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
;
774 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
775 struct buffer_head
*bh
, int create
)
777 return _ext4_get_block(inode
, iblock
, bh
,
778 create
? EXT4_GET_BLOCKS_CREATE
: 0);
782 * Get block function used when preparing for buffered write if we require
783 * creating an unwritten extent if blocks haven't been allocated. The extent
784 * will be converted to written after the IO is complete.
786 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
787 struct buffer_head
*bh_result
, int create
)
789 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
790 inode
->i_ino
, create
);
791 return _ext4_get_block(inode
, iblock
, bh_result
,
792 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
795 /* Maximum number of blocks we map for direct IO at once. */
796 #define DIO_MAX_BLOCKS 4096
799 * Get blocks function for the cases that need to start a transaction -
800 * generally difference cases of direct IO and DAX IO. It also handles retries
803 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
804 struct buffer_head
*bh_result
, int flags
)
811 /* Trim mapping request to maximum we can map at once for DIO */
812 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
813 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
814 dio_credits
= ext4_chunk_trans_blocks(inode
,
815 bh_result
->b_size
>> inode
->i_blkbits
);
817 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
819 return PTR_ERR(handle
);
821 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
822 ext4_journal_stop(handle
);
824 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
829 /* Get block function for DIO reads and writes to inodes without extents */
830 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
831 struct buffer_head
*bh
, int create
)
833 /* We don't expect handle for direct IO */
834 WARN_ON_ONCE(ext4_journal_current_handle());
837 return _ext4_get_block(inode
, iblock
, bh
, 0);
838 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
842 * Get block function for AIO DIO writes when we create unwritten extent if
843 * blocks are not allocated yet. The extent will be converted to written
844 * after IO is complete.
846 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
847 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
851 /* We don't expect handle for direct IO */
852 WARN_ON_ONCE(ext4_journal_current_handle());
854 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
855 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
858 * When doing DIO using unwritten extents, we need io_end to convert
859 * unwritten extents to written on IO completion. We allocate io_end
860 * once we spot unwritten extent and store it in b_private. Generic
861 * DIO code keeps b_private set and furthermore passes the value to
862 * our completion callback in 'private' argument.
864 if (!ret
&& buffer_unwritten(bh_result
)) {
865 if (!bh_result
->b_private
) {
866 ext4_io_end_t
*io_end
;
868 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
871 bh_result
->b_private
= io_end
;
872 ext4_set_io_unwritten_flag(inode
, io_end
);
874 set_buffer_defer_completion(bh_result
);
881 * Get block function for non-AIO DIO writes when we create unwritten extent if
882 * blocks are not allocated yet. The extent will be converted to written
883 * after IO is complete from ext4_ext_direct_IO() function.
885 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
886 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
890 /* We don't expect handle for direct IO */
891 WARN_ON_ONCE(ext4_journal_current_handle());
893 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
894 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
897 * Mark inode as having pending DIO writes to unwritten extents.
898 * ext4_ext_direct_IO() checks this flag and converts extents to
901 if (!ret
&& buffer_unwritten(bh_result
))
902 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
907 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
908 struct buffer_head
*bh_result
, int create
)
912 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
913 inode
->i_ino
, create
);
914 /* We don't expect handle for direct IO */
915 WARN_ON_ONCE(ext4_journal_current_handle());
917 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
919 * Blocks should have been preallocated! ext4_file_write_iter() checks
922 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
929 * `handle' can be NULL if create is zero
931 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
932 ext4_lblk_t block
, int map_flags
)
934 struct ext4_map_blocks map
;
935 struct buffer_head
*bh
;
936 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
939 J_ASSERT(handle
!= NULL
|| create
== 0);
943 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
946 return create
? ERR_PTR(-ENOSPC
) : NULL
;
950 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
952 return ERR_PTR(-ENOMEM
);
953 if (map
.m_flags
& EXT4_MAP_NEW
) {
954 J_ASSERT(create
!= 0);
955 J_ASSERT(handle
!= NULL
);
958 * Now that we do not always journal data, we should
959 * keep in mind whether this should always journal the
960 * new buffer as metadata. For now, regular file
961 * writes use ext4_get_block instead, so it's not a
965 BUFFER_TRACE(bh
, "call get_create_access");
966 err
= ext4_journal_get_create_access(handle
, bh
);
971 if (!buffer_uptodate(bh
)) {
972 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
973 set_buffer_uptodate(bh
);
976 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
977 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
981 BUFFER_TRACE(bh
, "not a new buffer");
988 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
989 ext4_lblk_t block
, int map_flags
)
991 struct buffer_head
*bh
;
993 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
996 if (!bh
|| buffer_uptodate(bh
))
998 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1000 if (buffer_uptodate(bh
))
1003 return ERR_PTR(-EIO
);
1006 int ext4_walk_page_buffers(handle_t
*handle
,
1007 struct buffer_head
*head
,
1011 int (*fn
)(handle_t
*handle
,
1012 struct buffer_head
*bh
))
1014 struct buffer_head
*bh
;
1015 unsigned block_start
, block_end
;
1016 unsigned blocksize
= head
->b_size
;
1018 struct buffer_head
*next
;
1020 for (bh
= head
, block_start
= 0;
1021 ret
== 0 && (bh
!= head
|| !block_start
);
1022 block_start
= block_end
, bh
= next
) {
1023 next
= bh
->b_this_page
;
1024 block_end
= block_start
+ blocksize
;
1025 if (block_end
<= from
|| block_start
>= to
) {
1026 if (partial
&& !buffer_uptodate(bh
))
1030 err
= (*fn
)(handle
, bh
);
1038 * To preserve ordering, it is essential that the hole instantiation and
1039 * the data write be encapsulated in a single transaction. We cannot
1040 * close off a transaction and start a new one between the ext4_get_block()
1041 * and the commit_write(). So doing the jbd2_journal_start at the start of
1042 * prepare_write() is the right place.
1044 * Also, this function can nest inside ext4_writepage(). In that case, we
1045 * *know* that ext4_writepage() has generated enough buffer credits to do the
1046 * whole page. So we won't block on the journal in that case, which is good,
1047 * because the caller may be PF_MEMALLOC.
1049 * By accident, ext4 can be reentered when a transaction is open via
1050 * quota file writes. If we were to commit the transaction while thus
1051 * reentered, there can be a deadlock - we would be holding a quota
1052 * lock, and the commit would never complete if another thread had a
1053 * transaction open and was blocking on the quota lock - a ranking
1056 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1057 * will _not_ run commit under these circumstances because handle->h_ref
1058 * is elevated. We'll still have enough credits for the tiny quotafile
1061 int do_journal_get_write_access(handle_t
*handle
,
1062 struct buffer_head
*bh
)
1064 int dirty
= buffer_dirty(bh
);
1067 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1070 * __block_write_begin() could have dirtied some buffers. Clean
1071 * the dirty bit as jbd2_journal_get_write_access() could complain
1072 * otherwise about fs integrity issues. Setting of the dirty bit
1073 * by __block_write_begin() isn't a real problem here as we clear
1074 * the bit before releasing a page lock and thus writeback cannot
1075 * ever write the buffer.
1078 clear_buffer_dirty(bh
);
1079 BUFFER_TRACE(bh
, "get write access");
1080 ret
= ext4_journal_get_write_access(handle
, bh
);
1082 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1086 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1087 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1088 get_block_t
*get_block
)
1090 unsigned from
= pos
& (PAGE_SIZE
- 1);
1091 unsigned to
= from
+ len
;
1092 struct inode
*inode
= page
->mapping
->host
;
1093 unsigned block_start
, block_end
;
1096 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1098 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1099 bool decrypt
= false;
1101 BUG_ON(!PageLocked(page
));
1102 BUG_ON(from
> PAGE_SIZE
);
1103 BUG_ON(to
> PAGE_SIZE
);
1106 if (!page_has_buffers(page
))
1107 create_empty_buffers(page
, blocksize
, 0);
1108 head
= page_buffers(page
);
1109 bbits
= ilog2(blocksize
);
1110 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1112 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1113 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1114 block_end
= block_start
+ blocksize
;
1115 if (block_end
<= from
|| block_start
>= to
) {
1116 if (PageUptodate(page
)) {
1117 if (!buffer_uptodate(bh
))
1118 set_buffer_uptodate(bh
);
1123 clear_buffer_new(bh
);
1124 if (!buffer_mapped(bh
)) {
1125 WARN_ON(bh
->b_size
!= blocksize
);
1126 err
= get_block(inode
, block
, bh
, 1);
1129 if (buffer_new(bh
)) {
1130 unmap_underlying_metadata(bh
->b_bdev
,
1132 if (PageUptodate(page
)) {
1133 clear_buffer_new(bh
);
1134 set_buffer_uptodate(bh
);
1135 mark_buffer_dirty(bh
);
1138 if (block_end
> to
|| block_start
< from
)
1139 zero_user_segments(page
, to
, block_end
,
1144 if (PageUptodate(page
)) {
1145 if (!buffer_uptodate(bh
))
1146 set_buffer_uptodate(bh
);
1149 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1150 !buffer_unwritten(bh
) &&
1151 (block_start
< from
|| block_end
> to
)) {
1152 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1154 decrypt
= ext4_encrypted_inode(inode
) &&
1155 S_ISREG(inode
->i_mode
);
1159 * If we issued read requests, let them complete.
1161 while (wait_bh
> wait
) {
1162 wait_on_buffer(*--wait_bh
);
1163 if (!buffer_uptodate(*wait_bh
))
1167 page_zero_new_buffers(page
, from
, to
);
1169 err
= fscrypt_decrypt_page(page
);
1174 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1175 loff_t pos
, unsigned len
, unsigned flags
,
1176 struct page
**pagep
, void **fsdata
)
1178 struct inode
*inode
= mapping
->host
;
1179 int ret
, needed_blocks
;
1186 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1188 * Reserve one block more for addition to orphan list in case
1189 * we allocate blocks but write fails for some reason
1191 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1192 index
= pos
>> PAGE_SHIFT
;
1193 from
= pos
& (PAGE_SIZE
- 1);
1196 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1197 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1206 * grab_cache_page_write_begin() can take a long time if the
1207 * system is thrashing due to memory pressure, or if the page
1208 * is being written back. So grab it first before we start
1209 * the transaction handle. This also allows us to allocate
1210 * the page (if needed) without using GFP_NOFS.
1213 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1219 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1220 if (IS_ERR(handle
)) {
1222 return PTR_ERR(handle
);
1226 if (page
->mapping
!= mapping
) {
1227 /* The page got truncated from under us */
1230 ext4_journal_stop(handle
);
1233 /* In case writeback began while the page was unlocked */
1234 wait_for_stable_page(page
);
1236 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1237 if (ext4_should_dioread_nolock(inode
))
1238 ret
= ext4_block_write_begin(page
, pos
, len
,
1239 ext4_get_block_unwritten
);
1241 ret
= ext4_block_write_begin(page
, pos
, len
,
1244 if (ext4_should_dioread_nolock(inode
))
1245 ret
= __block_write_begin(page
, pos
, len
,
1246 ext4_get_block_unwritten
);
1248 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1250 if (!ret
&& ext4_should_journal_data(inode
)) {
1251 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1253 do_journal_get_write_access
);
1259 * __block_write_begin may have instantiated a few blocks
1260 * outside i_size. Trim these off again. Don't need
1261 * i_size_read because we hold i_mutex.
1263 * Add inode to orphan list in case we crash before
1266 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1267 ext4_orphan_add(handle
, inode
);
1269 ext4_journal_stop(handle
);
1270 if (pos
+ len
> inode
->i_size
) {
1271 ext4_truncate_failed_write(inode
);
1273 * If truncate failed early the inode might
1274 * still be on the orphan list; we need to
1275 * make sure the inode is removed from the
1276 * orphan list in that case.
1279 ext4_orphan_del(NULL
, inode
);
1282 if (ret
== -ENOSPC
&&
1283 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1292 /* For write_end() in data=journal mode */
1293 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1296 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1298 set_buffer_uptodate(bh
);
1299 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1300 clear_buffer_meta(bh
);
1301 clear_buffer_prio(bh
);
1306 * We need to pick up the new inode size which generic_commit_write gave us
1307 * `file' can be NULL - eg, when called from page_symlink().
1309 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1310 * buffers are managed internally.
1312 static int ext4_write_end(struct file
*file
,
1313 struct address_space
*mapping
,
1314 loff_t pos
, unsigned len
, unsigned copied
,
1315 struct page
*page
, void *fsdata
)
1317 handle_t
*handle
= ext4_journal_current_handle();
1318 struct inode
*inode
= mapping
->host
;
1319 loff_t old_size
= inode
->i_size
;
1321 int i_size_changed
= 0;
1323 trace_ext4_write_end(inode
, pos
, len
, copied
);
1324 if (ext4_has_inline_data(inode
)) {
1325 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1331 copied
= block_write_end(file
, mapping
, pos
,
1332 len
, copied
, page
, fsdata
);
1334 * it's important to update i_size while still holding page lock:
1335 * page writeout could otherwise come in and zero beyond i_size.
1337 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1342 pagecache_isize_extended(inode
, old_size
, pos
);
1344 * Don't mark the inode dirty under page lock. First, it unnecessarily
1345 * makes the holding time of page lock longer. Second, it forces lock
1346 * ordering of page lock and transaction start for journaling
1350 ext4_mark_inode_dirty(handle
, inode
);
1352 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1353 /* if we have allocated more blocks and copied
1354 * less. We will have blocks allocated outside
1355 * inode->i_size. So truncate them
1357 ext4_orphan_add(handle
, inode
);
1359 ret2
= ext4_journal_stop(handle
);
1363 if (pos
+ len
> inode
->i_size
) {
1364 ext4_truncate_failed_write(inode
);
1366 * If truncate failed early the inode might still be
1367 * on the orphan list; we need to make sure the inode
1368 * is removed from the orphan list in that case.
1371 ext4_orphan_del(NULL
, inode
);
1374 return ret
? ret
: copied
;
1378 * This is a private version of page_zero_new_buffers() which doesn't
1379 * set the buffer to be dirty, since in data=journalled mode we need
1380 * to call ext4_handle_dirty_metadata() instead.
1382 static void zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1384 unsigned int block_start
= 0, block_end
;
1385 struct buffer_head
*head
, *bh
;
1387 bh
= head
= page_buffers(page
);
1389 block_end
= block_start
+ bh
->b_size
;
1390 if (buffer_new(bh
)) {
1391 if (block_end
> from
&& block_start
< to
) {
1392 if (!PageUptodate(page
)) {
1393 unsigned start
, size
;
1395 start
= max(from
, block_start
);
1396 size
= min(to
, block_end
) - start
;
1398 zero_user(page
, start
, size
);
1399 set_buffer_uptodate(bh
);
1401 clear_buffer_new(bh
);
1404 block_start
= block_end
;
1405 bh
= bh
->b_this_page
;
1406 } while (bh
!= head
);
1409 static int ext4_journalled_write_end(struct file
*file
,
1410 struct address_space
*mapping
,
1411 loff_t pos
, unsigned len
, unsigned copied
,
1412 struct page
*page
, void *fsdata
)
1414 handle_t
*handle
= ext4_journal_current_handle();
1415 struct inode
*inode
= mapping
->host
;
1416 loff_t old_size
= inode
->i_size
;
1420 int size_changed
= 0;
1422 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1423 from
= pos
& (PAGE_SIZE
- 1);
1426 BUG_ON(!ext4_handle_valid(handle
));
1428 if (ext4_has_inline_data(inode
))
1429 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1433 if (!PageUptodate(page
))
1435 zero_new_buffers(page
, from
+copied
, to
);
1438 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1439 to
, &partial
, write_end_fn
);
1441 SetPageUptodate(page
);
1443 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1444 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1445 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1450 pagecache_isize_extended(inode
, old_size
, pos
);
1453 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1458 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1459 /* if we have allocated more blocks and copied
1460 * less. We will have blocks allocated outside
1461 * inode->i_size. So truncate them
1463 ext4_orphan_add(handle
, inode
);
1465 ret2
= ext4_journal_stop(handle
);
1468 if (pos
+ len
> inode
->i_size
) {
1469 ext4_truncate_failed_write(inode
);
1471 * If truncate failed early the inode might still be
1472 * on the orphan list; we need to make sure the inode
1473 * is removed from the orphan list in that case.
1476 ext4_orphan_del(NULL
, inode
);
1479 return ret
? ret
: copied
;
1483 * Reserve space for a single cluster
1485 static int ext4_da_reserve_space(struct inode
*inode
)
1487 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1488 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1492 * We will charge metadata quota at writeout time; this saves
1493 * us from metadata over-estimation, though we may go over by
1494 * a small amount in the end. Here we just reserve for data.
1496 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1500 spin_lock(&ei
->i_block_reservation_lock
);
1501 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1502 spin_unlock(&ei
->i_block_reservation_lock
);
1503 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1506 ei
->i_reserved_data_blocks
++;
1507 trace_ext4_da_reserve_space(inode
);
1508 spin_unlock(&ei
->i_block_reservation_lock
);
1510 return 0; /* success */
1513 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1515 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1516 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1519 return; /* Nothing to release, exit */
1521 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1523 trace_ext4_da_release_space(inode
, to_free
);
1524 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1526 * if there aren't enough reserved blocks, then the
1527 * counter is messed up somewhere. Since this
1528 * function is called from invalidate page, it's
1529 * harmless to return without any action.
1531 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1532 "ino %lu, to_free %d with only %d reserved "
1533 "data blocks", inode
->i_ino
, to_free
,
1534 ei
->i_reserved_data_blocks
);
1536 to_free
= ei
->i_reserved_data_blocks
;
1538 ei
->i_reserved_data_blocks
-= to_free
;
1540 /* update fs dirty data blocks counter */
1541 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1543 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1545 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1548 static void ext4_da_page_release_reservation(struct page
*page
,
1549 unsigned int offset
,
1550 unsigned int length
)
1552 int to_release
= 0, contiguous_blks
= 0;
1553 struct buffer_head
*head
, *bh
;
1554 unsigned int curr_off
= 0;
1555 struct inode
*inode
= page
->mapping
->host
;
1556 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1557 unsigned int stop
= offset
+ length
;
1561 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1563 head
= page_buffers(page
);
1566 unsigned int next_off
= curr_off
+ bh
->b_size
;
1568 if (next_off
> stop
)
1571 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1574 clear_buffer_delay(bh
);
1575 } else if (contiguous_blks
) {
1576 lblk
= page
->index
<<
1577 (PAGE_SHIFT
- inode
->i_blkbits
);
1578 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1580 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1581 contiguous_blks
= 0;
1583 curr_off
= next_off
;
1584 } while ((bh
= bh
->b_this_page
) != head
);
1586 if (contiguous_blks
) {
1587 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1588 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1589 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1592 /* If we have released all the blocks belonging to a cluster, then we
1593 * need to release the reserved space for that cluster. */
1594 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1595 while (num_clusters
> 0) {
1596 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1597 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1598 if (sbi
->s_cluster_ratio
== 1 ||
1599 !ext4_find_delalloc_cluster(inode
, lblk
))
1600 ext4_da_release_space(inode
, 1);
1607 * Delayed allocation stuff
1610 struct mpage_da_data
{
1611 struct inode
*inode
;
1612 struct writeback_control
*wbc
;
1614 pgoff_t first_page
; /* The first page to write */
1615 pgoff_t next_page
; /* Current page to examine */
1616 pgoff_t last_page
; /* Last page to examine */
1618 * Extent to map - this can be after first_page because that can be
1619 * fully mapped. We somewhat abuse m_flags to store whether the extent
1620 * is delalloc or unwritten.
1622 struct ext4_map_blocks map
;
1623 struct ext4_io_submit io_submit
; /* IO submission data */
1626 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1631 struct pagevec pvec
;
1632 struct inode
*inode
= mpd
->inode
;
1633 struct address_space
*mapping
= inode
->i_mapping
;
1635 /* This is necessary when next_page == 0. */
1636 if (mpd
->first_page
>= mpd
->next_page
)
1639 index
= mpd
->first_page
;
1640 end
= mpd
->next_page
- 1;
1642 ext4_lblk_t start
, last
;
1643 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1644 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1645 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1648 pagevec_init(&pvec
, 0);
1649 while (index
<= end
) {
1650 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1653 for (i
= 0; i
< nr_pages
; i
++) {
1654 struct page
*page
= pvec
.pages
[i
];
1655 if (page
->index
> end
)
1657 BUG_ON(!PageLocked(page
));
1658 BUG_ON(PageWriteback(page
));
1660 if (page_mapped(page
))
1661 clear_page_dirty_for_io(page
);
1662 block_invalidatepage(page
, 0, PAGE_SIZE
);
1663 ClearPageUptodate(page
);
1667 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1668 pagevec_release(&pvec
);
1672 static void ext4_print_free_blocks(struct inode
*inode
)
1674 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1675 struct super_block
*sb
= inode
->i_sb
;
1676 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1678 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1679 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1680 ext4_count_free_clusters(sb
)));
1681 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1682 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1683 (long long) EXT4_C2B(EXT4_SB(sb
),
1684 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1685 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1686 (long long) EXT4_C2B(EXT4_SB(sb
),
1687 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1688 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1689 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1690 ei
->i_reserved_data_blocks
);
1694 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1696 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1700 * This function is grabs code from the very beginning of
1701 * ext4_map_blocks, but assumes that the caller is from delayed write
1702 * time. This function looks up the requested blocks and sets the
1703 * buffer delay bit under the protection of i_data_sem.
1705 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1706 struct ext4_map_blocks
*map
,
1707 struct buffer_head
*bh
)
1709 struct extent_status es
;
1711 sector_t invalid_block
= ~((sector_t
) 0xffff);
1712 #ifdef ES_AGGRESSIVE_TEST
1713 struct ext4_map_blocks orig_map
;
1715 memcpy(&orig_map
, map
, sizeof(*map
));
1718 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1722 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1723 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1724 (unsigned long) map
->m_lblk
);
1726 /* Lookup extent status tree firstly */
1727 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1728 if (ext4_es_is_hole(&es
)) {
1730 down_read(&EXT4_I(inode
)->i_data_sem
);
1735 * Delayed extent could be allocated by fallocate.
1736 * So we need to check it.
1738 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1739 map_bh(bh
, inode
->i_sb
, invalid_block
);
1741 set_buffer_delay(bh
);
1745 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1746 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1747 if (retval
> map
->m_len
)
1748 retval
= map
->m_len
;
1749 map
->m_len
= retval
;
1750 if (ext4_es_is_written(&es
))
1751 map
->m_flags
|= EXT4_MAP_MAPPED
;
1752 else if (ext4_es_is_unwritten(&es
))
1753 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1757 #ifdef ES_AGGRESSIVE_TEST
1758 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1764 * Try to see if we can get the block without requesting a new
1765 * file system block.
1767 down_read(&EXT4_I(inode
)->i_data_sem
);
1768 if (ext4_has_inline_data(inode
))
1770 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1771 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1773 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1779 * XXX: __block_prepare_write() unmaps passed block,
1783 * If the block was allocated from previously allocated cluster,
1784 * then we don't need to reserve it again. However we still need
1785 * to reserve metadata for every block we're going to write.
1787 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1788 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1789 ret
= ext4_da_reserve_space(inode
);
1791 /* not enough space to reserve */
1797 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1798 ~0, EXTENT_STATUS_DELAYED
);
1804 map_bh(bh
, inode
->i_sb
, invalid_block
);
1806 set_buffer_delay(bh
);
1807 } else if (retval
> 0) {
1809 unsigned int status
;
1811 if (unlikely(retval
!= map
->m_len
)) {
1812 ext4_warning(inode
->i_sb
,
1813 "ES len assertion failed for inode "
1814 "%lu: retval %d != map->m_len %d",
1815 inode
->i_ino
, retval
, map
->m_len
);
1819 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1820 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1821 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1822 map
->m_pblk
, status
);
1828 up_read((&EXT4_I(inode
)->i_data_sem
));
1834 * This is a special get_block_t callback which is used by
1835 * ext4_da_write_begin(). It will either return mapped block or
1836 * reserve space for a single block.
1838 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1839 * We also have b_blocknr = -1 and b_bdev initialized properly
1841 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1842 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1843 * initialized properly.
1845 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1846 struct buffer_head
*bh
, int create
)
1848 struct ext4_map_blocks map
;
1851 BUG_ON(create
== 0);
1852 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1854 map
.m_lblk
= iblock
;
1858 * first, we need to know whether the block is allocated already
1859 * preallocated blocks are unmapped but should treated
1860 * the same as allocated blocks.
1862 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1866 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1867 ext4_update_bh_state(bh
, map
.m_flags
);
1869 if (buffer_unwritten(bh
)) {
1870 /* A delayed write to unwritten bh should be marked
1871 * new and mapped. Mapped ensures that we don't do
1872 * get_block multiple times when we write to the same
1873 * offset and new ensures that we do proper zero out
1874 * for partial write.
1877 set_buffer_mapped(bh
);
1882 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1888 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1894 static int __ext4_journalled_writepage(struct page
*page
,
1897 struct address_space
*mapping
= page
->mapping
;
1898 struct inode
*inode
= mapping
->host
;
1899 struct buffer_head
*page_bufs
= NULL
;
1900 handle_t
*handle
= NULL
;
1901 int ret
= 0, err
= 0;
1902 int inline_data
= ext4_has_inline_data(inode
);
1903 struct buffer_head
*inode_bh
= NULL
;
1905 ClearPageChecked(page
);
1908 BUG_ON(page
->index
!= 0);
1909 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1910 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1911 if (inode_bh
== NULL
)
1914 page_bufs
= page_buffers(page
);
1919 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1923 * We need to release the page lock before we start the
1924 * journal, so grab a reference so the page won't disappear
1925 * out from under us.
1930 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1931 ext4_writepage_trans_blocks(inode
));
1932 if (IS_ERR(handle
)) {
1933 ret
= PTR_ERR(handle
);
1935 goto out_no_pagelock
;
1937 BUG_ON(!ext4_handle_valid(handle
));
1941 if (page
->mapping
!= mapping
) {
1942 /* The page got truncated from under us */
1943 ext4_journal_stop(handle
);
1949 BUFFER_TRACE(inode_bh
, "get write access");
1950 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1952 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1955 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1956 do_journal_get_write_access
);
1958 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1963 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1964 err
= ext4_journal_stop(handle
);
1968 if (!ext4_has_inline_data(inode
))
1969 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1971 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1980 * Note that we don't need to start a transaction unless we're journaling data
1981 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1982 * need to file the inode to the transaction's list in ordered mode because if
1983 * we are writing back data added by write(), the inode is already there and if
1984 * we are writing back data modified via mmap(), no one guarantees in which
1985 * transaction the data will hit the disk. In case we are journaling data, we
1986 * cannot start transaction directly because transaction start ranks above page
1987 * lock so we have to do some magic.
1989 * This function can get called via...
1990 * - ext4_writepages after taking page lock (have journal handle)
1991 * - journal_submit_inode_data_buffers (no journal handle)
1992 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1993 * - grab_page_cache when doing write_begin (have journal handle)
1995 * We don't do any block allocation in this function. If we have page with
1996 * multiple blocks we need to write those buffer_heads that are mapped. This
1997 * is important for mmaped based write. So if we do with blocksize 1K
1998 * truncate(f, 1024);
1999 * a = mmap(f, 0, 4096);
2001 * truncate(f, 4096);
2002 * we have in the page first buffer_head mapped via page_mkwrite call back
2003 * but other buffer_heads would be unmapped but dirty (dirty done via the
2004 * do_wp_page). So writepage should write the first block. If we modify
2005 * the mmap area beyond 1024 we will again get a page_fault and the
2006 * page_mkwrite callback will do the block allocation and mark the
2007 * buffer_heads mapped.
2009 * We redirty the page if we have any buffer_heads that is either delay or
2010 * unwritten in the page.
2012 * We can get recursively called as show below.
2014 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2017 * But since we don't do any block allocation we should not deadlock.
2018 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2020 static int ext4_writepage(struct page
*page
,
2021 struct writeback_control
*wbc
)
2026 struct buffer_head
*page_bufs
= NULL
;
2027 struct inode
*inode
= page
->mapping
->host
;
2028 struct ext4_io_submit io_submit
;
2029 bool keep_towrite
= false;
2031 trace_ext4_writepage(page
);
2032 size
= i_size_read(inode
);
2033 if (page
->index
== size
>> PAGE_SHIFT
)
2034 len
= size
& ~PAGE_MASK
;
2038 page_bufs
= page_buffers(page
);
2040 * We cannot do block allocation or other extent handling in this
2041 * function. If there are buffers needing that, we have to redirty
2042 * the page. But we may reach here when we do a journal commit via
2043 * journal_submit_inode_data_buffers() and in that case we must write
2044 * allocated buffers to achieve data=ordered mode guarantees.
2046 * Also, if there is only one buffer per page (the fs block
2047 * size == the page size), if one buffer needs block
2048 * allocation or needs to modify the extent tree to clear the
2049 * unwritten flag, we know that the page can't be written at
2050 * all, so we might as well refuse the write immediately.
2051 * Unfortunately if the block size != page size, we can't as
2052 * easily detect this case using ext4_walk_page_buffers(), but
2053 * for the extremely common case, this is an optimization that
2054 * skips a useless round trip through ext4_bio_write_page().
2056 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2057 ext4_bh_delay_or_unwritten
)) {
2058 redirty_page_for_writepage(wbc
, page
);
2059 if ((current
->flags
& PF_MEMALLOC
) ||
2060 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2062 * For memory cleaning there's no point in writing only
2063 * some buffers. So just bail out. Warn if we came here
2064 * from direct reclaim.
2066 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2071 keep_towrite
= true;
2074 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2076 * It's mmapped pagecache. Add buffers and journal it. There
2077 * doesn't seem much point in redirtying the page here.
2079 return __ext4_journalled_writepage(page
, len
);
2081 ext4_io_submit_init(&io_submit
, wbc
);
2082 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2083 if (!io_submit
.io_end
) {
2084 redirty_page_for_writepage(wbc
, page
);
2088 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2089 ext4_io_submit(&io_submit
);
2090 /* Drop io_end reference we got from init */
2091 ext4_put_io_end_defer(io_submit
.io_end
);
2095 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2098 loff_t size
= i_size_read(mpd
->inode
);
2101 BUG_ON(page
->index
!= mpd
->first_page
);
2102 if (page
->index
== size
>> PAGE_SHIFT
)
2103 len
= size
& ~PAGE_MASK
;
2106 clear_page_dirty_for_io(page
);
2107 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2109 mpd
->wbc
->nr_to_write
--;
2115 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2118 * mballoc gives us at most this number of blocks...
2119 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2120 * The rest of mballoc seems to handle chunks up to full group size.
2122 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2125 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2127 * @mpd - extent of blocks
2128 * @lblk - logical number of the block in the file
2129 * @bh - buffer head we want to add to the extent
2131 * The function is used to collect contig. blocks in the same state. If the
2132 * buffer doesn't require mapping for writeback and we haven't started the
2133 * extent of buffers to map yet, the function returns 'true' immediately - the
2134 * caller can write the buffer right away. Otherwise the function returns true
2135 * if the block has been added to the extent, false if the block couldn't be
2138 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2139 struct buffer_head
*bh
)
2141 struct ext4_map_blocks
*map
= &mpd
->map
;
2143 /* Buffer that doesn't need mapping for writeback? */
2144 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2145 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2146 /* So far no extent to map => we write the buffer right away */
2147 if (map
->m_len
== 0)
2152 /* First block in the extent? */
2153 if (map
->m_len
== 0) {
2156 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2160 /* Don't go larger than mballoc is willing to allocate */
2161 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2164 /* Can we merge the block to our big extent? */
2165 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2166 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2174 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2176 * @mpd - extent of blocks for mapping
2177 * @head - the first buffer in the page
2178 * @bh - buffer we should start processing from
2179 * @lblk - logical number of the block in the file corresponding to @bh
2181 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2182 * the page for IO if all buffers in this page were mapped and there's no
2183 * accumulated extent of buffers to map or add buffers in the page to the
2184 * extent of buffers to map. The function returns 1 if the caller can continue
2185 * by processing the next page, 0 if it should stop adding buffers to the
2186 * extent to map because we cannot extend it anymore. It can also return value
2187 * < 0 in case of error during IO submission.
2189 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2190 struct buffer_head
*head
,
2191 struct buffer_head
*bh
,
2194 struct inode
*inode
= mpd
->inode
;
2196 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2197 >> inode
->i_blkbits
;
2200 BUG_ON(buffer_locked(bh
));
2202 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2203 /* Found extent to map? */
2206 /* Everything mapped so far and we hit EOF */
2209 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2210 /* So far everything mapped? Submit the page for IO. */
2211 if (mpd
->map
.m_len
== 0) {
2212 err
= mpage_submit_page(mpd
, head
->b_page
);
2216 return lblk
< blocks
;
2220 * mpage_map_buffers - update buffers corresponding to changed extent and
2221 * submit fully mapped pages for IO
2223 * @mpd - description of extent to map, on return next extent to map
2225 * Scan buffers corresponding to changed extent (we expect corresponding pages
2226 * to be already locked) and update buffer state according to new extent state.
2227 * We map delalloc buffers to their physical location, clear unwritten bits,
2228 * and mark buffers as uninit when we perform writes to unwritten extents
2229 * and do extent conversion after IO is finished. If the last page is not fully
2230 * mapped, we update @map to the next extent in the last page that needs
2231 * mapping. Otherwise we submit the page for IO.
2233 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2235 struct pagevec pvec
;
2237 struct inode
*inode
= mpd
->inode
;
2238 struct buffer_head
*head
, *bh
;
2239 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2245 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2246 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2247 lblk
= start
<< bpp_bits
;
2248 pblock
= mpd
->map
.m_pblk
;
2250 pagevec_init(&pvec
, 0);
2251 while (start
<= end
) {
2252 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2256 for (i
= 0; i
< nr_pages
; i
++) {
2257 struct page
*page
= pvec
.pages
[i
];
2259 if (page
->index
> end
)
2261 /* Up to 'end' pages must be contiguous */
2262 BUG_ON(page
->index
!= start
);
2263 bh
= head
= page_buffers(page
);
2265 if (lblk
< mpd
->map
.m_lblk
)
2267 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2269 * Buffer after end of mapped extent.
2270 * Find next buffer in the page to map.
2273 mpd
->map
.m_flags
= 0;
2275 * FIXME: If dioread_nolock supports
2276 * blocksize < pagesize, we need to make
2277 * sure we add size mapped so far to
2278 * io_end->size as the following call
2279 * can submit the page for IO.
2281 err
= mpage_process_page_bufs(mpd
, head
,
2283 pagevec_release(&pvec
);
2288 if (buffer_delay(bh
)) {
2289 clear_buffer_delay(bh
);
2290 bh
->b_blocknr
= pblock
++;
2292 clear_buffer_unwritten(bh
);
2293 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2296 * FIXME: This is going to break if dioread_nolock
2297 * supports blocksize < pagesize as we will try to
2298 * convert potentially unmapped parts of inode.
2300 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2301 /* Page fully mapped - let IO run! */
2302 err
= mpage_submit_page(mpd
, page
);
2304 pagevec_release(&pvec
);
2309 pagevec_release(&pvec
);
2311 /* Extent fully mapped and matches with page boundary. We are done. */
2313 mpd
->map
.m_flags
= 0;
2317 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2319 struct inode
*inode
= mpd
->inode
;
2320 struct ext4_map_blocks
*map
= &mpd
->map
;
2321 int get_blocks_flags
;
2322 int err
, dioread_nolock
;
2324 trace_ext4_da_write_pages_extent(inode
, map
);
2326 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2327 * to convert an unwritten extent to be initialized (in the case
2328 * where we have written into one or more preallocated blocks). It is
2329 * possible that we're going to need more metadata blocks than
2330 * previously reserved. However we must not fail because we're in
2331 * writeback and there is nothing we can do about it so it might result
2332 * in data loss. So use reserved blocks to allocate metadata if
2335 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2336 * the blocks in question are delalloc blocks. This indicates
2337 * that the blocks and quotas has already been checked when
2338 * the data was copied into the page cache.
2340 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2341 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2342 EXT4_GET_BLOCKS_IO_SUBMIT
;
2343 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2345 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2346 if (map
->m_flags
& (1 << BH_Delay
))
2347 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2349 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2352 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2353 if (!mpd
->io_submit
.io_end
->handle
&&
2354 ext4_handle_valid(handle
)) {
2355 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2356 handle
->h_rsv_handle
= NULL
;
2358 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2361 BUG_ON(map
->m_len
== 0);
2362 if (map
->m_flags
& EXT4_MAP_NEW
) {
2363 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2366 for (i
= 0; i
< map
->m_len
; i
++)
2367 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2373 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2374 * mpd->len and submit pages underlying it for IO
2376 * @handle - handle for journal operations
2377 * @mpd - extent to map
2378 * @give_up_on_write - we set this to true iff there is a fatal error and there
2379 * is no hope of writing the data. The caller should discard
2380 * dirty pages to avoid infinite loops.
2382 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2383 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2384 * them to initialized or split the described range from larger unwritten
2385 * extent. Note that we need not map all the described range since allocation
2386 * can return less blocks or the range is covered by more unwritten extents. We
2387 * cannot map more because we are limited by reserved transaction credits. On
2388 * the other hand we always make sure that the last touched page is fully
2389 * mapped so that it can be written out (and thus forward progress is
2390 * guaranteed). After mapping we submit all mapped pages for IO.
2392 static int mpage_map_and_submit_extent(handle_t
*handle
,
2393 struct mpage_da_data
*mpd
,
2394 bool *give_up_on_write
)
2396 struct inode
*inode
= mpd
->inode
;
2397 struct ext4_map_blocks
*map
= &mpd
->map
;
2402 mpd
->io_submit
.io_end
->offset
=
2403 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2405 err
= mpage_map_one_extent(handle
, mpd
);
2407 struct super_block
*sb
= inode
->i_sb
;
2409 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2410 goto invalidate_dirty_pages
;
2412 * Let the uper layers retry transient errors.
2413 * In the case of ENOSPC, if ext4_count_free_blocks()
2414 * is non-zero, a commit should free up blocks.
2416 if ((err
== -ENOMEM
) ||
2417 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2419 goto update_disksize
;
2422 ext4_msg(sb
, KERN_CRIT
,
2423 "Delayed block allocation failed for "
2424 "inode %lu at logical offset %llu with"
2425 " max blocks %u with error %d",
2427 (unsigned long long)map
->m_lblk
,
2428 (unsigned)map
->m_len
, -err
);
2429 ext4_msg(sb
, KERN_CRIT
,
2430 "This should not happen!! Data will "
2433 ext4_print_free_blocks(inode
);
2434 invalidate_dirty_pages
:
2435 *give_up_on_write
= true;
2440 * Update buffer state, submit mapped pages, and get us new
2443 err
= mpage_map_and_submit_buffers(mpd
);
2445 goto update_disksize
;
2446 } while (map
->m_len
);
2450 * Update on-disk size after IO is submitted. Races with
2451 * truncate are avoided by checking i_size under i_data_sem.
2453 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2454 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2458 down_write(&EXT4_I(inode
)->i_data_sem
);
2459 i_size
= i_size_read(inode
);
2460 if (disksize
> i_size
)
2462 if (disksize
> EXT4_I(inode
)->i_disksize
)
2463 EXT4_I(inode
)->i_disksize
= disksize
;
2464 err2
= ext4_mark_inode_dirty(handle
, inode
);
2465 up_write(&EXT4_I(inode
)->i_data_sem
);
2467 ext4_error(inode
->i_sb
,
2468 "Failed to mark inode %lu dirty",
2477 * Calculate the total number of credits to reserve for one writepages
2478 * iteration. This is called from ext4_writepages(). We map an extent of
2479 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2480 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2481 * bpp - 1 blocks in bpp different extents.
2483 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2485 int bpp
= ext4_journal_blocks_per_page(inode
);
2487 return ext4_meta_trans_blocks(inode
,
2488 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2492 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2493 * and underlying extent to map
2495 * @mpd - where to look for pages
2497 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2498 * IO immediately. When we find a page which isn't mapped we start accumulating
2499 * extent of buffers underlying these pages that needs mapping (formed by
2500 * either delayed or unwritten buffers). We also lock the pages containing
2501 * these buffers. The extent found is returned in @mpd structure (starting at
2502 * mpd->lblk with length mpd->len blocks).
2504 * Note that this function can attach bios to one io_end structure which are
2505 * neither logically nor physically contiguous. Although it may seem as an
2506 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2507 * case as we need to track IO to all buffers underlying a page in one io_end.
2509 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2511 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2512 struct pagevec pvec
;
2513 unsigned int nr_pages
;
2514 long left
= mpd
->wbc
->nr_to_write
;
2515 pgoff_t index
= mpd
->first_page
;
2516 pgoff_t end
= mpd
->last_page
;
2519 int blkbits
= mpd
->inode
->i_blkbits
;
2521 struct buffer_head
*head
;
2523 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2524 tag
= PAGECACHE_TAG_TOWRITE
;
2526 tag
= PAGECACHE_TAG_DIRTY
;
2528 pagevec_init(&pvec
, 0);
2530 mpd
->next_page
= index
;
2531 while (index
<= end
) {
2532 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2533 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2537 for (i
= 0; i
< nr_pages
; i
++) {
2538 struct page
*page
= pvec
.pages
[i
];
2541 * At this point, the page may be truncated or
2542 * invalidated (changing page->mapping to NULL), or
2543 * even swizzled back from swapper_space to tmpfs file
2544 * mapping. However, page->index will not change
2545 * because we have a reference on the page.
2547 if (page
->index
> end
)
2551 * Accumulated enough dirty pages? This doesn't apply
2552 * to WB_SYNC_ALL mode. For integrity sync we have to
2553 * keep going because someone may be concurrently
2554 * dirtying pages, and we might have synced a lot of
2555 * newly appeared dirty pages, but have not synced all
2556 * of the old dirty pages.
2558 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2561 /* If we can't merge this page, we are done. */
2562 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2567 * If the page is no longer dirty, or its mapping no
2568 * longer corresponds to inode we are writing (which
2569 * means it has been truncated or invalidated), or the
2570 * page is already under writeback and we are not doing
2571 * a data integrity writeback, skip the page
2573 if (!PageDirty(page
) ||
2574 (PageWriteback(page
) &&
2575 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2576 unlikely(page
->mapping
!= mapping
)) {
2581 wait_on_page_writeback(page
);
2582 BUG_ON(PageWriteback(page
));
2584 if (mpd
->map
.m_len
== 0)
2585 mpd
->first_page
= page
->index
;
2586 mpd
->next_page
= page
->index
+ 1;
2587 /* Add all dirty buffers to mpd */
2588 lblk
= ((ext4_lblk_t
)page
->index
) <<
2589 (PAGE_SHIFT
- blkbits
);
2590 head
= page_buffers(page
);
2591 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2597 pagevec_release(&pvec
);
2602 pagevec_release(&pvec
);
2606 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2609 struct address_space
*mapping
= data
;
2610 int ret
= ext4_writepage(page
, wbc
);
2611 mapping_set_error(mapping
, ret
);
2615 static int ext4_writepages(struct address_space
*mapping
,
2616 struct writeback_control
*wbc
)
2618 pgoff_t writeback_index
= 0;
2619 long nr_to_write
= wbc
->nr_to_write
;
2620 int range_whole
= 0;
2622 handle_t
*handle
= NULL
;
2623 struct mpage_da_data mpd
;
2624 struct inode
*inode
= mapping
->host
;
2625 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2626 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2628 struct blk_plug plug
;
2629 bool give_up_on_write
= false;
2631 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2632 trace_ext4_writepages(inode
, wbc
);
2634 if (dax_mapping(mapping
)) {
2635 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2637 goto out_writepages
;
2641 * No pages to write? This is mainly a kludge to avoid starting
2642 * a transaction for special inodes like journal inode on last iput()
2643 * because that could violate lock ordering on umount
2645 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2646 goto out_writepages
;
2648 if (ext4_should_journal_data(inode
)) {
2649 struct blk_plug plug
;
2651 blk_start_plug(&plug
);
2652 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2653 blk_finish_plug(&plug
);
2654 goto out_writepages
;
2658 * If the filesystem has aborted, it is read-only, so return
2659 * right away instead of dumping stack traces later on that
2660 * will obscure the real source of the problem. We test
2661 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2662 * the latter could be true if the filesystem is mounted
2663 * read-only, and in that case, ext4_writepages should
2664 * *never* be called, so if that ever happens, we would want
2667 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2669 goto out_writepages
;
2672 if (ext4_should_dioread_nolock(inode
)) {
2674 * We may need to convert up to one extent per block in
2675 * the page and we may dirty the inode.
2677 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2681 * If we have inline data and arrive here, it means that
2682 * we will soon create the block for the 1st page, so
2683 * we'd better clear the inline data here.
2685 if (ext4_has_inline_data(inode
)) {
2686 /* Just inode will be modified... */
2687 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2688 if (IS_ERR(handle
)) {
2689 ret
= PTR_ERR(handle
);
2690 goto out_writepages
;
2692 BUG_ON(ext4_test_inode_state(inode
,
2693 EXT4_STATE_MAY_INLINE_DATA
));
2694 ext4_destroy_inline_data(handle
, inode
);
2695 ext4_journal_stop(handle
);
2698 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2701 if (wbc
->range_cyclic
) {
2702 writeback_index
= mapping
->writeback_index
;
2703 if (writeback_index
)
2705 mpd
.first_page
= writeback_index
;
2708 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2709 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2714 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2716 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2717 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2719 blk_start_plug(&plug
);
2720 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2721 /* For each extent of pages we use new io_end */
2722 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2723 if (!mpd
.io_submit
.io_end
) {
2729 * We have two constraints: We find one extent to map and we
2730 * must always write out whole page (makes a difference when
2731 * blocksize < pagesize) so that we don't block on IO when we
2732 * try to write out the rest of the page. Journalled mode is
2733 * not supported by delalloc.
2735 BUG_ON(ext4_should_journal_data(inode
));
2736 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2738 /* start a new transaction */
2739 handle
= ext4_journal_start_with_reserve(inode
,
2740 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2741 if (IS_ERR(handle
)) {
2742 ret
= PTR_ERR(handle
);
2743 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2744 "%ld pages, ino %lu; err %d", __func__
,
2745 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2746 /* Release allocated io_end */
2747 ext4_put_io_end(mpd
.io_submit
.io_end
);
2751 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2752 ret
= mpage_prepare_extent_to_map(&mpd
);
2755 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2759 * We scanned the whole range (or exhausted
2760 * nr_to_write), submitted what was mapped and
2761 * didn't find anything needing mapping. We are
2768 * Caution: If the handle is synchronous,
2769 * ext4_journal_stop() can wait for transaction commit
2770 * to finish which may depend on writeback of pages to
2771 * complete or on page lock to be released. In that
2772 * case, we have to wait until after after we have
2773 * submitted all the IO, released page locks we hold,
2774 * and dropped io_end reference (for extent conversion
2775 * to be able to complete) before stopping the handle.
2777 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2778 ext4_journal_stop(handle
);
2781 /* Submit prepared bio */
2782 ext4_io_submit(&mpd
.io_submit
);
2783 /* Unlock pages we didn't use */
2784 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2786 * Drop our io_end reference we got from init. We have
2787 * to be careful and use deferred io_end finishing if
2788 * we are still holding the transaction as we can
2789 * release the last reference to io_end which may end
2790 * up doing unwritten extent conversion.
2793 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2794 ext4_journal_stop(handle
);
2796 ext4_put_io_end(mpd
.io_submit
.io_end
);
2798 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2800 * Commit the transaction which would
2801 * free blocks released in the transaction
2804 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2808 /* Fatal error - ENOMEM, EIO... */
2812 blk_finish_plug(&plug
);
2813 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2815 mpd
.last_page
= writeback_index
- 1;
2821 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2823 * Set the writeback_index so that range_cyclic
2824 * mode will write it back later
2826 mapping
->writeback_index
= mpd
.first_page
;
2829 trace_ext4_writepages_result(inode
, wbc
, ret
,
2830 nr_to_write
- wbc
->nr_to_write
);
2831 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2835 static int ext4_nonda_switch(struct super_block
*sb
)
2837 s64 free_clusters
, dirty_clusters
;
2838 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2841 * switch to non delalloc mode if we are running low
2842 * on free block. The free block accounting via percpu
2843 * counters can get slightly wrong with percpu_counter_batch getting
2844 * accumulated on each CPU without updating global counters
2845 * Delalloc need an accurate free block accounting. So switch
2846 * to non delalloc when we are near to error range.
2849 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2851 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2853 * Start pushing delalloc when 1/2 of free blocks are dirty.
2855 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2856 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2858 if (2 * free_clusters
< 3 * dirty_clusters
||
2859 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2861 * free block count is less than 150% of dirty blocks
2862 * or free blocks is less than watermark
2869 /* We always reserve for an inode update; the superblock could be there too */
2870 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2872 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2875 if (pos
+ len
<= 0x7fffffffULL
)
2878 /* We might need to update the superblock to set LARGE_FILE */
2882 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2883 loff_t pos
, unsigned len
, unsigned flags
,
2884 struct page
**pagep
, void **fsdata
)
2886 int ret
, retries
= 0;
2889 struct inode
*inode
= mapping
->host
;
2892 index
= pos
>> PAGE_SHIFT
;
2894 if (ext4_nonda_switch(inode
->i_sb
)) {
2895 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2896 return ext4_write_begin(file
, mapping
, pos
,
2897 len
, flags
, pagep
, fsdata
);
2899 *fsdata
= (void *)0;
2900 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2902 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2903 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2913 * grab_cache_page_write_begin() can take a long time if the
2914 * system is thrashing due to memory pressure, or if the page
2915 * is being written back. So grab it first before we start
2916 * the transaction handle. This also allows us to allocate
2917 * the page (if needed) without using GFP_NOFS.
2920 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2926 * With delayed allocation, we don't log the i_disksize update
2927 * if there is delayed block allocation. But we still need
2928 * to journalling the i_disksize update if writes to the end
2929 * of file which has an already mapped buffer.
2932 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2933 ext4_da_write_credits(inode
, pos
, len
));
2934 if (IS_ERR(handle
)) {
2936 return PTR_ERR(handle
);
2940 if (page
->mapping
!= mapping
) {
2941 /* The page got truncated from under us */
2944 ext4_journal_stop(handle
);
2947 /* In case writeback began while the page was unlocked */
2948 wait_for_stable_page(page
);
2950 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2951 ret
= ext4_block_write_begin(page
, pos
, len
,
2952 ext4_da_get_block_prep
);
2954 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2958 ext4_journal_stop(handle
);
2960 * block_write_begin may have instantiated a few blocks
2961 * outside i_size. Trim these off again. Don't need
2962 * i_size_read because we hold i_mutex.
2964 if (pos
+ len
> inode
->i_size
)
2965 ext4_truncate_failed_write(inode
);
2967 if (ret
== -ENOSPC
&&
2968 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2980 * Check if we should update i_disksize
2981 * when write to the end of file but not require block allocation
2983 static int ext4_da_should_update_i_disksize(struct page
*page
,
2984 unsigned long offset
)
2986 struct buffer_head
*bh
;
2987 struct inode
*inode
= page
->mapping
->host
;
2991 bh
= page_buffers(page
);
2992 idx
= offset
>> inode
->i_blkbits
;
2994 for (i
= 0; i
< idx
; i
++)
2995 bh
= bh
->b_this_page
;
2997 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3002 static int ext4_da_write_end(struct file
*file
,
3003 struct address_space
*mapping
,
3004 loff_t pos
, unsigned len
, unsigned copied
,
3005 struct page
*page
, void *fsdata
)
3007 struct inode
*inode
= mapping
->host
;
3009 handle_t
*handle
= ext4_journal_current_handle();
3011 unsigned long start
, end
;
3012 int write_mode
= (int)(unsigned long)fsdata
;
3014 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3015 return ext4_write_end(file
, mapping
, pos
,
3016 len
, copied
, page
, fsdata
);
3018 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3019 start
= pos
& (PAGE_SIZE
- 1);
3020 end
= start
+ copied
- 1;
3023 * generic_write_end() will run mark_inode_dirty() if i_size
3024 * changes. So let's piggyback the i_disksize mark_inode_dirty
3027 new_i_size
= pos
+ copied
;
3028 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3029 if (ext4_has_inline_data(inode
) ||
3030 ext4_da_should_update_i_disksize(page
, end
)) {
3031 ext4_update_i_disksize(inode
, new_i_size
);
3032 /* We need to mark inode dirty even if
3033 * new_i_size is less that inode->i_size
3034 * bu greater than i_disksize.(hint delalloc)
3036 ext4_mark_inode_dirty(handle
, inode
);
3040 if (write_mode
!= CONVERT_INLINE_DATA
&&
3041 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3042 ext4_has_inline_data(inode
))
3043 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3046 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3052 ret2
= ext4_journal_stop(handle
);
3056 return ret
? ret
: copied
;
3059 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3060 unsigned int length
)
3063 * Drop reserved blocks
3065 BUG_ON(!PageLocked(page
));
3066 if (!page_has_buffers(page
))
3069 ext4_da_page_release_reservation(page
, offset
, length
);
3072 ext4_invalidatepage(page
, offset
, length
);
3078 * Force all delayed allocation blocks to be allocated for a given inode.
3080 int ext4_alloc_da_blocks(struct inode
*inode
)
3082 trace_ext4_alloc_da_blocks(inode
);
3084 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3088 * We do something simple for now. The filemap_flush() will
3089 * also start triggering a write of the data blocks, which is
3090 * not strictly speaking necessary (and for users of
3091 * laptop_mode, not even desirable). However, to do otherwise
3092 * would require replicating code paths in:
3094 * ext4_writepages() ->
3095 * write_cache_pages() ---> (via passed in callback function)
3096 * __mpage_da_writepage() -->
3097 * mpage_add_bh_to_extent()
3098 * mpage_da_map_blocks()
3100 * The problem is that write_cache_pages(), located in
3101 * mm/page-writeback.c, marks pages clean in preparation for
3102 * doing I/O, which is not desirable if we're not planning on
3105 * We could call write_cache_pages(), and then redirty all of
3106 * the pages by calling redirty_page_for_writepage() but that
3107 * would be ugly in the extreme. So instead we would need to
3108 * replicate parts of the code in the above functions,
3109 * simplifying them because we wouldn't actually intend to
3110 * write out the pages, but rather only collect contiguous
3111 * logical block extents, call the multi-block allocator, and
3112 * then update the buffer heads with the block allocations.
3114 * For now, though, we'll cheat by calling filemap_flush(),
3115 * which will map the blocks, and start the I/O, but not
3116 * actually wait for the I/O to complete.
3118 return filemap_flush(inode
->i_mapping
);
3122 * bmap() is special. It gets used by applications such as lilo and by
3123 * the swapper to find the on-disk block of a specific piece of data.
3125 * Naturally, this is dangerous if the block concerned is still in the
3126 * journal. If somebody makes a swapfile on an ext4 data-journaling
3127 * filesystem and enables swap, then they may get a nasty shock when the
3128 * data getting swapped to that swapfile suddenly gets overwritten by
3129 * the original zero's written out previously to the journal and
3130 * awaiting writeback in the kernel's buffer cache.
3132 * So, if we see any bmap calls here on a modified, data-journaled file,
3133 * take extra steps to flush any blocks which might be in the cache.
3135 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3137 struct inode
*inode
= mapping
->host
;
3142 * We can get here for an inline file via the FIBMAP ioctl
3144 if (ext4_has_inline_data(inode
))
3147 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3148 test_opt(inode
->i_sb
, DELALLOC
)) {
3150 * With delalloc we want to sync the file
3151 * so that we can make sure we allocate
3154 filemap_write_and_wait(mapping
);
3157 if (EXT4_JOURNAL(inode
) &&
3158 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3160 * This is a REALLY heavyweight approach, but the use of
3161 * bmap on dirty files is expected to be extremely rare:
3162 * only if we run lilo or swapon on a freshly made file
3163 * do we expect this to happen.
3165 * (bmap requires CAP_SYS_RAWIO so this does not
3166 * represent an unprivileged user DOS attack --- we'd be
3167 * in trouble if mortal users could trigger this path at
3170 * NB. EXT4_STATE_JDATA is not set on files other than
3171 * regular files. If somebody wants to bmap a directory
3172 * or symlink and gets confused because the buffer
3173 * hasn't yet been flushed to disk, they deserve
3174 * everything they get.
3177 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3178 journal
= EXT4_JOURNAL(inode
);
3179 jbd2_journal_lock_updates(journal
);
3180 err
= jbd2_journal_flush(journal
);
3181 jbd2_journal_unlock_updates(journal
);
3187 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3190 static int ext4_readpage(struct file
*file
, struct page
*page
)
3193 struct inode
*inode
= page
->mapping
->host
;
3195 trace_ext4_readpage(page
);
3197 if (ext4_has_inline_data(inode
))
3198 ret
= ext4_readpage_inline(inode
, page
);
3201 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3207 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3208 struct list_head
*pages
, unsigned nr_pages
)
3210 struct inode
*inode
= mapping
->host
;
3212 /* If the file has inline data, no need to do readpages. */
3213 if (ext4_has_inline_data(inode
))
3216 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3219 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3220 unsigned int length
)
3222 trace_ext4_invalidatepage(page
, offset
, length
);
3224 /* No journalling happens on data buffers when this function is used */
3225 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3227 block_invalidatepage(page
, offset
, length
);
3230 static int __ext4_journalled_invalidatepage(struct page
*page
,
3231 unsigned int offset
,
3232 unsigned int length
)
3234 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3236 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3239 * If it's a full truncate we just forget about the pending dirtying
3241 if (offset
== 0 && length
== PAGE_SIZE
)
3242 ClearPageChecked(page
);
3244 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3247 /* Wrapper for aops... */
3248 static void ext4_journalled_invalidatepage(struct page
*page
,
3249 unsigned int offset
,
3250 unsigned int length
)
3252 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3255 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3257 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3259 trace_ext4_releasepage(page
);
3261 /* Page has dirty journalled data -> cannot release */
3262 if (PageChecked(page
))
3265 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3267 return try_to_free_buffers(page
);
3270 #ifdef CONFIG_FS_DAX
3272 * Get block function for DAX IO and mmap faults. It takes care of converting
3273 * unwritten extents to written ones and initializes new / converted blocks
3276 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3277 struct buffer_head
*bh_result
, int create
)
3281 ext4_debug("inode %lu, create flag %d\n", inode
->i_ino
, create
);
3283 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
3285 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3286 EXT4_GET_BLOCKS_PRE_IO
|
3287 EXT4_GET_BLOCKS_CREATE_ZERO
);
3291 if (buffer_unwritten(bh_result
)) {
3293 * We are protected by i_mmap_sem or i_mutex so we know block
3294 * cannot go away from under us even though we dropped
3295 * i_data_sem. Convert extent to written and write zeros there.
3297 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3298 EXT4_GET_BLOCKS_CONVERT
|
3299 EXT4_GET_BLOCKS_CREATE_ZERO
);
3304 * At least for now we have to clear BH_New so that DAX code
3305 * doesn't attempt to zero blocks again in a racy way.
3307 clear_buffer_new(bh_result
);
3311 /* Just define empty function, it will never get called. */
3312 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3313 struct buffer_head
*bh_result
, int create
)
3320 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3321 ssize_t size
, void *private)
3323 ext4_io_end_t
*io_end
= private;
3325 /* if not async direct IO just return */
3329 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3330 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3331 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3334 * Error during AIO DIO. We cannot convert unwritten extents as the
3335 * data was not written. Just clear the unwritten flag and drop io_end.
3338 ext4_clear_io_unwritten_flag(io_end
);
3341 io_end
->offset
= offset
;
3342 io_end
->size
= size
;
3343 ext4_put_io_end(io_end
);
3349 * Handling of direct IO writes.
3351 * For ext4 extent files, ext4 will do direct-io write even to holes,
3352 * preallocated extents, and those write extend the file, no need to
3353 * fall back to buffered IO.
3355 * For holes, we fallocate those blocks, mark them as unwritten
3356 * If those blocks were preallocated, we mark sure they are split, but
3357 * still keep the range to write as unwritten.
3359 * The unwritten extents will be converted to written when DIO is completed.
3360 * For async direct IO, since the IO may still pending when return, we
3361 * set up an end_io call back function, which will do the conversion
3362 * when async direct IO completed.
3364 * If the O_DIRECT write will extend the file then add this inode to the
3365 * orphan list. So recovery will truncate it back to the original size
3366 * if the machine crashes during the write.
3369 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3371 struct file
*file
= iocb
->ki_filp
;
3372 struct inode
*inode
= file
->f_mapping
->host
;
3373 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3375 loff_t offset
= iocb
->ki_pos
;
3376 size_t count
= iov_iter_count(iter
);
3378 get_block_t
*get_block_func
= NULL
;
3380 loff_t final_size
= offset
+ count
;
3384 if (final_size
> inode
->i_size
) {
3385 /* Credits for sb + inode write */
3386 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3387 if (IS_ERR(handle
)) {
3388 ret
= PTR_ERR(handle
);
3391 ret
= ext4_orphan_add(handle
, inode
);
3393 ext4_journal_stop(handle
);
3397 ei
->i_disksize
= inode
->i_size
;
3398 ext4_journal_stop(handle
);
3401 BUG_ON(iocb
->private == NULL
);
3404 * Make all waiters for direct IO properly wait also for extent
3405 * conversion. This also disallows race between truncate() and
3406 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3408 inode_dio_begin(inode
);
3410 /* If we do a overwrite dio, i_mutex locking can be released */
3411 overwrite
= *((int *)iocb
->private);
3414 inode_unlock(inode
);
3417 * For extent mapped files we could direct write to holes and fallocate.
3419 * Allocated blocks to fill the hole are marked as unwritten to prevent
3420 * parallel buffered read to expose the stale data before DIO complete
3423 * As to previously fallocated extents, ext4 get_block will just simply
3424 * mark the buffer mapped but still keep the extents unwritten.
3426 * For non AIO case, we will convert those unwritten extents to written
3427 * after return back from blockdev_direct_IO. That way we save us from
3428 * allocating io_end structure and also the overhead of offloading
3429 * the extent convertion to a workqueue.
3431 * For async DIO, the conversion needs to be deferred when the
3432 * IO is completed. The ext4 end_io callback function will be
3433 * called to take care of the conversion work. Here for async
3434 * case, we allocate an io_end structure to hook to the iocb.
3436 iocb
->private = NULL
;
3438 get_block_func
= ext4_dio_get_block_overwrite
;
3439 else if (IS_DAX(inode
)) {
3441 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3442 * writes need zeroing either because they can race with page
3443 * faults or because they use partial blocks.
3445 if (round_down(offset
, 1<<inode
->i_blkbits
) >= inode
->i_size
&&
3446 ext4_aligned_io(inode
, offset
, count
))
3447 get_block_func
= ext4_dio_get_block
;
3449 get_block_func
= ext4_dax_get_block
;
3450 dio_flags
= DIO_LOCKING
;
3451 } else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3452 round_down(offset
, 1 << inode
->i_blkbits
) >= inode
->i_size
) {
3453 get_block_func
= ext4_dio_get_block
;
3454 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3455 } else if (is_sync_kiocb(iocb
)) {
3456 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3457 dio_flags
= DIO_LOCKING
;
3459 get_block_func
= ext4_dio_get_block_unwritten_async
;
3460 dio_flags
= DIO_LOCKING
;
3462 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3463 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3465 if (IS_DAX(inode
)) {
3466 ret
= dax_do_io(iocb
, inode
, iter
, get_block_func
,
3467 ext4_end_io_dio
, dio_flags
);
3469 ret
= __blockdev_direct_IO(iocb
, inode
,
3470 inode
->i_sb
->s_bdev
, iter
,
3472 ext4_end_io_dio
, NULL
, dio_flags
);
3474 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3475 EXT4_STATE_DIO_UNWRITTEN
)) {
3478 * for non AIO case, since the IO is already
3479 * completed, we could do the conversion right here
3481 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3485 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3488 inode_dio_end(inode
);
3489 /* take i_mutex locking again if we do a ovewrite dio */
3493 if (ret
< 0 && final_size
> inode
->i_size
)
3494 ext4_truncate_failed_write(inode
);
3496 /* Handle extending of i_size after direct IO write */
3500 /* Credits for sb + inode write */
3501 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3502 if (IS_ERR(handle
)) {
3503 /* This is really bad luck. We've written the data
3504 * but cannot extend i_size. Bail out and pretend
3505 * the write failed... */
3506 ret
= PTR_ERR(handle
);
3508 ext4_orphan_del(NULL
, inode
);
3513 ext4_orphan_del(handle
, inode
);
3515 loff_t end
= offset
+ ret
;
3516 if (end
> inode
->i_size
) {
3517 ei
->i_disksize
= end
;
3518 i_size_write(inode
, end
);
3520 * We're going to return a positive `ret'
3521 * here due to non-zero-length I/O, so there's
3522 * no way of reporting error returns from
3523 * ext4_mark_inode_dirty() to userspace. So
3526 ext4_mark_inode_dirty(handle
, inode
);
3529 err
= ext4_journal_stop(handle
);
3537 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3539 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3540 struct inode
*inode
= mapping
->host
;
3544 * Shared inode_lock is enough for us - it protects against concurrent
3545 * writes & truncates and since we take care of writing back page cache,
3546 * we are protected against page writeback as well.
3548 inode_lock_shared(inode
);
3549 if (IS_DAX(inode
)) {
3550 ret
= dax_do_io(iocb
, inode
, iter
, ext4_dio_get_block
, NULL
, 0);
3552 size_t count
= iov_iter_count(iter
);
3554 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3555 iocb
->ki_pos
+ count
);
3558 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3559 iter
, ext4_dio_get_block
,
3563 inode_unlock_shared(inode
);
3567 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3569 struct file
*file
= iocb
->ki_filp
;
3570 struct inode
*inode
= file
->f_mapping
->host
;
3571 size_t count
= iov_iter_count(iter
);
3572 loff_t offset
= iocb
->ki_pos
;
3575 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3576 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3581 * If we are doing data journalling we don't support O_DIRECT
3583 if (ext4_should_journal_data(inode
))
3586 /* Let buffer I/O handle the inline data case. */
3587 if (ext4_has_inline_data(inode
))
3590 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3591 if (iov_iter_rw(iter
) == READ
)
3592 ret
= ext4_direct_IO_read(iocb
, iter
);
3594 ret
= ext4_direct_IO_write(iocb
, iter
);
3595 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3600 * Pages can be marked dirty completely asynchronously from ext4's journalling
3601 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3602 * much here because ->set_page_dirty is called under VFS locks. The page is
3603 * not necessarily locked.
3605 * We cannot just dirty the page and leave attached buffers clean, because the
3606 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3607 * or jbddirty because all the journalling code will explode.
3609 * So what we do is to mark the page "pending dirty" and next time writepage
3610 * is called, propagate that into the buffers appropriately.
3612 static int ext4_journalled_set_page_dirty(struct page
*page
)
3614 SetPageChecked(page
);
3615 return __set_page_dirty_nobuffers(page
);
3618 static const struct address_space_operations ext4_aops
= {
3619 .readpage
= ext4_readpage
,
3620 .readpages
= ext4_readpages
,
3621 .writepage
= ext4_writepage
,
3622 .writepages
= ext4_writepages
,
3623 .write_begin
= ext4_write_begin
,
3624 .write_end
= ext4_write_end
,
3626 .invalidatepage
= ext4_invalidatepage
,
3627 .releasepage
= ext4_releasepage
,
3628 .direct_IO
= ext4_direct_IO
,
3629 .migratepage
= buffer_migrate_page
,
3630 .is_partially_uptodate
= block_is_partially_uptodate
,
3631 .error_remove_page
= generic_error_remove_page
,
3634 static const struct address_space_operations ext4_journalled_aops
= {
3635 .readpage
= ext4_readpage
,
3636 .readpages
= ext4_readpages
,
3637 .writepage
= ext4_writepage
,
3638 .writepages
= ext4_writepages
,
3639 .write_begin
= ext4_write_begin
,
3640 .write_end
= ext4_journalled_write_end
,
3641 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3643 .invalidatepage
= ext4_journalled_invalidatepage
,
3644 .releasepage
= ext4_releasepage
,
3645 .direct_IO
= ext4_direct_IO
,
3646 .is_partially_uptodate
= block_is_partially_uptodate
,
3647 .error_remove_page
= generic_error_remove_page
,
3650 static const struct address_space_operations ext4_da_aops
= {
3651 .readpage
= ext4_readpage
,
3652 .readpages
= ext4_readpages
,
3653 .writepage
= ext4_writepage
,
3654 .writepages
= ext4_writepages
,
3655 .write_begin
= ext4_da_write_begin
,
3656 .write_end
= ext4_da_write_end
,
3658 .invalidatepage
= ext4_da_invalidatepage
,
3659 .releasepage
= ext4_releasepage
,
3660 .direct_IO
= ext4_direct_IO
,
3661 .migratepage
= buffer_migrate_page
,
3662 .is_partially_uptodate
= block_is_partially_uptodate
,
3663 .error_remove_page
= generic_error_remove_page
,
3666 void ext4_set_aops(struct inode
*inode
)
3668 switch (ext4_inode_journal_mode(inode
)) {
3669 case EXT4_INODE_ORDERED_DATA_MODE
:
3670 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3672 case EXT4_INODE_JOURNAL_DATA_MODE
:
3673 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3678 if (test_opt(inode
->i_sb
, DELALLOC
))
3679 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3681 inode
->i_mapping
->a_ops
= &ext4_aops
;
3684 static int __ext4_block_zero_page_range(handle_t
*handle
,
3685 struct address_space
*mapping
, loff_t from
, loff_t length
)
3687 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3688 unsigned offset
= from
& (PAGE_SIZE
-1);
3689 unsigned blocksize
, pos
;
3691 struct inode
*inode
= mapping
->host
;
3692 struct buffer_head
*bh
;
3696 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3697 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3701 blocksize
= inode
->i_sb
->s_blocksize
;
3703 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3705 if (!page_has_buffers(page
))
3706 create_empty_buffers(page
, blocksize
, 0);
3708 /* Find the buffer that contains "offset" */
3709 bh
= page_buffers(page
);
3711 while (offset
>= pos
) {
3712 bh
= bh
->b_this_page
;
3716 if (buffer_freed(bh
)) {
3717 BUFFER_TRACE(bh
, "freed: skip");
3720 if (!buffer_mapped(bh
)) {
3721 BUFFER_TRACE(bh
, "unmapped");
3722 ext4_get_block(inode
, iblock
, bh
, 0);
3723 /* unmapped? It's a hole - nothing to do */
3724 if (!buffer_mapped(bh
)) {
3725 BUFFER_TRACE(bh
, "still unmapped");
3730 /* Ok, it's mapped. Make sure it's up-to-date */
3731 if (PageUptodate(page
))
3732 set_buffer_uptodate(bh
);
3734 if (!buffer_uptodate(bh
)) {
3736 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3738 /* Uhhuh. Read error. Complain and punt. */
3739 if (!buffer_uptodate(bh
))
3741 if (S_ISREG(inode
->i_mode
) &&
3742 ext4_encrypted_inode(inode
)) {
3743 /* We expect the key to be set. */
3744 BUG_ON(!fscrypt_has_encryption_key(inode
));
3745 BUG_ON(blocksize
!= PAGE_SIZE
);
3746 WARN_ON_ONCE(fscrypt_decrypt_page(page
));
3749 if (ext4_should_journal_data(inode
)) {
3750 BUFFER_TRACE(bh
, "get write access");
3751 err
= ext4_journal_get_write_access(handle
, bh
);
3755 zero_user(page
, offset
, length
);
3756 BUFFER_TRACE(bh
, "zeroed end of block");
3758 if (ext4_should_journal_data(inode
)) {
3759 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3762 mark_buffer_dirty(bh
);
3763 if (ext4_should_order_data(inode
))
3764 err
= ext4_jbd2_inode_add_write(handle
, inode
);
3774 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3775 * starting from file offset 'from'. The range to be zero'd must
3776 * be contained with in one block. If the specified range exceeds
3777 * the end of the block it will be shortened to end of the block
3778 * that cooresponds to 'from'
3780 static int ext4_block_zero_page_range(handle_t
*handle
,
3781 struct address_space
*mapping
, loff_t from
, loff_t length
)
3783 struct inode
*inode
= mapping
->host
;
3784 unsigned offset
= from
& (PAGE_SIZE
-1);
3785 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3786 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3789 * correct length if it does not fall between
3790 * 'from' and the end of the block
3792 if (length
> max
|| length
< 0)
3796 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3797 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3801 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3802 * up to the end of the block which corresponds to `from'.
3803 * This required during truncate. We need to physically zero the tail end
3804 * of that block so it doesn't yield old data if the file is later grown.
3806 static int ext4_block_truncate_page(handle_t
*handle
,
3807 struct address_space
*mapping
, loff_t from
)
3809 unsigned offset
= from
& (PAGE_SIZE
-1);
3812 struct inode
*inode
= mapping
->host
;
3814 blocksize
= inode
->i_sb
->s_blocksize
;
3815 length
= blocksize
- (offset
& (blocksize
- 1));
3817 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3820 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3821 loff_t lstart
, loff_t length
)
3823 struct super_block
*sb
= inode
->i_sb
;
3824 struct address_space
*mapping
= inode
->i_mapping
;
3825 unsigned partial_start
, partial_end
;
3826 ext4_fsblk_t start
, end
;
3827 loff_t byte_end
= (lstart
+ length
- 1);
3830 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3831 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3833 start
= lstart
>> sb
->s_blocksize_bits
;
3834 end
= byte_end
>> sb
->s_blocksize_bits
;
3836 /* Handle partial zero within the single block */
3838 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3839 err
= ext4_block_zero_page_range(handle
, mapping
,
3843 /* Handle partial zero out on the start of the range */
3844 if (partial_start
) {
3845 err
= ext4_block_zero_page_range(handle
, mapping
,
3846 lstart
, sb
->s_blocksize
);
3850 /* Handle partial zero out on the end of the range */
3851 if (partial_end
!= sb
->s_blocksize
- 1)
3852 err
= ext4_block_zero_page_range(handle
, mapping
,
3853 byte_end
- partial_end
,
3858 int ext4_can_truncate(struct inode
*inode
)
3860 if (S_ISREG(inode
->i_mode
))
3862 if (S_ISDIR(inode
->i_mode
))
3864 if (S_ISLNK(inode
->i_mode
))
3865 return !ext4_inode_is_fast_symlink(inode
);
3870 * We have to make sure i_disksize gets properly updated before we truncate
3871 * page cache due to hole punching or zero range. Otherwise i_disksize update
3872 * can get lost as it may have been postponed to submission of writeback but
3873 * that will never happen after we truncate page cache.
3875 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3879 loff_t size
= i_size_read(inode
);
3881 WARN_ON(!inode_is_locked(inode
));
3882 if (offset
> size
|| offset
+ len
< size
)
3885 if (EXT4_I(inode
)->i_disksize
>= size
)
3888 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3890 return PTR_ERR(handle
);
3891 ext4_update_i_disksize(inode
, size
);
3892 ext4_mark_inode_dirty(handle
, inode
);
3893 ext4_journal_stop(handle
);
3899 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3900 * associated with the given offset and length
3902 * @inode: File inode
3903 * @offset: The offset where the hole will begin
3904 * @len: The length of the hole
3906 * Returns: 0 on success or negative on failure
3909 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3911 struct super_block
*sb
= inode
->i_sb
;
3912 ext4_lblk_t first_block
, stop_block
;
3913 struct address_space
*mapping
= inode
->i_mapping
;
3914 loff_t first_block_offset
, last_block_offset
;
3916 unsigned int credits
;
3919 if (!S_ISREG(inode
->i_mode
))
3922 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3925 * Write out all dirty pages to avoid race conditions
3926 * Then release them.
3928 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3929 ret
= filemap_write_and_wait_range(mapping
, offset
,
3930 offset
+ length
- 1);
3937 /* No need to punch hole beyond i_size */
3938 if (offset
>= inode
->i_size
)
3942 * If the hole extends beyond i_size, set the hole
3943 * to end after the page that contains i_size
3945 if (offset
+ length
> inode
->i_size
) {
3946 length
= inode
->i_size
+
3947 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
3951 if (offset
& (sb
->s_blocksize
- 1) ||
3952 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3954 * Attach jinode to inode for jbd2 if we do any zeroing of
3957 ret
= ext4_inode_attach_jinode(inode
);
3963 /* Wait all existing dio workers, newcomers will block on i_mutex */
3964 ext4_inode_block_unlocked_dio(inode
);
3965 inode_dio_wait(inode
);
3968 * Prevent page faults from reinstantiating pages we have released from
3971 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3972 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3973 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3975 /* Now release the pages and zero block aligned part of pages*/
3976 if (last_block_offset
> first_block_offset
) {
3977 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
3980 truncate_pagecache_range(inode
, first_block_offset
,
3984 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3985 credits
= ext4_writepage_trans_blocks(inode
);
3987 credits
= ext4_blocks_for_truncate(inode
);
3988 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3989 if (IS_ERR(handle
)) {
3990 ret
= PTR_ERR(handle
);
3991 ext4_std_error(sb
, ret
);
3995 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4000 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4001 EXT4_BLOCK_SIZE_BITS(sb
);
4002 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4004 /* If there are no blocks to remove, return now */
4005 if (first_block
>= stop_block
)
4008 down_write(&EXT4_I(inode
)->i_data_sem
);
4009 ext4_discard_preallocations(inode
);
4011 ret
= ext4_es_remove_extent(inode
, first_block
,
4012 stop_block
- first_block
);
4014 up_write(&EXT4_I(inode
)->i_data_sem
);
4018 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4019 ret
= ext4_ext_remove_space(inode
, first_block
,
4022 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4025 up_write(&EXT4_I(inode
)->i_data_sem
);
4027 ext4_handle_sync(handle
);
4029 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4030 ext4_mark_inode_dirty(handle
, inode
);
4032 ext4_journal_stop(handle
);
4034 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4035 ext4_inode_resume_unlocked_dio(inode
);
4037 inode_unlock(inode
);
4041 int ext4_inode_attach_jinode(struct inode
*inode
)
4043 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4044 struct jbd2_inode
*jinode
;
4046 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4049 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4050 spin_lock(&inode
->i_lock
);
4053 spin_unlock(&inode
->i_lock
);
4056 ei
->jinode
= jinode
;
4057 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4060 spin_unlock(&inode
->i_lock
);
4061 if (unlikely(jinode
!= NULL
))
4062 jbd2_free_inode(jinode
);
4069 * We block out ext4_get_block() block instantiations across the entire
4070 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4071 * simultaneously on behalf of the same inode.
4073 * As we work through the truncate and commit bits of it to the journal there
4074 * is one core, guiding principle: the file's tree must always be consistent on
4075 * disk. We must be able to restart the truncate after a crash.
4077 * The file's tree may be transiently inconsistent in memory (although it
4078 * probably isn't), but whenever we close off and commit a journal transaction,
4079 * the contents of (the filesystem + the journal) must be consistent and
4080 * restartable. It's pretty simple, really: bottom up, right to left (although
4081 * left-to-right works OK too).
4083 * Note that at recovery time, journal replay occurs *before* the restart of
4084 * truncate against the orphan inode list.
4086 * The committed inode has the new, desired i_size (which is the same as
4087 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4088 * that this inode's truncate did not complete and it will again call
4089 * ext4_truncate() to have another go. So there will be instantiated blocks
4090 * to the right of the truncation point in a crashed ext4 filesystem. But
4091 * that's fine - as long as they are linked from the inode, the post-crash
4092 * ext4_truncate() run will find them and release them.
4094 void ext4_truncate(struct inode
*inode
)
4096 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4097 unsigned int credits
;
4099 struct address_space
*mapping
= inode
->i_mapping
;
4102 * There is a possibility that we're either freeing the inode
4103 * or it's a completely new inode. In those cases we might not
4104 * have i_mutex locked because it's not necessary.
4106 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4107 WARN_ON(!inode_is_locked(inode
));
4108 trace_ext4_truncate_enter(inode
);
4110 if (!ext4_can_truncate(inode
))
4113 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4115 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4116 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4118 if (ext4_has_inline_data(inode
)) {
4121 ext4_inline_data_truncate(inode
, &has_inline
);
4126 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4127 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4128 if (ext4_inode_attach_jinode(inode
) < 0)
4132 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4133 credits
= ext4_writepage_trans_blocks(inode
);
4135 credits
= ext4_blocks_for_truncate(inode
);
4137 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4138 if (IS_ERR(handle
)) {
4139 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
4143 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4144 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4147 * We add the inode to the orphan list, so that if this
4148 * truncate spans multiple transactions, and we crash, we will
4149 * resume the truncate when the filesystem recovers. It also
4150 * marks the inode dirty, to catch the new size.
4152 * Implication: the file must always be in a sane, consistent
4153 * truncatable state while each transaction commits.
4155 if (ext4_orphan_add(handle
, inode
))
4158 down_write(&EXT4_I(inode
)->i_data_sem
);
4160 ext4_discard_preallocations(inode
);
4162 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4163 ext4_ext_truncate(handle
, inode
);
4165 ext4_ind_truncate(handle
, inode
);
4167 up_write(&ei
->i_data_sem
);
4170 ext4_handle_sync(handle
);
4174 * If this was a simple ftruncate() and the file will remain alive,
4175 * then we need to clear up the orphan record which we created above.
4176 * However, if this was a real unlink then we were called by
4177 * ext4_evict_inode(), and we allow that function to clean up the
4178 * orphan info for us.
4181 ext4_orphan_del(handle
, inode
);
4183 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4184 ext4_mark_inode_dirty(handle
, inode
);
4185 ext4_journal_stop(handle
);
4187 trace_ext4_truncate_exit(inode
);
4191 * ext4_get_inode_loc returns with an extra refcount against the inode's
4192 * underlying buffer_head on success. If 'in_mem' is true, we have all
4193 * data in memory that is needed to recreate the on-disk version of this
4196 static int __ext4_get_inode_loc(struct inode
*inode
,
4197 struct ext4_iloc
*iloc
, int in_mem
)
4199 struct ext4_group_desc
*gdp
;
4200 struct buffer_head
*bh
;
4201 struct super_block
*sb
= inode
->i_sb
;
4203 int inodes_per_block
, inode_offset
;
4206 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4207 return -EFSCORRUPTED
;
4209 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4210 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4215 * Figure out the offset within the block group inode table
4217 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4218 inode_offset
= ((inode
->i_ino
- 1) %
4219 EXT4_INODES_PER_GROUP(sb
));
4220 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4221 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4223 bh
= sb_getblk(sb
, block
);
4226 if (!buffer_uptodate(bh
)) {
4230 * If the buffer has the write error flag, we have failed
4231 * to write out another inode in the same block. In this
4232 * case, we don't have to read the block because we may
4233 * read the old inode data successfully.
4235 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4236 set_buffer_uptodate(bh
);
4238 if (buffer_uptodate(bh
)) {
4239 /* someone brought it uptodate while we waited */
4245 * If we have all information of the inode in memory and this
4246 * is the only valid inode in the block, we need not read the
4250 struct buffer_head
*bitmap_bh
;
4253 start
= inode_offset
& ~(inodes_per_block
- 1);
4255 /* Is the inode bitmap in cache? */
4256 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4257 if (unlikely(!bitmap_bh
))
4261 * If the inode bitmap isn't in cache then the
4262 * optimisation may end up performing two reads instead
4263 * of one, so skip it.
4265 if (!buffer_uptodate(bitmap_bh
)) {
4269 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4270 if (i
== inode_offset
)
4272 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4276 if (i
== start
+ inodes_per_block
) {
4277 /* all other inodes are free, so skip I/O */
4278 memset(bh
->b_data
, 0, bh
->b_size
);
4279 set_buffer_uptodate(bh
);
4287 * If we need to do any I/O, try to pre-readahead extra
4288 * blocks from the inode table.
4290 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4291 ext4_fsblk_t b
, end
, table
;
4293 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4295 table
= ext4_inode_table(sb
, gdp
);
4296 /* s_inode_readahead_blks is always a power of 2 */
4297 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4301 num
= EXT4_INODES_PER_GROUP(sb
);
4302 if (ext4_has_group_desc_csum(sb
))
4303 num
-= ext4_itable_unused_count(sb
, gdp
);
4304 table
+= num
/ inodes_per_block
;
4308 sb_breadahead(sb
, b
++);
4312 * There are other valid inodes in the buffer, this inode
4313 * has in-inode xattrs, or we don't have this inode in memory.
4314 * Read the block from disk.
4316 trace_ext4_load_inode(inode
);
4318 bh
->b_end_io
= end_buffer_read_sync
;
4319 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4321 if (!buffer_uptodate(bh
)) {
4322 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4323 "unable to read itable block");
4333 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4335 /* We have all inode data except xattrs in memory here. */
4336 return __ext4_get_inode_loc(inode
, iloc
,
4337 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4340 void ext4_set_inode_flags(struct inode
*inode
)
4342 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4343 unsigned int new_fl
= 0;
4345 if (flags
& EXT4_SYNC_FL
)
4347 if (flags
& EXT4_APPEND_FL
)
4349 if (flags
& EXT4_IMMUTABLE_FL
)
4350 new_fl
|= S_IMMUTABLE
;
4351 if (flags
& EXT4_NOATIME_FL
)
4352 new_fl
|= S_NOATIME
;
4353 if (flags
& EXT4_DIRSYNC_FL
)
4354 new_fl
|= S_DIRSYNC
;
4355 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
))
4357 inode_set_flags(inode
, new_fl
,
4358 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4361 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4362 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4364 unsigned int vfs_fl
;
4365 unsigned long old_fl
, new_fl
;
4368 vfs_fl
= ei
->vfs_inode
.i_flags
;
4369 old_fl
= ei
->i_flags
;
4370 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4371 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4373 if (vfs_fl
& S_SYNC
)
4374 new_fl
|= EXT4_SYNC_FL
;
4375 if (vfs_fl
& S_APPEND
)
4376 new_fl
|= EXT4_APPEND_FL
;
4377 if (vfs_fl
& S_IMMUTABLE
)
4378 new_fl
|= EXT4_IMMUTABLE_FL
;
4379 if (vfs_fl
& S_NOATIME
)
4380 new_fl
|= EXT4_NOATIME_FL
;
4381 if (vfs_fl
& S_DIRSYNC
)
4382 new_fl
|= EXT4_DIRSYNC_FL
;
4383 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4386 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4387 struct ext4_inode_info
*ei
)
4390 struct inode
*inode
= &(ei
->vfs_inode
);
4391 struct super_block
*sb
= inode
->i_sb
;
4393 if (ext4_has_feature_huge_file(sb
)) {
4394 /* we are using combined 48 bit field */
4395 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4396 le32_to_cpu(raw_inode
->i_blocks_lo
);
4397 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4398 /* i_blocks represent file system block size */
4399 return i_blocks
<< (inode
->i_blkbits
- 9);
4404 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4408 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4409 struct ext4_inode
*raw_inode
,
4410 struct ext4_inode_info
*ei
)
4412 __le32
*magic
= (void *)raw_inode
+
4413 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4414 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4415 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4416 ext4_find_inline_data_nolock(inode
);
4418 EXT4_I(inode
)->i_inline_off
= 0;
4421 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4423 if (!ext4_has_feature_project(inode
->i_sb
))
4425 *projid
= EXT4_I(inode
)->i_projid
;
4429 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4431 struct ext4_iloc iloc
;
4432 struct ext4_inode
*raw_inode
;
4433 struct ext4_inode_info
*ei
;
4434 struct inode
*inode
;
4435 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4442 inode
= iget_locked(sb
, ino
);
4444 return ERR_PTR(-ENOMEM
);
4445 if (!(inode
->i_state
& I_NEW
))
4451 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4454 raw_inode
= ext4_raw_inode(&iloc
);
4456 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4457 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4458 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4459 EXT4_INODE_SIZE(inode
->i_sb
)) {
4460 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4461 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4462 EXT4_INODE_SIZE(inode
->i_sb
));
4463 ret
= -EFSCORRUPTED
;
4467 ei
->i_extra_isize
= 0;
4469 /* Precompute checksum seed for inode metadata */
4470 if (ext4_has_metadata_csum(sb
)) {
4471 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4473 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4474 __le32 gen
= raw_inode
->i_generation
;
4475 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4477 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4481 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4482 EXT4_ERROR_INODE(inode
, "checksum invalid");
4487 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4488 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4489 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4490 if (ext4_has_feature_project(sb
) &&
4491 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4492 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4493 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4495 i_projid
= EXT4_DEF_PROJID
;
4497 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4498 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4499 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4501 i_uid_write(inode
, i_uid
);
4502 i_gid_write(inode
, i_gid
);
4503 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4504 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4506 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4507 ei
->i_inline_off
= 0;
4508 ei
->i_dir_start_lookup
= 0;
4509 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4510 /* We now have enough fields to check if the inode was active or not.
4511 * This is needed because nfsd might try to access dead inodes
4512 * the test is that same one that e2fsck uses
4513 * NeilBrown 1999oct15
4515 if (inode
->i_nlink
== 0) {
4516 if ((inode
->i_mode
== 0 ||
4517 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4518 ino
!= EXT4_BOOT_LOADER_INO
) {
4519 /* this inode is deleted */
4523 /* The only unlinked inodes we let through here have
4524 * valid i_mode and are being read by the orphan
4525 * recovery code: that's fine, we're about to complete
4526 * the process of deleting those.
4527 * OR it is the EXT4_BOOT_LOADER_INO which is
4528 * not initialized on a new filesystem. */
4530 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4531 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4532 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4533 if (ext4_has_feature_64bit(sb
))
4535 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4536 inode
->i_size
= ext4_isize(raw_inode
);
4537 ei
->i_disksize
= inode
->i_size
;
4539 ei
->i_reserved_quota
= 0;
4541 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4542 ei
->i_block_group
= iloc
.block_group
;
4543 ei
->i_last_alloc_group
= ~0;
4545 * NOTE! The in-memory inode i_data array is in little-endian order
4546 * even on big-endian machines: we do NOT byteswap the block numbers!
4548 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4549 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4550 INIT_LIST_HEAD(&ei
->i_orphan
);
4553 * Set transaction id's of transactions that have to be committed
4554 * to finish f[data]sync. We set them to currently running transaction
4555 * as we cannot be sure that the inode or some of its metadata isn't
4556 * part of the transaction - the inode could have been reclaimed and
4557 * now it is reread from disk.
4560 transaction_t
*transaction
;
4563 read_lock(&journal
->j_state_lock
);
4564 if (journal
->j_running_transaction
)
4565 transaction
= journal
->j_running_transaction
;
4567 transaction
= journal
->j_committing_transaction
;
4569 tid
= transaction
->t_tid
;
4571 tid
= journal
->j_commit_sequence
;
4572 read_unlock(&journal
->j_state_lock
);
4573 ei
->i_sync_tid
= tid
;
4574 ei
->i_datasync_tid
= tid
;
4577 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4578 if (ei
->i_extra_isize
== 0) {
4579 /* The extra space is currently unused. Use it. */
4580 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4581 EXT4_GOOD_OLD_INODE_SIZE
;
4583 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4587 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4588 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4589 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4590 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4592 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4593 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4594 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4595 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4597 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4602 if (ei
->i_file_acl
&&
4603 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4604 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4606 ret
= -EFSCORRUPTED
;
4608 } else if (!ext4_has_inline_data(inode
)) {
4609 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4610 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4611 (S_ISLNK(inode
->i_mode
) &&
4612 !ext4_inode_is_fast_symlink(inode
))))
4613 /* Validate extent which is part of inode */
4614 ret
= ext4_ext_check_inode(inode
);
4615 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4616 (S_ISLNK(inode
->i_mode
) &&
4617 !ext4_inode_is_fast_symlink(inode
))) {
4618 /* Validate block references which are part of inode */
4619 ret
= ext4_ind_check_inode(inode
);
4625 if (S_ISREG(inode
->i_mode
)) {
4626 inode
->i_op
= &ext4_file_inode_operations
;
4627 inode
->i_fop
= &ext4_file_operations
;
4628 ext4_set_aops(inode
);
4629 } else if (S_ISDIR(inode
->i_mode
)) {
4630 inode
->i_op
= &ext4_dir_inode_operations
;
4631 inode
->i_fop
= &ext4_dir_operations
;
4632 } else if (S_ISLNK(inode
->i_mode
)) {
4633 if (ext4_encrypted_inode(inode
)) {
4634 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4635 ext4_set_aops(inode
);
4636 } else if (ext4_inode_is_fast_symlink(inode
)) {
4637 inode
->i_link
= (char *)ei
->i_data
;
4638 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4639 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4640 sizeof(ei
->i_data
) - 1);
4642 inode
->i_op
= &ext4_symlink_inode_operations
;
4643 ext4_set_aops(inode
);
4645 inode_nohighmem(inode
);
4646 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4647 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4648 inode
->i_op
= &ext4_special_inode_operations
;
4649 if (raw_inode
->i_block
[0])
4650 init_special_inode(inode
, inode
->i_mode
,
4651 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4653 init_special_inode(inode
, inode
->i_mode
,
4654 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4655 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4656 make_bad_inode(inode
);
4658 ret
= -EFSCORRUPTED
;
4659 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4663 ext4_set_inode_flags(inode
);
4664 unlock_new_inode(inode
);
4670 return ERR_PTR(ret
);
4673 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4675 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4676 return ERR_PTR(-EFSCORRUPTED
);
4677 return ext4_iget(sb
, ino
);
4680 static int ext4_inode_blocks_set(handle_t
*handle
,
4681 struct ext4_inode
*raw_inode
,
4682 struct ext4_inode_info
*ei
)
4684 struct inode
*inode
= &(ei
->vfs_inode
);
4685 u64 i_blocks
= inode
->i_blocks
;
4686 struct super_block
*sb
= inode
->i_sb
;
4688 if (i_blocks
<= ~0U) {
4690 * i_blocks can be represented in a 32 bit variable
4691 * as multiple of 512 bytes
4693 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4694 raw_inode
->i_blocks_high
= 0;
4695 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4698 if (!ext4_has_feature_huge_file(sb
))
4701 if (i_blocks
<= 0xffffffffffffULL
) {
4703 * i_blocks can be represented in a 48 bit variable
4704 * as multiple of 512 bytes
4706 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4707 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4708 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4710 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4711 /* i_block is stored in file system block size */
4712 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4713 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4714 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4719 struct other_inode
{
4720 unsigned long orig_ino
;
4721 struct ext4_inode
*raw_inode
;
4724 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4727 struct other_inode
*oi
= (struct other_inode
*) data
;
4729 if ((inode
->i_ino
!= ino
) ||
4730 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4731 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4732 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4734 spin_lock(&inode
->i_lock
);
4735 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4736 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4737 (inode
->i_state
& I_DIRTY_TIME
)) {
4738 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4740 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4741 spin_unlock(&inode
->i_lock
);
4743 spin_lock(&ei
->i_raw_lock
);
4744 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4745 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4746 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4747 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4748 spin_unlock(&ei
->i_raw_lock
);
4749 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4752 spin_unlock(&inode
->i_lock
);
4757 * Opportunistically update the other time fields for other inodes in
4758 * the same inode table block.
4760 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4761 unsigned long orig_ino
, char *buf
)
4763 struct other_inode oi
;
4765 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4766 int inode_size
= EXT4_INODE_SIZE(sb
);
4768 oi
.orig_ino
= orig_ino
;
4770 * Calculate the first inode in the inode table block. Inode
4771 * numbers are one-based. That is, the first inode in a block
4772 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4774 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4775 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4776 if (ino
== orig_ino
)
4778 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4779 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4784 * Post the struct inode info into an on-disk inode location in the
4785 * buffer-cache. This gobbles the caller's reference to the
4786 * buffer_head in the inode location struct.
4788 * The caller must have write access to iloc->bh.
4790 static int ext4_do_update_inode(handle_t
*handle
,
4791 struct inode
*inode
,
4792 struct ext4_iloc
*iloc
)
4794 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4795 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4796 struct buffer_head
*bh
= iloc
->bh
;
4797 struct super_block
*sb
= inode
->i_sb
;
4798 int err
= 0, rc
, block
;
4799 int need_datasync
= 0, set_large_file
= 0;
4804 spin_lock(&ei
->i_raw_lock
);
4806 /* For fields not tracked in the in-memory inode,
4807 * initialise them to zero for new inodes. */
4808 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4809 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4811 ext4_get_inode_flags(ei
);
4812 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4813 i_uid
= i_uid_read(inode
);
4814 i_gid
= i_gid_read(inode
);
4815 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4816 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4817 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4818 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4820 * Fix up interoperability with old kernels. Otherwise, old inodes get
4821 * re-used with the upper 16 bits of the uid/gid intact
4823 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4824 raw_inode
->i_uid_high
= 0;
4825 raw_inode
->i_gid_high
= 0;
4827 raw_inode
->i_uid_high
=
4828 cpu_to_le16(high_16_bits(i_uid
));
4829 raw_inode
->i_gid_high
=
4830 cpu_to_le16(high_16_bits(i_gid
));
4833 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4834 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4835 raw_inode
->i_uid_high
= 0;
4836 raw_inode
->i_gid_high
= 0;
4838 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4840 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4841 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4842 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4843 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4845 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4847 spin_unlock(&ei
->i_raw_lock
);
4850 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4851 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4852 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4853 raw_inode
->i_file_acl_high
=
4854 cpu_to_le16(ei
->i_file_acl
>> 32);
4855 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4856 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4857 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4860 if (ei
->i_disksize
> 0x7fffffffULL
) {
4861 if (!ext4_has_feature_large_file(sb
) ||
4862 EXT4_SB(sb
)->s_es
->s_rev_level
==
4863 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4866 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4867 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4868 if (old_valid_dev(inode
->i_rdev
)) {
4869 raw_inode
->i_block
[0] =
4870 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4871 raw_inode
->i_block
[1] = 0;
4873 raw_inode
->i_block
[0] = 0;
4874 raw_inode
->i_block
[1] =
4875 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4876 raw_inode
->i_block
[2] = 0;
4878 } else if (!ext4_has_inline_data(inode
)) {
4879 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4880 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4883 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4884 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4885 if (ei
->i_extra_isize
) {
4886 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4887 raw_inode
->i_version_hi
=
4888 cpu_to_le32(inode
->i_version
>> 32);
4889 raw_inode
->i_extra_isize
=
4890 cpu_to_le16(ei
->i_extra_isize
);
4894 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
4895 i_projid
!= EXT4_DEF_PROJID
);
4897 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4898 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4899 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4901 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4902 spin_unlock(&ei
->i_raw_lock
);
4903 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4904 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4907 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4908 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4911 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4912 if (set_large_file
) {
4913 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4914 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4917 ext4_update_dynamic_rev(sb
);
4918 ext4_set_feature_large_file(sb
);
4919 ext4_handle_sync(handle
);
4920 err
= ext4_handle_dirty_super(handle
, sb
);
4922 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4925 ext4_std_error(inode
->i_sb
, err
);
4930 * ext4_write_inode()
4932 * We are called from a few places:
4934 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4935 * Here, there will be no transaction running. We wait for any running
4936 * transaction to commit.
4938 * - Within flush work (sys_sync(), kupdate and such).
4939 * We wait on commit, if told to.
4941 * - Within iput_final() -> write_inode_now()
4942 * We wait on commit, if told to.
4944 * In all cases it is actually safe for us to return without doing anything,
4945 * because the inode has been copied into a raw inode buffer in
4946 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4949 * Note that we are absolutely dependent upon all inode dirtiers doing the
4950 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4951 * which we are interested.
4953 * It would be a bug for them to not do this. The code:
4955 * mark_inode_dirty(inode)
4957 * inode->i_size = expr;
4959 * is in error because write_inode() could occur while `stuff()' is running,
4960 * and the new i_size will be lost. Plus the inode will no longer be on the
4961 * superblock's dirty inode list.
4963 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4967 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4970 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4971 if (ext4_journal_current_handle()) {
4972 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4978 * No need to force transaction in WB_SYNC_NONE mode. Also
4979 * ext4_sync_fs() will force the commit after everything is
4982 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4985 err
= ext4_force_commit(inode
->i_sb
);
4987 struct ext4_iloc iloc
;
4989 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4993 * sync(2) will flush the whole buffer cache. No need to do
4994 * it here separately for each inode.
4996 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4997 sync_dirty_buffer(iloc
.bh
);
4998 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4999 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5000 "IO error syncing inode");
5009 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5010 * buffers that are attached to a page stradding i_size and are undergoing
5011 * commit. In that case we have to wait for commit to finish and try again.
5013 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5017 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5018 tid_t commit_tid
= 0;
5021 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5023 * All buffers in the last page remain valid? Then there's nothing to
5024 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5027 if (offset
> PAGE_SIZE
- (1 << inode
->i_blkbits
))
5030 page
= find_lock_page(inode
->i_mapping
,
5031 inode
->i_size
>> PAGE_SHIFT
);
5034 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5035 PAGE_SIZE
- offset
);
5041 read_lock(&journal
->j_state_lock
);
5042 if (journal
->j_committing_transaction
)
5043 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5044 read_unlock(&journal
->j_state_lock
);
5046 jbd2_log_wait_commit(journal
, commit_tid
);
5053 * Called from notify_change.
5055 * We want to trap VFS attempts to truncate the file as soon as
5056 * possible. In particular, we want to make sure that when the VFS
5057 * shrinks i_size, we put the inode on the orphan list and modify
5058 * i_disksize immediately, so that during the subsequent flushing of
5059 * dirty pages and freeing of disk blocks, we can guarantee that any
5060 * commit will leave the blocks being flushed in an unused state on
5061 * disk. (On recovery, the inode will get truncated and the blocks will
5062 * be freed, so we have a strong guarantee that no future commit will
5063 * leave these blocks visible to the user.)
5065 * Another thing we have to assure is that if we are in ordered mode
5066 * and inode is still attached to the committing transaction, we must
5067 * we start writeout of all the dirty pages which are being truncated.
5068 * This way we are sure that all the data written in the previous
5069 * transaction are already on disk (truncate waits for pages under
5072 * Called with inode->i_mutex down.
5074 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5076 struct inode
*inode
= d_inode(dentry
);
5079 const unsigned int ia_valid
= attr
->ia_valid
;
5081 error
= setattr_prepare(dentry
, attr
);
5085 if (is_quota_modification(inode
, attr
)) {
5086 error
= dquot_initialize(inode
);
5090 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5091 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5094 /* (user+group)*(old+new) structure, inode write (sb,
5095 * inode block, ? - but truncate inode update has it) */
5096 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5097 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5098 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5099 if (IS_ERR(handle
)) {
5100 error
= PTR_ERR(handle
);
5103 error
= dquot_transfer(inode
, attr
);
5105 ext4_journal_stop(handle
);
5108 /* Update corresponding info in inode so that everything is in
5109 * one transaction */
5110 if (attr
->ia_valid
& ATTR_UID
)
5111 inode
->i_uid
= attr
->ia_uid
;
5112 if (attr
->ia_valid
& ATTR_GID
)
5113 inode
->i_gid
= attr
->ia_gid
;
5114 error
= ext4_mark_inode_dirty(handle
, inode
);
5115 ext4_journal_stop(handle
);
5118 if (attr
->ia_valid
& ATTR_SIZE
) {
5120 loff_t oldsize
= inode
->i_size
;
5121 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5123 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5124 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5126 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5129 if (!S_ISREG(inode
->i_mode
))
5132 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5133 inode_inc_iversion(inode
);
5135 if (ext4_should_order_data(inode
) &&
5136 (attr
->ia_size
< inode
->i_size
)) {
5137 error
= ext4_begin_ordered_truncate(inode
,
5142 if (attr
->ia_size
!= inode
->i_size
) {
5143 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5144 if (IS_ERR(handle
)) {
5145 error
= PTR_ERR(handle
);
5148 if (ext4_handle_valid(handle
) && shrink
) {
5149 error
= ext4_orphan_add(handle
, inode
);
5153 * Update c/mtime on truncate up, ext4_truncate() will
5154 * update c/mtime in shrink case below
5157 inode
->i_mtime
= ext4_current_time(inode
);
5158 inode
->i_ctime
= inode
->i_mtime
;
5160 down_write(&EXT4_I(inode
)->i_data_sem
);
5161 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5162 rc
= ext4_mark_inode_dirty(handle
, inode
);
5166 * We have to update i_size under i_data_sem together
5167 * with i_disksize to avoid races with writeback code
5168 * running ext4_wb_update_i_disksize().
5171 i_size_write(inode
, attr
->ia_size
);
5172 up_write(&EXT4_I(inode
)->i_data_sem
);
5173 ext4_journal_stop(handle
);
5176 ext4_orphan_del(NULL
, inode
);
5181 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5184 * Blocks are going to be removed from the inode. Wait
5185 * for dio in flight. Temporarily disable
5186 * dioread_nolock to prevent livelock.
5189 if (!ext4_should_journal_data(inode
)) {
5190 ext4_inode_block_unlocked_dio(inode
);
5191 inode_dio_wait(inode
);
5192 ext4_inode_resume_unlocked_dio(inode
);
5194 ext4_wait_for_tail_page_commit(inode
);
5196 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5198 * Truncate pagecache after we've waited for commit
5199 * in data=journal mode to make pages freeable.
5201 truncate_pagecache(inode
, inode
->i_size
);
5203 ext4_truncate(inode
);
5204 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5208 setattr_copy(inode
, attr
);
5209 mark_inode_dirty(inode
);
5213 * If the call to ext4_truncate failed to get a transaction handle at
5214 * all, we need to clean up the in-core orphan list manually.
5216 if (orphan
&& inode
->i_nlink
)
5217 ext4_orphan_del(NULL
, inode
);
5219 if (!rc
&& (ia_valid
& ATTR_MODE
))
5220 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5223 ext4_std_error(inode
->i_sb
, error
);
5229 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5232 struct inode
*inode
;
5233 unsigned long long delalloc_blocks
;
5235 inode
= d_inode(dentry
);
5236 generic_fillattr(inode
, stat
);
5239 * If there is inline data in the inode, the inode will normally not
5240 * have data blocks allocated (it may have an external xattr block).
5241 * Report at least one sector for such files, so tools like tar, rsync,
5242 * others doen't incorrectly think the file is completely sparse.
5244 if (unlikely(ext4_has_inline_data(inode
)))
5245 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5248 * We can't update i_blocks if the block allocation is delayed
5249 * otherwise in the case of system crash before the real block
5250 * allocation is done, we will have i_blocks inconsistent with
5251 * on-disk file blocks.
5252 * We always keep i_blocks updated together with real
5253 * allocation. But to not confuse with user, stat
5254 * will return the blocks that include the delayed allocation
5255 * blocks for this file.
5257 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5258 EXT4_I(inode
)->i_reserved_data_blocks
);
5259 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5263 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5266 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5267 return ext4_ind_trans_blocks(inode
, lblocks
);
5268 return ext4_ext_index_trans_blocks(inode
, pextents
);
5272 * Account for index blocks, block groups bitmaps and block group
5273 * descriptor blocks if modify datablocks and index blocks
5274 * worse case, the indexs blocks spread over different block groups
5276 * If datablocks are discontiguous, they are possible to spread over
5277 * different block groups too. If they are contiguous, with flexbg,
5278 * they could still across block group boundary.
5280 * Also account for superblock, inode, quota and xattr blocks
5282 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5285 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5291 * How many index blocks need to touch to map @lblocks logical blocks
5292 * to @pextents physical extents?
5294 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5299 * Now let's see how many group bitmaps and group descriptors need
5302 groups
= idxblocks
+ pextents
;
5304 if (groups
> ngroups
)
5306 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5307 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5309 /* bitmaps and block group descriptor blocks */
5310 ret
+= groups
+ gdpblocks
;
5312 /* Blocks for super block, inode, quota and xattr blocks */
5313 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5319 * Calculate the total number of credits to reserve to fit
5320 * the modification of a single pages into a single transaction,
5321 * which may include multiple chunks of block allocations.
5323 * This could be called via ext4_write_begin()
5325 * We need to consider the worse case, when
5326 * one new block per extent.
5328 int ext4_writepage_trans_blocks(struct inode
*inode
)
5330 int bpp
= ext4_journal_blocks_per_page(inode
);
5333 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5335 /* Account for data blocks for journalled mode */
5336 if (ext4_should_journal_data(inode
))
5342 * Calculate the journal credits for a chunk of data modification.
5344 * This is called from DIO, fallocate or whoever calling
5345 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5347 * journal buffers for data blocks are not included here, as DIO
5348 * and fallocate do no need to journal data buffers.
5350 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5352 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5356 * The caller must have previously called ext4_reserve_inode_write().
5357 * Give this, we know that the caller already has write access to iloc->bh.
5359 int ext4_mark_iloc_dirty(handle_t
*handle
,
5360 struct inode
*inode
, struct ext4_iloc
*iloc
)
5364 if (IS_I_VERSION(inode
))
5365 inode_inc_iversion(inode
);
5367 /* the do_update_inode consumes one bh->b_count */
5370 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5371 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5377 * On success, We end up with an outstanding reference count against
5378 * iloc->bh. This _must_ be cleaned up later.
5382 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5383 struct ext4_iloc
*iloc
)
5387 err
= ext4_get_inode_loc(inode
, iloc
);
5389 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5390 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5396 ext4_std_error(inode
->i_sb
, err
);
5401 * Expand an inode by new_extra_isize bytes.
5402 * Returns 0 on success or negative error number on failure.
5404 static int ext4_expand_extra_isize(struct inode
*inode
,
5405 unsigned int new_extra_isize
,
5406 struct ext4_iloc iloc
,
5409 struct ext4_inode
*raw_inode
;
5410 struct ext4_xattr_ibody_header
*header
;
5412 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5415 raw_inode
= ext4_raw_inode(&iloc
);
5417 header
= IHDR(inode
, raw_inode
);
5419 /* No extended attributes present */
5420 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5421 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5422 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5424 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5428 /* try to expand with EAs present */
5429 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5434 * What we do here is to mark the in-core inode as clean with respect to inode
5435 * dirtiness (it may still be data-dirty).
5436 * This means that the in-core inode may be reaped by prune_icache
5437 * without having to perform any I/O. This is a very good thing,
5438 * because *any* task may call prune_icache - even ones which
5439 * have a transaction open against a different journal.
5441 * Is this cheating? Not really. Sure, we haven't written the
5442 * inode out, but prune_icache isn't a user-visible syncing function.
5443 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5444 * we start and wait on commits.
5446 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5448 struct ext4_iloc iloc
;
5449 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5450 static unsigned int mnt_count
;
5454 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5455 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5458 if (ext4_handle_valid(handle
) &&
5459 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5460 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5462 * We need extra buffer credits since we may write into EA block
5463 * with this same handle. If journal_extend fails, then it will
5464 * only result in a minor loss of functionality for that inode.
5465 * If this is felt to be critical, then e2fsck should be run to
5466 * force a large enough s_min_extra_isize.
5468 if ((jbd2_journal_extend(handle
,
5469 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5470 ret
= ext4_expand_extra_isize(inode
,
5471 sbi
->s_want_extra_isize
,
5475 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5476 ext4_warning(inode
->i_sb
,
5477 "Unable to expand inode %lu. Delete"
5478 " some EAs or run e2fsck.",
5481 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5486 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5490 * ext4_dirty_inode() is called from __mark_inode_dirty()
5492 * We're really interested in the case where a file is being extended.
5493 * i_size has been changed by generic_commit_write() and we thus need
5494 * to include the updated inode in the current transaction.
5496 * Also, dquot_alloc_block() will always dirty the inode when blocks
5497 * are allocated to the file.
5499 * If the inode is marked synchronous, we don't honour that here - doing
5500 * so would cause a commit on atime updates, which we don't bother doing.
5501 * We handle synchronous inodes at the highest possible level.
5503 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5504 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5505 * to copy into the on-disk inode structure are the timestamp files.
5507 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5511 if (flags
== I_DIRTY_TIME
)
5513 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5517 ext4_mark_inode_dirty(handle
, inode
);
5519 ext4_journal_stop(handle
);
5526 * Bind an inode's backing buffer_head into this transaction, to prevent
5527 * it from being flushed to disk early. Unlike
5528 * ext4_reserve_inode_write, this leaves behind no bh reference and
5529 * returns no iloc structure, so the caller needs to repeat the iloc
5530 * lookup to mark the inode dirty later.
5532 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5534 struct ext4_iloc iloc
;
5538 err
= ext4_get_inode_loc(inode
, &iloc
);
5540 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5541 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5543 err
= ext4_handle_dirty_metadata(handle
,
5549 ext4_std_error(inode
->i_sb
, err
);
5554 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5559 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5562 * We have to be very careful here: changing a data block's
5563 * journaling status dynamically is dangerous. If we write a
5564 * data block to the journal, change the status and then delete
5565 * that block, we risk forgetting to revoke the old log record
5566 * from the journal and so a subsequent replay can corrupt data.
5567 * So, first we make sure that the journal is empty and that
5568 * nobody is changing anything.
5571 journal
= EXT4_JOURNAL(inode
);
5574 if (is_journal_aborted(journal
))
5577 /* Wait for all existing dio workers */
5578 ext4_inode_block_unlocked_dio(inode
);
5579 inode_dio_wait(inode
);
5582 * Before flushing the journal and switching inode's aops, we have
5583 * to flush all dirty data the inode has. There can be outstanding
5584 * delayed allocations, there can be unwritten extents created by
5585 * fallocate or buffered writes in dioread_nolock mode covered by
5586 * dirty data which can be converted only after flushing the dirty
5587 * data (and journalled aops don't know how to handle these cases).
5590 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5591 err
= filemap_write_and_wait(inode
->i_mapping
);
5593 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5594 ext4_inode_resume_unlocked_dio(inode
);
5599 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5600 jbd2_journal_lock_updates(journal
);
5603 * OK, there are no updates running now, and all cached data is
5604 * synced to disk. We are now in a completely consistent state
5605 * which doesn't have anything in the journal, and we know that
5606 * no filesystem updates are running, so it is safe to modify
5607 * the inode's in-core data-journaling state flag now.
5611 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5613 err
= jbd2_journal_flush(journal
);
5615 jbd2_journal_unlock_updates(journal
);
5616 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5617 ext4_inode_resume_unlocked_dio(inode
);
5620 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5622 ext4_set_aops(inode
);
5624 jbd2_journal_unlock_updates(journal
);
5625 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5628 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5629 ext4_inode_resume_unlocked_dio(inode
);
5631 /* Finally we can mark the inode as dirty. */
5633 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5635 return PTR_ERR(handle
);
5637 err
= ext4_mark_inode_dirty(handle
, inode
);
5638 ext4_handle_sync(handle
);
5639 ext4_journal_stop(handle
);
5640 ext4_std_error(inode
->i_sb
, err
);
5645 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5647 return !buffer_mapped(bh
);
5650 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5652 struct page
*page
= vmf
->page
;
5656 struct file
*file
= vma
->vm_file
;
5657 struct inode
*inode
= file_inode(file
);
5658 struct address_space
*mapping
= inode
->i_mapping
;
5660 get_block_t
*get_block
;
5663 sb_start_pagefault(inode
->i_sb
);
5664 file_update_time(vma
->vm_file
);
5666 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5667 /* Delalloc case is easy... */
5668 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5669 !ext4_should_journal_data(inode
) &&
5670 !ext4_nonda_switch(inode
->i_sb
)) {
5672 ret
= block_page_mkwrite(vma
, vmf
,
5673 ext4_da_get_block_prep
);
5674 } while (ret
== -ENOSPC
&&
5675 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5680 size
= i_size_read(inode
);
5681 /* Page got truncated from under us? */
5682 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5684 ret
= VM_FAULT_NOPAGE
;
5688 if (page
->index
== size
>> PAGE_SHIFT
)
5689 len
= size
& ~PAGE_MASK
;
5693 * Return if we have all the buffers mapped. This avoids the need to do
5694 * journal_start/journal_stop which can block and take a long time
5696 if (page_has_buffers(page
)) {
5697 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5699 ext4_bh_unmapped
)) {
5700 /* Wait so that we don't change page under IO */
5701 wait_for_stable_page(page
);
5702 ret
= VM_FAULT_LOCKED
;
5707 /* OK, we need to fill the hole... */
5708 if (ext4_should_dioread_nolock(inode
))
5709 get_block
= ext4_get_block_unwritten
;
5711 get_block
= ext4_get_block
;
5713 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5714 ext4_writepage_trans_blocks(inode
));
5715 if (IS_ERR(handle
)) {
5716 ret
= VM_FAULT_SIGBUS
;
5719 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5720 if (!ret
&& ext4_should_journal_data(inode
)) {
5721 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5722 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
5724 ret
= VM_FAULT_SIGBUS
;
5725 ext4_journal_stop(handle
);
5728 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5730 ext4_journal_stop(handle
);
5731 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5734 ret
= block_page_mkwrite_return(ret
);
5736 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5737 sb_end_pagefault(inode
->i_sb
);
5741 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5743 struct inode
*inode
= file_inode(vma
->vm_file
);
5746 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5747 err
= filemap_fault(vma
, vmf
);
5748 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5754 * Find the first extent at or after @lblk in an inode that is not a hole.
5755 * Search for @map_len blocks at most. The extent is returned in @result.
5757 * The function returns 1 if we found an extent. The function returns 0 in
5758 * case there is no extent at or after @lblk and in that case also sets
5759 * @result->es_len to 0. In case of error, the error code is returned.
5761 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
5762 unsigned int map_len
, struct extent_status
*result
)
5764 struct ext4_map_blocks map
;
5765 struct extent_status es
= {};
5769 map
.m_len
= map_len
;
5772 * For non-extent based files this loop may iterate several times since
5773 * we do not determine full hole size.
5775 while (map
.m_len
> 0) {
5776 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
5779 /* There's extent covering m_lblk? Just return it. */
5783 ext4_es_store_pblock(result
, map
.m_pblk
);
5784 result
->es_lblk
= map
.m_lblk
;
5785 result
->es_len
= map
.m_len
;
5786 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
5787 status
= EXTENT_STATUS_UNWRITTEN
;
5789 status
= EXTENT_STATUS_WRITTEN
;
5790 ext4_es_store_status(result
, status
);
5793 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
5794 map
.m_lblk
+ map
.m_len
- 1,
5796 /* Is delalloc data before next block in extent tree? */
5797 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
5798 ext4_lblk_t offset
= 0;
5800 if (es
.es_lblk
< lblk
)
5801 offset
= lblk
- es
.es_lblk
;
5802 result
->es_lblk
= es
.es_lblk
+ offset
;
5803 ext4_es_store_pblock(result
,
5804 ext4_es_pblock(&es
) + offset
);
5805 result
->es_len
= es
.es_len
- offset
;
5806 ext4_es_store_status(result
, ext4_es_status(&es
));
5810 /* There's a hole at m_lblk, advance us after it */
5811 map
.m_lblk
+= map
.m_len
;
5812 map_len
-= map
.m_len
;
5813 map
.m_len
= map_len
;