop when it
269 * restarts journal transactions; therefore there may not be
270 * enough credits left in the handle to remove the inode from
271 * the orphan list and set the dtime field.
273 if (!ext4_handle_has_enough_credits(handle
, 3)) {
274 err
= ext4_journal_extend(handle
, 3);
276 err
= ext4_journal_restart(handle
, 3);
278 ext4_warning(inode
->i_sb
,
279 "couldn't extend journal (err %d)", err
);
281 ext4_journal_stop(handle
);
282 ext4_orphan_del(NULL
, inode
);
283 sb_end_intwrite(inode
->i_sb
);
289 * Kill off the orphan record which ext4_truncate created.
290 * AKPM: I think this can be inside the above `if'.
291 * Note that ext4_orphan_del() has to be able to cope with the
292 * deletion of a non-existent orphan - this is because we don't
293 * know if ext4_truncate() actually created an orphan record.
294 * (Well, we could do this if we need to, but heck - it works)
296 ext4_orphan_del(handle
, inode
);
297 EXT4_I(inode
)->i_dtime
= get_seconds();
300 * One subtle ordering requirement: if anything has gone wrong
301 * (transaction abort, IO errors, whatever), then we can still
302 * do these next steps (the fs will already have been marked as
303 * having errors), but we can't free the inode if the mark_dirty
306 if (ext4_mark_inode_dirty(handle
, inode
))
307 /* If that failed, just do the required in-core inode clear. */
308 ext4_clear_inode(inode
);
310 ext4_free_inode(handle
, inode
);
311 ext4_journal_stop(handle
);
312 sb_end_intwrite(inode
->i_sb
);
315 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
319 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
321 return &EXT4_I(inode
)->i_reserved_quota
;
326 * Called with i_data_sem down, which is important since we can call
327 * ext4_discard_preallocations() from here.
329 void ext4_da_update_reserve_space(struct inode
*inode
,
330 int used
, int quota_claim
)
332 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
333 struct ext4_inode_info
*ei
= EXT4_I(inode
);
335 spin_lock(&ei
->i_block_reservation_lock
);
336 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
337 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
338 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
339 "with only %d reserved data blocks",
340 __func__
, inode
->i_ino
, used
,
341 ei
->i_reserved_data_blocks
);
343 used
= ei
->i_reserved_data_blocks
;
346 /* Update per-inode reservations */
347 ei
->i_reserved_data_blocks
-= used
;
348 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
350 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
352 /* Update quota subsystem for data blocks */
354 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
357 * We did fallocate with an offset that is already delayed
358 * allocated. So on delayed allocated writeback we should
359 * not re-claim the quota for fallocated blocks.
361 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
365 * If we have done all the pending block allocations and if
366 * there aren't any writers on the inode, we can discard the
367 * inode's preallocations.
369 if ((ei
->i_reserved_data_blocks
== 0) &&
370 (atomic_read(&inode
->i_writecount
) == 0))
371 ext4_discard_preallocations(inode
);
374 static int __check_block_validity(struct inode
*inode
, const char *func
,
376 struct ext4_map_blocks
*map
)
378 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
380 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
381 "lblock %lu mapped to illegal pblock "
382 "(length %d)", (unsigned long) map
->m_lblk
,
384 return -EFSCORRUPTED
;
389 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
394 if (ext4_encrypted_inode(inode
))
395 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
397 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
404 #define check_block_validity(inode, map) \
405 __check_block_validity((inode), __func__, __LINE__, (map))
407 #ifdef ES_AGGRESSIVE_TEST
408 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
410 struct ext4_map_blocks
*es_map
,
411 struct ext4_map_blocks
*map
,
418 * There is a race window that the result is not the same.
419 * e.g. xfstests #223 when dioread_nolock enables. The reason
420 * is that we lookup a block mapping in extent status tree with
421 * out taking i_data_sem. So at the time the unwritten extent
422 * could be converted.
424 down_read(&EXT4_I(inode
)->i_data_sem
);
425 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
426 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
427 EXT4_GET_BLOCKS_KEEP_SIZE
);
429 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
430 EXT4_GET_BLOCKS_KEEP_SIZE
);
432 up_read((&EXT4_I(inode
)->i_data_sem
));
435 * We don't check m_len because extent will be collpased in status
436 * tree. So the m_len might not equal.
438 if (es_map
->m_lblk
!= map
->m_lblk
||
439 es_map
->m_flags
!= map
->m_flags
||
440 es_map
->m_pblk
!= map
->m_pblk
) {
441 printk("ES cache assertion failed for inode: %lu "
442 "es_cached ex [%d/%d/%llu/%x] != "
443 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
444 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
445 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
446 map
->m_len
, map
->m_pblk
, map
->m_flags
,
450 #endif /* ES_AGGRESSIVE_TEST */
453 * The ext4_map_blocks() function tries to look up the requested blocks,
454 * and returns if the blocks are already mapped.
456 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
457 * and store the allocated blocks in the result buffer head and mark it
460 * If file type is extents based, it will call ext4_ext_map_blocks(),
461 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
464 * On success, it returns the number of blocks being mapped or allocated. if
465 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
466 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
468 * It returns 0 if plain look up failed (blocks have not been allocated), in
469 * that case, @map is returned as unmapped but we still do fill map->m_len to
470 * indicate the length of a hole starting at map->m_lblk.
472 * It returns the error in case of allocation failure.
474 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
475 struct ext4_map_blocks
*map
, int flags
)
477 struct extent_status es
;
480 #ifdef ES_AGGRESSIVE_TEST
481 struct ext4_map_blocks orig_map
;
483 memcpy(&orig_map
, map
, sizeof(*map
));
487 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
488 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
489 (unsigned long) map
->m_lblk
);
492 * ext4_map_blocks returns an int, and m_len is an unsigned int
494 if (unlikely(map
->m_len
> INT_MAX
))
495 map
->m_len
= INT_MAX
;
497 /* We can handle the block number less than EXT_MAX_BLOCKS */
498 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
499 return -EFSCORRUPTED
;
501 /* Lookup extent status tree firstly */
502 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
503 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
504 map
->m_pblk
= ext4_es_pblock(&es
) +
505 map
->m_lblk
- es
.es_lblk
;
506 map
->m_flags
|= ext4_es_is_written(&es
) ?
507 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
508 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
509 if (retval
> map
->m_len
)
512 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
514 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
515 if (retval
> map
->m_len
)
522 #ifdef ES_AGGRESSIVE_TEST
523 ext4_map_blocks_es_recheck(handle
, inode
, map
,
530 * Try to see if we can get the block without requesting a new
533 down_read(&EXT4_I(inode
)->i_data_sem
);
534 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
535 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
536 EXT4_GET_BLOCKS_KEEP_SIZE
);
538 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
539 EXT4_GET_BLOCKS_KEEP_SIZE
);
544 if (unlikely(retval
!= map
->m_len
)) {
545 ext4_warning(inode
->i_sb
,
546 "ES len assertion failed for inode "
547 "%lu: retval %d != map->m_len %d",
548 inode
->i_ino
, retval
, map
->m_len
);
552 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
553 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
554 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
555 !(status
& EXTENT_STATUS_WRITTEN
) &&
556 ext4_find_delalloc_range(inode
, map
->m_lblk
,
557 map
->m_lblk
+ map
->m_len
- 1))
558 status
|= EXTENT_STATUS_DELAYED
;
559 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
560 map
->m_len
, map
->m_pblk
, status
);
564 up_read((&EXT4_I(inode
)->i_data_sem
));
567 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
568 ret
= check_block_validity(inode
, map
);
573 /* If it is only a block(s) look up */
574 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
578 * Returns if the blocks have already allocated
580 * Note that if blocks have been preallocated
581 * ext4_ext_get_block() returns the create = 0
582 * with buffer head unmapped.
584 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
586 * If we need to convert extent to unwritten
587 * we continue and do the actual work in
588 * ext4_ext_map_blocks()
590 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
594 * Here we clear m_flags because after allocating an new extent,
595 * it will be set again.
597 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
600 * New blocks allocate and/or writing to unwritten extent
601 * will possibly result in updating i_data, so we take
602 * the write lock of i_data_sem, and call get_block()
603 * with create == 1 flag.
605 down_write(&EXT4_I(inode
)->i_data_sem
);
608 * We need to check for EXT4 here because migrate
609 * could have changed the inode type in between
611 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
612 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
614 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
616 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
618 * We allocated new blocks which will result in
619 * i_data's format changing. Force the migrate
620 * to fail by clearing migrate flags
622 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
626 * Update reserved blocks/metadata blocks after successful
627 * block allocation which had been deferred till now. We don't
628 * support fallocate for non extent files. So we can update
629 * reserve space here.
632 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
633 ext4_da_update_reserve_space(inode
, retval
, 1);
639 if (unlikely(retval
!= map
->m_len
)) {
640 ext4_warning(inode
->i_sb
,
641 "ES len assertion failed for inode "
642 "%lu: retval %d != map->m_len %d",
643 inode
->i_ino
, retval
, map
->m_len
);
648 * We have to zeroout blocks before inserting them into extent
649 * status tree. Otherwise someone could look them up there and
650 * use them before they are really zeroed. We also have to
651 * unmap metadata before zeroing as otherwise writeback can
652 * overwrite zeros with stale data from block device.
654 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
655 map
->m_flags
& EXT4_MAP_MAPPED
&&
656 map
->m_flags
& EXT4_MAP_NEW
) {
659 for (i
= 0; i
< map
->m_len
; i
++) {
660 unmap_underlying_metadata(inode
->i_sb
->s_bdev
,
663 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
664 map
->m_pblk
, map
->m_len
);
672 * If the extent has been zeroed out, we don't need to update
673 * extent status tree.
675 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
676 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
677 if (ext4_es_is_written(&es
))
680 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
681 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
682 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
683 !(status
& EXTENT_STATUS_WRITTEN
) &&
684 ext4_find_delalloc_range(inode
, map
->m_lblk
,
685 map
->m_lblk
+ map
->m_len
- 1))
686 status
|= EXTENT_STATUS_DELAYED
;
687 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
688 map
->m_pblk
, status
);
696 up_write((&EXT4_I(inode
)->i_data_sem
));
697 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
698 ret
= check_block_validity(inode
, map
);
703 * Inodes with freshly allocated blocks where contents will be
704 * visible after transaction commit must be on transaction's
707 if (map
->m_flags
& EXT4_MAP_NEW
&&
708 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
709 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
710 !IS_NOQUOTA(inode
) &&
711 ext4_should_order_data(inode
)) {
712 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
713 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
715 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
724 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
725 * we have to be careful as someone else may be manipulating b_state as well.
727 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
729 unsigned long old_state
;
730 unsigned long new_state
;
732 flags
&= EXT4_MAP_FLAGS
;
734 /* Dummy buffer_head? Set non-atomically. */
736 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
740 * Someone else may be modifying b_state. Be careful! This is ugly but
741 * once we get rid of using bh as a container for mapping information
742 * to pass to / from get_block functions, this can go away.
745 old_state
= READ_ONCE(bh
->b_state
);
746 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
748 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
751 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
752 struct buffer_head
*bh
, int flags
)
754 struct ext4_map_blocks map
;
757 if (ext4_has_inline_data(inode
))
761 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
763 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
766 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
767 ext4_update_bh_state(bh
, map
.m_flags
);
768 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
770 } else if (ret
== 0) {
771 /* hole case, need to fill in bh->b_size */
772 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
777 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
778 struct buffer_head
*bh
, int create
)
780 return _ext4_get_block(inode
, iblock
, bh
,
781 create
? EXT4_GET_BLOCKS_CREATE
: 0);
785 * Get block function used when preparing for buffered write if we require
786 * creating an unwritten extent if blocks haven't been allocated. The extent
787 * will be converted to written after the IO is complete.
789 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
790 struct buffer_head
*bh_result
, int create
)
792 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
793 inode
->i_ino
, create
);
794 return _ext4_get_block(inode
, iblock
, bh_result
,
795 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
798 /* Maximum number of blocks we map for direct IO at once. */
799 #define DIO_MAX_BLOCKS 4096
802 * Get blocks function for the cases that need to start a transaction -
803 * generally difference cases of direct IO and DAX IO. It also handles retries
806 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
807 struct buffer_head
*bh_result
, int flags
)
814 /* Trim mapping request to maximum we can map at once for DIO */
815 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
816 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
817 dio_credits
= ext4_chunk_trans_blocks(inode
,
818 bh_result
->b_size
>> inode
->i_blkbits
);
820 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
822 return PTR_ERR(handle
);
824 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
825 ext4_journal_stop(handle
);
827 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
832 /* Get block function for DIO reads and writes to inodes without extents */
833 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
834 struct buffer_head
*bh
, int create
)
836 /* We don't expect handle for direct IO */
837 WARN_ON_ONCE(ext4_journal_current_handle());
840 return _ext4_get_block(inode
, iblock
, bh
, 0);
841 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
845 * Get block function for AIO DIO writes when we create unwritten extent if
846 * blocks are not allocated yet. The extent will be converted to written
847 * after IO is complete.
849 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
850 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
854 /* We don't expect handle for direct IO */
855 WARN_ON_ONCE(ext4_journal_current_handle());
857 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
858 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
861 * When doing DIO using unwritten extents, we need io_end to convert
862 * unwritten extents to written on IO completion. We allocate io_end
863 * once we spot unwritten extent and store it in b_private. Generic
864 * DIO code keeps b_private set and furthermore passes the value to
865 * our completion callback in 'private' argument.
867 if (!ret
&& buffer_unwritten(bh_result
)) {
868 if (!bh_result
->b_private
) {
869 ext4_io_end_t
*io_end
;
871 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
874 bh_result
->b_private
= io_end
;
875 ext4_set_io_unwritten_flag(inode
, io_end
);
877 set_buffer_defer_completion(bh_result
);
884 * Get block function for non-AIO DIO writes when we create unwritten extent if
885 * blocks are not allocated yet. The extent will be converted to written
886 * after IO is complete from ext4_ext_direct_IO() function.
888 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
889 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
893 /* We don't expect handle for direct IO */
894 WARN_ON_ONCE(ext4_journal_current_handle());
896 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
897 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
900 * Mark inode as having pending DIO writes to unwritten extents.
901 * ext4_ext_direct_IO() checks this flag and converts extents to
904 if (!ret
&& buffer_unwritten(bh_result
))
905 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
910 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
911 struct buffer_head
*bh_result
, int create
)
915 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
916 inode
->i_ino
, create
);
917 /* We don't expect handle for direct IO */
918 WARN_ON_ONCE(ext4_journal_current_handle());
920 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
922 * Blocks should have been preallocated! ext4_file_write_iter() checks
925 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
932 * `handle' can be NULL if create is zero
934 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
935 ext4_lblk_t block
, int map_flags
)
937 struct ext4_map_blocks map
;
938 struct buffer_head
*bh
;
939 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
942 J_ASSERT(handle
!= NULL
|| create
== 0);
946 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
949 return create
? ERR_PTR(-ENOSPC
) : NULL
;
953 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
955 return ERR_PTR(-ENOMEM
);
956 if (map
.m_flags
& EXT4_MAP_NEW
) {
957 J_ASSERT(create
!= 0);
958 J_ASSERT(handle
!= NULL
);
961 * Now that we do not always journal data, we should
962 * keep in mind whether this should always journal the
963 * new buffer as metadata. For now, regular file
964 * writes use ext4_get_block instead, so it's not a
968 BUFFER_TRACE(bh
, "call get_create_access");
969 err
= ext4_journal_get_create_access(handle
, bh
);
974 if (!buffer_uptodate(bh
)) {
975 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
976 set_buffer_uptodate(bh
);
979 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
980 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
984 BUFFER_TRACE(bh
, "not a new buffer");
991 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
992 ext4_lblk_t block
, int map_flags
)
994 struct buffer_head
*bh
;
996 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
999 if (!bh
|| buffer_uptodate(bh
))
1001 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1003 if (buffer_uptodate(bh
))
1006 return ERR_PTR(-EIO
);
1009 int ext4_walk_page_buffers(handle_t
*handle
,
1010 struct buffer_head
*head
,
1014 int (*fn
)(handle_t
*handle
,
1015 struct buffer_head
*bh
))
1017 struct buffer_head
*bh
;
1018 unsigned block_start
, block_end
;
1019 unsigned blocksize
= head
->b_size
;
1021 struct buffer_head
*next
;
1023 for (bh
= head
, block_start
= 0;
1024 ret
== 0 && (bh
!= head
|| !block_start
);
1025 block_start
= block_end
, bh
= next
) {
1026 next
= bh
->b_this_page
;
1027 block_end
= block_start
+ blocksize
;
1028 if (block_end
<= from
|| block_start
>= to
) {
1029 if (partial
&& !buffer_uptodate(bh
))
1033 err
= (*fn
)(handle
, bh
);
1041 * To preserve ordering, it is essential that the hole instantiation and
1042 * the data write be encapsulated in a single transaction. We cannot
1043 * close off a transaction and start a new one between the ext4_get_block()
1044 * and the commit_write(). So doing the jbd2_journal_start at the start of
1045 * prepare_write() is the right place.
1047 * Also, this function can nest inside ext4_writepage(). In that case, we
1048 * *know* that ext4_writepage() has generated enough buffer credits to do the
1049 * whole page. So we won't block on the journal in that case, which is good,
1050 * because the caller may be PF_MEMALLOC.
1052 * By accident, ext4 can be reentered when a transaction is open via
1053 * quota file writes. If we were to commit the transaction while thus
1054 * reentered, there can be a deadlock - we would be holding a quota
1055 * lock, and the commit would never complete if another thread had a
1056 * transaction open and was blocking on the quota lock - a ranking
1059 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1060 * will _not_ run commit under these circumstances because handle->h_ref
1061 * is elevated. We'll still have enough credits for the tiny quotafile
1064 int do_journal_get_write_access(handle_t
*handle
,
1065 struct buffer_head
*bh
)
1067 int dirty
= buffer_dirty(bh
);
1070 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1073 * __block_write_begin() could have dirtied some buffers. Clean
1074 * the dirty bit as jbd2_journal_get_write_access() could complain
1075 * otherwise about fs integrity issues. Setting of the dirty bit
1076 * by __block_write_begin() isn't a real problem here as we clear
1077 * the bit before releasing a page lock and thus writeback cannot
1078 * ever write the buffer.
1081 clear_buffer_dirty(bh
);
1082 BUFFER_TRACE(bh
, "get write access");
1083 ret
= ext4_journal_get_write_access(handle
, bh
);
1085 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1089 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1090 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1091 get_block_t
*get_block
)
1093 unsigned from
= pos
& (PAGE_SIZE
- 1);
1094 unsigned to
= from
+ len
;
1095 struct inode
*inode
= page
->mapping
->host
;
1096 unsigned block_start
, block_end
;
1099 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1101 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1102 bool decrypt
= false;
1104 BUG_ON(!PageLocked(page
));
1105 BUG_ON(from
> PAGE_SIZE
);
1106 BUG_ON(to
> PAGE_SIZE
);
1109 if (!page_has_buffers(page
))
1110 create_empty_buffers(page
, blocksize
, 0);
1111 head
= page_buffers(page
);
1112 bbits
= ilog2(blocksize
);
1113 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1115 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1116 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1117 block_end
= block_start
+ blocksize
;
1118 if (block_end
<= from
|| block_start
>= to
) {
1119 if (PageUptodate(page
)) {
1120 if (!buffer_uptodate(bh
))
1121 set_buffer_uptodate(bh
);
1126 clear_buffer_new(bh
);
1127 if (!buffer_mapped(bh
)) {
1128 WARN_ON(bh
->b_size
!= blocksize
);
1129 err
= get_block(inode
, block
, bh
, 1);
1132 if (buffer_new(bh
)) {
1133 unmap_underlying_metadata(bh
->b_bdev
,
1135 if (PageUptodate(page
)) {
1136 clear_buffer_new(bh
);
1137 set_buffer_uptodate(bh
);
1138 mark_buffer_dirty(bh
);
1141 if (block_end
> to
|| block_start
< from
)
1142 zero_user_segments(page
, to
, block_end
,
1147 if (PageUptodate(page
)) {
1148 if (!buffer_uptodate(bh
))
1149 set_buffer_uptodate(bh
);
1152 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1153 !buffer_unwritten(bh
) &&
1154 (block_start
< from
|| block_end
> to
)) {
1155 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1157 decrypt
= ext4_encrypted_inode(inode
) &&
1158 S_ISREG(inode
->i_mode
);
1162 * If we issued read requests, let them complete.
1164 while (wait_bh
> wait
) {
1165 wait_on_buffer(*--wait_bh
);
1166 if (!buffer_uptodate(*wait_bh
))
1170 page_zero_new_buffers(page
, from
, to
);
1172 err
= fscrypt_decrypt_page(page
);
1177 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1178 loff_t pos
, unsigned len
, unsigned flags
,
1179 struct page
**pagep
, void **fsdata
)
1181 struct inode
*inode
= mapping
->host
;
1182 int ret
, needed_blocks
;
1189 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1191 * Reserve one block more for addition to orphan list in case
1192 * we allocate blocks but write fails for some reason
1194 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1195 index
= pos
>> PAGE_SHIFT
;
1196 from
= pos
& (PAGE_SIZE
- 1);
1199 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1200 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1209 * grab_cache_page_write_begin() can take a long time if the
1210 * system is thrashing due to memory pressure, or if the page
1211 * is being written back. So grab it first before we start
1212 * the transaction handle. This also allows us to allocate
1213 * the page (if needed) without using GFP_NOFS.
1216 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1222 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1223 if (IS_ERR(handle
)) {
1225 return PTR_ERR(handle
);
1229 if (page
->mapping
!= mapping
) {
1230 /* The page got truncated from under us */
1233 ext4_journal_stop(handle
);
1236 /* In case writeback began while the page was unlocked */
1237 wait_for_stable_page(page
);
1239 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1240 if (ext4_should_dioread_nolock(inode
))
1241 ret
= ext4_block_write_begin(page
, pos
, len
,
1242 ext4_get_block_unwritten
);
1244 ret
= ext4_block_write_begin(page
, pos
, len
,
1247 if (ext4_should_dioread_nolock(inode
))
1248 ret
= __block_write_begin(page
, pos
, len
,
1249 ext4_get_block_unwritten
);
1251 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1253 if (!ret
&& ext4_should_journal_data(inode
)) {
1254 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1256 do_journal_get_write_access
);
1262 * __block_write_begin may have instantiated a few blocks
1263 * outside i_size. Trim these off again. Don't need
1264 * i_size_read because we hold i_mutex.
1266 * Add inode to orphan list in case we crash before
1269 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1270 ext4_orphan_add(handle
, inode
);
1272 ext4_journal_stop(handle
);
1273 if (pos
+ len
> inode
->i_size
) {
1274 ext4_truncate_failed_write(inode
);
1276 * If truncate failed early the inode might
1277 * still be on the orphan list; we need to
1278 * make sure the inode is removed from the
1279 * orphan list in that case.
1282 ext4_orphan_del(NULL
, inode
);
1285 if (ret
== -ENOSPC
&&
1286 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1295 /* For write_end() in data=journal mode */
1296 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1299 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1301 set_buffer_uptodate(bh
);
1302 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1303 clear_buffer_meta(bh
);
1304 clear_buffer_prio(bh
);
1309 * We need to pick up the new inode size which generic_commit_write gave us
1310 * `file' can be NULL - eg, when called from page_symlink().
1312 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1313 * buffers are managed internally.
1315 static int ext4_write_end(struct file
*file
,
1316 struct address_space
*mapping
,
1317 loff_t pos
, unsigned len
, unsigned copied
,
1318 struct page
*page
, void *fsdata
)
1320 handle_t
*handle
= ext4_journal_current_handle();
1321 struct inode
*inode
= mapping
->host
;
1322 loff_t old_size
= inode
->i_size
;
1324 int i_size_changed
= 0;
1326 trace_ext4_write_end(inode
, pos
, len
, copied
);
1327 if (ext4_has_inline_data(inode
)) {
1328 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1334 copied
= block_write_end(file
, mapping
, pos
,
1335 len
, copied
, page
, fsdata
);
1337 * it's important to update i_size while still holding page lock:
1338 * page writeout could otherwise come in and zero beyond i_size.
1340 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1345 pagecache_isize_extended(inode
, old_size
, pos
);
1347 * Don't mark the inode dirty under page lock. First, it unnecessarily
1348 * makes the holding time of page lock longer. Second, it forces lock
1349 * ordering of page lock and transaction start for journaling
1353 ext4_mark_inode_dirty(handle
, inode
);
1355 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1356 /* if we have allocated more blocks and copied
1357 * less. We will have blocks allocated outside
1358 * inode->i_size. So truncate them
1360 ext4_orphan_add(handle
, inode
);
1362 ret2
= ext4_journal_stop(handle
);
1366 if (pos
+ len
> inode
->i_size
) {
1367 ext4_truncate_failed_write(inode
);
1369 * If truncate failed early the inode might still be
1370 * on the orphan list; we need to make sure the inode
1371 * is removed from the orphan list in that case.
1374 ext4_orphan_del(NULL
, inode
);
1377 return ret
? ret
: copied
;
1381 * This is a private version of page_zero_new_buffers() which doesn't
1382 * set the buffer to be dirty, since in data=journalled mode we need
1383 * to call ext4_handle_dirty_metadata() instead.
1385 static void zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1387 unsigned int block_start
= 0, block_end
;
1388 struct buffer_head
*head
, *bh
;
1390 bh
= head
= page_buffers(page
);
1392 block_end
= block_start
+ bh
->b_size
;
1393 if (buffer_new(bh
)) {
1394 if (block_end
> from
&& block_start
< to
) {
1395 if (!PageUptodate(page
)) {
1396 unsigned start
, size
;
1398 start
= max(from
, block_start
);
1399 size
= min(to
, block_end
) - start
;
1401 zero_user(page
, start
, size
);
1402 set_buffer_uptodate(bh
);
1404 clear_buffer_new(bh
);
1407 block_start
= block_end
;
1408 bh
= bh
->b_this_page
;
1409 } while (bh
!= head
);
1412 static int ext4_journalled_write_end(struct file
*file
,
1413 struct address_space
*mapping
,
1414 loff_t pos
, unsigned len
, unsigned copied
,
1415 struct page
*page
, void *fsdata
)
1417 handle_t
*handle
= ext4_journal_current_handle();
1418 struct inode
*inode
= mapping
->host
;
1419 loff_t old_size
= inode
->i_size
;
1423 int size_changed
= 0;
1425 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1426 from
= pos
& (PAGE_SIZE
- 1);
1429 BUG_ON(!ext4_handle_valid(handle
));
1431 if (ext4_has_inline_data(inode
))
1432 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1436 if (!PageUptodate(page
))
1438 zero_new_buffers(page
, from
+copied
, to
);
1441 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1442 to
, &partial
, write_end_fn
);
1444 SetPageUptodate(page
);
1446 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1447 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1448 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1453 pagecache_isize_extended(inode
, old_size
, pos
);
1456 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1461 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1462 /* if we have allocated more blocks and copied
1463 * less. We will have blocks allocated outside
1464 * inode->i_size. So truncate them
1466 ext4_orphan_add(handle
, inode
);
1468 ret2
= ext4_journal_stop(handle
);
1471 if (pos
+ len
> inode
->i_size
) {
1472 ext4_truncate_failed_write(inode
);
1474 * If truncate failed early the inode might still be
1475 * on the orphan list; we need to make sure the inode
1476 * is removed from the orphan list in that case.
1479 ext4_orphan_del(NULL
, inode
);
1482 return ret
? ret
: copied
;
1486 * Reserve space for a single cluster
1488 static int ext4_da_reserve_space(struct inode
*inode
)
1490 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1491 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1495 * We will charge metadata quota at writeout time; this saves
1496 * us from metadata over-estimation, though we may go over by
1497 * a small amount in the end. Here we just reserve for data.
1499 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1503 spin_lock(&ei
->i_block_reservation_lock
);
1504 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1505 spin_unlock(&ei
->i_block_reservation_lock
);
1506 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1509 ei
->i_reserved_data_blocks
++;
1510 trace_ext4_da_reserve_space(inode
);
1511 spin_unlock(&ei
->i_block_reservation_lock
);
1513 return 0; /* success */
1516 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1518 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1519 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1522 return; /* Nothing to release, exit */
1524 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1526 trace_ext4_da_release_space(inode
, to_free
);
1527 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1529 * if there aren't enough reserved blocks, then the
1530 * counter is messed up somewhere. Since this
1531 * function is called from invalidate page, it's
1532 * harmless to return without any action.
1534 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1535 "ino %lu, to_free %d with only %d reserved "
1536 "data blocks", inode
->i_ino
, to_free
,
1537 ei
->i_reserved_data_blocks
);
1539 to_free
= ei
->i_reserved_data_blocks
;
1541 ei
->i_reserved_data_blocks
-= to_free
;
1543 /* update fs dirty data blocks counter */
1544 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1546 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1548 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1551 static void ext4_da_page_release_reservation(struct page
*page
,
1552 unsigned int offset
,
1553 unsigned int length
)
1555 int to_release
= 0, contiguous_blks
= 0;
1556 struct buffer_head
*head
, *bh
;
1557 unsigned int curr_off
= 0;
1558 struct inode
*inode
= page
->mapping
->host
;
1559 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1560 unsigned int stop
= offset
+ length
;
1564 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1566 head
= page_buffers(page
);
1569 unsigned int next_off
= curr_off
+ bh
->b_size
;
1571 if (next_off
> stop
)
1574 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1577 clear_buffer_delay(bh
);
1578 } else if (contiguous_blks
) {
1579 lblk
= page
->index
<<
1580 (PAGE_SHIFT
- inode
->i_blkbits
);
1581 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1583 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1584 contiguous_blks
= 0;
1586 curr_off
= next_off
;
1587 } while ((bh
= bh
->b_this_page
) != head
);
1589 if (contiguous_blks
) {
1590 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1591 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1592 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1595 /* If we have released all the blocks belonging to a cluster, then we
1596 * need to release the reserved space for that cluster. */
1597 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1598 while (num_clusters
> 0) {
1599 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1600 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1601 if (sbi
->s_cluster_ratio
== 1 ||
1602 !ext4_find_delalloc_cluster(inode
, lblk
))
1603 ext4_da_release_space(inode
, 1);
1610 * Delayed allocation stuff
1613 struct mpage_da_data
{
1614 struct inode
*inode
;
1615 struct writeback_control
*wbc
;
1617 pgoff_t first_page
; /* The first page to write */
1618 pgoff_t next_page
; /* Current page to examine */
1619 pgoff_t last_page
; /* Last page to examine */
1621 * Extent to map - this can be after first_page because that can be
1622 * fully mapped. We somewhat abuse m_flags to store whether the extent
1623 * is delalloc or unwritten.
1625 struct ext4_map_blocks map
;
1626 struct ext4_io_submit io_submit
; /* IO submission data */
1629 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1634 struct pagevec pvec
;
1635 struct inode
*inode
= mpd
->inode
;
1636 struct address_space
*mapping
= inode
->i_mapping
;
1638 /* This is necessary when next_page == 0. */
1639 if (mpd
->first_page
>= mpd
->next_page
)
1642 index
= mpd
->first_page
;
1643 end
= mpd
->next_page
- 1;
1645 ext4_lblk_t start
, last
;
1646 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1647 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1648 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1651 pagevec_init(&pvec
, 0);
1652 while (index
<= end
) {
1653 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1656 for (i
= 0; i
< nr_pages
; i
++) {
1657 struct page
*page
= pvec
.pages
[i
];
1658 if (page
->index
> end
)
1660 BUG_ON(!PageLocked(page
));
1661 BUG_ON(PageWriteback(page
));
1663 if (page_mapped(page
))
1664 clear_page_dirty_for_io(page
);
1665 block_invalidatepage(page
, 0, PAGE_SIZE
);
1666 ClearPageUptodate(page
);
1670 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1671 pagevec_release(&pvec
);
1675 static void ext4_print_free_blocks(struct inode
*inode
)
1677 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1678 struct super_block
*sb
= inode
->i_sb
;
1679 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1681 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1682 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1683 ext4_count_free_clusters(sb
)));
1684 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1685 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1686 (long long) EXT4_C2B(EXT4_SB(sb
),
1687 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1688 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1689 (long long) EXT4_C2B(EXT4_SB(sb
),
1690 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1691 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1692 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1693 ei
->i_reserved_data_blocks
);
1697 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1699 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1703 * This function is grabs code from the very beginning of
1704 * ext4_map_blocks, but assumes that the caller is from delayed write
1705 * time. This function looks up the requested blocks and sets the
1706 * buffer delay bit under the protection of i_data_sem.
1708 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1709 struct ext4_map_blocks
*map
,
1710 struct buffer_head
*bh
)
1712 struct extent_status es
;
1714 sector_t invalid_block
= ~((sector_t
) 0xffff);
1715 #ifdef ES_AGGRESSIVE_TEST
1716 struct ext4_map_blocks orig_map
;
1718 memcpy(&orig_map
, map
, sizeof(*map
));
1721 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1725 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1726 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1727 (unsigned long) map
->m_lblk
);
1729 /* Lookup extent status tree firstly */
1730 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1731 if (ext4_es_is_hole(&es
)) {
1733 down_read(&EXT4_I(inode
)->i_data_sem
);
1738 * Delayed extent could be allocated by fallocate.
1739 * So we need to check it.
1741 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1742 map_bh(bh
, inode
->i_sb
, invalid_block
);
1744 set_buffer_delay(bh
);
1748 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1749 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1750 if (retval
> map
->m_len
)
1751 retval
= map
->m_len
;
1752 map
->m_len
= retval
;
1753 if (ext4_es_is_written(&es
))
1754 map
->m_flags
|= EXT4_MAP_MAPPED
;
1755 else if (ext4_es_is_unwritten(&es
))
1756 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1760 #ifdef ES_AGGRESSIVE_TEST
1761 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1767 * Try to see if we can get the block without requesting a new
1768 * file system block.
1770 down_read(&EXT4_I(inode
)->i_data_sem
);
1771 if (ext4_has_inline_data(inode
))
1773 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1774 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1776 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1782 * XXX: __block_prepare_write() unmaps passed block,
1786 * If the block was allocated from previously allocated cluster,
1787 * then we don't need to reserve it again. However we still need
1788 * to reserve metadata for every block we're going to write.
1790 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1791 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1792 ret
= ext4_da_reserve_space(inode
);
1794 /* not enough space to reserve */
1800 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1801 ~0, EXTENT_STATUS_DELAYED
);
1807 map_bh(bh
, inode
->i_sb
, invalid_block
);
1809 set_buffer_delay(bh
);
1810 } else if (retval
> 0) {
1812 unsigned int status
;
1814 if (unlikely(retval
!= map
->m_len
)) {
1815 ext4_warning(inode
->i_sb
,
1816 "ES len assertion failed for inode "
1817 "%lu: retval %d != map->m_len %d",
1818 inode
->i_ino
, retval
, map
->m_len
);
1822 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1823 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1824 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1825 map
->m_pblk
, status
);
1831 up_read((&EXT4_I(inode
)->i_data_sem
));
1837 * This is a special get_block_t callback which is used by
1838 * ext4_da_write_begin(). It will either return mapped block or
1839 * reserve space for a single block.
1841 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1842 * We also have b_blocknr = -1 and b_bdev initialized properly
1844 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1845 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1846 * initialized properly.
1848 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1849 struct buffer_head
*bh
, int create
)
1851 struct ext4_map_blocks map
;
1854 BUG_ON(create
== 0);
1855 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1857 map
.m_lblk
= iblock
;
1861 * first, we need to know whether the block is allocated already
1862 * preallocated blocks are unmapped but should treated
1863 * the same as allocated blocks.
1865 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1869 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1870 ext4_update_bh_state(bh
, map
.m_flags
);
1872 if (buffer_unwritten(bh
)) {
1873 /* A delayed write to unwritten bh should be marked
1874 * new and mapped. Mapped ensures that we don't do
1875 * get_block multiple times when we write to the same
1876 * offset and new ensures that we do proper zero out
1877 * for partial write.
1880 set_buffer_mapped(bh
);
1885 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1891 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1897 static int __ext4_journalled_writepage(struct page
*page
,
1900 struct address_space
*mapping
= page
->mapping
;
1901 struct inode
*inode
= mapping
->host
;
1902 struct buffer_head
*page_bufs
= NULL
;
1903 handle_t
*handle
= NULL
;
1904 int ret
= 0, err
= 0;
1905 int inline_data
= ext4_has_inline_data(inode
);
1906 struct buffer_head
*inode_bh
= NULL
;
1908 ClearPageChecked(page
);
1911 BUG_ON(page
->index
!= 0);
1912 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1913 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1914 if (inode_bh
== NULL
)
1917 page_bufs
= page_buffers(page
);
1922 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1926 * We need to release the page lock before we start the
1927 * journal, so grab a reference so the page won't disappear
1928 * out from under us.
1933 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1934 ext4_writepage_trans_blocks(inode
));
1935 if (IS_ERR(handle
)) {
1936 ret
= PTR_ERR(handle
);
1938 goto out_no_pagelock
;
1940 BUG_ON(!ext4_handle_valid(handle
));
1944 if (page
->mapping
!= mapping
) {
1945 /* The page got truncated from under us */
1946 ext4_journal_stop(handle
);
1952 BUFFER_TRACE(inode_bh
, "get write access");
1953 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1955 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1958 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1959 do_journal_get_write_access
);
1961 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1966 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1967 err
= ext4_journal_stop(handle
);
1971 if (!ext4_has_inline_data(inode
))
1972 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1974 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1983 * Note that we don't need to start a transaction unless we're journaling data
1984 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1985 * need to file the inode to the transaction's list in ordered mode because if
1986 * we are writing back data added by write(), the inode is already there and if
1987 * we are writing back data modified via mmap(), no one guarantees in which
1988 * transaction the data will hit the disk. In case we are journaling data, we
1989 * cannot start transaction directly because transaction start ranks above page
1990 * lock so we have to do some magic.
1992 * This function can get called via...
1993 * - ext4_writepages after taking page lock (have journal handle)
1994 * - journal_submit_inode_data_buffers (no journal handle)
1995 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1996 * - grab_page_cache when doing write_begin (have journal handle)
1998 * We don't do any block allocation in this function. If we have page with
1999 * multiple blocks we need to write those buffer_heads that are mapped. This
2000 * is important for mmaped based write. So if we do with blocksize 1K
2001 * truncate(f, 1024);
2002 * a = mmap(f, 0, 4096);
2004 * truncate(f, 4096);
2005 * we have in the page first buffer_head mapped via page_mkwrite call back
2006 * but other buffer_heads would be unmapped but dirty (dirty done via the
2007 * do_wp_page). So writepage should write the first block. If we modify
2008 * the mmap area beyond 1024 we will again get a page_fault and the
2009 * page_mkwrite callback will do the block allocation and mark the
2010 * buffer_heads mapped.
2012 * We redirty the page if we have any buffer_heads that is either delay or
2013 * unwritten in the page.
2015 * We can get recursively called as show below.
2017 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2020 * But since we don't do any block allocation we should not deadlock.
2021 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2023 static int ext4_writepage(struct page
*page
,
2024 struct writeback_control
*wbc
)
2029 struct buffer_head
*page_bufs
= NULL
;
2030 struct inode
*inode
= page
->mapping
->host
;
2031 struct ext4_io_submit io_submit
;
2032 bool keep_towrite
= false;
2034 trace_ext4_writepage(page
);
2035 size
= i_size_read(inode
);
2036 if (page
->index
== size
>> PAGE_SHIFT
)
2037 len
= size
& ~PAGE_MASK
;
2041 page_bufs
= page_buffers(page
);
2043 * We cannot do block allocation or other extent handling in this
2044 * function. If there are buffers needing that, we have to redirty
2045 * the page. But we may reach here when we do a journal commit via
2046 * journal_submit_inode_data_buffers() and in that case we must write
2047 * allocated buffers to achieve data=ordered mode guarantees.
2049 * Also, if there is only one buffer per page (the fs block
2050 * size == the page size), if one buffer needs block
2051 * allocation or needs to modify the extent tree to clear the
2052 * unwritten flag, we know that the page can't be written at
2053 * all, so we might as well refuse the write immediately.
2054 * Unfortunately if the block size != page size, we can't as
2055 * easily detect this case using ext4_walk_page_buffers(), but
2056 * for the extremely common case, this is an optimization that
2057 * skips a useless round trip through ext4_bio_write_page().
2059 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2060 ext4_bh_delay_or_unwritten
)) {
2061 redirty_page_for_writepage(wbc
, page
);
2062 if ((current
->flags
& PF_MEMALLOC
) ||
2063 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2065 * For memory cleaning there's no point in writing only
2066 * some buffers. So just bail out. Warn if we came here
2067 * from direct reclaim.
2069 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2074 keep_towrite
= true;
2077 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2079 * It's mmapped pagecache. Add buffers and journal it. There
2080 * doesn't seem much point in redirtying the page here.
2082 return __ext4_journalled_writepage(page
, len
);
2084 ext4_io_submit_init(&io_submit
, wbc
);
2085 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2086 if (!io_submit
.io_end
) {
2087 redirty_page_for_writepage(wbc
, page
);
2091 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2092 ext4_io_submit(&io_submit
);
2093 /* Drop io_end reference we got from init */
2094 ext4_put_io_end_defer(io_submit
.io_end
);
2098 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2101 loff_t size
= i_size_read(mpd
->inode
);
2104 BUG_ON(page
->index
!= mpd
->first_page
);
2105 if (page
->index
== size
>> PAGE_SHIFT
)
2106 len
= size
& ~PAGE_MASK
;
2109 clear_page_dirty_for_io(page
);
2110 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2112 mpd
->wbc
->nr_to_write
--;
2118 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2121 * mballoc gives us at most this number of blocks...
2122 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2123 * The rest of mballoc seems to handle chunks up to full group size.
2125 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2128 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2130 * @mpd - extent of blocks
2131 * @lblk - logical number of the block in the file
2132 * @bh - buffer head we want to add to the extent
2134 * The function is used to collect contig. blocks in the same state. If the
2135 * buffer doesn't require mapping for writeback and we haven't started the
2136 * extent of buffers to map yet, the function returns 'true' immediately - the
2137 * caller can write the buffer right away. Otherwise the function returns true
2138 * if the block has been added to the extent, false if the block couldn't be
2141 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2142 struct buffer_head
*bh
)
2144 struct ext4_map_blocks
*map
= &mpd
->map
;
2146 /* Buffer that doesn't need mapping for writeback? */
2147 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2148 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2149 /* So far no extent to map => we write the buffer right away */
2150 if (map
->m_len
== 0)
2155 /* First block in the extent? */
2156 if (map
->m_len
== 0) {
2159 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2163 /* Don't go larger than mballoc is willing to allocate */
2164 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2167 /* Can we merge the block to our big extent? */
2168 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2169 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2177 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2179 * @mpd - extent of blocks for mapping
2180 * @head - the first buffer in the page
2181 * @bh - buffer we should start processing from
2182 * @lblk - logical number of the block in the file corresponding to @bh
2184 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2185 * the page for IO if all buffers in this page were mapped and there's no
2186 * accumulated extent of buffers to map or add buffers in the page to the
2187 * extent of buffers to map. The function returns 1 if the caller can continue
2188 * by processing the next page, 0 if it should stop adding buffers to the
2189 * extent to map because we cannot extend it anymore. It can also return value
2190 * < 0 in case of error during IO submission.
2192 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2193 struct buffer_head
*head
,
2194 struct buffer_head
*bh
,
2197 struct inode
*inode
= mpd
->inode
;
2199 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2200 >> inode
->i_blkbits
;
2203 BUG_ON(buffer_locked(bh
));
2205 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2206 /* Found extent to map? */
2209 /* Everything mapped so far and we hit EOF */
2212 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2213 /* So far everything mapped? Submit the page for IO. */
2214 if (mpd
->map
.m_len
== 0) {
2215 err
= mpage_submit_page(mpd
, head
->b_page
);
2219 return lblk
< blocks
;
2223 * mpage_map_buffers - update buffers corresponding to changed extent and
2224 * submit fully mapped pages for IO
2226 * @mpd - description of extent to map, on return next extent to map
2228 * Scan buffers corresponding to changed extent (we expect corresponding pages
2229 * to be already locked) and update buffer state according to new extent state.
2230 * We map delalloc buffers to their physical location, clear unwritten bits,
2231 * and mark buffers as uninit when we perform writes to unwritten extents
2232 * and do extent conversion after IO is finished. If the last page is not fully
2233 * mapped, we update @map to the next extent in the last page that needs
2234 * mapping. Otherwise we submit the page for IO.
2236 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2238 struct pagevec pvec
;
2240 struct inode
*inode
= mpd
->inode
;
2241 struct buffer_head
*head
, *bh
;
2242 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2248 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2249 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2250 lblk
= start
<< bpp_bits
;
2251 pblock
= mpd
->map
.m_pblk
;
2253 pagevec_init(&pvec
, 0);
2254 while (start
<= end
) {
2255 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2259 for (i
= 0; i
< nr_pages
; i
++) {
2260 struct page
*page
= pvec
.pages
[i
];
2262 if (page
->index
> end
)
2264 /* Up to 'end' pages must be contiguous */
2265 BUG_ON(page
->index
!= start
);
2266 bh
= head
= page_buffers(page
);
2268 if (lblk
< mpd
->map
.m_lblk
)
2270 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2272 * Buffer after end of mapped extent.
2273 * Find next buffer in the page to map.
2276 mpd
->map
.m_flags
= 0;
2278 * FIXME: If dioread_nolock supports
2279 * blocksize < pagesize, we need to make
2280 * sure we add size mapped so far to
2281 * io_end->size as the following call
2282 * can submit the page for IO.
2284 err
= mpage_process_page_bufs(mpd
, head
,
2286 pagevec_release(&pvec
);
2291 if (buffer_delay(bh
)) {
2292 clear_buffer_delay(bh
);
2293 bh
->b_blocknr
= pblock
++;
2295 clear_buffer_unwritten(bh
);
2296 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2299 * FIXME: This is going to break if dioread_nolock
2300 * supports blocksize < pagesize as we will try to
2301 * convert potentially unmapped parts of inode.
2303 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2304 /* Page fully mapped - let IO run! */
2305 err
= mpage_submit_page(mpd
, page
);
2307 pagevec_release(&pvec
);
2312 pagevec_release(&pvec
);
2314 /* Extent fully mapped and matches with page boundary. We are done. */
2316 mpd
->map
.m_flags
= 0;
2320 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2322 struct inode
*inode
= mpd
->inode
;
2323 struct ext4_map_blocks
*map
= &mpd
->map
;
2324 int get_blocks_flags
;
2325 int err
, dioread_nolock
;
2327 trace_ext4_da_write_pages_extent(inode
, map
);
2329 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2330 * to convert an unwritten extent to be initialized (in the case
2331 * where we have written into one or more preallocated blocks). It is
2332 * possible that we're going to need more metadata blocks than
2333 * previously reserved. However we must not fail because we're in
2334 * writeback and there is nothing we can do about it so it might result
2335 * in data loss. So use reserved blocks to allocate metadata if
2338 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2339 * the blocks in question are delalloc blocks. This indicates
2340 * that the blocks and quotas has already been checked when
2341 * the data was copied into the page cache.
2343 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2344 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2345 EXT4_GET_BLOCKS_IO_SUBMIT
;
2346 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2348 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2349 if (map
->m_flags
& (1 << BH_Delay
))
2350 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2352 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2355 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2356 if (!mpd
->io_submit
.io_end
->handle
&&
2357 ext4_handle_valid(handle
)) {
2358 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2359 handle
->h_rsv_handle
= NULL
;
2361 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2364 BUG_ON(map
->m_len
== 0);
2365 if (map
->m_flags
& EXT4_MAP_NEW
) {
2366 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2369 for (i
= 0; i
< map
->m_len
; i
++)
2370 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2376 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2377 * mpd->len and submit pages underlying it for IO
2379 * @handle - handle for journal operations
2380 * @mpd - extent to map
2381 * @give_up_on_write - we set this to true iff there is a fatal error and there
2382 * is no hope of writing the data. The caller should discard
2383 * dirty pages to avoid infinite loops.
2385 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2386 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2387 * them to initialized or split the described range from larger unwritten
2388 * extent. Note that we need not map all the described range since allocation
2389 * can return less blocks or the range is covered by more unwritten extents. We
2390 * cannot map more because we are limited by reserved transaction credits. On
2391 * the other hand we always make sure that the last touched page is fully
2392 * mapped so that it can be written out (and thus forward progress is
2393 * guaranteed). After mapping we submit all mapped pages for IO.
2395 static int mpage_map_and_submit_extent(handle_t
*handle
,
2396 struct mpage_da_data
*mpd
,
2397 bool *give_up_on_write
)
2399 struct inode
*inode
= mpd
->inode
;
2400 struct ext4_map_blocks
*map
= &mpd
->map
;
2405 mpd
->io_submit
.io_end
->offset
=
2406 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2408 err
= mpage_map_one_extent(handle
, mpd
);
2410 struct super_block
*sb
= inode
->i_sb
;
2412 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2413 goto invalidate_dirty_pages
;
2415 * Let the uper layers retry transient errors.
2416 * In the case of ENOSPC, if ext4_count_free_blocks()
2417 * is non-zero, a commit should free up blocks.
2419 if ((err
== -ENOMEM
) ||
2420 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2422 goto update_disksize
;
2425 ext4_msg(sb
, KERN_CRIT
,
2426 "Delayed block allocation failed for "
2427 "inode %lu at logical offset %llu with"
2428 " max blocks %u with error %d",
2430 (unsigned long long)map
->m_lblk
,
2431 (unsigned)map
->m_len
, -err
);
2432 ext4_msg(sb
, KERN_CRIT
,
2433 "This should not happen!! Data will "
2436 ext4_print_free_blocks(inode
);
2437 invalidate_dirty_pages
:
2438 *give_up_on_write
= true;
2443 * Update buffer state, submit mapped pages, and get us new
2446 err
= mpage_map_and_submit_buffers(mpd
);
2448 goto update_disksize
;
2449 } while (map
->m_len
);
2453 * Update on-disk size after IO is submitted. Races with
2454 * truncate are avoided by checking i_size under i_data_sem.
2456 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2457 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2461 down_write(&EXT4_I(inode
)->i_data_sem
);
2462 i_size
= i_size_read(inode
);
2463 if (disksize
> i_size
)
2465 if (disksize
> EXT4_I(inode
)->i_disksize
)
2466 EXT4_I(inode
)->i_disksize
= disksize
;
2467 err2
= ext4_mark_inode_dirty(handle
, inode
);
2468 up_write(&EXT4_I(inode
)->i_data_sem
);
2470 ext4_error(inode
->i_sb
,
2471 "Failed to mark inode %lu dirty",
2480 * Calculate the total number of credits to reserve for one writepages
2481 * iteration. This is called from ext4_writepages(). We map an extent of
2482 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2483 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2484 * bpp - 1 blocks in bpp different extents.
2486 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2488 int bpp
= ext4_journal_blocks_per_page(inode
);
2490 return ext4_meta_trans_blocks(inode
,
2491 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2495 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2496 * and underlying extent to map
2498 * @mpd - where to look for pages
2500 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2501 * IO immediately. When we find a page which isn't mapped we start accumulating
2502 * extent of buffers underlying these pages that needs mapping (formed by
2503 * either delayed or unwritten buffers). We also lock the pages containing
2504 * these buffers. The extent found is returned in @mpd structure (starting at
2505 * mpd->lblk with length mpd->len blocks).
2507 * Note that this function can attach bios to one io_end structure which are
2508 * neither logically nor physically contiguous. Although it may seem as an
2509 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2510 * case as we need to track IO to all buffers underlying a page in one io_end.
2512 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2514 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2515 struct pagevec pvec
;
2516 unsigned int nr_pages
;
2517 long left
= mpd
->wbc
->nr_to_write
;
2518 pgoff_t index
= mpd
->first_page
;
2519 pgoff_t end
= mpd
->last_page
;
2522 int blkbits
= mpd
->inode
->i_blkbits
;
2524 struct buffer_head
*head
;
2526 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2527 tag
= PAGECACHE_TAG_TOWRITE
;
2529 tag
= PAGECACHE_TAG_DIRTY
;
2531 pagevec_init(&pvec
, 0);
2533 mpd
->next_page
= index
;
2534 while (index
<= end
) {
2535 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2536 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2540 for (i
= 0; i
< nr_pages
; i
++) {
2541 struct page
*page
= pvec
.pages
[i
];
2544 * At this point, the page may be truncated or
2545 * invalidated (changing page->mapping to NULL), or
2546 * even swizzled back from swapper_space to tmpfs file
2547 * mapping. However, page->index will not change
2548 * because we have a reference on the page.
2550 if (page
->index
> end
)
2554 * Accumulated enough dirty pages? This doesn't apply
2555 * to WB_SYNC_ALL mode. For integrity sync we have to
2556 * keep going because someone may be concurrently
2557 * dirtying pages, and we might have synced a lot of
2558 * newly appeared dirty pages, but have not synced all
2559 * of the old dirty pages.
2561 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2564 /* If we can't merge this page, we are done. */
2565 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2570 * If the page is no longer dirty, or its mapping no
2571 * longer corresponds to inode we are writing (which
2572 * means it has been truncated or invalidated), or the
2573 * page is already under writeback and we are not doing
2574 * a data integrity writeback, skip the page
2576 if (!PageDirty(page
) ||
2577 (PageWriteback(page
) &&
2578 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2579 unlikely(page
->mapping
!= mapping
)) {
2584 wait_on_page_writeback(page
);
2585 BUG_ON(PageWriteback(page
));
2587 if (mpd
->map
.m_len
== 0)
2588 mpd
->first_page
= page
->index
;
2589 mpd
->next_page
= page
->index
+ 1;
2590 /* Add all dirty buffers to mpd */
2591 lblk
= ((ext4_lblk_t
)page
->index
) <<
2592 (PAGE_SHIFT
- blkbits
);
2593 head
= page_buffers(page
);
2594 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2600 pagevec_release(&pvec
);
2605 pagevec_release(&pvec
);
2609 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2612 struct address_space
*mapping
= data
;
2613 int ret
= ext4_writepage(page
, wbc
);
2614 mapping_set_error(mapping
, ret
);
2618 static int ext4_writepages(struct address_space
*mapping
,
2619 struct writeback_control
*wbc
)
2621 pgoff_t writeback_index
= 0;
2622 long nr_to_write
= wbc
->nr_to_write
;
2623 int range_whole
= 0;
2625 handle_t
*handle
= NULL
;
2626 struct mpage_da_data mpd
;
2627 struct inode
*inode
= mapping
->host
;
2628 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2629 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2631 struct blk_plug plug
;
2632 bool give_up_on_write
= false;
2634 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2635 trace_ext4_writepages(inode
, wbc
);
2637 if (dax_mapping(mapping
)) {
2638 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2640 goto out_writepages
;
2644 * No pages to write? This is mainly a kludge to avoid starting
2645 * a transaction for special inodes like journal inode on last iput()
2646 * because that could violate lock ordering on umount
2648 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2649 goto out_writepages
;
2651 if (ext4_should_journal_data(inode
)) {
2652 struct blk_plug plug
;
2654 blk_start_plug(&plug
);
2655 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2656 blk_finish_plug(&plug
);
2657 goto out_writepages
;
2661 * If the filesystem has aborted, it is read-only, so return
2662 * right away instead of dumping stack traces later on that
2663 * will obscure the real source of the problem. We test
2664 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2665 * the latter could be true if the filesystem is mounted
2666 * read-only, and in that case, ext4_writepages should
2667 * *never* be called, so if that ever happens, we would want
2670 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2672 goto out_writepages
;
2675 if (ext4_should_dioread_nolock(inode
)) {
2677 * We may need to convert up to one extent per block in
2678 * the page and we may dirty the inode.
2680 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2684 * If we have inline data and arrive here, it means that
2685 * we will soon create the block for the 1st page, so
2686 * we'd better clear the inline data here.
2688 if (ext4_has_inline_data(inode
)) {
2689 /* Just inode will be modified... */
2690 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2691 if (IS_ERR(handle
)) {
2692 ret
= PTR_ERR(handle
);
2693 goto out_writepages
;
2695 BUG_ON(ext4_test_inode_state(inode
,
2696 EXT4_STATE_MAY_INLINE_DATA
));
2697 ext4_destroy_inline_data(handle
, inode
);
2698 ext4_journal_stop(handle
);
2701 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2704 if (wbc
->range_cyclic
) {
2705 writeback_index
= mapping
->writeback_index
;
2706 if (writeback_index
)
2708 mpd
.first_page
= writeback_index
;
2711 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2712 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2717 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2719 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2720 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2722 blk_start_plug(&plug
);
2723 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2724 /* For each extent of pages we use new io_end */
2725 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2726 if (!mpd
.io_submit
.io_end
) {
2732 * We have two constraints: We find one extent to map and we
2733 * must always write out whole page (makes a difference when
2734 * blocksize < pagesize) so that we don't block on IO when we
2735 * try to write out the rest of the page. Journalled mode is
2736 * not supported by delalloc.
2738 BUG_ON(ext4_should_journal_data(inode
));
2739 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2741 /* start a new transaction */
2742 handle
= ext4_journal_start_with_reserve(inode
,
2743 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2744 if (IS_ERR(handle
)) {
2745 ret
= PTR_ERR(handle
);
2746 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2747 "%ld pages, ino %lu; err %d", __func__
,
2748 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2749 /* Release allocated io_end */
2750 ext4_put_io_end(mpd
.io_submit
.io_end
);
2754 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2755 ret
= mpage_prepare_extent_to_map(&mpd
);
2758 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2762 * We scanned the whole range (or exhausted
2763 * nr_to_write), submitted what was mapped and
2764 * didn't find anything needing mapping. We are
2771 * Caution: If the handle is synchronous,
2772 * ext4_journal_stop() can wait for transaction commit
2773 * to finish which may depend on writeback of pages to
2774 * complete or on page lock to be released. In that
2775 * case, we have to wait until after after we have
2776 * submitted all the IO, released page locks we hold,
2777 * and dropped io_end reference (for extent conversion
2778 * to be able to complete) before stopping the handle.
2780 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2781 ext4_journal_stop(handle
);
2784 /* Submit prepared bio */
2785 ext4_io_submit(&mpd
.io_submit
);
2786 /* Unlock pages we didn't use */
2787 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2789 * Drop our io_end reference we got from init. We have
2790 * to be careful and use deferred io_end finishing if
2791 * we are still holding the transaction as we can
2792 * release the last reference to io_end which may end
2793 * up doing unwritten extent conversion.
2796 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2797 ext4_journal_stop(handle
);
2799 ext4_put_io_end(mpd
.io_submit
.io_end
);
2801 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2803 * Commit the transaction which would
2804 * free blocks released in the transaction
2807 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2811 /* Fatal error - ENOMEM, EIO... */
2815 blk_finish_plug(&plug
);
2816 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2818 mpd
.last_page
= writeback_index
- 1;
2824 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2826 * Set the writeback_index so that range_cyclic
2827 * mode will write it back later
2829 mapping
->writeback_index
= mpd
.first_page
;
2832 trace_ext4_writepages_result(inode
, wbc
, ret
,
2833 nr_to_write
- wbc
->nr_to_write
);
2834 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2838 static int ext4_nonda_switch(struct super_block
*sb
)
2840 s64 free_clusters
, dirty_clusters
;
2841 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2844 * switch to non delalloc mode if we are running low
2845 * on free block. The free block accounting via percpu
2846 * counters can get slightly wrong with percpu_counter_batch getting
2847 * accumulated on each CPU without updating global counters
2848 * Delalloc need an accurate free block accounting. So switch
2849 * to non delalloc when we are near to error range.
2852 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2854 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2856 * Start pushing delalloc when 1/2 of free blocks are dirty.
2858 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2859 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2861 if (2 * free_clusters
< 3 * dirty_clusters
||
2862 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2864 * free block count is less than 150% of dirty blocks
2865 * or free blocks is less than watermark
2872 /* We always reserve for an inode update; the superblock could be there too */
2873 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2875 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2878 if (pos
+ len
<= 0x7fffffffULL
)
2881 /* We might need to update the superblock to set LARGE_FILE */
2885 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2886 loff_t pos
, unsigned len
, unsigned flags
,
2887 struct page
**pagep
, void **fsdata
)
2889 int ret
, retries
= 0;
2892 struct inode
*inode
= mapping
->host
;
2895 index
= pos
>> PAGE_SHIFT
;
2897 if (ext4_nonda_switch(inode
->i_sb
)) {
2898 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2899 return ext4_write_begin(file
, mapping
, pos
,
2900 len
, flags
, pagep
, fsdata
);
2902 *fsdata
= (void *)0;
2903 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2905 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2906 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2916 * grab_cache_page_write_begin() can take a long time if the
2917 * system is thrashing due to memory pressure, or if the page
2918 * is being written back. So grab it first before we start
2919 * the transaction handle. This also allows us to allocate
2920 * the page (if needed) without using GFP_NOFS.
2923 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2929 * With delayed allocation, we don't log the i_disksize update
2930 * if there is delayed block allocation. But we still need
2931 * to journalling the i_disksize update if writes to the end
2932 * of file which has an already mapped buffer.
2935 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2936 ext4_da_write_credits(inode
, pos
, len
));
2937 if (IS_ERR(handle
)) {
2939 return PTR_ERR(handle
);
2943 if (page
->mapping
!= mapping
) {
2944 /* The page got truncated from under us */
2947 ext4_journal_stop(handle
);
2950 /* In case writeback began while the page was unlocked */
2951 wait_for_stable_page(page
);
2953 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2954 ret
= ext4_block_write_begin(page
, pos
, len
,
2955 ext4_da_get_block_prep
);
2957 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2961 ext4_journal_stop(handle
);
2963 * block_write_begin may have instantiated a few blocks
2964 * outside i_size. Trim these off again. Don't need
2965 * i_size_read because we hold i_mutex.
2967 if (pos
+ len
> inode
->i_size
)
2968 ext4_truncate_failed_write(inode
);
2970 if (ret
== -ENOSPC
&&
2971 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2983 * Check if we should update i_disksize
2984 * when write to the end of file but not require block allocation
2986 static int ext4_da_should_update_i_disksize(struct page
*page
,
2987 unsigned long offset
)
2989 struct buffer_head
*bh
;
2990 struct inode
*inode
= page
->mapping
->host
;
2994 bh
= page_buffers(page
);
2995 idx
= offset
>> inode
->i_blkbits
;
2997 for (i
= 0; i
< idx
; i
++)
2998 bh
= bh
->b_this_page
;
3000 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3005 static int ext4_da_write_end(struct file
*file
,
3006 struct address_space
*mapping
,
3007 loff_t pos
, unsigned len
, unsigned copied
,
3008 struct page
*page
, void *fsdata
)
3010 struct inode
*inode
= mapping
->host
;
3012 handle_t
*handle
= ext4_journal_current_handle();
3014 unsigned long start
, end
;
3015 int write_mode
= (int)(unsigned long)fsdata
;
3017 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3018 return ext4_write_end(file
, mapping
, pos
,
3019 len
, copied
, page
, fsdata
);
3021 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3022 start
= pos
& (PAGE_SIZE
- 1);
3023 end
= start
+ copied
- 1;
3026 * generic_write_end() will run mark_inode_dirty() if i_size
3027 * changes. So let's piggyback the i_disksize mark_inode_dirty
3030 new_i_size
= pos
+ copied
;
3031 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3032 if (ext4_has_inline_data(inode
) ||
3033 ext4_da_should_update_i_disksize(page
, end
)) {
3034 ext4_update_i_disksize(inode
, new_i_size
);
3035 /* We need to mark inode dirty even if
3036 * new_i_size is less that inode->i_size
3037 * bu greater than i_disksize.(hint delalloc)
3039 ext4_mark_inode_dirty(handle
, inode
);
3043 if (write_mode
!= CONVERT_INLINE_DATA
&&
3044 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3045 ext4_has_inline_data(inode
))
3046 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3049 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3055 ret2
= ext4_journal_stop(handle
);
3059 return ret
? ret
: copied
;
3062 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3063 unsigned int length
)
3066 * Drop reserved blocks
3068 BUG_ON(!PageLocked(page
));
3069 if (!page_has_buffers(page
))
3072 ext4_da_page_release_reservation(page
, offset
, length
);
3075 ext4_invalidatepage(page
, offset
, length
);
3081 * Force all delayed allocation blocks to be allocated for a given inode.
3083 int ext4_alloc_da_blocks(struct inode
*inode
)
3085 trace_ext4_alloc_da_blocks(inode
);
3087 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3091 * We do something simple for now. The filemap_flush() will
3092 * also start triggering a write of the data blocks, which is
3093 * not strictly speaking necessary (and for users of
3094 * laptop_mode, not even desirable). However, to do otherwise
3095 * would require replicating code paths in:
3097 * ext4_writepages() ->
3098 * write_cache_pages() ---> (via passed in callback function)
3099 * __mpage_da_writepage() -->
3100 * mpage_add_bh_to_extent()
3101 * mpage_da_map_blocks()
3103 * The problem is that write_cache_pages(), located in
3104 * mm/page-writeback.c, marks pages clean in preparation for
3105 * doing I/O, which is not desirable if we're not planning on
3108 * We could call write_cache_pages(), and then redirty all of
3109 * the pages by calling redirty_page_for_writepage() but that
3110 * would be ugly in the extreme. So instead we would need to
3111 * replicate parts of the code in the above functions,
3112 * simplifying them because we wouldn't actually intend to
3113 * write out the pages, but rather only collect contiguous
3114 * logical block extents, call the multi-block allocator, and
3115 * then update the buffer heads with the block allocations.
3117 * For now, though, we'll cheat by calling filemap_flush(),
3118 * which will map the blocks, and start the I/O, but not
3119 * actually wait for the I/O to complete.
3121 return filemap_flush(inode
->i_mapping
);
3125 * bmap() is special. It gets used by applications such as lilo and by
3126 * the swapper to find the on-disk block of a specific piece of data.
3128 * Naturally, this is dangerous if the block concerned is still in the
3129 * journal. If somebody makes a swapfile on an ext4 data-journaling
3130 * filesystem and enables swap, then they may get a nasty shock when the
3131 * data getting swapped to that swapfile suddenly gets overwritten by
3132 * the original zero's written out previously to the journal and
3133 * awaiting writeback in the kernel's buffer cache.
3135 * So, if we see any bmap calls here on a modified, data-journaled file,
3136 * take extra steps to flush any blocks which might be in the cache.
3138 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3140 struct inode
*inode
= mapping
->host
;
3145 * We can get here for an inline file via the FIBMAP ioctl
3147 if (ext4_has_inline_data(inode
))
3150 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3151 test_opt(inode
->i_sb
, DELALLOC
)) {
3153 * With delalloc we want to sync the file
3154 * so that we can make sure we allocate
3157 filemap_write_and_wait(mapping
);
3160 if (EXT4_JOURNAL(inode
) &&
3161 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3163 * This is a REALLY heavyweight approach, but the use of
3164 * bmap on dirty files is expected to be extremely rare:
3165 * only if we run lilo or swapon on a freshly made file
3166 * do we expect this to happen.
3168 * (bmap requires CAP_SYS_RAWIO so this does not
3169 * represent an unprivileged user DOS attack --- we'd be
3170 * in trouble if mortal users could trigger this path at
3173 * NB. EXT4_STATE_JDATA is not set on files other than
3174 * regular files. If somebody wants to bmap a directory
3175 * or symlink and gets confused because the buffer
3176 * hasn't yet been flushed to disk, they deserve
3177 * everything they get.
3180 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3181 journal
= EXT4_JOURNAL(inode
);
3182 jbd2_journal_lock_updates(journal
);
3183 err
= jbd2_journal_flush(journal
);
3184 jbd2_journal_unlock_updates(journal
);
3190 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3193 static int ext4_readpage(struct file
*file
, struct page
*page
)
3196 struct inode
*inode
= page
->mapping
->host
;
3198 trace_ext4_readpage(page
);
3200 if (ext4_has_inline_data(inode
))
3201 ret
= ext4_readpage_inline(inode
, page
);
3204 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3210 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3211 struct list_head
*pages
, unsigned nr_pages
)
3213 struct inode
*inode
= mapping
->host
;
3215 /* If the file has inline data, no need to do readpages. */
3216 if (ext4_has_inline_data(inode
))
3219 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3222 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3223 unsigned int length
)
3225 trace_ext4_invalidatepage(page
, offset
, length
);
3227 /* No journalling happens on data buffers when this function is used */
3228 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3230 block_invalidatepage(page
, offset
, length
);
3233 static int __ext4_journalled_invalidatepage(struct page
*page
,
3234 unsigned int offset
,
3235 unsigned int length
)
3237 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3239 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3242 * If it's a full truncate we just forget about the pending dirtying
3244 if (offset
== 0 && length
== PAGE_SIZE
)
3245 ClearPageChecked(page
);
3247 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3250 /* Wrapper for aops... */
3251 static void ext4_journalled_invalidatepage(struct page
*page
,
3252 unsigned int offset
,
3253 unsigned int length
)
3255 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3258 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3260 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3262 trace_ext4_releasepage(page
);
3264 /* Page has dirty journalled data -> cannot release */
3265 if (PageChecked(page
))
3268 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3270 return try_to_free_buffers(page
);
3273 #ifdef CONFIG_FS_DAX
3275 * Get block function for DAX IO and mmap faults. It takes care of converting
3276 * unwritten extents to written ones and initializes new / converted blocks
3279 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3280 struct buffer_head
*bh_result
, int create
)
3284 ext4_debug("inode %lu, create flag %d\n", inode
->i_ino
, create
);
3286 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
3288 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3289 EXT4_GET_BLOCKS_PRE_IO
|
3290 EXT4_GET_BLOCKS_CREATE_ZERO
);
3294 if (buffer_unwritten(bh_result
)) {
3296 * We are protected by i_mmap_sem or i_mutex so we know block
3297 * cannot go away from under us even though we dropped
3298 * i_data_sem. Convert extent to written and write zeros there.
3300 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3301 EXT4_GET_BLOCKS_CONVERT
|
3302 EXT4_GET_BLOCKS_CREATE_ZERO
);
3307 * At least for now we have to clear BH_New so that DAX code
3308 * doesn't attempt to zero blocks again in a racy way.
3310 clear_buffer_new(bh_result
);
3314 /* Just define empty function, it will never get called. */
3315 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3316 struct buffer_head
*bh_result
, int create
)
3323 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3324 ssize_t size
, void *private)
3326 ext4_io_end_t
*io_end
= private;
3328 /* if not async direct IO just return */
3332 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3333 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3334 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3337 * Error during AIO DIO. We cannot convert unwritten extents as the
3338 * data was not written. Just clear the unwritten flag and drop io_end.
3341 ext4_clear_io_unwritten_flag(io_end
);
3344 io_end
->offset
= offset
;
3345 io_end
->size
= size
;
3346 ext4_put_io_end(io_end
);
3352 * Handling of direct IO writes.
3354 * For ext4 extent files, ext4 will do direct-io write even to holes,
3355 * preallocated extents, and those write extend the file, no need to
3356 * fall back to buffered IO.
3358 * For holes, we fallocate those blocks, mark them as unwritten
3359 * If those blocks were preallocated, we mark sure they are split, but
3360 * still keep the range to write as unwritten.
3362 * The unwritten extents will be converted to written when DIO is completed.
3363 * For async direct IO, since the IO may still pending when return, we
3364 * set up an end_io call back function, which will do the conversion
3365 * when async direct IO completed.
3367 * If the O_DIRECT write will extend the file then add this inode to the
3368 * orphan list. So recovery will truncate it back to the original size
3369 * if the machine crashes during the write.
3372 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3374 struct file
*file
= iocb
->ki_filp
;
3375 struct inode
*inode
= file
->f_mapping
->host
;
3376 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3378 loff_t offset
= iocb
->ki_pos
;
3379 size_t count
= iov_iter_count(iter
);
3381 get_block_t
*get_block_func
= NULL
;
3383 loff_t final_size
= offset
+ count
;
3387 if (final_size
> inode
->i_size
) {
3388 /* Credits for sb + inode write */
3389 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3390 if (IS_ERR(handle
)) {
3391 ret
= PTR_ERR(handle
);
3394 ret
= ext4_orphan_add(handle
, inode
);
3396 ext4_journal_stop(handle
);
3400 ei
->i_disksize
= inode
->i_size
;
3401 ext4_journal_stop(handle
);
3404 BUG_ON(iocb
->private == NULL
);
3407 * Make all waiters for direct IO properly wait also for extent
3408 * conversion. This also disallows race between truncate() and
3409 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3411 inode_dio_begin(inode
);
3413 /* If we do a overwrite dio, i_mutex locking can be released */
3414 overwrite
= *((int *)iocb
->private);
3417 inode_unlock(inode
);
3420 * For extent mapped files we could direct write to holes and fallocate.
3422 * Allocated blocks to fill the hole are marked as unwritten to prevent
3423 * parallel buffered read to expose the stale data before DIO complete
3426 * As to previously fallocated extents, ext4 get_block will just simply
3427 * mark the buffer mapped but still keep the extents unwritten.
3429 * For non AIO case, we will convert those unwritten extents to written
3430 * after return back from blockdev_direct_IO. That way we save us from
3431 * allocating io_end structure and also the overhead of offloading
3432 * the extent convertion to a workqueue.
3434 * For async DIO, the conversion needs to be deferred when the
3435 * IO is completed. The ext4 end_io callback function will be
3436 * called to take care of the conversion work. Here for async
3437 * case, we allocate an io_end structure to hook to the iocb.
3439 iocb
->private = NULL
;
3441 get_block_func
= ext4_dio_get_block_overwrite
;
3442 else if (IS_DAX(inode
)) {
3444 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3445 * writes need zeroing either because they can race with page
3446 * faults or because they use partial blocks.
3448 if (round_down(offset
, 1<<inode
->i_blkbits
) >= inode
->i_size
&&
3449 ext4_aligned_io(inode
, offset
, count
))
3450 get_block_func
= ext4_dio_get_block
;
3452 get_block_func
= ext4_dax_get_block
;
3453 dio_flags
= DIO_LOCKING
;
3454 } else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3455 round_down(offset
, 1 << inode
->i_blkbits
) >= inode
->i_size
) {
3456 get_block_func
= ext4_dio_get_block
;
3457 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3458 } else if (is_sync_kiocb(iocb
)) {
3459 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3460 dio_flags
= DIO_LOCKING
;
3462 get_block_func
= ext4_dio_get_block_unwritten_async
;
3463 dio_flags
= DIO_LOCKING
;
3465 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3466 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3468 if (IS_DAX(inode
)) {
3469 ret
= dax_do_io(iocb
, inode
, iter
, get_block_func
,
3470 ext4_end_io_dio
, dio_flags
);
3472 ret
= __blockdev_direct_IO(iocb
, inode
,
3473 inode
->i_sb
->s_bdev
, iter
,
3475 ext4_end_io_dio
, NULL
, dio_flags
);
3477 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3478 EXT4_STATE_DIO_UNWRITTEN
)) {
3481 * for non AIO case, since the IO is already
3482 * completed, we could do the conversion right here
3484 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3488 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3491 inode_dio_end(inode
);
3492 /* take i_mutex locking again if we do a ovewrite dio */
3496 if (ret
< 0 && final_size
> inode
->i_size
)
3497 ext4_truncate_failed_write(inode
);
3499 /* Handle extending of i_size after direct IO write */
3503 /* Credits for sb + inode write */
3504 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3505 if (IS_ERR(handle
)) {
3506 /* This is really bad luck. We've written the data
3507 * but cannot extend i_size. Bail out and pretend
3508 * the write failed... */
3509 ret
= PTR_ERR(handle
);
3511 ext4_orphan_del(NULL
, inode
);
3516 ext4_orphan_del(handle
, inode
);
3518 loff_t end
= offset
+ ret
;
3519 if (end
> inode
->i_size
) {
3520 ei
->i_disksize
= end
;
3521 i_size_write(inode
, end
);
3523 * We're going to return a positive `ret'
3524 * here due to non-zero-length I/O, so there's
3525 * no way of reporting error returns from
3526 * ext4_mark_inode_dirty() to userspace. So
3529 ext4_mark_inode_dirty(handle
, inode
);
3532 err
= ext4_journal_stop(handle
);
3540 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3542 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3543 struct inode
*inode
= mapping
->host
;
3547 * Shared inode_lock is enough for us - it protects against concurrent
3548 * writes & truncates and since we take care of writing back page cache,
3549 * we are protected against page writeback as well.
3551 inode_lock_shared(inode
);
3552 if (IS_DAX(inode
)) {
3553 ret
= dax_do_io(iocb
, inode
, iter
, ext4_dio_get_block
, NULL
, 0);
3555 size_t count
= iov_iter_count(iter
);
3557 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3558 iocb
->ki_pos
+ count
);
3561 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3562 iter
, ext4_dio_get_block
,
3566 inode_unlock_shared(inode
);
3570 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3572 struct file
*file
= iocb
->ki_filp
;
3573 struct inode
*inode
= file
->f_mapping
->host
;
3574 size_t count
= iov_iter_count(iter
);
3575 loff_t offset
= iocb
->ki_pos
;
3578 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3579 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3584 * If we are doing data journalling we don't support O_DIRECT
3586 if (ext4_should_journal_data(inode
))
3589 /* Let buffer I/O handle the inline data case. */
3590 if (ext4_has_inline_data(inode
))
3593 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3594 if (iov_iter_rw(iter
) == READ
)
3595 ret
= ext4_direct_IO_read(iocb
, iter
);
3597 ret
= ext4_direct_IO_write(iocb
, iter
);
3598 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3603 * Pages can be marked dirty completely asynchronously from ext4's journalling
3604 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3605 * much here because ->set_page_dirty is called under VFS locks. The page is
3606 * not necessarily locked.
3608 * We cannot just dirty the page and leave attached buffers clean, because the
3609 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3610 * or jbddirty because all the journalling code will explode.
3612 * So what we do is to mark the page "pending dirty" and next time writepage
3613 * is called, propagate that into the buffers appropriately.
3615 static int ext4_journalled_set_page_dirty(struct page
*page
)
3617 SetPageChecked(page
);
3618 return __set_page_dirty_nobuffers(page
);
3621 static const struct address_space_operations ext4_aops
= {
3622 .readpage
= ext4_readpage
,
3623 .readpages
= ext4_readpages
,
3624 .writepage
= ext4_writepage
,
3625 .writepages
= ext4_writepages
,
3626 .write_begin
= ext4_write_begin
,
3627 .write_end
= ext4_write_end
,
3629 .invalidatepage
= ext4_invalidatepage
,
3630 .releasepage
= ext4_releasepage
,
3631 .direct_IO
= ext4_direct_IO
,
3632 .migratepage
= buffer_migrate_page
,
3633 .is_partially_uptodate
= block_is_partially_uptodate
,
3634 .error_remove_page
= generic_error_remove_page
,
3637 static const struct address_space_operations ext4_journalled_aops
= {
3638 .readpage
= ext4_readpage
,
3639 .readpages
= ext4_readpages
,
3640 .writepage
= ext4_writepage
,
3641 .writepages
= ext4_writepages
,
3642 .write_begin
= ext4_write_begin
,
3643 .write_end
= ext4_journalled_write_end
,
3644 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3646 .invalidatepage
= ext4_journalled_invalidatepage
,
3647 .releasepage
= ext4_releasepage
,
3648 .direct_IO
= ext4_direct_IO
,
3649 .is_partially_uptodate
= block_is_partially_uptodate
,
3650 .error_remove_page
= generic_error_remove_page
,
3653 static const struct address_space_operations ext4_da_aops
= {
3654 .readpage
= ext4_readpage
,
3655 .readpages
= ext4_readpages
,
3656 .writepage
= ext4_writepage
,
3657 .writepages
= ext4_writepages
,
3658 .write_begin
= ext4_da_write_begin
,
3659 .write_end
= ext4_da_write_end
,
3661 .invalidatepage
= ext4_da_invalidatepage
,
3662 .releasepage
= ext4_releasepage
,
3663 .direct_IO
= ext4_direct_IO
,
3664 .migratepage
= buffer_migrate_page
,
3665 .is_partially_uptodate
= block_is_partially_uptodate
,
3666 .error_remove_page
= generic_error_remove_page
,
3669 void ext4_set_aops(struct inode
*inode
)
3671 switch (ext4_inode_journal_mode(inode
)) {
3672 case EXT4_INODE_ORDERED_DATA_MODE
:
3673 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3675 case EXT4_INODE_JOURNAL_DATA_MODE
:
3676 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3681 if (test_opt(inode
->i_sb
, DELALLOC
))
3682 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3684 inode
->i_mapping
->a_ops
= &ext4_aops
;
3687 static int __ext4_block_zero_page_range(handle_t
*handle
,
3688 struct address_space
*mapping
, loff_t from
, loff_t length
)
3690 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3691 unsigned offset
= from
& (PAGE_SIZE
-1);
3692 unsigned blocksize
, pos
;
3694 struct inode
*inode
= mapping
->host
;
3695 struct buffer_head
*bh
;
3699 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3700 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3704 blocksize
= inode
->i_sb
->s_blocksize
;
3706 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3708 if (!page_has_buffers(page
))
3709 create_empty_buffers(page
, blocksize
, 0);
3711 /* Find the buffer that contains "offset" */
3712 bh
= page_buffers(page
);
3714 while (offset
>= pos
) {
3715 bh
= bh
->b_this_page
;
3719 if (buffer_freed(bh
)) {
3720 BUFFER_TRACE(bh
, "freed: skip");
3723 if (!buffer_mapped(bh
)) {
3724 BUFFER_TRACE(bh
, "unmapped");
3725 ext4_get_block(inode
, iblock
, bh
, 0);
3726 /* unmapped? It's a hole - nothing to do */
3727 if (!buffer_mapped(bh
)) {
3728 BUFFER_TRACE(bh
, "still unmapped");
3733 /* Ok, it's mapped. Make sure it's up-to-date */
3734 if (PageUptodate(page
))
3735 set_buffer_uptodate(bh
);
3737 if (!buffer_uptodate(bh
)) {
3739 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3741 /* Uhhuh. Read error. Complain and punt. */
3742 if (!buffer_uptodate(bh
))
3744 if (S_ISREG(inode
->i_mode
) &&
3745 ext4_encrypted_inode(inode
)) {
3746 /* We expect the key to be set. */
3747 BUG_ON(!fscrypt_has_encryption_key(inode
));
3748 BUG_ON(blocksize
!= PAGE_SIZE
);
3749 WARN_ON_ONCE(fscrypt_decrypt_page(page
));
3752 if (ext4_should_journal_data(inode
)) {
3753 BUFFER_TRACE(bh
, "get write access");
3754 err
= ext4_journal_get_write_access(handle
, bh
);
3758 zero_user(page
, offset
, length
);
3759 BUFFER_TRACE(bh
, "zeroed end of block");
3761 if (ext4_should_journal_data(inode
)) {
3762 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3765 mark_buffer_dirty(bh
);
3766 if (ext4_should_order_data(inode
))
3767 err
= ext4_jbd2_inode_add_write(handle
, inode
);
3777 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3778 * starting from file offset 'from'. The range to be zero'd must
3779 * be contained with in one block. If the specified range exceeds
3780 * the end of the block it will be shortened to end of the block
3781 * that cooresponds to 'from'
3783 static int ext4_block_zero_page_range(handle_t
*handle
,
3784 struct address_space
*mapping
, loff_t from
, loff_t length
)
3786 struct inode
*inode
= mapping
->host
;
3787 unsigned offset
= from
& (PAGE_SIZE
-1);
3788 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3789 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3792 * correct length if it does not fall between
3793 * 'from' and the end of the block
3795 if (length
> max
|| length
< 0)
3799 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3800 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3804 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3805 * up to the end of the block which corresponds to `from'.
3806 * This required during truncate. We need to physically zero the tail end
3807 * of that block so it doesn't yield old data if the file is later grown.
3809 static int ext4_block_truncate_page(handle_t
*handle
,
3810 struct address_space
*mapping
, loff_t from
)
3812 unsigned offset
= from
& (PAGE_SIZE
-1);
3815 struct inode
*inode
= mapping
->host
;
3817 blocksize
= inode
->i_sb
->s_blocksize
;
3818 length
= blocksize
- (offset
& (blocksize
- 1));
3820 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3823 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3824 loff_t lstart
, loff_t length
)
3826 struct super_block
*sb
= inode
->i_sb
;
3827 struct address_space
*mapping
= inode
->i_mapping
;
3828 unsigned partial_start
, partial_end
;
3829 ext4_fsblk_t start
, end
;
3830 loff_t byte_end
= (lstart
+ length
- 1);
3833 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3834 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3836 start
= lstart
>> sb
->s_blocksize_bits
;
3837 end
= byte_end
>> sb
->s_blocksize_bits
;
3839 /* Handle partial zero within the single block */
3841 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3842 err
= ext4_block_zero_page_range(handle
, mapping
,
3846 /* Handle partial zero out on the start of the range */
3847 if (partial_start
) {
3848 err
= ext4_block_zero_page_range(handle
, mapping
,
3849 lstart
, sb
->s_blocksize
);
3853 /* Handle partial zero out on the end of the range */
3854 if (partial_end
!= sb
->s_blocksize
- 1)
3855 err
= ext4_block_zero_page_range(handle
, mapping
,
3856 byte_end
- partial_end
,
3861 int ext4_can_truncate(struct inode
*inode
)
3863 if (S_ISREG(inode
->i_mode
))
3865 if (S_ISDIR(inode
->i_mode
))
3867 if (S_ISLNK(inode
->i_mode
))
3868 return !ext4_inode_is_fast_symlink(inode
);
3873 * We have to make sure i_disksize gets properly updated before we truncate
3874 * page cache due to hole punching or zero range. Otherwise i_disksize update
3875 * can get lost as it may have been postponed to submission of writeback but
3876 * that will never happen after we truncate page cache.
3878 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3882 loff_t size
= i_size_read(inode
);
3884 WARN_ON(!inode_is_locked(inode
));
3885 if (offset
> size
|| offset
+ len
< size
)
3888 if (EXT4_I(inode
)->i_disksize
>= size
)
3891 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3893 return PTR_ERR(handle
);
3894 ext4_update_i_disksize(inode
, size
);
3895 ext4_mark_inode_dirty(handle
, inode
);
3896 ext4_journal_stop(handle
);
3902 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3903 * associated with the given offset and length
3905 * @inode: File inode
3906 * @offset: The offset where the hole will begin
3907 * @len: The length of the hole
3909 * Returns: 0 on success or negative on failure
3912 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3914 struct super_block
*sb
= inode
->i_sb
;
3915 ext4_lblk_t first_block
, stop_block
;
3916 struct address_space
*mapping
= inode
->i_mapping
;
3917 loff_t first_block_offset
, last_block_offset
;
3919 unsigned int credits
;
3922 if (!S_ISREG(inode
->i_mode
))
3925 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3928 * Write out all dirty pages to avoid race conditions
3929 * Then release them.
3931 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3932 ret
= filemap_write_and_wait_range(mapping
, offset
,
3933 offset
+ length
- 1);
3940 /* No need to punch hole beyond i_size */
3941 if (offset
>= inode
->i_size
)
3945 * If the hole extends beyond i_size, set the hole
3946 * to end after the page that contains i_size
3948 if (offset
+ length
> inode
->i_size
) {
3949 length
= inode
->i_size
+
3950 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
3954 if (offset
& (sb
->s_blocksize
- 1) ||
3955 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3957 * Attach jinode to inode for jbd2 if we do any zeroing of
3960 ret
= ext4_inode_attach_jinode(inode
);
3966 /* Wait all existing dio workers, newcomers will block on i_mutex */
3967 ext4_inode_block_unlocked_dio(inode
);
3968 inode_dio_wait(inode
);
3971 * Prevent page faults from reinstantiating pages we have released from
3974 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3975 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3976 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3978 /* Now release the pages and zero block aligned part of pages*/
3979 if (last_block_offset
> first_block_offset
) {
3980 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
3983 truncate_pagecache_range(inode
, first_block_offset
,
3987 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3988 credits
= ext4_writepage_trans_blocks(inode
);
3990 credits
= ext4_blocks_for_truncate(inode
);
3991 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3992 if (IS_ERR(handle
)) {
3993 ret
= PTR_ERR(handle
);
3994 ext4_std_error(sb
, ret
);
3998 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4003 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4004 EXT4_BLOCK_SIZE_BITS(sb
);
4005 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4007 /* If there are no blocks to remove, return now */
4008 if (first_block
>= stop_block
)
4011 down_write(&EXT4_I(inode
)->i_data_sem
);
4012 ext4_discard_preallocations(inode
);
4014 ret
= ext4_es_remove_extent(inode
, first_block
,
4015 stop_block
- first_block
);
4017 up_write(&EXT4_I(inode
)->i_data_sem
);
4021 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4022 ret
= ext4_ext_remove_space(inode
, first_block
,
4025 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4028 up_write(&EXT4_I(inode
)->i_data_sem
);
4030 ext4_handle_sync(handle
);
4032 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4033 ext4_mark_inode_dirty(handle
, inode
);
4035 ext4_journal_stop(handle
);
4037 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4038 ext4_inode_resume_unlocked_dio(inode
);
4040 inode_unlock(inode
);
4044 int ext4_inode_attach_jinode(struct inode
*inode
)
4046 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4047 struct jbd2_inode
*jinode
;
4049 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4052 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4053 spin_lock(&inode
->i_lock
);
4056 spin_unlock(&inode
->i_lock
);
4059 ei
->jinode
= jinode
;
4060 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4063 spin_unlock(&inode
->i_lock
);
4064 if (unlikely(jinode
!= NULL
))
4065 jbd2_free_inode(jinode
);
4072 * We block out ext4_get_block() block instantiations across the entire
4073 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4074 * simultaneously on behalf of the same inode.
4076 * As we work through the truncate and commit bits of it to the journal there
4077 * is one core, guiding principle: the file's tree must always be consistent on
4078 * disk. We must be able to restart the truncate after a crash.
4080 * The file's tree may be transiently inconsistent in memory (although it
4081 * probably isn't), but whenever we close off and commit a journal transaction,
4082 * the contents of (the filesystem + the journal) must be consistent and
4083 * restartable. It's pretty simple, really: bottom up, right to left (although
4084 * left-to-right works OK too).
4086 * Note that at recovery time, journal replay occurs *before* the restart of
4087 * truncate against the orphan inode list.
4089 * The committed inode has the new, desired i_size (which is the same as
4090 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4091 * that this inode's truncate did not complete and it will again call
4092 * ext4_truncate() to have another go. So there will be instantiated blocks
4093 * to the right of the truncation point in a crashed ext4 filesystem. But
4094 * that's fine - as long as they are linked from the inode, the post-crash
4095 * ext4_truncate() run will find them and release them.
4097 void ext4_truncate(struct inode
*inode
)
4099 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4100 unsigned int credits
;
4102 struct address_space
*mapping
= inode
->i_mapping
;
4105 * There is a possibility that we're either freeing the inode
4106 * or it's a completely new inode. In those cases we might not
4107 * have i_mutex locked because it's not necessary.
4109 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4110 WARN_ON(!inode_is_locked(inode
));
4111 trace_ext4_truncate_enter(inode
);
4113 if (!ext4_can_truncate(inode
))
4116 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4118 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4119 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4121 if (ext4_has_inline_data(inode
)) {
4124 ext4_inline_data_truncate(inode
, &has_inline
);
4129 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4130 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4131 if (ext4_inode_attach_jinode(inode
) < 0)
4135 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4136 credits
= ext4_writepage_trans_blocks(inode
);
4138 credits
= ext4_blocks_for_truncate(inode
);
4140 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4141 if (IS_ERR(handle
)) {
4142 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
4146 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4147 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4150 * We add the inode to the orphan list, so that if this
4151 * truncate spans multiple transactions, and we crash, we will
4152 * resume the truncate when the filesystem recovers. It also
4153 * marks the inode dirty, to catch the new size.
4155 * Implication: the file must always be in a sane, consistent
4156 * truncatable state while each transaction commits.
4158 if (ext4_orphan_add(handle
, inode
))
4161 down_write(&EXT4_I(inode
)->i_data_sem
);
4163 ext4_discard_preallocations(inode
);
4165 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4166 ext4_ext_truncate(handle
, inode
);
4168 ext4_ind_truncate(handle
, inode
);
4170 up_write(&ei
->i_data_sem
);
4173 ext4_handle_sync(handle
);
4177 * If this was a simple ftruncate() and the file will remain alive,
4178 * then we need to clear up the orphan record which we created above.
4179 * However, if this was a real unlink then we were called by
4180 * ext4_evict_inode(), and we allow that function to clean up the
4181 * orphan info for us.
4184 ext4_orphan_del(handle
, inode
);
4186 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4187 ext4_mark_inode_dirty(handle
, inode
);
4188 ext4_journal_stop(handle
);
4190 trace_ext4_truncate_exit(inode
);
4194 * ext4_get_inode_loc returns with an extra refcount against the inode's
4195 * underlying buffer_head on success. If 'in_mem' is true, we have all
4196 * data in memory that is needed to recreate the on-disk version of this
4199 static int __ext4_get_inode_loc(struct inode
*inode
,
4200 struct ext4_iloc
*iloc
, int in_mem
)
4202 struct ext4_group_desc
*gdp
;
4203 struct buffer_head
*bh
;
4204 struct super_block
*sb
= inode
->i_sb
;
4206 int inodes_per_block
, inode_offset
;
4209 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4210 return -EFSCORRUPTED
;
4212 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4213 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4218 * Figure out the offset within the block group inode table
4220 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4221 inode_offset
= ((inode
->i_ino
- 1) %
4222 EXT4_INODES_PER_GROUP(sb
));
4223 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4224 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4226 bh
= sb_getblk(sb
, block
);
4229 if (!buffer_uptodate(bh
)) {
4233 * If the buffer has the write error flag, we have failed
4234 * to write out another inode in the same block. In this
4235 * case, we don't have to read the block because we may
4236 * read the old inode data successfully.
4238 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4239 set_buffer_uptodate(bh
);
4241 if (buffer_uptodate(bh
)) {
4242 /* someone brought it uptodate while we waited */
4248 * If we have all information of the inode in memory and this
4249 * is the only valid inode in the block, we need not read the
4253 struct buffer_head
*bitmap_bh
;
4256 start
= inode_offset
& ~(inodes_per_block
- 1);
4258 /* Is the inode bitmap in cache? */
4259 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4260 if (unlikely(!bitmap_bh
))
4264 * If the inode bitmap isn't in cache then the
4265 * optimisation may end up performing two reads instead
4266 * of one, so skip it.
4268 if (!buffer_uptodate(bitmap_bh
)) {
4272 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4273 if (i
== inode_offset
)
4275 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4279 if (i
== start
+ inodes_per_block
) {
4280 /* all other inodes are free, so skip I/O */
4281 memset(bh
->b_data
, 0, bh
->b_size
);
4282 set_buffer_uptodate(bh
);
4290 * If we need to do any I/O, try to pre-readahead extra
4291 * blocks from the inode table.
4293 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4294 ext4_fsblk_t b
, end
, table
;
4296 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4298 table
= ext4_inode_table(sb
, gdp
);
4299 /* s_inode_readahead_blks is always a power of 2 */
4300 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4304 num
= EXT4_INODES_PER_GROUP(sb
);
4305 if (ext4_has_group_desc_csum(sb
))
4306 num
-= ext4_itable_unused_count(sb
, gdp
);
4307 table
+= num
/ inodes_per_block
;
4311 sb_breadahead(sb
, b
++);
4315 * There are other valid inodes in the buffer, this inode
4316 * has in-inode xattrs, or we don't have this inode in memory.
4317 * Read the block from disk.
4319 trace_ext4_load_inode(inode
);
4321 bh
->b_end_io
= end_buffer_read_sync
;
4322 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4324 if (!buffer_uptodate(bh
)) {
4325 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4326 "unable to read itable block");
4336 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4338 /* We have all inode data except xattrs in memory here. */
4339 return __ext4_get_inode_loc(inode
, iloc
,
4340 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4343 void ext4_set_inode_flags(struct inode
*inode
)
4345 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4346 unsigned int new_fl
= 0;
4348 if (flags
& EXT4_SYNC_FL
)
4350 if (flags
& EXT4_APPEND_FL
)
4352 if (flags
& EXT4_IMMUTABLE_FL
)
4353 new_fl
|= S_IMMUTABLE
;
4354 if (flags
& EXT4_NOATIME_FL
)
4355 new_fl
|= S_NOATIME
;
4356 if (flags
& EXT4_DIRSYNC_FL
)
4357 new_fl
|= S_DIRSYNC
;
4358 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
))
4360 inode_set_flags(inode
, new_fl
,
4361 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4364 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4365 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4367 unsigned int vfs_fl
;
4368 unsigned long old_fl
, new_fl
;
4371 vfs_fl
= ei
->vfs_inode
.i_flags
;
4372 old_fl
= ei
->i_flags
;
4373 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4374 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4376 if (vfs_fl
& S_SYNC
)
4377 new_fl
|= EXT4_SYNC_FL
;
4378 if (vfs_fl
& S_APPEND
)
4379 new_fl
|= EXT4_APPEND_FL
;
4380 if (vfs_fl
& S_IMMUTABLE
)
4381 new_fl
|= EXT4_IMMUTABLE_FL
;
4382 if (vfs_fl
& S_NOATIME
)
4383 new_fl
|= EXT4_NOATIME_FL
;
4384 if (vfs_fl
& S_DIRSYNC
)
4385 new_fl
|= EXT4_DIRSYNC_FL
;
4386 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4389 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4390 struct ext4_inode_info
*ei
)
4393 struct inode
*inode
= &(ei
->vfs_inode
);
4394 struct super_block
*sb
= inode
->i_sb
;
4396 if (ext4_has_feature_huge_file(sb
)) {
4397 /* we are using combined 48 bit field */
4398 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4399 le32_to_cpu(raw_inode
->i_blocks_lo
);
4400 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4401 /* i_blocks represent file system block size */
4402 return i_blocks
<< (inode
->i_blkbits
- 9);
4407 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4411 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4412 struct ext4_inode
*raw_inode
,
4413 struct ext4_inode_info
*ei
)
4415 __le32
*magic
= (void *)raw_inode
+
4416 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4417 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4418 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4419 ext4_find_inline_data_nolock(inode
);
4421 EXT4_I(inode
)->i_inline_off
= 0;
4424 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4426 if (!ext4_has_feature_project(inode
->i_sb
))
4428 *projid
= EXT4_I(inode
)->i_projid
;
4432 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4434 struct ext4_iloc iloc
;
4435 struct ext4_inode
*raw_inode
;
4436 struct ext4_inode_info
*ei
;
4437 struct inode
*inode
;
4438 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4445 inode
= iget_locked(sb
, ino
);
4447 return ERR_PTR(-ENOMEM
);
4448 if (!(inode
->i_state
& I_NEW
))
4454 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4457 raw_inode
= ext4_raw_inode(&iloc
);
4459 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4460 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4461 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4462 EXT4_INODE_SIZE(inode
->i_sb
)) {
4463 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4464 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4465 EXT4_INODE_SIZE(inode
->i_sb
));
4466 ret
= -EFSCORRUPTED
;
4470 ei
->i_extra_isize
= 0;
4472 /* Precompute checksum seed for inode metadata */
4473 if (ext4_has_metadata_csum(sb
)) {
4474 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4476 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4477 __le32 gen
= raw_inode
->i_generation
;
4478 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4480 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4484 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4485 EXT4_ERROR_INODE(inode
, "checksum invalid");
4490 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4491 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4492 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4493 if (ext4_has_feature_project(sb
) &&
4494 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4495 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4496 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4498 i_projid
= EXT4_DEF_PROJID
;
4500 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4501 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4502 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4504 i_uid_write(inode
, i_uid
);
4505 i_gid_write(inode
, i_gid
);
4506 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4507 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4509 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4510 ei
->i_inline_off
= 0;
4511 ei
->i_dir_start_lookup
= 0;
4512 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4513 /* We now have enough fields to check if the inode was active or not.
4514 * This is needed because nfsd might try to access dead inodes
4515 * the test is that same one that e2fsck uses
4516 * NeilBrown 1999oct15
4518 if (inode
->i_nlink
== 0) {
4519 if ((inode
->i_mode
== 0 ||
4520 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4521 ino
!= EXT4_BOOT_LOADER_INO
) {
4522 /* this inode is deleted */
4526 /* The only unlinked inodes we let through here have
4527 * valid i_mode and are being read by the orphan
4528 * recovery code: that's fine, we're about to complete
4529 * the process of deleting those.
4530 * OR it is the EXT4_BOOT_LOADER_INO which is
4531 * not initialized on a new filesystem. */
4533 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4534 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4535 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4536 if (ext4_has_feature_64bit(sb
))
4538 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4539 inode
->i_size
= ext4_isize(raw_inode
);
4540 ei
->i_disksize
= inode
->i_size
;
4542 ei
->i_reserved_quota
= 0;
4544 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4545 ei
->i_block_group
= iloc
.block_group
;
4546 ei
->i_last_alloc_group
= ~0;
4548 * NOTE! The in-memory inode i_data array is in little-endian order
4549 * even on big-endian machines: we do NOT byteswap the block numbers!
4551 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4552 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4553 INIT_LIST_HEAD(&ei
->i_orphan
);
4556 * Set transaction id's of transactions that have to be committed
4557 * to finish f[data]sync. We set them to currently running transaction
4558 * as we cannot be sure that the inode or some of its metadata isn't
4559 * part of the transaction - the inode could have been reclaimed and
4560 * now it is reread from disk.
4563 transaction_t
*transaction
;
4566 read_lock(&journal
->j_state_lock
);
4567 if (journal
->j_running_transaction
)
4568 transaction
= journal
->j_running_transaction
;
4570 transaction
= journal
->j_committing_transaction
;
4572 tid
= transaction
->t_tid
;
4574 tid
= journal
->j_commit_sequence
;
4575 read_unlock(&journal
->j_state_lock
);
4576 ei
->i_sync_tid
= tid
;
4577 ei
->i_datasync_tid
= tid
;
4580 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4581 if (ei
->i_extra_isize
== 0) {
4582 /* The extra space is currently unused. Use it. */
4583 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4584 EXT4_GOOD_OLD_INODE_SIZE
;
4586 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4590 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4591 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4592 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4593 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4595 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4596 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4597 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4598 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4600 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4605 if (ei
->i_file_acl
&&
4606 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4607 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4609 ret
= -EFSCORRUPTED
;
4611 } else if (!ext4_has_inline_data(inode
)) {
4612 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4613 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4614 (S_ISLNK(inode
->i_mode
) &&
4615 !ext4_inode_is_fast_symlink(inode
))))
4616 /* Validate extent which is part of inode */
4617 ret
= ext4_ext_check_inode(inode
);
4618 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4619 (S_ISLNK(inode
->i_mode
) &&
4620 !ext4_inode_is_fast_symlink(inode
))) {
4621 /* Validate block references which are part of inode */
4622 ret
= ext4_ind_check_inode(inode
);
4628 if (S_ISREG(inode
->i_mode
)) {
4629 inode
->i_op
= &ext4_file_inode_operations
;
4630 inode
->i_fop
= &ext4_file_operations
;
4631 ext4_set_aops(inode
);
4632 } else if (S_ISDIR(inode
->i_mode
)) {
4633 inode
->i_op
= &ext4_dir_inode_operations
;
4634 inode
->i_fop
= &ext4_dir_operations
;
4635 } else if (S_ISLNK(inode
->i_mode
)) {
4636 if (ext4_encrypted_inode(inode
)) {
4637 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4638 ext4_set_aops(inode
);
4639 } else if (ext4_inode_is_fast_symlink(inode
)) {
4640 inode
->i_link
= (char *)ei
->i_data
;
4641 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4642 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4643 sizeof(ei
->i_data
) - 1);
4645 inode
->i_op
= &ext4_symlink_inode_operations
;
4646 ext4_set_aops(inode
);
4648 inode_nohighmem(inode
);
4649 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4650 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4651 inode
->i_op
= &ext4_special_inode_operations
;
4652 if (raw_inode
->i_block
[0])
4653 init_special_inode(inode
, inode
->i_mode
,
4654 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4656 init_special_inode(inode
, inode
->i_mode
,
4657 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4658 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4659 make_bad_inode(inode
);
4661 ret
= -EFSCORRUPTED
;
4662 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4666 ext4_set_inode_flags(inode
);
4667 unlock_new_inode(inode
);
4673 return ERR_PTR(ret
);
4676 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4678 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4679 return ERR_PTR(-EFSCORRUPTED
);
4680 return ext4_iget(sb
, ino
);
4683 static int ext4_inode_blocks_set(handle_t
*handle
,
4684 struct ext4_inode
*raw_inode
,
4685 struct ext4_inode_info
*ei
)
4687 struct inode
*inode
= &(ei
->vfs_inode
);
4688 u64 i_blocks
= inode
->i_blocks
;
4689 struct super_block
*sb
= inode
->i_sb
;
4691 if (i_blocks
<= ~0U) {
4693 * i_blocks can be represented in a 32 bit variable
4694 * as multiple of 512 bytes
4696 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4697 raw_inode
->i_blocks_high
= 0;
4698 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4701 if (!ext4_has_feature_huge_file(sb
))
4704 if (i_blocks
<= 0xffffffffffffULL
) {
4706 * i_blocks can be represented in a 48 bit variable
4707 * as multiple of 512 bytes
4709 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4710 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4711 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4713 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4714 /* i_block is stored in file system block size */
4715 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4716 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4717 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4722 struct other_inode
{
4723 unsigned long orig_ino
;
4724 struct ext4_inode
*raw_inode
;
4727 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4730 struct other_inode
*oi
= (struct other_inode
*) data
;
4732 if ((inode
->i_ino
!= ino
) ||
4733 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4734 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4735 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4737 spin_lock(&inode
->i_lock
);
4738 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4739 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4740 (inode
->i_state
& I_DIRTY_TIME
)) {
4741 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4743 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4744 spin_unlock(&inode
->i_lock
);
4746 spin_lock(&ei
->i_raw_lock
);
4747 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4748 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4749 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4750 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4751 spin_unlock(&ei
->i_raw_lock
);
4752 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4755 spin_unlock(&inode
->i_lock
);
4760 * Opportunistically update the other time fields for other inodes in
4761 * the same inode table block.
4763 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4764 unsigned long orig_ino
, char *buf
)
4766 struct other_inode oi
;
4768 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4769 int inode_size
= EXT4_INODE_SIZE(sb
);
4771 oi
.orig_ino
= orig_ino
;
4773 * Calculate the first inode in the inode table block. Inode
4774 * numbers are one-based. That is, the first inode in a block
4775 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4777 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4778 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4779 if (ino
== orig_ino
)
4781 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4782 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4787 * Post the struct inode info into an on-disk inode location in the
4788 * buffer-cache. This gobbles the caller's reference to the
4789 * buffer_head in the inode location struct.
4791 * The caller must have write access to iloc->bh.
4793 static int ext4_do_update_inode(handle_t
*handle
,
4794 struct inode
*inode
,
4795 struct ext4_iloc
*iloc
)
4797 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4798 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4799 struct buffer_head
*bh
= iloc
->bh
;
4800 struct super_block
*sb
= inode
->i_sb
;
4801 int err
= 0, rc
, block
;
4802 int need_datasync
= 0, set_large_file
= 0;
4807 spin_lock(&ei
->i_raw_lock
);
4809 /* For fields not tracked in the in-memory inode,
4810 * initialise them to zero for new inodes. */
4811 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4812 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4814 ext4_get_inode_flags(ei
);
4815 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4816 i_uid
= i_uid_read(inode
);
4817 i_gid
= i_gid_read(inode
);
4818 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4819 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4820 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4821 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4823 * Fix up interoperability with old kernels. Otherwise, old inodes get
4824 * re-used with the upper 16 bits of the uid/gid intact
4826 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4827 raw_inode
->i_uid_high
= 0;
4828 raw_inode
->i_gid_high
= 0;
4830 raw_inode
->i_uid_high
=
4831 cpu_to_le16(high_16_bits(i_uid
));
4832 raw_inode
->i_gid_high
=
4833 cpu_to_le16(high_16_bits(i_gid
));
4836 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4837 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4838 raw_inode
->i_uid_high
= 0;
4839 raw_inode
->i_gid_high
= 0;
4841 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4843 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4844 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4845 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4846 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4848 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4850 spin_unlock(&ei
->i_raw_lock
);
4853 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4854 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4855 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4856 raw_inode
->i_file_acl_high
=
4857 cpu_to_le16(ei
->i_file_acl
>> 32);
4858 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4859 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4860 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4863 if (ei
->i_disksize
> 0x7fffffffULL
) {
4864 if (!ext4_has_feature_large_file(sb
) ||
4865 EXT4_SB(sb
)->s_es
->s_rev_level
==
4866 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4869 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4870 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4871 if (old_valid_dev(inode
->i_rdev
)) {
4872 raw_inode
->i_block
[0] =
4873 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4874 raw_inode
->i_block
[1] = 0;
4876 raw_inode
->i_block
[0] = 0;
4877 raw_inode
->i_block
[1] =
4878 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4879 raw_inode
->i_block
[2] = 0;
4881 } else if (!ext4_has_inline_data(inode
)) {
4882 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4883 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4886 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4887 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4888 if (ei
->i_extra_isize
) {
4889 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4890 raw_inode
->i_version_hi
=
4891 cpu_to_le32(inode
->i_version
>> 32);
4892 raw_inode
->i_extra_isize
=
4893 cpu_to_le16(ei
->i_extra_isize
);
4897 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
4898 i_projid
!= EXT4_DEF_PROJID
);
4900 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4901 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4902 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4904 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4905 spin_unlock(&ei
->i_raw_lock
);
4906 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4907 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4910 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4911 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4914 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4915 if (set_large_file
) {
4916 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4917 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4920 ext4_update_dynamic_rev(sb
);
4921 ext4_set_feature_large_file(sb
);
4922 ext4_handle_sync(handle
);
4923 err
= ext4_handle_dirty_super(handle
, sb
);
4925 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4928 ext4_std_error(inode
->i_sb
, err
);
4933 * ext4_write_inode()
4935 * We are called from a few places:
4937 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4938 * Here, there will be no transaction running. We wait for any running
4939 * transaction to commit.
4941 * - Within flush work (sys_sync(), kupdate and such).
4942 * We wait on commit, if told to.
4944 * - Within iput_final() -> write_inode_now()
4945 * We wait on commit, if told to.
4947 * In all cases it is actually safe for us to return without doing anything,
4948 * because the inode has been copied into a raw inode buffer in
4949 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4952 * Note that we are absolutely dependent upon all inode dirtiers doing the
4953 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4954 * which we are interested.
4956 * It would be a bug for them to not do this. The code:
4958 * mark_inode_dirty(inode)
4960 * inode->i_size = expr;
4962 * is in error because write_inode() could occur while `stuff()' is running,
4963 * and the new i_size will be lost. Plus the inode will no longer be on the
4964 * superblock's dirty inode list.
4966 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4970 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4973 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4974 if (ext4_journal_current_handle()) {
4975 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4981 * No need to force transaction in WB_SYNC_NONE mode. Also
4982 * ext4_sync_fs() will force the commit after everything is
4985 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4988 err
= ext4_force_commit(inode
->i_sb
);
4990 struct ext4_iloc iloc
;
4992 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4996 * sync(2) will flush the whole buffer cache. No need to do
4997 * it here separately for each inode.
4999 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5000 sync_dirty_buffer(iloc
.bh
);
5001 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5002 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5003 "IO error syncing inode");
5012 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5013 * buffers that are attached to a page stradding i_size and are undergoing
5014 * commit. In that case we have to wait for commit to finish and try again.
5016 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5020 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5021 tid_t commit_tid
= 0;
5024 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5026 * All buffers in the last page remain valid? Then there's nothing to
5027 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5030 if (offset
> PAGE_SIZE
- (1 << inode
->i_blkbits
))
5033 page
= find_lock_page(inode
->i_mapping
,
5034 inode
->i_size
>> PAGE_SHIFT
);
5037 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5038 PAGE_SIZE
- offset
);
5044 read_lock(&journal
->j_state_lock
);
5045 if (journal
->j_committing_transaction
)
5046 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5047 read_unlock(&journal
->j_state_lock
);
5049 jbd2_log_wait_commit(journal
, commit_tid
);
5056 * Called from notify_change.
5058 * We want to trap VFS attempts to truncate the file as soon as
5059 * possible. In particular, we want to make sure that when the VFS
5060 * shrinks i_size, we put the inode on the orphan list and modify
5061 * i_disksize immediately, so that during the subsequent flushing of
5062 * dirty pages and freeing of disk blocks, we can guarantee that any
5063 * commit will leave the blocks being flushed in an unused state on
5064 * disk. (On recovery, the inode will get truncated and the blocks will
5065 * be freed, so we have a strong guarantee that no future commit will
5066 * leave these blocks visible to the user.)
5068 * Another thing we have to assure is that if we are in ordered mode
5069 * and inode is still attached to the committing transaction, we must
5070 * we start writeout of all the dirty pages which are being truncated.
5071 * This way we are sure that all the data written in the previous
5072 * transaction are already on disk (truncate waits for pages under
5075 * Called with inode->i_mutex down.
5077 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5079 struct inode
*inode
= d_inode(dentry
);
5082 const unsigned int ia_valid
= attr
->ia_valid
;
5084 error
= setattr_prepare(dentry
, attr
);
5088 if (is_quota_modification(inode
, attr
)) {
5089 error
= dquot_initialize(inode
);
5093 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5094 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5097 /* (user+group)*(old+new) structure, inode write (sb,
5098 * inode block, ? - but truncate inode update has it) */
5099 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5100 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5101 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5102 if (IS_ERR(handle
)) {
5103 error
= PTR_ERR(handle
);
5106 error
= dquot_transfer(inode
, attr
);
5108 ext4_journal_stop(handle
);
5111 /* Update corresponding info in inode so that everything is in
5112 * one transaction */
5113 if (attr
->ia_valid
& ATTR_UID
)
5114 inode
->i_uid
= attr
->ia_uid
;
5115 if (attr
->ia_valid
& ATTR_GID
)
5116 inode
->i_gid
= attr
->ia_gid
;
5117 error
= ext4_mark_inode_dirty(handle
, inode
);
5118 ext4_journal_stop(handle
);
5121 if (attr
->ia_valid
& ATTR_SIZE
) {
5123 loff_t oldsize
= inode
->i_size
;
5124 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5126 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5127 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5129 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5132 if (!S_ISREG(inode
->i_mode
))
5135 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5136 inode_inc_iversion(inode
);
5138 if (ext4_should_order_data(inode
) &&
5139 (attr
->ia_size
< inode
->i_size
)) {
5140 error
= ext4_begin_ordered_truncate(inode
,
5145 if (attr
->ia_size
!= inode
->i_size
) {
5146 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5147 if (IS_ERR(handle
)) {
5148 error
= PTR_ERR(handle
);
5151 if (ext4_handle_valid(handle
) && shrink
) {
5152 error
= ext4_orphan_add(handle
, inode
);
5156 * Update c/mtime on truncate up, ext4_truncate() will
5157 * update c/mtime in shrink case below
5160 inode
->i_mtime
= ext4_current_time(inode
);
5161 inode
->i_ctime
= inode
->i_mtime
;
5163 down_write(&EXT4_I(inode
)->i_data_sem
);
5164 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5165 rc
= ext4_mark_inode_dirty(handle
, inode
);
5169 * We have to update i_size under i_data_sem together
5170 * with i_disksize to avoid races with writeback code
5171 * running ext4_wb_update_i_disksize().
5174 i_size_write(inode
, attr
->ia_size
);
5175 up_write(&EXT4_I(inode
)->i_data_sem
);
5176 ext4_journal_stop(handle
);
5179 ext4_orphan_del(NULL
, inode
);
5184 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5187 * Blocks are going to be removed from the inode. Wait
5188 * for dio in flight. Temporarily disable
5189 * dioread_nolock to prevent livelock.
5192 if (!ext4_should_journal_data(inode
)) {
5193 ext4_inode_block_unlocked_dio(inode
);
5194 inode_dio_wait(inode
);
5195 ext4_inode_resume_unlocked_dio(inode
);
5197 ext4_wait_for_tail_page_commit(inode
);
5199 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5201 * Truncate pagecache after we've waited for commit
5202 * in data=journal mode to make pages freeable.
5204 truncate_pagecache(inode
, inode
->i_size
);
5206 ext4_truncate(inode
);
5207 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5211 setattr_copy(inode
, attr
);
5212 mark_inode_dirty(inode
);
5216 * If the call to ext4_truncate failed to get a transaction handle at
5217 * all, we need to clean up the in-core orphan list manually.
5219 if (orphan
&& inode
->i_nlink
)
5220 ext4_orphan_del(NULL
, inode
);
5222 if (!rc
&& (ia_valid
& ATTR_MODE
))
5223 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5226 ext4_std_error(inode
->i_sb
, error
);
5232 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5235 struct inode
*inode
;
5236 unsigned long long delalloc_blocks
;
5238 inode
= d_inode(dentry
);
5239 generic_fillattr(inode
, stat
);
5242 * If there is inline data in the inode, the inode will normally not
5243 * have data blocks allocated (it may have an external xattr block).
5244 * Report at least one sector for such files, so tools like tar, rsync,
5245 * others doen't incorrectly think the file is completely sparse.
5247 if (unlikely(ext4_has_inline_data(inode
)))
5248 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5251 * We can't update i_blocks if the block allocation is delayed
5252 * otherwise in the case of system crash before the real block
5253 * allocation is done, we will have i_blocks inconsistent with
5254 * on-disk file blocks.
5255 * We always keep i_blocks updated together with real
5256 * allocation. But to not confuse with user, stat
5257 * will return the blocks that include the delayed allocation
5258 * blocks for this file.
5260 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5261 EXT4_I(inode
)->i_reserved_data_blocks
);
5262 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5266 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5269 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5270 return ext4_ind_trans_blocks(inode
, lblocks
);
5271 return ext4_ext_index_trans_blocks(inode
, pextents
);
5275 * Account for index blocks, block groups bitmaps and block group
5276 * descriptor blocks if modify datablocks and index blocks
5277 * worse case, the indexs blocks spread over different block groups
5279 * If datablocks are discontiguous, they are possible to spread over
5280 * different block groups too. If they are contiguous, with flexbg,
5281 * they could still across block group boundary.
5283 * Also account for superblock, inode, quota and xattr blocks
5285 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5288 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5294 * How many index blocks need to touch to map @lblocks logical blocks
5295 * to @pextents physical extents?
5297 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5302 * Now let's see how many group bitmaps and group descriptors need
5305 groups
= idxblocks
+ pextents
;
5307 if (groups
> ngroups
)
5309 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5310 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5312 /* bitmaps and block group descriptor blocks */
5313 ret
+= groups
+ gdpblocks
;
5315 /* Blocks for super block, inode, quota and xattr blocks */
5316 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5322 * Calculate the total number of credits to reserve to fit
5323 * the modification of a single pages into a single transaction,
5324 * which may include multiple chunks of block allocations.
5326 * This could be called via ext4_write_begin()
5328 * We need to consider the worse case, when
5329 * one new block per extent.
5331 int ext4_writepage_trans_blocks(struct inode
*inode
)
5333 int bpp
= ext4_journal_blocks_per_page(inode
);
5336 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5338 /* Account for data blocks for journalled mode */
5339 if (ext4_should_journal_data(inode
))
5345 * Calculate the journal credits for a chunk of data modification.
5347 * This is called from DIO, fallocate or whoever calling
5348 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5350 * journal buffers for data blocks are not included here, as DIO
5351 * and fallocate do no need to journal data buffers.
5353 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5355 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5359 * The caller must have previously called ext4_reserve_inode_write().
5360 * Give this, we know that the caller already has write access to iloc->bh.
5362 int ext4_mark_iloc_dirty(handle_t
*handle
,
5363 struct inode
*inode
, struct ext4_iloc
*iloc
)
5367 if (IS_I_VERSION(inode
))
5368 inode_inc_iversion(inode
);
5370 /* the do_update_inode consumes one bh->b_count */
5373 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5374 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5380 * On success, We end up with an outstanding reference count against
5381 * iloc->bh. This _must_ be cleaned up later.
5385 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5386 struct ext4_iloc
*iloc
)
5390 err
= ext4_get_inode_loc(inode
, iloc
);
5392 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5393 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5399 ext4_std_error(inode
->i_sb
, err
);
5404 * Expand an inode by new_extra_isize bytes.
5405 * Returns 0 on success or negative error number on failure.
5407 static int ext4_expand_extra_isize(struct inode
*inode
,
5408 unsigned int new_extra_isize
,
5409 struct ext4_iloc iloc
,
5412 struct ext4_inode
*raw_inode
;
5413 struct ext4_xattr_ibody_header
*header
;
5415 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5418 raw_inode
= ext4_raw_inode(&iloc
);
5420 header
= IHDR(inode
, raw_inode
);
5422 /* No extended attributes present */
5423 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5424 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5425 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5427 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5431 /* try to expand with EAs present */
5432 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5437 * What we do here is to mark the in-core inode as clean with respect to inode
5438 * dirtiness (it may still be data-dirty).
5439 * This means that the in-core inode may be reaped by prune_icache
5440 * without having to perform any I/O. This is a very good thing,
5441 * because *any* task may call prune_icache - even ones which
5442 * have a transaction open against a different journal.
5444 * Is this cheating? Not really. Sure, we haven't written the
5445 * inode out, but prune_icache isn't a user-visible syncing function.
5446 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5447 * we start and wait on commits.
5449 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5451 struct ext4_iloc iloc
;
5452 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5453 static unsigned int mnt_count
;
5457 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5458 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5461 if (ext4_handle_valid(handle
) &&
5462 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5463 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5465 * We need extra buffer credits since we may write into EA block
5466 * with this same handle. If journal_extend fails, then it will
5467 * only result in a minor loss of functionality for that inode.
5468 * If this is felt to be critical, then e2fsck should be run to
5469 * force a large enough s_min_extra_isize.
5471 if ((jbd2_journal_extend(handle
,
5472 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5473 ret
= ext4_expand_extra_isize(inode
,
5474 sbi
->s_want_extra_isize
,
5478 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5479 ext4_warning(inode
->i_sb
,
5480 "Unable to expand inode %lu. Delete"
5481 " some EAs or run e2fsck.",
5484 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5489 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5493 * ext4_dirty_inode() is called from __mark_inode_dirty()
5495 * We're really interested in the case where a file is being extended.
5496 * i_size has been changed by generic_commit_write() and we thus need
5497 * to include the updated inode in the current transaction.
5499 * Also, dquot_alloc_block() will always dirty the inode when blocks
5500 * are allocated to the file.
5502 * If the inode is marked synchronous, we don't honour that here - doing
5503 * so would cause a commit on atime updates, which we don't bother doing.
5504 * We handle synchronous inodes at the highest possible level.
5506 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5507 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5508 * to copy into the on-disk inode structure are the timestamp files.
5510 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5514 if (flags
== I_DIRTY_TIME
)
5516 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5520 ext4_mark_inode_dirty(handle
, inode
);
5522 ext4_journal_stop(handle
);
5529 * Bind an inode's backing buffer_head into this transaction, to prevent
5530 * it from being flushed to disk early. Unlike
5531 * ext4_reserve_inode_write, this leaves behind no bh reference and
5532 * returns no iloc structure, so the caller needs to repeat the iloc
5533 * lookup to mark the inode dirty later.
5535 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5537 struct ext4_iloc iloc
;
5541 err
= ext4_get_inode_loc(inode
, &iloc
);
5543 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5544 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5546 err
= ext4_handle_dirty_metadata(handle
,
5552 ext4_std_error(inode
->i_sb
, err
);
5557 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5562 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5565 * We have to be very careful here: changing a data block's
5566 * journaling status dynamically is dangerous. If we write a
5567 * data block to the journal, change the status and then delete
5568 * that block, we risk forgetting to revoke the old log record
5569 * from the journal and so a subsequent replay can corrupt data.
5570 * So, first we make sure that the journal is empty and that
5571 * nobody is changing anything.
5574 journal
= EXT4_JOURNAL(inode
);
5577 if (is_journal_aborted(journal
))
5580 /* Wait for all existing dio workers */
5581 ext4_inode_block_unlocked_dio(inode
);
5582 inode_dio_wait(inode
);
5585 * Before flushing the journal and switching inode's aops, we have
5586 * to flush all dirty data the inode has. There can be outstanding
5587 * delayed allocations, there can be unwritten extents created by
5588 * fallocate or buffered writes in dioread_nolock mode covered by
5589 * dirty data which can be converted only after flushing the dirty
5590 * data (and journalled aops don't know how to handle these cases).
5593 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5594 err
= filemap_write_and_wait(inode
->i_mapping
);
5596 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5597 ext4_inode_resume_unlocked_dio(inode
);
5602 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5603 jbd2_journal_lock_updates(journal
);
5606 * OK, there are no updates running now, and all cached data is
5607 * synced to disk. We are now in a completely consistent state
5608 * which doesn't have anything in the journal, and we know that
5609 * no filesystem updates are running, so it is safe to modify
5610 * the inode's in-core data-journaling state flag now.
5614 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5616 err
= jbd2_journal_flush(journal
);
5618 jbd2_journal_unlock_updates(journal
);
5619 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5620 ext4_inode_resume_unlocked_dio(inode
);
5623 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5625 ext4_set_aops(inode
);
5627 jbd2_journal_unlock_updates(journal
);
5628 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5631 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5632 ext4_inode_resume_unlocked_dio(inode
);
5634 /* Finally we can mark the inode as dirty. */
5636 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5638 return PTR_ERR(handle
);
5640 err
= ext4_mark_inode_dirty(handle
, inode
);
5641 ext4_handle_sync(handle
);
5642 ext4_journal_stop(handle
);
5643 ext4_std_error(inode
->i_sb
, err
);
5648 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5650 return !buffer_mapped(bh
);
5653 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5655 struct page
*page
= vmf
->page
;
5659 struct file
*file
= vma
->vm_file
;
5660 struct inode
*inode
= file_inode(file
);
5661 struct address_space
*mapping
= inode
->i_mapping
;
5663 get_block_t
*get_block
;
5666 sb_start_pagefault(inode
->i_sb
);
5667 file_update_time(vma
->vm_file
);
5669 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5670 /* Delalloc case is easy... */
5671 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5672 !ext4_should_journal_data(inode
) &&
5673 !ext4_nonda_switch(inode
->i_sb
)) {
5675 ret
= block_page_mkwrite(vma
, vmf
,
5676 ext4_da_get_block_prep
);
5677 } while (ret
== -ENOSPC
&&
5678 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5683 size
= i_size_read(inode
);
5684 /* Page got truncated from under us? */
5685 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5687 ret
= VM_FAULT_NOPAGE
;
5691 if (page
->index
== size
>> PAGE_SHIFT
)
5692 len
= size
& ~PAGE_MASK
;
5696 * Return if we have all the buffers mapped. This avoids the need to do
5697 * journal_start/journal_stop which can block and take a long time
5699 if (page_has_buffers(page
)) {
5700 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5702 ext4_bh_unmapped
)) {
5703 /* Wait so that we don't change page under IO */
5704 wait_for_stable_page(page
);
5705 ret
= VM_FAULT_LOCKED
;
5710 /* OK, we need to fill the hole... */
5711 if (ext4_should_dioread_nolock(inode
))
5712 get_block
= ext4_get_block_unwritten
;
5714 get_block
= ext4_get_block
;
5716 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5717 ext4_writepage_trans_blocks(inode
));
5718 if (IS_ERR(handle
)) {
5719 ret
= VM_FAULT_SIGBUS
;
5722 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5723 if (!ret
&& ext4_should_journal_data(inode
)) {
5724 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5725 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
5727 ret
= VM_FAULT_SIGBUS
;
5728 ext4_journal_stop(handle
);
5731 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5733 ext4_journal_stop(handle
);
5734 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5737 ret
= block_page_mkwrite_return(ret
);
5739 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5740 sb_end_pagefault(inode
->i_sb
);
5744 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5746 struct inode
*inode
= file_inode(vma
->vm_file
);
5749 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5750 err
= filemap_fault(vma
, vmf
);
5751 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5757 * Find the first extent at or after @lblk in an inode that is not a hole.
5758 * Search for @map_len blocks at most. The extent is returned in @result.
5760 * The function returns 1 if we found an extent. The function returns 0 in
5761 * case there is no extent at or after @lblk and in that case also sets
5762 * @result->es_len to 0. In case of error, the error code is returned.
5764 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
5765 unsigned int map_len
, struct extent_status
*result
)
5767 struct ext4_map_blocks map
;
5768 struct extent_status es
= {};
5772 map
.m_len
= map_len
;
5775 * For non-extent based files this loop may iterate several times since
5776 * we do not determine full hole size.
5778 while (map
.m_len
> 0) {
5779 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
5782 /* There's extent covering m_lblk? Just return it. */
5786 ext4_es_store_pblock(result
, map
.m_pblk
);
5787 result
->es_lblk
= map
.m_lblk
;
5788 result
->es_len
= map
.m_len
;
5789 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
5790 status
= EXTENT_STATUS_UNWRITTEN
;
5792 status
= EXTENT_STATUS_WRITTEN
;
5793 ext4_es_store_status(result
, status
);
5796 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
5797 map
.m_lblk
+ map
.m_len
- 1,
5799 /* Is delalloc data before next block in extent tree? */
5800 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
5801 ext4_lblk_t offset
= 0;
5803 if (es
.es_lblk
< lblk
)
5804 offset
= lblk
- es
.es_lblk
;
5805 result
->es_lblk
= es
.es_lblk
+ offset
;
5806 ext4_es_store_pblock(result
,
5807 ext4_es_pblock(&es
) + offset
);
5808 result
->es_len
= es
.es_len
- offset
;
5809 ext4_es_store_status(result
, ext4_es_status(&es
));
5813 /* There's a hole at m_lblk, advance us after it */
5814 map
.m_lblk
+= map
.m_len
;
5815 map_len
-= map
.m_len
;
5816 map
.m_len
= map_len
;