2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
51 struct ext4_inode_info
*ei
)
53 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
59 raw
->i_checksum_lo
= 0;
60 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
61 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
62 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
63 raw
->i_checksum_hi
= 0;
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
67 EXT4_INODE_SIZE(inode
->i_sb
));
69 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
70 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
71 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
72 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
77 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
78 struct ext4_inode_info
*ei
)
80 __u32 provided
, calculated
;
82 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
83 cpu_to_le32(EXT4_OS_LINUX
) ||
84 !ext4_has_metadata_csum(inode
->i_sb
))
87 provided
= le16_to_cpu(raw
->i_checksum_lo
);
88 calculated
= ext4_inode_csum(inode
, raw
, ei
);
89 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
90 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
91 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
95 return provided
== calculated
;
98 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
99 struct ext4_inode_info
*ei
)
103 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
104 cpu_to_le32(EXT4_OS_LINUX
) ||
105 !ext4_has_metadata_csum(inode
->i_sb
))
108 csum
= ext4_inode_csum(inode
, raw
, ei
);
109 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
110 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
111 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
112 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
118 trace_ext4_begin_ordered_truncate(inode
, new_size
);
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
125 if (!EXT4_I(inode
)->jinode
)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
128 EXT4_I(inode
)->jinode
,
132 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
133 unsigned int length
);
134 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
135 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
136 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode
*inode
)
144 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
145 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
147 if (ext4_has_inline_data(inode
))
150 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
170 jbd_debug(2, "restarting handle %p\n", handle
);
171 up_write(&EXT4_I(inode
)->i_data_sem
);
172 ret
= ext4_journal_restart(handle
, nblocks
);
173 down_write(&EXT4_I(inode
)->i_data_sem
);
174 ext4_discard_preallocations(inode
);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode
*inode
)
187 trace_ext4_evict_inode(inode
);
189 if (inode
->i_nlink
) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode
) &&
209 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
210 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
211 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
212 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
214 jbd2_complete_transaction(journal
, commit_tid
);
215 filemap_write_and_wait(&inode
->i_data
);
217 truncate_inode_pages_final(&inode
->i_data
);
219 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
223 if (is_bad_inode(inode
))
225 dquot_initialize(inode
);
227 if (ext4_should_order_data(inode
))
228 ext4_begin_ordered_truncate(inode
, 0);
229 truncate_inode_pages_final(&inode
->i_data
);
231 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode
->i_sb
);
238 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
239 ext4_blocks_for_truncate(inode
)+3);
240 if (IS_ERR(handle
)) {
241 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL
, inode
);
248 sb_end_intwrite(inode
->i_sb
);
253 ext4_handle_sync(handle
);
255 err
= ext4_mark_inode_dirty(handle
, inode
);
257 ext4_warning(inode
->i_sb
,
258 "couldn't mark inode dirty (err %d)", err
);
262 ext4_truncate(inode
);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle
, 3)) {
271 err
= ext4_journal_extend(handle
, 3);
273 err
= ext4_journal_restart(handle
, 3);
275 ext4_warning(inode
->i_sb
,
276 "couldn't extend journal (err %d)", err
);
278 ext4_journal_stop(handle
);
279 ext4_orphan_del(NULL
, inode
);
280 sb_end_intwrite(inode
->i_sb
);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle
, inode
);
294 EXT4_I(inode
)->i_dtime
= get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle
, inode
))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode
);
307 ext4_free_inode(handle
, inode
);
308 ext4_journal_stop(handle
);
309 sb_end_intwrite(inode
->i_sb
);
312 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
316 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
318 return &EXT4_I(inode
)->i_reserved_quota
;
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
326 void ext4_da_update_reserve_space(struct inode
*inode
,
327 int used
, int quota_claim
)
329 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
330 struct ext4_inode_info
*ei
= EXT4_I(inode
);
332 spin_lock(&ei
->i_block_reservation_lock
);
333 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
334 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
335 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
336 "with only %d reserved data blocks",
337 __func__
, inode
->i_ino
, used
,
338 ei
->i_reserved_data_blocks
);
340 used
= ei
->i_reserved_data_blocks
;
343 /* Update per-inode reservations */
344 ei
->i_reserved_data_blocks
-= used
;
345 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
347 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
349 /* Update quota subsystem for data blocks */
351 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
354 * We did fallocate with an offset that is already delayed
355 * allocated. So on delayed allocated writeback we should
356 * not re-claim the quota for fallocated blocks.
358 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
362 * If we have done all the pending block allocations and if
363 * there aren't any writers on the inode, we can discard the
364 * inode's preallocations.
366 if ((ei
->i_reserved_data_blocks
== 0) &&
367 (atomic_read(&inode
->i_writecount
) == 0))
368 ext4_discard_preallocations(inode
);
371 static int __check_block_validity(struct inode
*inode
, const char *func
,
373 struct ext4_map_blocks
*map
)
375 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
377 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
378 "lblock %lu mapped to illegal pblock "
379 "(length %d)", (unsigned long) map
->m_lblk
,
381 return -EFSCORRUPTED
;
386 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
391 if (ext4_encrypted_inode(inode
))
392 return ext4_encrypted_zeroout(inode
, lblk
, pblk
, len
);
394 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
401 #define check_block_validity(inode, map) \
402 __check_block_validity((inode), __func__, __LINE__, (map))
404 #ifdef ES_AGGRESSIVE_TEST
405 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
407 struct ext4_map_blocks
*es_map
,
408 struct ext4_map_blocks
*map
,
415 * There is a race window that the result is not the same.
416 * e.g. xfstests #223 when dioread_nolock enables. The reason
417 * is that we lookup a block mapping in extent status tree with
418 * out taking i_data_sem. So at the time the unwritten extent
419 * could be converted.
421 down_read(&EXT4_I(inode
)->i_data_sem
);
422 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
423 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
424 EXT4_GET_BLOCKS_KEEP_SIZE
);
426 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
427 EXT4_GET_BLOCKS_KEEP_SIZE
);
429 up_read((&EXT4_I(inode
)->i_data_sem
));
432 * We don't check m_len because extent will be collpased in status
433 * tree. So the m_len might not equal.
435 if (es_map
->m_lblk
!= map
->m_lblk
||
436 es_map
->m_flags
!= map
->m_flags
||
437 es_map
->m_pblk
!= map
->m_pblk
) {
438 printk("ES cache assertion failed for inode: %lu "
439 "es_cached ex [%d/%d/%llu/%x] != "
440 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
441 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
442 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
443 map
->m_len
, map
->m_pblk
, map
->m_flags
,
447 #endif /* ES_AGGRESSIVE_TEST */
450 * The ext4_map_blocks() function tries to look up the requested blocks,
451 * and returns if the blocks are already mapped.
453 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
454 * and store the allocated blocks in the result buffer head and mark it
457 * If file type is extents based, it will call ext4_ext_map_blocks(),
458 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
461 * On success, it returns the number of blocks being mapped or allocated.
462 * if create==0 and the blocks are pre-allocated and unwritten block,
463 * the result buffer head is unmapped. If the create ==1, it will make sure
464 * the buffer head is mapped.
466 * It returns 0 if plain look up failed (blocks have not been allocated), in
467 * that case, buffer head is unmapped
469 * It returns the error in case of allocation failure.
471 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
472 struct ext4_map_blocks
*map
, int flags
)
474 struct extent_status es
;
477 #ifdef ES_AGGRESSIVE_TEST
478 struct ext4_map_blocks orig_map
;
480 memcpy(&orig_map
, map
, sizeof(*map
));
484 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
485 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
486 (unsigned long) map
->m_lblk
);
489 * ext4_map_blocks returns an int, and m_len is an unsigned int
491 if (unlikely(map
->m_len
> INT_MAX
))
492 map
->m_len
= INT_MAX
;
494 /* We can handle the block number less than EXT_MAX_BLOCKS */
495 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
496 return -EFSCORRUPTED
;
498 /* Lookup extent status tree firstly */
499 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
500 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
501 map
->m_pblk
= ext4_es_pblock(&es
) +
502 map
->m_lblk
- es
.es_lblk
;
503 map
->m_flags
|= ext4_es_is_written(&es
) ?
504 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
505 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
506 if (retval
> map
->m_len
)
509 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
514 #ifdef ES_AGGRESSIVE_TEST
515 ext4_map_blocks_es_recheck(handle
, inode
, map
,
522 * Try to see if we can get the block without requesting a new
525 down_read(&EXT4_I(inode
)->i_data_sem
);
526 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
527 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
528 EXT4_GET_BLOCKS_KEEP_SIZE
);
530 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
531 EXT4_GET_BLOCKS_KEEP_SIZE
);
536 if (unlikely(retval
!= map
->m_len
)) {
537 ext4_warning(inode
->i_sb
,
538 "ES len assertion failed for inode "
539 "%lu: retval %d != map->m_len %d",
540 inode
->i_ino
, retval
, map
->m_len
);
544 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
545 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
546 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
547 !(status
& EXTENT_STATUS_WRITTEN
) &&
548 ext4_find_delalloc_range(inode
, map
->m_lblk
,
549 map
->m_lblk
+ map
->m_len
- 1))
550 status
|= EXTENT_STATUS_DELAYED
;
551 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
552 map
->m_len
, map
->m_pblk
, status
);
556 up_read((&EXT4_I(inode
)->i_data_sem
));
559 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
560 ret
= check_block_validity(inode
, map
);
565 /* If it is only a block(s) look up */
566 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
570 * Returns if the blocks have already allocated
572 * Note that if blocks have been preallocated
573 * ext4_ext_get_block() returns the create = 0
574 * with buffer head unmapped.
576 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
578 * If we need to convert extent to unwritten
579 * we continue and do the actual work in
580 * ext4_ext_map_blocks()
582 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
586 * Here we clear m_flags because after allocating an new extent,
587 * it will be set again.
589 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
592 * New blocks allocate and/or writing to unwritten extent
593 * will possibly result in updating i_data, so we take
594 * the write lock of i_data_sem, and call get_block()
595 * with create == 1 flag.
597 down_write(&EXT4_I(inode
)->i_data_sem
);
600 * We need to check for EXT4 here because migrate
601 * could have changed the inode type in between
603 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
604 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
606 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
608 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
610 * We allocated new blocks which will result in
611 * i_data's format changing. Force the migrate
612 * to fail by clearing migrate flags
614 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
618 * Update reserved blocks/metadata blocks after successful
619 * block allocation which had been deferred till now. We don't
620 * support fallocate for non extent files. So we can update
621 * reserve space here.
624 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
625 ext4_da_update_reserve_space(inode
, retval
, 1);
631 if (unlikely(retval
!= map
->m_len
)) {
632 ext4_warning(inode
->i_sb
,
633 "ES len assertion failed for inode "
634 "%lu: retval %d != map->m_len %d",
635 inode
->i_ino
, retval
, map
->m_len
);
640 * We have to zeroout blocks before inserting them into extent
641 * status tree. Otherwise someone could look them up there and
642 * use them before they are really zeroed.
644 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
645 map
->m_flags
& EXT4_MAP_MAPPED
&&
646 map
->m_flags
& EXT4_MAP_NEW
) {
647 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
648 map
->m_pblk
, map
->m_len
);
656 * If the extent has been zeroed out, we don't need to update
657 * extent status tree.
659 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
660 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
661 if (ext4_es_is_written(&es
))
664 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
665 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
666 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
667 !(status
& EXTENT_STATUS_WRITTEN
) &&
668 ext4_find_delalloc_range(inode
, map
->m_lblk
,
669 map
->m_lblk
+ map
->m_len
- 1))
670 status
|= EXTENT_STATUS_DELAYED
;
671 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
672 map
->m_pblk
, status
);
680 up_write((&EXT4_I(inode
)->i_data_sem
));
681 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
682 ret
= check_block_validity(inode
, map
);
690 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
691 * we have to be careful as someone else may be manipulating b_state as well.
693 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
695 unsigned long old_state
;
696 unsigned long new_state
;
698 flags
&= EXT4_MAP_FLAGS
;
700 /* Dummy buffer_head? Set non-atomically. */
702 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
706 * Someone else may be modifying b_state. Be careful! This is ugly but
707 * once we get rid of using bh as a container for mapping information
708 * to pass to / from get_block functions, this can go away.
711 old_state
= READ_ONCE(bh
->b_state
);
712 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
714 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
717 /* Maximum number of blocks we map for direct IO at once. */
718 #define DIO_MAX_BLOCKS 4096
720 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
721 struct buffer_head
*bh
, int flags
)
723 handle_t
*handle
= ext4_journal_current_handle();
724 struct ext4_map_blocks map
;
725 int ret
= 0, started
= 0;
728 if (ext4_has_inline_data(inode
))
732 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
734 if (flags
&& !handle
) {
735 /* Direct IO write... */
736 if (map
.m_len
> DIO_MAX_BLOCKS
)
737 map
.m_len
= DIO_MAX_BLOCKS
;
738 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
739 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
741 if (IS_ERR(handle
)) {
742 ret
= PTR_ERR(handle
);
748 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
750 ext4_io_end_t
*io_end
= ext4_inode_aio(inode
);
752 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
753 ext4_update_bh_state(bh
, map
.m_flags
);
754 if (io_end
&& io_end
->flag
& EXT4_IO_END_UNWRITTEN
)
755 set_buffer_defer_completion(bh
);
756 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
760 ext4_journal_stop(handle
);
764 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
765 struct buffer_head
*bh
, int create
)
767 return _ext4_get_block(inode
, iblock
, bh
,
768 create
? EXT4_GET_BLOCKS_CREATE
: 0);
772 * `handle' can be NULL if create is zero
774 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
775 ext4_lblk_t block
, int map_flags
)
777 struct ext4_map_blocks map
;
778 struct buffer_head
*bh
;
779 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
782 J_ASSERT(handle
!= NULL
|| create
== 0);
786 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
789 return create
? ERR_PTR(-ENOSPC
) : NULL
;
793 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
795 return ERR_PTR(-ENOMEM
);
796 if (map
.m_flags
& EXT4_MAP_NEW
) {
797 J_ASSERT(create
!= 0);
798 J_ASSERT(handle
!= NULL
);
801 * Now that we do not always journal data, we should
802 * keep in mind whether this should always journal the
803 * new buffer as metadata. For now, regular file
804 * writes use ext4_get_block instead, so it's not a
808 BUFFER_TRACE(bh
, "call get_create_access");
809 err
= ext4_journal_get_create_access(handle
, bh
);
814 if (!buffer_uptodate(bh
)) {
815 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
816 set_buffer_uptodate(bh
);
819 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
820 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
824 BUFFER_TRACE(bh
, "not a new buffer");
831 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
832 ext4_lblk_t block
, int map_flags
)
834 struct buffer_head
*bh
;
836 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
839 if (!bh
|| buffer_uptodate(bh
))
841 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
843 if (buffer_uptodate(bh
))
846 return ERR_PTR(-EIO
);
849 int ext4_walk_page_buffers(handle_t
*handle
,
850 struct buffer_head
*head
,
854 int (*fn
)(handle_t
*handle
,
855 struct buffer_head
*bh
))
857 struct buffer_head
*bh
;
858 unsigned block_start
, block_end
;
859 unsigned blocksize
= head
->b_size
;
861 struct buffer_head
*next
;
863 for (bh
= head
, block_start
= 0;
864 ret
== 0 && (bh
!= head
|| !block_start
);
865 block_start
= block_end
, bh
= next
) {
866 next
= bh
->b_this_page
;
867 block_end
= block_start
+ blocksize
;
868 if (block_end
<= from
|| block_start
>= to
) {
869 if (partial
&& !buffer_uptodate(bh
))
873 err
= (*fn
)(handle
, bh
);
881 * To preserve ordering, it is essential that the hole instantiation and
882 * the data write be encapsulated in a single transaction. We cannot
883 * close off a transaction and start a new one between the ext4_get_block()
884 * and the commit_write(). So doing the jbd2_journal_start at the start of
885 * prepare_write() is the right place.
887 * Also, this function can nest inside ext4_writepage(). In that case, we
888 * *know* that ext4_writepage() has generated enough buffer credits to do the
889 * whole page. So we won't block on the journal in that case, which is good,
890 * because the caller may be PF_MEMALLOC.
892 * By accident, ext4 can be reentered when a transaction is open via
893 * quota file writes. If we were to commit the transaction while thus
894 * reentered, there can be a deadlock - we would be holding a quota
895 * lock, and the commit would never complete if another thread had a
896 * transaction open and was blocking on the quota lock - a ranking
899 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
900 * will _not_ run commit under these circumstances because handle->h_ref
901 * is elevated. We'll still have enough credits for the tiny quotafile
904 int do_journal_get_write_access(handle_t
*handle
,
905 struct buffer_head
*bh
)
907 int dirty
= buffer_dirty(bh
);
910 if (!buffer_mapped(bh
) || buffer_freed(bh
))
913 * __block_write_begin() could have dirtied some buffers. Clean
914 * the dirty bit as jbd2_journal_get_write_access() could complain
915 * otherwise about fs integrity issues. Setting of the dirty bit
916 * by __block_write_begin() isn't a real problem here as we clear
917 * the bit before releasing a page lock and thus writeback cannot
918 * ever write the buffer.
921 clear_buffer_dirty(bh
);
922 BUFFER_TRACE(bh
, "get write access");
923 ret
= ext4_journal_get_write_access(handle
, bh
);
925 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
929 #ifdef CONFIG_EXT4_FS_ENCRYPTION
930 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
931 get_block_t
*get_block
)
933 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
934 unsigned to
= from
+ len
;
935 struct inode
*inode
= page
->mapping
->host
;
936 unsigned block_start
, block_end
;
939 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
941 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
942 bool decrypt
= false;
944 BUG_ON(!PageLocked(page
));
945 BUG_ON(from
> PAGE_CACHE_SIZE
);
946 BUG_ON(to
> PAGE_CACHE_SIZE
);
949 if (!page_has_buffers(page
))
950 create_empty_buffers(page
, blocksize
, 0);
951 head
= page_buffers(page
);
952 bbits
= ilog2(blocksize
);
953 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
955 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
956 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
957 block_end
= block_start
+ blocksize
;
958 if (block_end
<= from
|| block_start
>= to
) {
959 if (PageUptodate(page
)) {
960 if (!buffer_uptodate(bh
))
961 set_buffer_uptodate(bh
);
966 clear_buffer_new(bh
);
967 if (!buffer_mapped(bh
)) {
968 WARN_ON(bh
->b_size
!= blocksize
);
969 err
= get_block(inode
, block
, bh
, 1);
972 if (buffer_new(bh
)) {
973 unmap_underlying_metadata(bh
->b_bdev
,
975 if (PageUptodate(page
)) {
976 clear_buffer_new(bh
);
977 set_buffer_uptodate(bh
);
978 mark_buffer_dirty(bh
);
981 if (block_end
> to
|| block_start
< from
)
982 zero_user_segments(page
, to
, block_end
,
987 if (PageUptodate(page
)) {
988 if (!buffer_uptodate(bh
))
989 set_buffer_uptodate(bh
);
992 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
993 !buffer_unwritten(bh
) &&
994 (block_start
< from
|| block_end
> to
)) {
995 ll_rw_block(READ
, 1, &bh
);
997 decrypt
= ext4_encrypted_inode(inode
) &&
998 S_ISREG(inode
->i_mode
);
1002 * If we issued read requests, let them complete.
1004 while (wait_bh
> wait
) {
1005 wait_on_buffer(*--wait_bh
);
1006 if (!buffer_uptodate(*wait_bh
))
1010 page_zero_new_buffers(page
, from
, to
);
1012 err
= ext4_decrypt(page
);
1017 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1018 loff_t pos
, unsigned len
, unsigned flags
,
1019 struct page
**pagep
, void **fsdata
)
1021 struct inode
*inode
= mapping
->host
;
1022 int ret
, needed_blocks
;
1029 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1031 * Reserve one block more for addition to orphan list in case
1032 * we allocate blocks but write fails for some reason
1034 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1035 index
= pos
>> PAGE_CACHE_SHIFT
;
1036 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1039 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1040 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1049 * grab_cache_page_write_begin() can take a long time if the
1050 * system is thrashing due to memory pressure, or if the page
1051 * is being written back. So grab it first before we start
1052 * the transaction handle. This also allows us to allocate
1053 * the page (if needed) without using GFP_NOFS.
1056 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1062 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1063 if (IS_ERR(handle
)) {
1064 page_cache_release(page
);
1065 return PTR_ERR(handle
);
1069 if (page
->mapping
!= mapping
) {
1070 /* The page got truncated from under us */
1072 page_cache_release(page
);
1073 ext4_journal_stop(handle
);
1076 /* In case writeback began while the page was unlocked */
1077 wait_for_stable_page(page
);
1079 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1080 if (ext4_should_dioread_nolock(inode
))
1081 ret
= ext4_block_write_begin(page
, pos
, len
,
1082 ext4_get_block_write
);
1084 ret
= ext4_block_write_begin(page
, pos
, len
,
1087 if (ext4_should_dioread_nolock(inode
))
1088 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1090 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1092 if (!ret
&& ext4_should_journal_data(inode
)) {
1093 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1095 do_journal_get_write_access
);
1101 * __block_write_begin may have instantiated a few blocks
1102 * outside i_size. Trim these off again. Don't need
1103 * i_size_read because we hold i_mutex.
1105 * Add inode to orphan list in case we crash before
1108 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1109 ext4_orphan_add(handle
, inode
);
1111 ext4_journal_stop(handle
);
1112 if (pos
+ len
> inode
->i_size
) {
1113 ext4_truncate_failed_write(inode
);
1115 * If truncate failed early the inode might
1116 * still be on the orphan list; we need to
1117 * make sure the inode is removed from the
1118 * orphan list in that case.
1121 ext4_orphan_del(NULL
, inode
);
1124 if (ret
== -ENOSPC
&&
1125 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1127 page_cache_release(page
);
1134 /* For write_end() in data=journal mode */
1135 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1138 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1140 set_buffer_uptodate(bh
);
1141 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1142 clear_buffer_meta(bh
);
1143 clear_buffer_prio(bh
);
1148 * We need to pick up the new inode size which generic_commit_write gave us
1149 * `file' can be NULL - eg, when called from page_symlink().
1151 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1152 * buffers are managed internally.
1154 static int ext4_write_end(struct file
*file
,
1155 struct address_space
*mapping
,
1156 loff_t pos
, unsigned len
, unsigned copied
,
1157 struct page
*page
, void *fsdata
)
1159 handle_t
*handle
= ext4_journal_current_handle();
1160 struct inode
*inode
= mapping
->host
;
1161 loff_t old_size
= inode
->i_size
;
1163 int i_size_changed
= 0;
1165 trace_ext4_write_end(inode
, pos
, len
, copied
);
1166 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1167 ret
= ext4_jbd2_file_inode(handle
, inode
);
1170 page_cache_release(page
);
1175 if (ext4_has_inline_data(inode
)) {
1176 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1182 copied
= block_write_end(file
, mapping
, pos
,
1183 len
, copied
, page
, fsdata
);
1185 * it's important to update i_size while still holding page lock:
1186 * page writeout could otherwise come in and zero beyond i_size.
1188 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1190 page_cache_release(page
);
1193 pagecache_isize_extended(inode
, old_size
, pos
);
1195 * Don't mark the inode dirty under page lock. First, it unnecessarily
1196 * makes the holding time of page lock longer. Second, it forces lock
1197 * ordering of page lock and transaction start for journaling
1201 ext4_mark_inode_dirty(handle
, inode
);
1203 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1204 /* if we have allocated more blocks and copied
1205 * less. We will have blocks allocated outside
1206 * inode->i_size. So truncate them
1208 ext4_orphan_add(handle
, inode
);
1210 ret2
= ext4_journal_stop(handle
);
1214 if (pos
+ len
> inode
->i_size
) {
1215 ext4_truncate_failed_write(inode
);
1217 * If truncate failed early the inode might still be
1218 * on the orphan list; we need to make sure the inode
1219 * is removed from the orphan list in that case.
1222 ext4_orphan_del(NULL
, inode
);
1225 return ret
? ret
: copied
;
1229 * This is a private version of page_zero_new_buffers() which doesn't
1230 * set the buffer to be dirty, since in data=journalled mode we need
1231 * to call ext4_handle_dirty_metadata() instead.
1233 static void zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1235 unsigned int block_start
= 0, block_end
;
1236 struct buffer_head
*head
, *bh
;
1238 bh
= head
= page_buffers(page
);
1240 block_end
= block_start
+ bh
->b_size
;
1241 if (buffer_new(bh
)) {
1242 if (block_end
> from
&& block_start
< to
) {
1243 if (!PageUptodate(page
)) {
1244 unsigned start
, size
;
1246 start
= max(from
, block_start
);
1247 size
= min(to
, block_end
) - start
;
1249 zero_user(page
, start
, size
);
1250 set_buffer_uptodate(bh
);
1252 clear_buffer_new(bh
);
1255 block_start
= block_end
;
1256 bh
= bh
->b_this_page
;
1257 } while (bh
!= head
);
1260 static int ext4_journalled_write_end(struct file
*file
,
1261 struct address_space
*mapping
,
1262 loff_t pos
, unsigned len
, unsigned copied
,
1263 struct page
*page
, void *fsdata
)
1265 handle_t
*handle
= ext4_journal_current_handle();
1266 struct inode
*inode
= mapping
->host
;
1267 loff_t old_size
= inode
->i_size
;
1271 int size_changed
= 0;
1273 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1274 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1277 BUG_ON(!ext4_handle_valid(handle
));
1279 if (ext4_has_inline_data(inode
))
1280 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1284 if (!PageUptodate(page
))
1286 zero_new_buffers(page
, from
+copied
, to
);
1289 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1290 to
, &partial
, write_end_fn
);
1292 SetPageUptodate(page
);
1294 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1295 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1296 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1298 page_cache_release(page
);
1301 pagecache_isize_extended(inode
, old_size
, pos
);
1304 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1309 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1310 /* if we have allocated more blocks and copied
1311 * less. We will have blocks allocated outside
1312 * inode->i_size. So truncate them
1314 ext4_orphan_add(handle
, inode
);
1316 ret2
= ext4_journal_stop(handle
);
1319 if (pos
+ len
> inode
->i_size
) {
1320 ext4_truncate_failed_write(inode
);
1322 * If truncate failed early the inode might still be
1323 * on the orphan list; we need to make sure the inode
1324 * is removed from the orphan list in that case.
1327 ext4_orphan_del(NULL
, inode
);
1330 return ret
? ret
: copied
;
1334 * Reserve space for a single cluster
1336 static int ext4_da_reserve_space(struct inode
*inode
)
1338 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1339 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1343 * We will charge metadata quota at writeout time; this saves
1344 * us from metadata over-estimation, though we may go over by
1345 * a small amount in the end. Here we just reserve for data.
1347 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1351 spin_lock(&ei
->i_block_reservation_lock
);
1352 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1353 spin_unlock(&ei
->i_block_reservation_lock
);
1354 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1357 ei
->i_reserved_data_blocks
++;
1358 trace_ext4_da_reserve_space(inode
);
1359 spin_unlock(&ei
->i_block_reservation_lock
);
1361 return 0; /* success */
1364 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1366 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1367 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1370 return; /* Nothing to release, exit */
1372 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1374 trace_ext4_da_release_space(inode
, to_free
);
1375 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1377 * if there aren't enough reserved blocks, then the
1378 * counter is messed up somewhere. Since this
1379 * function is called from invalidate page, it's
1380 * harmless to return without any action.
1382 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1383 "ino %lu, to_free %d with only %d reserved "
1384 "data blocks", inode
->i_ino
, to_free
,
1385 ei
->i_reserved_data_blocks
);
1387 to_free
= ei
->i_reserved_data_blocks
;
1389 ei
->i_reserved_data_blocks
-= to_free
;
1391 /* update fs dirty data blocks counter */
1392 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1394 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1396 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1399 static void ext4_da_page_release_reservation(struct page
*page
,
1400 unsigned int offset
,
1401 unsigned int length
)
1403 int to_release
= 0, contiguous_blks
= 0;
1404 struct buffer_head
*head
, *bh
;
1405 unsigned int curr_off
= 0;
1406 struct inode
*inode
= page
->mapping
->host
;
1407 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1408 unsigned int stop
= offset
+ length
;
1412 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1414 head
= page_buffers(page
);
1417 unsigned int next_off
= curr_off
+ bh
->b_size
;
1419 if (next_off
> stop
)
1422 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1425 clear_buffer_delay(bh
);
1426 } else if (contiguous_blks
) {
1427 lblk
= page
->index
<<
1428 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1429 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1431 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1432 contiguous_blks
= 0;
1434 curr_off
= next_off
;
1435 } while ((bh
= bh
->b_this_page
) != head
);
1437 if (contiguous_blks
) {
1438 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1439 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1440 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1443 /* If we have released all the blocks belonging to a cluster, then we
1444 * need to release the reserved space for that cluster. */
1445 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1446 while (num_clusters
> 0) {
1447 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1448 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1449 if (sbi
->s_cluster_ratio
== 1 ||
1450 !ext4_find_delalloc_cluster(inode
, lblk
))
1451 ext4_da_release_space(inode
, 1);
1458 * Delayed allocation stuff
1461 struct mpage_da_data
{
1462 struct inode
*inode
;
1463 struct writeback_control
*wbc
;
1465 pgoff_t first_page
; /* The first page to write */
1466 pgoff_t next_page
; /* Current page to examine */
1467 pgoff_t last_page
; /* Last page to examine */
1469 * Extent to map - this can be after first_page because that can be
1470 * fully mapped. We somewhat abuse m_flags to store whether the extent
1471 * is delalloc or unwritten.
1473 struct ext4_map_blocks map
;
1474 struct ext4_io_submit io_submit
; /* IO submission data */
1477 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1482 struct pagevec pvec
;
1483 struct inode
*inode
= mpd
->inode
;
1484 struct address_space
*mapping
= inode
->i_mapping
;
1486 /* This is necessary when next_page == 0. */
1487 if (mpd
->first_page
>= mpd
->next_page
)
1490 index
= mpd
->first_page
;
1491 end
= mpd
->next_page
- 1;
1493 ext4_lblk_t start
, last
;
1494 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1495 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1496 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1499 pagevec_init(&pvec
, 0);
1500 while (index
<= end
) {
1501 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1504 for (i
= 0; i
< nr_pages
; i
++) {
1505 struct page
*page
= pvec
.pages
[i
];
1506 if (page
->index
> end
)
1508 BUG_ON(!PageLocked(page
));
1509 BUG_ON(PageWriteback(page
));
1511 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1512 ClearPageUptodate(page
);
1516 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1517 pagevec_release(&pvec
);
1521 static void ext4_print_free_blocks(struct inode
*inode
)
1523 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1524 struct super_block
*sb
= inode
->i_sb
;
1525 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1527 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1528 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1529 ext4_count_free_clusters(sb
)));
1530 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1531 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1532 (long long) EXT4_C2B(EXT4_SB(sb
),
1533 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1534 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1535 (long long) EXT4_C2B(EXT4_SB(sb
),
1536 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1537 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1538 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1539 ei
->i_reserved_data_blocks
);
1543 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1545 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1549 * This function is grabs code from the very beginning of
1550 * ext4_map_blocks, but assumes that the caller is from delayed write
1551 * time. This function looks up the requested blocks and sets the
1552 * buffer delay bit under the protection of i_data_sem.
1554 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1555 struct ext4_map_blocks
*map
,
1556 struct buffer_head
*bh
)
1558 struct extent_status es
;
1560 sector_t invalid_block
= ~((sector_t
) 0xffff);
1561 #ifdef ES_AGGRESSIVE_TEST
1562 struct ext4_map_blocks orig_map
;
1564 memcpy(&orig_map
, map
, sizeof(*map
));
1567 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1571 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1572 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1573 (unsigned long) map
->m_lblk
);
1575 /* Lookup extent status tree firstly */
1576 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1577 if (ext4_es_is_hole(&es
)) {
1579 down_read(&EXT4_I(inode
)->i_data_sem
);
1584 * Delayed extent could be allocated by fallocate.
1585 * So we need to check it.
1587 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1588 map_bh(bh
, inode
->i_sb
, invalid_block
);
1590 set_buffer_delay(bh
);
1594 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1595 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1596 if (retval
> map
->m_len
)
1597 retval
= map
->m_len
;
1598 map
->m_len
= retval
;
1599 if (ext4_es_is_written(&es
))
1600 map
->m_flags
|= EXT4_MAP_MAPPED
;
1601 else if (ext4_es_is_unwritten(&es
))
1602 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1606 #ifdef ES_AGGRESSIVE_TEST
1607 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1613 * Try to see if we can get the block without requesting a new
1614 * file system block.
1616 down_read(&EXT4_I(inode
)->i_data_sem
);
1617 if (ext4_has_inline_data(inode
))
1619 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1620 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1622 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1628 * XXX: __block_prepare_write() unmaps passed block,
1632 * If the block was allocated from previously allocated cluster,
1633 * then we don't need to reserve it again. However we still need
1634 * to reserve metadata for every block we're going to write.
1636 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1637 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1638 ret
= ext4_da_reserve_space(inode
);
1640 /* not enough space to reserve */
1646 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1647 ~0, EXTENT_STATUS_DELAYED
);
1653 map_bh(bh
, inode
->i_sb
, invalid_block
);
1655 set_buffer_delay(bh
);
1656 } else if (retval
> 0) {
1658 unsigned int status
;
1660 if (unlikely(retval
!= map
->m_len
)) {
1661 ext4_warning(inode
->i_sb
,
1662 "ES len assertion failed for inode "
1663 "%lu: retval %d != map->m_len %d",
1664 inode
->i_ino
, retval
, map
->m_len
);
1668 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1669 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1670 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1671 map
->m_pblk
, status
);
1677 up_read((&EXT4_I(inode
)->i_data_sem
));
1683 * This is a special get_block_t callback which is used by
1684 * ext4_da_write_begin(). It will either return mapped block or
1685 * reserve space for a single block.
1687 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1688 * We also have b_blocknr = -1 and b_bdev initialized properly
1690 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1691 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1692 * initialized properly.
1694 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1695 struct buffer_head
*bh
, int create
)
1697 struct ext4_map_blocks map
;
1700 BUG_ON(create
== 0);
1701 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1703 map
.m_lblk
= iblock
;
1707 * first, we need to know whether the block is allocated already
1708 * preallocated blocks are unmapped but should treated
1709 * the same as allocated blocks.
1711 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1715 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1716 ext4_update_bh_state(bh
, map
.m_flags
);
1718 if (buffer_unwritten(bh
)) {
1719 /* A delayed write to unwritten bh should be marked
1720 * new and mapped. Mapped ensures that we don't do
1721 * get_block multiple times when we write to the same
1722 * offset and new ensures that we do proper zero out
1723 * for partial write.
1726 set_buffer_mapped(bh
);
1731 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1737 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1743 static int __ext4_journalled_writepage(struct page
*page
,
1746 struct address_space
*mapping
= page
->mapping
;
1747 struct inode
*inode
= mapping
->host
;
1748 struct buffer_head
*page_bufs
= NULL
;
1749 handle_t
*handle
= NULL
;
1750 int ret
= 0, err
= 0;
1751 int inline_data
= ext4_has_inline_data(inode
);
1752 struct buffer_head
*inode_bh
= NULL
;
1754 ClearPageChecked(page
);
1757 BUG_ON(page
->index
!= 0);
1758 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1759 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1760 if (inode_bh
== NULL
)
1763 page_bufs
= page_buffers(page
);
1768 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1772 * We need to release the page lock before we start the
1773 * journal, so grab a reference so the page won't disappear
1774 * out from under us.
1779 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1780 ext4_writepage_trans_blocks(inode
));
1781 if (IS_ERR(handle
)) {
1782 ret
= PTR_ERR(handle
);
1784 goto out_no_pagelock
;
1786 BUG_ON(!ext4_handle_valid(handle
));
1790 if (page
->mapping
!= mapping
) {
1791 /* The page got truncated from under us */
1792 ext4_journal_stop(handle
);
1798 BUFFER_TRACE(inode_bh
, "get write access");
1799 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1801 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1804 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1805 do_journal_get_write_access
);
1807 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1812 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1813 err
= ext4_journal_stop(handle
);
1817 if (!ext4_has_inline_data(inode
))
1818 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1820 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1829 * Note that we don't need to start a transaction unless we're journaling data
1830 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1831 * need to file the inode to the transaction's list in ordered mode because if
1832 * we are writing back data added by write(), the inode is already there and if
1833 * we are writing back data modified via mmap(), no one guarantees in which
1834 * transaction the data will hit the disk. In case we are journaling data, we
1835 * cannot start transaction directly because transaction start ranks above page
1836 * lock so we have to do some magic.
1838 * This function can get called via...
1839 * - ext4_writepages after taking page lock (have journal handle)
1840 * - journal_submit_inode_data_buffers (no journal handle)
1841 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1842 * - grab_page_cache when doing write_begin (have journal handle)
1844 * We don't do any block allocation in this function. If we have page with
1845 * multiple blocks we need to write those buffer_heads that are mapped. This
1846 * is important for mmaped based write. So if we do with blocksize 1K
1847 * truncate(f, 1024);
1848 * a = mmap(f, 0, 4096);
1850 * truncate(f, 4096);
1851 * we have in the page first buffer_head mapped via page_mkwrite call back
1852 * but other buffer_heads would be unmapped but dirty (dirty done via the
1853 * do_wp_page). So writepage should write the first block. If we modify
1854 * the mmap area beyond 1024 we will again get a page_fault and the
1855 * page_mkwrite callback will do the block allocation and mark the
1856 * buffer_heads mapped.
1858 * We redirty the page if we have any buffer_heads that is either delay or
1859 * unwritten in the page.
1861 * We can get recursively called as show below.
1863 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1866 * But since we don't do any block allocation we should not deadlock.
1867 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1869 static int ext4_writepage(struct page
*page
,
1870 struct writeback_control
*wbc
)
1875 struct buffer_head
*page_bufs
= NULL
;
1876 struct inode
*inode
= page
->mapping
->host
;
1877 struct ext4_io_submit io_submit
;
1878 bool keep_towrite
= false;
1880 trace_ext4_writepage(page
);
1881 size
= i_size_read(inode
);
1882 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1883 len
= size
& ~PAGE_CACHE_MASK
;
1885 len
= PAGE_CACHE_SIZE
;
1887 page_bufs
= page_buffers(page
);
1889 * We cannot do block allocation or other extent handling in this
1890 * function. If there are buffers needing that, we have to redirty
1891 * the page. But we may reach here when we do a journal commit via
1892 * journal_submit_inode_data_buffers() and in that case we must write
1893 * allocated buffers to achieve data=ordered mode guarantees.
1895 * Also, if there is only one buffer per page (the fs block
1896 * size == the page size), if one buffer needs block
1897 * allocation or needs to modify the extent tree to clear the
1898 * unwritten flag, we know that the page can't be written at
1899 * all, so we might as well refuse the write immediately.
1900 * Unfortunately if the block size != page size, we can't as
1901 * easily detect this case using ext4_walk_page_buffers(), but
1902 * for the extremely common case, this is an optimization that
1903 * skips a useless round trip through ext4_bio_write_page().
1905 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1906 ext4_bh_delay_or_unwritten
)) {
1907 redirty_page_for_writepage(wbc
, page
);
1908 if ((current
->flags
& PF_MEMALLOC
) ||
1909 (inode
->i_sb
->s_blocksize
== PAGE_CACHE_SIZE
)) {
1911 * For memory cleaning there's no point in writing only
1912 * some buffers. So just bail out. Warn if we came here
1913 * from direct reclaim.
1915 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1920 keep_towrite
= true;
1923 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1925 * It's mmapped pagecache. Add buffers and journal it. There
1926 * doesn't seem much point in redirtying the page here.
1928 return __ext4_journalled_writepage(page
, len
);
1930 ext4_io_submit_init(&io_submit
, wbc
);
1931 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1932 if (!io_submit
.io_end
) {
1933 redirty_page_for_writepage(wbc
, page
);
1937 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
1938 ext4_io_submit(&io_submit
);
1939 /* Drop io_end reference we got from init */
1940 ext4_put_io_end_defer(io_submit
.io_end
);
1944 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1947 loff_t size
= i_size_read(mpd
->inode
);
1950 BUG_ON(page
->index
!= mpd
->first_page
);
1951 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1952 len
= size
& ~PAGE_CACHE_MASK
;
1954 len
= PAGE_CACHE_SIZE
;
1955 clear_page_dirty_for_io(page
);
1956 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
1958 mpd
->wbc
->nr_to_write
--;
1964 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1967 * mballoc gives us at most this number of blocks...
1968 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1969 * The rest of mballoc seems to handle chunks up to full group size.
1971 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1974 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1976 * @mpd - extent of blocks
1977 * @lblk - logical number of the block in the file
1978 * @bh - buffer head we want to add to the extent
1980 * The function is used to collect contig. blocks in the same state. If the
1981 * buffer doesn't require mapping for writeback and we haven't started the
1982 * extent of buffers to map yet, the function returns 'true' immediately - the
1983 * caller can write the buffer right away. Otherwise the function returns true
1984 * if the block has been added to the extent, false if the block couldn't be
1987 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
1988 struct buffer_head
*bh
)
1990 struct ext4_map_blocks
*map
= &mpd
->map
;
1992 /* Buffer that doesn't need mapping for writeback? */
1993 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
1994 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
1995 /* So far no extent to map => we write the buffer right away */
1996 if (map
->m_len
== 0)
2001 /* First block in the extent? */
2002 if (map
->m_len
== 0) {
2005 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2009 /* Don't go larger than mballoc is willing to allocate */
2010 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2013 /* Can we merge the block to our big extent? */
2014 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2015 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2023 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2025 * @mpd - extent of blocks for mapping
2026 * @head - the first buffer in the page
2027 * @bh - buffer we should start processing from
2028 * @lblk - logical number of the block in the file corresponding to @bh
2030 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2031 * the page for IO if all buffers in this page were mapped and there's no
2032 * accumulated extent of buffers to map or add buffers in the page to the
2033 * extent of buffers to map. The function returns 1 if the caller can continue
2034 * by processing the next page, 0 if it should stop adding buffers to the
2035 * extent to map because we cannot extend it anymore. It can also return value
2036 * < 0 in case of error during IO submission.
2038 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2039 struct buffer_head
*head
,
2040 struct buffer_head
*bh
,
2043 struct inode
*inode
= mpd
->inode
;
2045 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2046 >> inode
->i_blkbits
;
2049 BUG_ON(buffer_locked(bh
));
2051 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2052 /* Found extent to map? */
2055 /* Everything mapped so far and we hit EOF */
2058 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2059 /* So far everything mapped? Submit the page for IO. */
2060 if (mpd
->map
.m_len
== 0) {
2061 err
= mpage_submit_page(mpd
, head
->b_page
);
2065 return lblk
< blocks
;
2069 * mpage_map_buffers - update buffers corresponding to changed extent and
2070 * submit fully mapped pages for IO
2072 * @mpd - description of extent to map, on return next extent to map
2074 * Scan buffers corresponding to changed extent (we expect corresponding pages
2075 * to be already locked) and update buffer state according to new extent state.
2076 * We map delalloc buffers to their physical location, clear unwritten bits,
2077 * and mark buffers as uninit when we perform writes to unwritten extents
2078 * and do extent conversion after IO is finished. If the last page is not fully
2079 * mapped, we update @map to the next extent in the last page that needs
2080 * mapping. Otherwise we submit the page for IO.
2082 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2084 struct pagevec pvec
;
2086 struct inode
*inode
= mpd
->inode
;
2087 struct buffer_head
*head
, *bh
;
2088 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
2094 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2095 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2096 lblk
= start
<< bpp_bits
;
2097 pblock
= mpd
->map
.m_pblk
;
2099 pagevec_init(&pvec
, 0);
2100 while (start
<= end
) {
2101 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2105 for (i
= 0; i
< nr_pages
; i
++) {
2106 struct page
*page
= pvec
.pages
[i
];
2108 if (page
->index
> end
)
2110 /* Up to 'end' pages must be contiguous */
2111 BUG_ON(page
->index
!= start
);
2112 bh
= head
= page_buffers(page
);
2114 if (lblk
< mpd
->map
.m_lblk
)
2116 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2118 * Buffer after end of mapped extent.
2119 * Find next buffer in the page to map.
2122 mpd
->map
.m_flags
= 0;
2124 * FIXME: If dioread_nolock supports
2125 * blocksize < pagesize, we need to make
2126 * sure we add size mapped so far to
2127 * io_end->size as the following call
2128 * can submit the page for IO.
2130 err
= mpage_process_page_bufs(mpd
, head
,
2132 pagevec_release(&pvec
);
2137 if (buffer_delay(bh
)) {
2138 clear_buffer_delay(bh
);
2139 bh
->b_blocknr
= pblock
++;
2141 clear_buffer_unwritten(bh
);
2142 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2145 * FIXME: This is going to break if dioread_nolock
2146 * supports blocksize < pagesize as we will try to
2147 * convert potentially unmapped parts of inode.
2149 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
2150 /* Page fully mapped - let IO run! */
2151 err
= mpage_submit_page(mpd
, page
);
2153 pagevec_release(&pvec
);
2158 pagevec_release(&pvec
);
2160 /* Extent fully mapped and matches with page boundary. We are done. */
2162 mpd
->map
.m_flags
= 0;
2166 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2168 struct inode
*inode
= mpd
->inode
;
2169 struct ext4_map_blocks
*map
= &mpd
->map
;
2170 int get_blocks_flags
;
2171 int err
, dioread_nolock
;
2173 trace_ext4_da_write_pages_extent(inode
, map
);
2175 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2176 * to convert an unwritten extent to be initialized (in the case
2177 * where we have written into one or more preallocated blocks). It is
2178 * possible that we're going to need more metadata blocks than
2179 * previously reserved. However we must not fail because we're in
2180 * writeback and there is nothing we can do about it so it might result
2181 * in data loss. So use reserved blocks to allocate metadata if
2184 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2185 * the blocks in question are delalloc blocks. This indicates
2186 * that the blocks and quotas has already been checked when
2187 * the data was copied into the page cache.
2189 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2190 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
2191 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2193 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2194 if (map
->m_flags
& (1 << BH_Delay
))
2195 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2197 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2200 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2201 if (!mpd
->io_submit
.io_end
->handle
&&
2202 ext4_handle_valid(handle
)) {
2203 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2204 handle
->h_rsv_handle
= NULL
;
2206 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2209 BUG_ON(map
->m_len
== 0);
2210 if (map
->m_flags
& EXT4_MAP_NEW
) {
2211 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2214 for (i
= 0; i
< map
->m_len
; i
++)
2215 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2221 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2222 * mpd->len and submit pages underlying it for IO
2224 * @handle - handle for journal operations
2225 * @mpd - extent to map
2226 * @give_up_on_write - we set this to true iff there is a fatal error and there
2227 * is no hope of writing the data. The caller should discard
2228 * dirty pages to avoid infinite loops.
2230 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2231 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2232 * them to initialized or split the described range from larger unwritten
2233 * extent. Note that we need not map all the described range since allocation
2234 * can return less blocks or the range is covered by more unwritten extents. We
2235 * cannot map more because we are limited by reserved transaction credits. On
2236 * the other hand we always make sure that the last touched page is fully
2237 * mapped so that it can be written out (and thus forward progress is
2238 * guaranteed). After mapping we submit all mapped pages for IO.
2240 static int mpage_map_and_submit_extent(handle_t
*handle
,
2241 struct mpage_da_data
*mpd
,
2242 bool *give_up_on_write
)
2244 struct inode
*inode
= mpd
->inode
;
2245 struct ext4_map_blocks
*map
= &mpd
->map
;
2250 mpd
->io_submit
.io_end
->offset
=
2251 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2253 err
= mpage_map_one_extent(handle
, mpd
);
2255 struct super_block
*sb
= inode
->i_sb
;
2257 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2258 goto invalidate_dirty_pages
;
2260 * Let the uper layers retry transient errors.
2261 * In the case of ENOSPC, if ext4_count_free_blocks()
2262 * is non-zero, a commit should free up blocks.
2264 if ((err
== -ENOMEM
) ||
2265 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2267 goto update_disksize
;
2270 ext4_msg(sb
, KERN_CRIT
,
2271 "Delayed block allocation failed for "
2272 "inode %lu at logical offset %llu with"
2273 " max blocks %u with error %d",
2275 (unsigned long long)map
->m_lblk
,
2276 (unsigned)map
->m_len
, -err
);
2277 ext4_msg(sb
, KERN_CRIT
,
2278 "This should not happen!! Data will "
2281 ext4_print_free_blocks(inode
);
2282 invalidate_dirty_pages
:
2283 *give_up_on_write
= true;
2288 * Update buffer state, submit mapped pages, and get us new
2291 err
= mpage_map_and_submit_buffers(mpd
);
2293 goto update_disksize
;
2294 } while (map
->m_len
);
2298 * Update on-disk size after IO is submitted. Races with
2299 * truncate are avoided by checking i_size under i_data_sem.
2301 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2302 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2306 down_write(&EXT4_I(inode
)->i_data_sem
);
2307 i_size
= i_size_read(inode
);
2308 if (disksize
> i_size
)
2310 if (disksize
> EXT4_I(inode
)->i_disksize
)
2311 EXT4_I(inode
)->i_disksize
= disksize
;
2312 err2
= ext4_mark_inode_dirty(handle
, inode
);
2313 up_write(&EXT4_I(inode
)->i_data_sem
);
2315 ext4_error(inode
->i_sb
,
2316 "Failed to mark inode %lu dirty",
2325 * Calculate the total number of credits to reserve for one writepages
2326 * iteration. This is called from ext4_writepages(). We map an extent of
2327 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2328 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2329 * bpp - 1 blocks in bpp different extents.
2331 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2333 int bpp
= ext4_journal_blocks_per_page(inode
);
2335 return ext4_meta_trans_blocks(inode
,
2336 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2340 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2341 * and underlying extent to map
2343 * @mpd - where to look for pages
2345 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2346 * IO immediately. When we find a page which isn't mapped we start accumulating
2347 * extent of buffers underlying these pages that needs mapping (formed by
2348 * either delayed or unwritten buffers). We also lock the pages containing
2349 * these buffers. The extent found is returned in @mpd structure (starting at
2350 * mpd->lblk with length mpd->len blocks).
2352 * Note that this function can attach bios to one io_end structure which are
2353 * neither logically nor physically contiguous. Although it may seem as an
2354 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2355 * case as we need to track IO to all buffers underlying a page in one io_end.
2357 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2359 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2360 struct pagevec pvec
;
2361 unsigned int nr_pages
;
2362 long left
= mpd
->wbc
->nr_to_write
;
2363 pgoff_t index
= mpd
->first_page
;
2364 pgoff_t end
= mpd
->last_page
;
2367 int blkbits
= mpd
->inode
->i_blkbits
;
2369 struct buffer_head
*head
;
2371 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2372 tag
= PAGECACHE_TAG_TOWRITE
;
2374 tag
= PAGECACHE_TAG_DIRTY
;
2376 pagevec_init(&pvec
, 0);
2378 mpd
->next_page
= index
;
2379 while (index
<= end
) {
2380 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2381 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2385 for (i
= 0; i
< nr_pages
; i
++) {
2386 struct page
*page
= pvec
.pages
[i
];
2389 * At this point, the page may be truncated or
2390 * invalidated (changing page->mapping to NULL), or
2391 * even swizzled back from swapper_space to tmpfs file
2392 * mapping. However, page->index will not change
2393 * because we have a reference on the page.
2395 if (page
->index
> end
)
2399 * Accumulated enough dirty pages? This doesn't apply
2400 * to WB_SYNC_ALL mode. For integrity sync we have to
2401 * keep going because someone may be concurrently
2402 * dirtying pages, and we might have synced a lot of
2403 * newly appeared dirty pages, but have not synced all
2404 * of the old dirty pages.
2406 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2409 /* If we can't merge this page, we are done. */
2410 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2415 * If the page is no longer dirty, or its mapping no
2416 * longer corresponds to inode we are writing (which
2417 * means it has been truncated or invalidated), or the
2418 * page is already under writeback and we are not doing
2419 * a data integrity writeback, skip the page
2421 if (!PageDirty(page
) ||
2422 (PageWriteback(page
) &&
2423 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2424 unlikely(page
->mapping
!= mapping
)) {
2429 wait_on_page_writeback(page
);
2430 BUG_ON(PageWriteback(page
));
2432 if (mpd
->map
.m_len
== 0)
2433 mpd
->first_page
= page
->index
;
2434 mpd
->next_page
= page
->index
+ 1;
2435 /* Add all dirty buffers to mpd */
2436 lblk
= ((ext4_lblk_t
)page
->index
) <<
2437 (PAGE_CACHE_SHIFT
- blkbits
);
2438 head
= page_buffers(page
);
2439 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2445 pagevec_release(&pvec
);
2450 pagevec_release(&pvec
);
2454 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2457 struct address_space
*mapping
= data
;
2458 int ret
= ext4_writepage(page
, wbc
);
2459 mapping_set_error(mapping
, ret
);
2463 static int ext4_writepages(struct address_space
*mapping
,
2464 struct writeback_control
*wbc
)
2466 pgoff_t writeback_index
= 0;
2467 long nr_to_write
= wbc
->nr_to_write
;
2468 int range_whole
= 0;
2470 handle_t
*handle
= NULL
;
2471 struct mpage_da_data mpd
;
2472 struct inode
*inode
= mapping
->host
;
2473 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2474 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2476 struct blk_plug plug
;
2477 bool give_up_on_write
= false;
2479 trace_ext4_writepages(inode
, wbc
);
2482 * No pages to write? This is mainly a kludge to avoid starting
2483 * a transaction for special inodes like journal inode on last iput()
2484 * because that could violate lock ordering on umount
2486 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2487 goto out_writepages
;
2489 if (ext4_should_journal_data(inode
)) {
2490 struct blk_plug plug
;
2492 blk_start_plug(&plug
);
2493 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2494 blk_finish_plug(&plug
);
2495 goto out_writepages
;
2499 * If the filesystem has aborted, it is read-only, so return
2500 * right away instead of dumping stack traces later on that
2501 * will obscure the real source of the problem. We test
2502 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2503 * the latter could be true if the filesystem is mounted
2504 * read-only, and in that case, ext4_writepages should
2505 * *never* be called, so if that ever happens, we would want
2508 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2510 goto out_writepages
;
2513 if (ext4_should_dioread_nolock(inode
)) {
2515 * We may need to convert up to one extent per block in
2516 * the page and we may dirty the inode.
2518 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2522 * If we have inline data and arrive here, it means that
2523 * we will soon create the block for the 1st page, so
2524 * we'd better clear the inline data here.
2526 if (ext4_has_inline_data(inode
)) {
2527 /* Just inode will be modified... */
2528 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2529 if (IS_ERR(handle
)) {
2530 ret
= PTR_ERR(handle
);
2531 goto out_writepages
;
2533 BUG_ON(ext4_test_inode_state(inode
,
2534 EXT4_STATE_MAY_INLINE_DATA
));
2535 ext4_destroy_inline_data(handle
, inode
);
2536 ext4_journal_stop(handle
);
2539 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2542 if (wbc
->range_cyclic
) {
2543 writeback_index
= mapping
->writeback_index
;
2544 if (writeback_index
)
2546 mpd
.first_page
= writeback_index
;
2549 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2550 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2555 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2557 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2558 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2560 blk_start_plug(&plug
);
2561 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2562 /* For each extent of pages we use new io_end */
2563 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2564 if (!mpd
.io_submit
.io_end
) {
2570 * We have two constraints: We find one extent to map and we
2571 * must always write out whole page (makes a difference when
2572 * blocksize < pagesize) so that we don't block on IO when we
2573 * try to write out the rest of the page. Journalled mode is
2574 * not supported by delalloc.
2576 BUG_ON(ext4_should_journal_data(inode
));
2577 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2579 /* start a new transaction */
2580 handle
= ext4_journal_start_with_reserve(inode
,
2581 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2582 if (IS_ERR(handle
)) {
2583 ret
= PTR_ERR(handle
);
2584 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2585 "%ld pages, ino %lu; err %d", __func__
,
2586 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2587 /* Release allocated io_end */
2588 ext4_put_io_end(mpd
.io_submit
.io_end
);
2592 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2593 ret
= mpage_prepare_extent_to_map(&mpd
);
2596 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2600 * We scanned the whole range (or exhausted
2601 * nr_to_write), submitted what was mapped and
2602 * didn't find anything needing mapping. We are
2608 ext4_journal_stop(handle
);
2609 /* Submit prepared bio */
2610 ext4_io_submit(&mpd
.io_submit
);
2611 /* Unlock pages we didn't use */
2612 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2613 /* Drop our io_end reference we got from init */
2614 ext4_put_io_end(mpd
.io_submit
.io_end
);
2616 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2618 * Commit the transaction which would
2619 * free blocks released in the transaction
2622 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2626 /* Fatal error - ENOMEM, EIO... */
2630 blk_finish_plug(&plug
);
2631 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2633 mpd
.last_page
= writeback_index
- 1;
2639 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2641 * Set the writeback_index so that range_cyclic
2642 * mode will write it back later
2644 mapping
->writeback_index
= mpd
.first_page
;
2647 trace_ext4_writepages_result(inode
, wbc
, ret
,
2648 nr_to_write
- wbc
->nr_to_write
);
2652 static int ext4_nonda_switch(struct super_block
*sb
)
2654 s64 free_clusters
, dirty_clusters
;
2655 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2658 * switch to non delalloc mode if we are running low
2659 * on free block. The free block accounting via percpu
2660 * counters can get slightly wrong with percpu_counter_batch getting
2661 * accumulated on each CPU without updating global counters
2662 * Delalloc need an accurate free block accounting. So switch
2663 * to non delalloc when we are near to error range.
2666 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2668 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2670 * Start pushing delalloc when 1/2 of free blocks are dirty.
2672 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2673 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2675 if (2 * free_clusters
< 3 * dirty_clusters
||
2676 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2678 * free block count is less than 150% of dirty blocks
2679 * or free blocks is less than watermark
2686 /* We always reserve for an inode update; the superblock could be there too */
2687 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2689 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2692 if (pos
+ len
<= 0x7fffffffULL
)
2695 /* We might need to update the superblock to set LARGE_FILE */
2699 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2700 loff_t pos
, unsigned len
, unsigned flags
,
2701 struct page
**pagep
, void **fsdata
)
2703 int ret
, retries
= 0;
2706 struct inode
*inode
= mapping
->host
;
2709 index
= pos
>> PAGE_CACHE_SHIFT
;
2711 if (ext4_nonda_switch(inode
->i_sb
)) {
2712 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2713 return ext4_write_begin(file
, mapping
, pos
,
2714 len
, flags
, pagep
, fsdata
);
2716 *fsdata
= (void *)0;
2717 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2719 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2720 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2730 * grab_cache_page_write_begin() can take a long time if the
2731 * system is thrashing due to memory pressure, or if the page
2732 * is being written back. So grab it first before we start
2733 * the transaction handle. This also allows us to allocate
2734 * the page (if needed) without using GFP_NOFS.
2737 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2743 * With delayed allocation, we don't log the i_disksize update
2744 * if there is delayed block allocation. But we still need
2745 * to journalling the i_disksize update if writes to the end
2746 * of file which has an already mapped buffer.
2749 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2750 ext4_da_write_credits(inode
, pos
, len
));
2751 if (IS_ERR(handle
)) {
2752 page_cache_release(page
);
2753 return PTR_ERR(handle
);
2757 if (page
->mapping
!= mapping
) {
2758 /* The page got truncated from under us */
2760 page_cache_release(page
);
2761 ext4_journal_stop(handle
);
2764 /* In case writeback began while the page was unlocked */
2765 wait_for_stable_page(page
);
2767 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2768 ret
= ext4_block_write_begin(page
, pos
, len
,
2769 ext4_da_get_block_prep
);
2771 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2775 ext4_journal_stop(handle
);
2777 * block_write_begin may have instantiated a few blocks
2778 * outside i_size. Trim these off again. Don't need
2779 * i_size_read because we hold i_mutex.
2781 if (pos
+ len
> inode
->i_size
)
2782 ext4_truncate_failed_write(inode
);
2784 if (ret
== -ENOSPC
&&
2785 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2788 page_cache_release(page
);
2797 * Check if we should update i_disksize
2798 * when write to the end of file but not require block allocation
2800 static int ext4_da_should_update_i_disksize(struct page
*page
,
2801 unsigned long offset
)
2803 struct buffer_head
*bh
;
2804 struct inode
*inode
= page
->mapping
->host
;
2808 bh
= page_buffers(page
);
2809 idx
= offset
>> inode
->i_blkbits
;
2811 for (i
= 0; i
< idx
; i
++)
2812 bh
= bh
->b_this_page
;
2814 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2819 static int ext4_da_write_end(struct file
*file
,
2820 struct address_space
*mapping
,
2821 loff_t pos
, unsigned len
, unsigned copied
,
2822 struct page
*page
, void *fsdata
)
2824 struct inode
*inode
= mapping
->host
;
2826 handle_t
*handle
= ext4_journal_current_handle();
2828 unsigned long start
, end
;
2829 int write_mode
= (int)(unsigned long)fsdata
;
2831 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2832 return ext4_write_end(file
, mapping
, pos
,
2833 len
, copied
, page
, fsdata
);
2835 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2836 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2837 end
= start
+ copied
- 1;
2840 * generic_write_end() will run mark_inode_dirty() if i_size
2841 * changes. So let's piggyback the i_disksize mark_inode_dirty
2844 new_i_size
= pos
+ copied
;
2845 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2846 if (ext4_has_inline_data(inode
) ||
2847 ext4_da_should_update_i_disksize(page
, end
)) {
2848 ext4_update_i_disksize(inode
, new_i_size
);
2849 /* We need to mark inode dirty even if
2850 * new_i_size is less that inode->i_size
2851 * bu greater than i_disksize.(hint delalloc)
2853 ext4_mark_inode_dirty(handle
, inode
);
2857 if (write_mode
!= CONVERT_INLINE_DATA
&&
2858 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2859 ext4_has_inline_data(inode
))
2860 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2863 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2869 ret2
= ext4_journal_stop(handle
);
2873 return ret
? ret
: copied
;
2876 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2877 unsigned int length
)
2880 * Drop reserved blocks
2882 BUG_ON(!PageLocked(page
));
2883 if (!page_has_buffers(page
))
2886 ext4_da_page_release_reservation(page
, offset
, length
);
2889 ext4_invalidatepage(page
, offset
, length
);
2895 * Force all delayed allocation blocks to be allocated for a given inode.
2897 int ext4_alloc_da_blocks(struct inode
*inode
)
2899 trace_ext4_alloc_da_blocks(inode
);
2901 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
2905 * We do something simple for now. The filemap_flush() will
2906 * also start triggering a write of the data blocks, which is
2907 * not strictly speaking necessary (and for users of
2908 * laptop_mode, not even desirable). However, to do otherwise
2909 * would require replicating code paths in:
2911 * ext4_writepages() ->
2912 * write_cache_pages() ---> (via passed in callback function)
2913 * __mpage_da_writepage() -->
2914 * mpage_add_bh_to_extent()
2915 * mpage_da_map_blocks()
2917 * The problem is that write_cache_pages(), located in
2918 * mm/page-writeback.c, marks pages clean in preparation for
2919 * doing I/O, which is not desirable if we're not planning on
2922 * We could call write_cache_pages(), and then redirty all of
2923 * the pages by calling redirty_page_for_writepage() but that
2924 * would be ugly in the extreme. So instead we would need to
2925 * replicate parts of the code in the above functions,
2926 * simplifying them because we wouldn't actually intend to
2927 * write out the pages, but rather only collect contiguous
2928 * logical block extents, call the multi-block allocator, and
2929 * then update the buffer heads with the block allocations.
2931 * For now, though, we'll cheat by calling filemap_flush(),
2932 * which will map the blocks, and start the I/O, but not
2933 * actually wait for the I/O to complete.
2935 return filemap_flush(inode
->i_mapping
);
2939 * bmap() is special. It gets used by applications such as lilo and by
2940 * the swapper to find the on-disk block of a specific piece of data.
2942 * Naturally, this is dangerous if the block concerned is still in the
2943 * journal. If somebody makes a swapfile on an ext4 data-journaling
2944 * filesystem and enables swap, then they may get a nasty shock when the
2945 * data getting swapped to that swapfile suddenly gets overwritten by
2946 * the original zero's written out previously to the journal and
2947 * awaiting writeback in the kernel's buffer cache.
2949 * So, if we see any bmap calls here on a modified, data-journaled file,
2950 * take extra steps to flush any blocks which might be in the cache.
2952 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2954 struct inode
*inode
= mapping
->host
;
2959 * We can get here for an inline file via the FIBMAP ioctl
2961 if (ext4_has_inline_data(inode
))
2964 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2965 test_opt(inode
->i_sb
, DELALLOC
)) {
2967 * With delalloc we want to sync the file
2968 * so that we can make sure we allocate
2971 filemap_write_and_wait(mapping
);
2974 if (EXT4_JOURNAL(inode
) &&
2975 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2977 * This is a REALLY heavyweight approach, but the use of
2978 * bmap on dirty files is expected to be extremely rare:
2979 * only if we run lilo or swapon on a freshly made file
2980 * do we expect this to happen.
2982 * (bmap requires CAP_SYS_RAWIO so this does not
2983 * represent an unprivileged user DOS attack --- we'd be
2984 * in trouble if mortal users could trigger this path at
2987 * NB. EXT4_STATE_JDATA is not set on files other than
2988 * regular files. If somebody wants to bmap a directory
2989 * or symlink and gets confused because the buffer
2990 * hasn't yet been flushed to disk, they deserve
2991 * everything they get.
2994 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2995 journal
= EXT4_JOURNAL(inode
);
2996 jbd2_journal_lock_updates(journal
);
2997 err
= jbd2_journal_flush(journal
);
2998 jbd2_journal_unlock_updates(journal
);
3004 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3007 static int ext4_readpage(struct file
*file
, struct page
*page
)
3010 struct inode
*inode
= page
->mapping
->host
;
3012 trace_ext4_readpage(page
);
3014 if (ext4_has_inline_data(inode
))
3015 ret
= ext4_readpage_inline(inode
, page
);
3018 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3024 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3025 struct list_head
*pages
, unsigned nr_pages
)
3027 struct inode
*inode
= mapping
->host
;
3029 /* If the file has inline data, no need to do readpages. */
3030 if (ext4_has_inline_data(inode
))
3033 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3036 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3037 unsigned int length
)
3039 trace_ext4_invalidatepage(page
, offset
, length
);
3041 /* No journalling happens on data buffers when this function is used */
3042 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3044 block_invalidatepage(page
, offset
, length
);
3047 static int __ext4_journalled_invalidatepage(struct page
*page
,
3048 unsigned int offset
,
3049 unsigned int length
)
3051 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3053 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3056 * If it's a full truncate we just forget about the pending dirtying
3058 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
3059 ClearPageChecked(page
);
3061 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3064 /* Wrapper for aops... */
3065 static void ext4_journalled_invalidatepage(struct page
*page
,
3066 unsigned int offset
,
3067 unsigned int length
)
3069 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3072 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3074 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3076 trace_ext4_releasepage(page
);
3078 /* Page has dirty journalled data -> cannot release */
3079 if (PageChecked(page
))
3082 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3084 return try_to_free_buffers(page
);
3088 * ext4_get_block used when preparing for a DIO write or buffer write.
3089 * We allocate an uinitialized extent if blocks haven't been allocated.
3090 * The extent will be converted to initialized after the IO is complete.
3092 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3093 struct buffer_head
*bh_result
, int create
)
3095 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3096 inode
->i_ino
, create
);
3097 return _ext4_get_block(inode
, iblock
, bh_result
,
3098 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3101 static int ext4_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
3102 struct buffer_head
*bh_result
, int create
)
3106 ext4_debug("ext4_get_block_overwrite: inode %lu, create flag %d\n",
3107 inode
->i_ino
, create
);
3108 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
3110 * Blocks should have been preallocated! ext4_file_write_iter() checks
3113 WARN_ON_ONCE(!buffer_mapped(bh_result
));
3118 #ifdef CONFIG_FS_DAX
3119 int ext4_dax_mmap_get_block(struct inode
*inode
, sector_t iblock
,
3120 struct buffer_head
*bh_result
, int create
)
3124 struct ext4_map_blocks map
;
3125 handle_t
*handle
= NULL
;
3128 ext4_debug("ext4_dax_mmap_get_block: inode %lu, create flag %d\n",
3129 inode
->i_ino
, create
);
3130 map
.m_lblk
= iblock
;
3131 map
.m_len
= bh_result
->b_size
>> inode
->i_blkbits
;
3132 credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3134 flags
|= EXT4_GET_BLOCKS_PRE_IO
| EXT4_GET_BLOCKS_CREATE_ZERO
;
3135 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, credits
);
3136 if (IS_ERR(handle
)) {
3137 ret
= PTR_ERR(handle
);
3142 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
3144 err
= ext4_journal_stop(handle
);
3145 if (ret
>= 0 && err
< 0)
3150 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3154 * We are protected by i_mmap_sem so we know block cannot go
3155 * away from under us even though we dropped i_data_sem.
3156 * Convert extent to written and write zeros there.
3158 * Note: We may get here even when create == 0.
3160 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, credits
);
3161 if (IS_ERR(handle
)) {
3162 ret
= PTR_ERR(handle
);
3166 err
= ext4_map_blocks(handle
, inode
, &map
,
3167 EXT4_GET_BLOCKS_CONVERT
| EXT4_GET_BLOCKS_CREATE_ZERO
);
3170 err2
= ext4_journal_stop(handle
);
3171 if (err2
< 0 && ret
> 0)
3175 WARN_ON_ONCE(ret
== 0 && create
);
3177 map_bh(bh_result
, inode
->i_sb
, map
.m_pblk
);
3178 bh_result
->b_state
= (bh_result
->b_state
& ~EXT4_MAP_FLAGS
) |
3181 * At least for now we have to clear BH_New so that DAX code
3182 * doesn't attempt to zero blocks again in a racy way.
3184 bh_result
->b_state
&= ~(1 << BH_New
);
3185 bh_result
->b_size
= map
.m_len
<< inode
->i_blkbits
;
3192 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3193 ssize_t size
, void *private)
3195 ext4_io_end_t
*io_end
= iocb
->private;
3197 /* if not async direct IO just return */
3201 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3202 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3203 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3206 iocb
->private = NULL
;
3207 io_end
->offset
= offset
;
3208 io_end
->size
= size
;
3209 ext4_put_io_end(io_end
);
3213 * For ext4 extent files, ext4 will do direct-io write to holes,
3214 * preallocated extents, and those write extend the file, no need to
3215 * fall back to buffered IO.
3217 * For holes, we fallocate those blocks, mark them as unwritten
3218 * If those blocks were preallocated, we mark sure they are split, but
3219 * still keep the range to write as unwritten.
3221 * The unwritten extents will be converted to written when DIO is completed.
3222 * For async direct IO, since the IO may still pending when return, we
3223 * set up an end_io call back function, which will do the conversion
3224 * when async direct IO completed.
3226 * If the O_DIRECT write will extend the file then add this inode to the
3227 * orphan list. So recovery will truncate it back to the original size
3228 * if the machine crashes during the write.
3231 static ssize_t
ext4_ext_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
3234 struct file
*file
= iocb
->ki_filp
;
3235 struct inode
*inode
= file
->f_mapping
->host
;
3237 size_t count
= iov_iter_count(iter
);
3239 get_block_t
*get_block_func
= NULL
;
3241 loff_t final_size
= offset
+ count
;
3242 ext4_io_end_t
*io_end
= NULL
;
3244 /* Use the old path for reads and writes beyond i_size. */
3245 if (iov_iter_rw(iter
) != WRITE
|| final_size
> inode
->i_size
)
3246 return ext4_ind_direct_IO(iocb
, iter
, offset
);
3248 BUG_ON(iocb
->private == NULL
);
3251 * Make all waiters for direct IO properly wait also for extent
3252 * conversion. This also disallows race between truncate() and
3253 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3255 if (iov_iter_rw(iter
) == WRITE
)
3256 inode_dio_begin(inode
);
3258 /* If we do a overwrite dio, i_mutex locking can be released */
3259 overwrite
= *((int *)iocb
->private);
3262 inode_unlock(inode
);
3265 * We could direct write to holes and fallocate.
3267 * Allocated blocks to fill the hole are marked as
3268 * unwritten to prevent parallel buffered read to expose
3269 * the stale data before DIO complete the data IO.
3271 * As to previously fallocated extents, ext4 get_block will
3272 * just simply mark the buffer mapped but still keep the
3273 * extents unwritten.
3275 * For non AIO case, we will convert those unwritten extents
3276 * to written after return back from blockdev_direct_IO.
3278 * For async DIO, the conversion needs to be deferred when the
3279 * IO is completed. The ext4 end_io callback function will be
3280 * called to take care of the conversion work. Here for async
3281 * case, we allocate an io_end structure to hook to the iocb.
3283 iocb
->private = NULL
;
3285 get_block_func
= ext4_get_block_overwrite
;
3287 ext4_inode_aio_set(inode
, NULL
);
3288 if (!is_sync_kiocb(iocb
)) {
3289 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3295 * Grab reference for DIO. Will be dropped in
3298 iocb
->private = ext4_get_io_end(io_end
);
3300 * we save the io structure for current async direct
3301 * IO, so that later ext4_map_blocks() could flag the
3302 * io structure whether there is a unwritten extents
3303 * needs to be converted when IO is completed.
3305 ext4_inode_aio_set(inode
, io_end
);
3307 get_block_func
= ext4_get_block_write
;
3308 dio_flags
= DIO_LOCKING
;
3310 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3311 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3314 ret
= dax_do_io(iocb
, inode
, iter
, offset
, get_block_func
,
3315 ext4_end_io_dio
, dio_flags
);
3317 ret
= __blockdev_direct_IO(iocb
, inode
,
3318 inode
->i_sb
->s_bdev
, iter
, offset
,
3320 ext4_end_io_dio
, NULL
, dio_flags
);
3323 * Put our reference to io_end. This can free the io_end structure e.g.
3324 * in sync IO case or in case of error. It can even perform extent
3325 * conversion if all bios we submitted finished before we got here.
3326 * Note that in that case iocb->private can be already set to NULL
3330 ext4_inode_aio_set(inode
, NULL
);
3331 ext4_put_io_end(io_end
);
3333 * When no IO was submitted ext4_end_io_dio() was not
3334 * called so we have to put iocb's reference.
3336 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3337 WARN_ON(iocb
->private != io_end
);
3338 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3339 ext4_put_io_end(io_end
);
3340 iocb
->private = NULL
;
3343 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3344 EXT4_STATE_DIO_UNWRITTEN
)) {
3347 * for non AIO case, since the IO is already
3348 * completed, we could do the conversion right here
3350 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3354 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3358 if (iov_iter_rw(iter
) == WRITE
)
3359 inode_dio_end(inode
);
3360 /* take i_mutex locking again if we do a ovewrite dio */
3367 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
3370 struct file
*file
= iocb
->ki_filp
;
3371 struct inode
*inode
= file
->f_mapping
->host
;
3372 size_t count
= iov_iter_count(iter
);
3375 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3376 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3381 * If we are doing data journalling we don't support O_DIRECT
3383 if (ext4_should_journal_data(inode
))
3386 /* Let buffer I/O handle the inline data case. */
3387 if (ext4_has_inline_data(inode
))
3390 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3391 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3392 ret
= ext4_ext_direct_IO(iocb
, iter
, offset
);
3394 ret
= ext4_ind_direct_IO(iocb
, iter
, offset
);
3395 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3400 * Pages can be marked dirty completely asynchronously from ext4's journalling
3401 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3402 * much here because ->set_page_dirty is called under VFS locks. The page is
3403 * not necessarily locked.
3405 * We cannot just dirty the page and leave attached buffers clean, because the
3406 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3407 * or jbddirty because all the journalling code will explode.
3409 * So what we do is to mark the page "pending dirty" and next time writepage
3410 * is called, propagate that into the buffers appropriately.
3412 static int ext4_journalled_set_page_dirty(struct page
*page
)
3414 SetPageChecked(page
);
3415 return __set_page_dirty_nobuffers(page
);
3418 static const struct address_space_operations ext4_aops
= {
3419 .readpage
= ext4_readpage
,
3420 .readpages
= ext4_readpages
,
3421 .writepage
= ext4_writepage
,
3422 .writepages
= ext4_writepages
,
3423 .write_begin
= ext4_write_begin
,
3424 .write_end
= ext4_write_end
,
3426 .invalidatepage
= ext4_invalidatepage
,
3427 .releasepage
= ext4_releasepage
,
3428 .direct_IO
= ext4_direct_IO
,
3429 .migratepage
= buffer_migrate_page
,
3430 .is_partially_uptodate
= block_is_partially_uptodate
,
3431 .error_remove_page
= generic_error_remove_page
,
3434 static const struct address_space_operations ext4_journalled_aops
= {
3435 .readpage
= ext4_readpage
,
3436 .readpages
= ext4_readpages
,
3437 .writepage
= ext4_writepage
,
3438 .writepages
= ext4_writepages
,
3439 .write_begin
= ext4_write_begin
,
3440 .write_end
= ext4_journalled_write_end
,
3441 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3443 .invalidatepage
= ext4_journalled_invalidatepage
,
3444 .releasepage
= ext4_releasepage
,
3445 .direct_IO
= ext4_direct_IO
,
3446 .is_partially_uptodate
= block_is_partially_uptodate
,
3447 .error_remove_page
= generic_error_remove_page
,
3450 static const struct address_space_operations ext4_da_aops
= {
3451 .readpage
= ext4_readpage
,
3452 .readpages
= ext4_readpages
,
3453 .writepage
= ext4_writepage
,
3454 .writepages
= ext4_writepages
,
3455 .write_begin
= ext4_da_write_begin
,
3456 .write_end
= ext4_da_write_end
,
3458 .invalidatepage
= ext4_da_invalidatepage
,
3459 .releasepage
= ext4_releasepage
,
3460 .direct_IO
= ext4_direct_IO
,
3461 .migratepage
= buffer_migrate_page
,
3462 .is_partially_uptodate
= block_is_partially_uptodate
,
3463 .error_remove_page
= generic_error_remove_page
,
3466 void ext4_set_aops(struct inode
*inode
)
3468 switch (ext4_inode_journal_mode(inode
)) {
3469 case EXT4_INODE_ORDERED_DATA_MODE
:
3470 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3472 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3473 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3475 case EXT4_INODE_JOURNAL_DATA_MODE
:
3476 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3481 if (test_opt(inode
->i_sb
, DELALLOC
))
3482 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3484 inode
->i_mapping
->a_ops
= &ext4_aops
;
3487 static int __ext4_block_zero_page_range(handle_t
*handle
,
3488 struct address_space
*mapping
, loff_t from
, loff_t length
)
3490 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3491 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3492 unsigned blocksize
, pos
;
3494 struct inode
*inode
= mapping
->host
;
3495 struct buffer_head
*bh
;
3499 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3500 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3504 blocksize
= inode
->i_sb
->s_blocksize
;
3506 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3508 if (!page_has_buffers(page
))
3509 create_empty_buffers(page
, blocksize
, 0);
3511 /* Find the buffer that contains "offset" */
3512 bh
= page_buffers(page
);
3514 while (offset
>= pos
) {
3515 bh
= bh
->b_this_page
;
3519 if (buffer_freed(bh
)) {
3520 BUFFER_TRACE(bh
, "freed: skip");
3523 if (!buffer_mapped(bh
)) {
3524 BUFFER_TRACE(bh
, "unmapped");
3525 ext4_get_block(inode
, iblock
, bh
, 0);
3526 /* unmapped? It's a hole - nothing to do */
3527 if (!buffer_mapped(bh
)) {
3528 BUFFER_TRACE(bh
, "still unmapped");
3533 /* Ok, it's mapped. Make sure it's up-to-date */
3534 if (PageUptodate(page
))
3535 set_buffer_uptodate(bh
);
3537 if (!buffer_uptodate(bh
)) {
3539 ll_rw_block(READ
, 1, &bh
);
3541 /* Uhhuh. Read error. Complain and punt. */
3542 if (!buffer_uptodate(bh
))
3544 if (S_ISREG(inode
->i_mode
) &&
3545 ext4_encrypted_inode(inode
)) {
3546 /* We expect the key to be set. */
3547 BUG_ON(!ext4_has_encryption_key(inode
));
3548 BUG_ON(blocksize
!= PAGE_CACHE_SIZE
);
3549 WARN_ON_ONCE(ext4_decrypt(page
));
3552 if (ext4_should_journal_data(inode
)) {
3553 BUFFER_TRACE(bh
, "get write access");
3554 err
= ext4_journal_get_write_access(handle
, bh
);
3558 zero_user(page
, offset
, length
);
3559 BUFFER_TRACE(bh
, "zeroed end of block");
3561 if (ext4_should_journal_data(inode
)) {
3562 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3565 mark_buffer_dirty(bh
);
3566 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3567 err
= ext4_jbd2_file_inode(handle
, inode
);
3572 page_cache_release(page
);
3577 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3578 * starting from file offset 'from'. The range to be zero'd must
3579 * be contained with in one block. If the specified range exceeds
3580 * the end of the block it will be shortened to end of the block
3581 * that cooresponds to 'from'
3583 static int ext4_block_zero_page_range(handle_t
*handle
,
3584 struct address_space
*mapping
, loff_t from
, loff_t length
)
3586 struct inode
*inode
= mapping
->host
;
3587 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3588 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3589 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3592 * correct length if it does not fall between
3593 * 'from' and the end of the block
3595 if (length
> max
|| length
< 0)
3599 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3600 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3604 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3605 * up to the end of the block which corresponds to `from'.
3606 * This required during truncate. We need to physically zero the tail end
3607 * of that block so it doesn't yield old data if the file is later grown.
3609 static int ext4_block_truncate_page(handle_t
*handle
,
3610 struct address_space
*mapping
, loff_t from
)
3612 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3615 struct inode
*inode
= mapping
->host
;
3617 blocksize
= inode
->i_sb
->s_blocksize
;
3618 length
= blocksize
- (offset
& (blocksize
- 1));
3620 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3623 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3624 loff_t lstart
, loff_t length
)
3626 struct super_block
*sb
= inode
->i_sb
;
3627 struct address_space
*mapping
= inode
->i_mapping
;
3628 unsigned partial_start
, partial_end
;
3629 ext4_fsblk_t start
, end
;
3630 loff_t byte_end
= (lstart
+ length
- 1);
3633 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3634 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3636 start
= lstart
>> sb
->s_blocksize_bits
;
3637 end
= byte_end
>> sb
->s_blocksize_bits
;
3639 /* Handle partial zero within the single block */
3641 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3642 err
= ext4_block_zero_page_range(handle
, mapping
,
3646 /* Handle partial zero out on the start of the range */
3647 if (partial_start
) {
3648 err
= ext4_block_zero_page_range(handle
, mapping
,
3649 lstart
, sb
->s_blocksize
);
3653 /* Handle partial zero out on the end of the range */
3654 if (partial_end
!= sb
->s_blocksize
- 1)
3655 err
= ext4_block_zero_page_range(handle
, mapping
,
3656 byte_end
- partial_end
,
3661 int ext4_can_truncate(struct inode
*inode
)
3663 if (S_ISREG(inode
->i_mode
))
3665 if (S_ISDIR(inode
->i_mode
))
3667 if (S_ISLNK(inode
->i_mode
))
3668 return !ext4_inode_is_fast_symlink(inode
);
3673 * We have to make sure i_disksize gets properly updated before we truncate
3674 * page cache due to hole punching or zero range. Otherwise i_disksize update
3675 * can get lost as it may have been postponed to submission of writeback but
3676 * that will never happen after we truncate page cache.
3678 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3682 loff_t size
= i_size_read(inode
);
3684 WARN_ON(!inode_is_locked(inode
));
3685 if (offset
> size
|| offset
+ len
< size
)
3688 if (EXT4_I(inode
)->i_disksize
>= size
)
3691 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3693 return PTR_ERR(handle
);
3694 ext4_update_i_disksize(inode
, size
);
3695 ext4_mark_inode_dirty(handle
, inode
);
3696 ext4_journal_stop(handle
);
3702 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3703 * associated with the given offset and length
3705 * @inode: File inode
3706 * @offset: The offset where the hole will begin
3707 * @len: The length of the hole
3709 * Returns: 0 on success or negative on failure
3712 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3714 struct super_block
*sb
= inode
->i_sb
;
3715 ext4_lblk_t first_block
, stop_block
;
3716 struct address_space
*mapping
= inode
->i_mapping
;
3717 loff_t first_block_offset
, last_block_offset
;
3719 unsigned int credits
;
3722 if (!S_ISREG(inode
->i_mode
))
3725 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3728 * Write out all dirty pages to avoid race conditions
3729 * Then release them.
3731 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3732 ret
= filemap_write_and_wait_range(mapping
, offset
,
3733 offset
+ length
- 1);
3740 /* No need to punch hole beyond i_size */
3741 if (offset
>= inode
->i_size
)
3745 * If the hole extends beyond i_size, set the hole
3746 * to end after the page that contains i_size
3748 if (offset
+ length
> inode
->i_size
) {
3749 length
= inode
->i_size
+
3750 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3754 if (offset
& (sb
->s_blocksize
- 1) ||
3755 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3757 * Attach jinode to inode for jbd2 if we do any zeroing of
3760 ret
= ext4_inode_attach_jinode(inode
);
3766 /* Wait all existing dio workers, newcomers will block on i_mutex */
3767 ext4_inode_block_unlocked_dio(inode
);
3768 inode_dio_wait(inode
);
3771 * Prevent page faults from reinstantiating pages we have released from
3774 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3775 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3776 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3778 /* Now release the pages and zero block aligned part of pages*/
3779 if (last_block_offset
> first_block_offset
) {
3780 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
3783 truncate_pagecache_range(inode
, first_block_offset
,
3787 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3788 credits
= ext4_writepage_trans_blocks(inode
);
3790 credits
= ext4_blocks_for_truncate(inode
);
3791 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3792 if (IS_ERR(handle
)) {
3793 ret
= PTR_ERR(handle
);
3794 ext4_std_error(sb
, ret
);
3798 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3803 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3804 EXT4_BLOCK_SIZE_BITS(sb
);
3805 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3807 /* If there are no blocks to remove, return now */
3808 if (first_block
>= stop_block
)
3811 down_write(&EXT4_I(inode
)->i_data_sem
);
3812 ext4_discard_preallocations(inode
);
3814 ret
= ext4_es_remove_extent(inode
, first_block
,
3815 stop_block
- first_block
);
3817 up_write(&EXT4_I(inode
)->i_data_sem
);
3821 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3822 ret
= ext4_ext_remove_space(inode
, first_block
,
3825 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3828 up_write(&EXT4_I(inode
)->i_data_sem
);
3830 ext4_handle_sync(handle
);
3832 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3833 ext4_mark_inode_dirty(handle
, inode
);
3835 ext4_journal_stop(handle
);
3837 up_write(&EXT4_I(inode
)->i_mmap_sem
);
3838 ext4_inode_resume_unlocked_dio(inode
);
3840 inode_unlock(inode
);
3844 int ext4_inode_attach_jinode(struct inode
*inode
)
3846 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3847 struct jbd2_inode
*jinode
;
3849 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3852 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3853 spin_lock(&inode
->i_lock
);
3856 spin_unlock(&inode
->i_lock
);
3859 ei
->jinode
= jinode
;
3860 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3863 spin_unlock(&inode
->i_lock
);
3864 if (unlikely(jinode
!= NULL
))
3865 jbd2_free_inode(jinode
);
3872 * We block out ext4_get_block() block instantiations across the entire
3873 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3874 * simultaneously on behalf of the same inode.
3876 * As we work through the truncate and commit bits of it to the journal there
3877 * is one core, guiding principle: the file's tree must always be consistent on
3878 * disk. We must be able to restart the truncate after a crash.
3880 * The file's tree may be transiently inconsistent in memory (although it
3881 * probably isn't), but whenever we close off and commit a journal transaction,
3882 * the contents of (the filesystem + the journal) must be consistent and
3883 * restartable. It's pretty simple, really: bottom up, right to left (although
3884 * left-to-right works OK too).
3886 * Note that at recovery time, journal replay occurs *before* the restart of
3887 * truncate against the orphan inode list.
3889 * The committed inode has the new, desired i_size (which is the same as
3890 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3891 * that this inode's truncate did not complete and it will again call
3892 * ext4_truncate() to have another go. So there will be instantiated blocks
3893 * to the right of the truncation point in a crashed ext4 filesystem. But
3894 * that's fine - as long as they are linked from the inode, the post-crash
3895 * ext4_truncate() run will find them and release them.
3897 void ext4_truncate(struct inode
*inode
)
3899 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3900 unsigned int credits
;
3902 struct address_space
*mapping
= inode
->i_mapping
;
3905 * There is a possibility that we're either freeing the inode
3906 * or it's a completely new inode. In those cases we might not
3907 * have i_mutex locked because it's not necessary.
3909 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3910 WARN_ON(!inode_is_locked(inode
));
3911 trace_ext4_truncate_enter(inode
);
3913 if (!ext4_can_truncate(inode
))
3916 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3918 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3919 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3921 if (ext4_has_inline_data(inode
)) {
3924 ext4_inline_data_truncate(inode
, &has_inline
);
3929 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3930 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3931 if (ext4_inode_attach_jinode(inode
) < 0)
3935 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3936 credits
= ext4_writepage_trans_blocks(inode
);
3938 credits
= ext4_blocks_for_truncate(inode
);
3940 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3941 if (IS_ERR(handle
)) {
3942 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3946 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3947 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3950 * We add the inode to the orphan list, so that if this
3951 * truncate spans multiple transactions, and we crash, we will
3952 * resume the truncate when the filesystem recovers. It also
3953 * marks the inode dirty, to catch the new size.
3955 * Implication: the file must always be in a sane, consistent
3956 * truncatable state while each transaction commits.
3958 if (ext4_orphan_add(handle
, inode
))
3961 down_write(&EXT4_I(inode
)->i_data_sem
);
3963 ext4_discard_preallocations(inode
);
3965 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3966 ext4_ext_truncate(handle
, inode
);
3968 ext4_ind_truncate(handle
, inode
);
3970 up_write(&ei
->i_data_sem
);
3973 ext4_handle_sync(handle
);
3977 * If this was a simple ftruncate() and the file will remain alive,
3978 * then we need to clear up the orphan record which we created above.
3979 * However, if this was a real unlink then we were called by
3980 * ext4_evict_inode(), and we allow that function to clean up the
3981 * orphan info for us.
3984 ext4_orphan_del(handle
, inode
);
3986 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3987 ext4_mark_inode_dirty(handle
, inode
);
3988 ext4_journal_stop(handle
);
3990 trace_ext4_truncate_exit(inode
);
3994 * ext4_get_inode_loc returns with an extra refcount against the inode's
3995 * underlying buffer_head on success. If 'in_mem' is true, we have all
3996 * data in memory that is needed to recreate the on-disk version of this
3999 static int __ext4_get_inode_loc(struct inode
*inode
,
4000 struct ext4_iloc
*iloc
, int in_mem
)
4002 struct ext4_group_desc
*gdp
;
4003 struct buffer_head
*bh
;
4004 struct super_block
*sb
= inode
->i_sb
;
4006 int inodes_per_block
, inode_offset
;
4009 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4010 return -EFSCORRUPTED
;
4012 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4013 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4018 * Figure out the offset within the block group inode table
4020 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4021 inode_offset
= ((inode
->i_ino
- 1) %
4022 EXT4_INODES_PER_GROUP(sb
));
4023 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4024 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4026 bh
= sb_getblk(sb
, block
);
4029 if (!buffer_uptodate(bh
)) {
4033 * If the buffer has the write error flag, we have failed
4034 * to write out another inode in the same block. In this
4035 * case, we don't have to read the block because we may
4036 * read the old inode data successfully.
4038 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4039 set_buffer_uptodate(bh
);
4041 if (buffer_uptodate(bh
)) {
4042 /* someone brought it uptodate while we waited */
4048 * If we have all information of the inode in memory and this
4049 * is the only valid inode in the block, we need not read the
4053 struct buffer_head
*bitmap_bh
;
4056 start
= inode_offset
& ~(inodes_per_block
- 1);
4058 /* Is the inode bitmap in cache? */
4059 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4060 if (unlikely(!bitmap_bh
))
4064 * If the inode bitmap isn't in cache then the
4065 * optimisation may end up performing two reads instead
4066 * of one, so skip it.
4068 if (!buffer_uptodate(bitmap_bh
)) {
4072 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4073 if (i
== inode_offset
)
4075 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4079 if (i
== start
+ inodes_per_block
) {
4080 /* all other inodes are free, so skip I/O */
4081 memset(bh
->b_data
, 0, bh
->b_size
);
4082 set_buffer_uptodate(bh
);
4090 * If we need to do any I/O, try to pre-readahead extra
4091 * blocks from the inode table.
4093 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4094 ext4_fsblk_t b
, end
, table
;
4096 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4098 table
= ext4_inode_table(sb
, gdp
);
4099 /* s_inode_readahead_blks is always a power of 2 */
4100 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4104 num
= EXT4_INODES_PER_GROUP(sb
);
4105 if (ext4_has_group_desc_csum(sb
))
4106 num
-= ext4_itable_unused_count(sb
, gdp
);
4107 table
+= num
/ inodes_per_block
;
4111 sb_breadahead(sb
, b
++);
4115 * There are other valid inodes in the buffer, this inode
4116 * has in-inode xattrs, or we don't have this inode in memory.
4117 * Read the block from disk.
4119 trace_ext4_load_inode(inode
);
4121 bh
->b_end_io
= end_buffer_read_sync
;
4122 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4124 if (!buffer_uptodate(bh
)) {
4125 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4126 "unable to read itable block");
4136 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4138 /* We have all inode data except xattrs in memory here. */
4139 return __ext4_get_inode_loc(inode
, iloc
,
4140 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4143 void ext4_set_inode_flags(struct inode
*inode
)
4145 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4146 unsigned int new_fl
= 0;
4148 if (flags
& EXT4_SYNC_FL
)
4150 if (flags
& EXT4_APPEND_FL
)
4152 if (flags
& EXT4_IMMUTABLE_FL
)
4153 new_fl
|= S_IMMUTABLE
;
4154 if (flags
& EXT4_NOATIME_FL
)
4155 new_fl
|= S_NOATIME
;
4156 if (flags
& EXT4_DIRSYNC_FL
)
4157 new_fl
|= S_DIRSYNC
;
4158 if (test_opt(inode
->i_sb
, DAX
))
4160 inode_set_flags(inode
, new_fl
,
4161 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4164 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4165 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4167 unsigned int vfs_fl
;
4168 unsigned long old_fl
, new_fl
;
4171 vfs_fl
= ei
->vfs_inode
.i_flags
;
4172 old_fl
= ei
->i_flags
;
4173 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4174 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4176 if (vfs_fl
& S_SYNC
)
4177 new_fl
|= EXT4_SYNC_FL
;
4178 if (vfs_fl
& S_APPEND
)
4179 new_fl
|= EXT4_APPEND_FL
;
4180 if (vfs_fl
& S_IMMUTABLE
)
4181 new_fl
|= EXT4_IMMUTABLE_FL
;
4182 if (vfs_fl
& S_NOATIME
)
4183 new_fl
|= EXT4_NOATIME_FL
;
4184 if (vfs_fl
& S_DIRSYNC
)
4185 new_fl
|= EXT4_DIRSYNC_FL
;
4186 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4189 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4190 struct ext4_inode_info
*ei
)
4193 struct inode
*inode
= &(ei
->vfs_inode
);
4194 struct super_block
*sb
= inode
->i_sb
;
4196 if (ext4_has_feature_huge_file(sb
)) {
4197 /* we are using combined 48 bit field */
4198 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4199 le32_to_cpu(raw_inode
->i_blocks_lo
);
4200 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4201 /* i_blocks represent file system block size */
4202 return i_blocks
<< (inode
->i_blkbits
- 9);
4207 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4211 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4212 struct ext4_inode
*raw_inode
,
4213 struct ext4_inode_info
*ei
)
4215 __le32
*magic
= (void *)raw_inode
+
4216 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4217 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4218 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4219 ext4_find_inline_data_nolock(inode
);
4221 EXT4_I(inode
)->i_inline_off
= 0;
4224 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4226 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
, EXT4_FEATURE_RO_COMPAT_PROJECT
))
4228 *projid
= EXT4_I(inode
)->i_projid
;
4232 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4234 struct ext4_iloc iloc
;
4235 struct ext4_inode
*raw_inode
;
4236 struct ext4_inode_info
*ei
;
4237 struct inode
*inode
;
4238 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4245 inode
= iget_locked(sb
, ino
);
4247 return ERR_PTR(-ENOMEM
);
4248 if (!(inode
->i_state
& I_NEW
))
4254 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4257 raw_inode
= ext4_raw_inode(&iloc
);
4259 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4260 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4261 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4262 EXT4_INODE_SIZE(inode
->i_sb
)) {
4263 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4264 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4265 EXT4_INODE_SIZE(inode
->i_sb
));
4266 ret
= -EFSCORRUPTED
;
4270 ei
->i_extra_isize
= 0;
4272 /* Precompute checksum seed for inode metadata */
4273 if (ext4_has_metadata_csum(sb
)) {
4274 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4276 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4277 __le32 gen
= raw_inode
->i_generation
;
4278 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4280 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4284 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4285 EXT4_ERROR_INODE(inode
, "checksum invalid");
4290 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4291 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4292 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4293 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_PROJECT
) &&
4294 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4295 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4296 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4298 i_projid
= EXT4_DEF_PROJID
;
4300 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4301 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4302 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4304 i_uid_write(inode
, i_uid
);
4305 i_gid_write(inode
, i_gid
);
4306 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4307 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4309 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4310 ei
->i_inline_off
= 0;
4311 ei
->i_dir_start_lookup
= 0;
4312 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4313 /* We now have enough fields to check if the inode was active or not.
4314 * This is needed because nfsd might try to access dead inodes
4315 * the test is that same one that e2fsck uses
4316 * NeilBrown 1999oct15
4318 if (inode
->i_nlink
== 0) {
4319 if ((inode
->i_mode
== 0 ||
4320 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4321 ino
!= EXT4_BOOT_LOADER_INO
) {
4322 /* this inode is deleted */
4326 /* The only unlinked inodes we let through here have
4327 * valid i_mode and are being read by the orphan
4328 * recovery code: that's fine, we're about to complete
4329 * the process of deleting those.
4330 * OR it is the EXT4_BOOT_LOADER_INO which is
4331 * not initialized on a new filesystem. */
4333 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4334 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4335 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4336 if (ext4_has_feature_64bit(sb
))
4338 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4339 inode
->i_size
= ext4_isize(raw_inode
);
4340 ei
->i_disksize
= inode
->i_size
;
4342 ei
->i_reserved_quota
= 0;
4344 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4345 ei
->i_block_group
= iloc
.block_group
;
4346 ei
->i_last_alloc_group
= ~0;
4348 * NOTE! The in-memory inode i_data array is in little-endian order
4349 * even on big-endian machines: we do NOT byteswap the block numbers!
4351 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4352 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4353 INIT_LIST_HEAD(&ei
->i_orphan
);
4356 * Set transaction id's of transactions that have to be committed
4357 * to finish f[data]sync. We set them to currently running transaction
4358 * as we cannot be sure that the inode or some of its metadata isn't
4359 * part of the transaction - the inode could have been reclaimed and
4360 * now it is reread from disk.
4363 transaction_t
*transaction
;
4366 read_lock(&journal
->j_state_lock
);
4367 if (journal
->j_running_transaction
)
4368 transaction
= journal
->j_running_transaction
;
4370 transaction
= journal
->j_committing_transaction
;
4372 tid
= transaction
->t_tid
;
4374 tid
= journal
->j_commit_sequence
;
4375 read_unlock(&journal
->j_state_lock
);
4376 ei
->i_sync_tid
= tid
;
4377 ei
->i_datasync_tid
= tid
;
4380 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4381 if (ei
->i_extra_isize
== 0) {
4382 /* The extra space is currently unused. Use it. */
4383 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4384 EXT4_GOOD_OLD_INODE_SIZE
;
4386 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4390 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4391 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4392 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4393 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4395 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4396 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4397 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4398 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4400 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4405 if (ei
->i_file_acl
&&
4406 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4407 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4409 ret
= -EFSCORRUPTED
;
4411 } else if (!ext4_has_inline_data(inode
)) {
4412 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4413 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4414 (S_ISLNK(inode
->i_mode
) &&
4415 !ext4_inode_is_fast_symlink(inode
))))
4416 /* Validate extent which is part of inode */
4417 ret
= ext4_ext_check_inode(inode
);
4418 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4419 (S_ISLNK(inode
->i_mode
) &&
4420 !ext4_inode_is_fast_symlink(inode
))) {
4421 /* Validate block references which are part of inode */
4422 ret
= ext4_ind_check_inode(inode
);
4428 if (S_ISREG(inode
->i_mode
)) {
4429 inode
->i_op
= &ext4_file_inode_operations
;
4430 inode
->i_fop
= &ext4_file_operations
;
4431 ext4_set_aops(inode
);
4432 } else if (S_ISDIR(inode
->i_mode
)) {
4433 inode
->i_op
= &ext4_dir_inode_operations
;
4434 inode
->i_fop
= &ext4_dir_operations
;
4435 } else if (S_ISLNK(inode
->i_mode
)) {
4436 if (ext4_encrypted_inode(inode
)) {
4437 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4438 ext4_set_aops(inode
);
4439 } else if (ext4_inode_is_fast_symlink(inode
)) {
4440 inode
->i_link
= (char *)ei
->i_data
;
4441 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4442 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4443 sizeof(ei
->i_data
) - 1);
4445 inode
->i_op
= &ext4_symlink_inode_operations
;
4446 ext4_set_aops(inode
);
4448 inode_nohighmem(inode
);
4449 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4450 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4451 inode
->i_op
= &ext4_special_inode_operations
;
4452 if (raw_inode
->i_block
[0])
4453 init_special_inode(inode
, inode
->i_mode
,
4454 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4456 init_special_inode(inode
, inode
->i_mode
,
4457 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4458 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4459 make_bad_inode(inode
);
4461 ret
= -EFSCORRUPTED
;
4462 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4466 ext4_set_inode_flags(inode
);
4467 unlock_new_inode(inode
);
4473 return ERR_PTR(ret
);
4476 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4478 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4479 return ERR_PTR(-EFSCORRUPTED
);
4480 return ext4_iget(sb
, ino
);
4483 static int ext4_inode_blocks_set(handle_t
*handle
,
4484 struct ext4_inode
*raw_inode
,
4485 struct ext4_inode_info
*ei
)
4487 struct inode
*inode
= &(ei
->vfs_inode
);
4488 u64 i_blocks
= inode
->i_blocks
;
4489 struct super_block
*sb
= inode
->i_sb
;
4491 if (i_blocks
<= ~0U) {
4493 * i_blocks can be represented in a 32 bit variable
4494 * as multiple of 512 bytes
4496 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4497 raw_inode
->i_blocks_high
= 0;
4498 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4501 if (!ext4_has_feature_huge_file(sb
))
4504 if (i_blocks
<= 0xffffffffffffULL
) {
4506 * i_blocks can be represented in a 48 bit variable
4507 * as multiple of 512 bytes
4509 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4510 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4511 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4513 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4514 /* i_block is stored in file system block size */
4515 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4516 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4517 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4522 struct other_inode
{
4523 unsigned long orig_ino
;
4524 struct ext4_inode
*raw_inode
;
4527 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4530 struct other_inode
*oi
= (struct other_inode
*) data
;
4532 if ((inode
->i_ino
!= ino
) ||
4533 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4534 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4535 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4537 spin_lock(&inode
->i_lock
);
4538 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4539 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4540 (inode
->i_state
& I_DIRTY_TIME
)) {
4541 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4543 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4544 spin_unlock(&inode
->i_lock
);
4546 spin_lock(&ei
->i_raw_lock
);
4547 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4548 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4549 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4550 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4551 spin_unlock(&ei
->i_raw_lock
);
4552 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4555 spin_unlock(&inode
->i_lock
);
4560 * Opportunistically update the other time fields for other inodes in
4561 * the same inode table block.
4563 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4564 unsigned long orig_ino
, char *buf
)
4566 struct other_inode oi
;
4568 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4569 int inode_size
= EXT4_INODE_SIZE(sb
);
4571 oi
.orig_ino
= orig_ino
;
4573 * Calculate the first inode in the inode table block. Inode
4574 * numbers are one-based. That is, the first inode in a block
4575 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4577 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4578 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4579 if (ino
== orig_ino
)
4581 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4582 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4587 * Post the struct inode info into an on-disk inode location in the
4588 * buffer-cache. This gobbles the caller's reference to the
4589 * buffer_head in the inode location struct.
4591 * The caller must have write access to iloc->bh.
4593 static int ext4_do_update_inode(handle_t
*handle
,
4594 struct inode
*inode
,
4595 struct ext4_iloc
*iloc
)
4597 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4598 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4599 struct buffer_head
*bh
= iloc
->bh
;
4600 struct super_block
*sb
= inode
->i_sb
;
4601 int err
= 0, rc
, block
;
4602 int need_datasync
= 0, set_large_file
= 0;
4607 spin_lock(&ei
->i_raw_lock
);
4609 /* For fields not tracked in the in-memory inode,
4610 * initialise them to zero for new inodes. */
4611 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4612 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4614 ext4_get_inode_flags(ei
);
4615 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4616 i_uid
= i_uid_read(inode
);
4617 i_gid
= i_gid_read(inode
);
4618 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4619 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4620 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4621 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4623 * Fix up interoperability with old kernels. Otherwise, old inodes get
4624 * re-used with the upper 16 bits of the uid/gid intact
4627 raw_inode
->i_uid_high
=
4628 cpu_to_le16(high_16_bits(i_uid
));
4629 raw_inode
->i_gid_high
=
4630 cpu_to_le16(high_16_bits(i_gid
));
4632 raw_inode
->i_uid_high
= 0;
4633 raw_inode
->i_gid_high
= 0;
4636 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4637 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4638 raw_inode
->i_uid_high
= 0;
4639 raw_inode
->i_gid_high
= 0;
4641 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4643 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4644 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4645 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4646 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4648 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4650 spin_unlock(&ei
->i_raw_lock
);
4653 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4654 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4655 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4656 raw_inode
->i_file_acl_high
=
4657 cpu_to_le16(ei
->i_file_acl
>> 32);
4658 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4659 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4660 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4663 if (ei
->i_disksize
> 0x7fffffffULL
) {
4664 if (!ext4_has_feature_large_file(sb
) ||
4665 EXT4_SB(sb
)->s_es
->s_rev_level
==
4666 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4669 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4670 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4671 if (old_valid_dev(inode
->i_rdev
)) {
4672 raw_inode
->i_block
[0] =
4673 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4674 raw_inode
->i_block
[1] = 0;
4676 raw_inode
->i_block
[0] = 0;
4677 raw_inode
->i_block
[1] =
4678 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4679 raw_inode
->i_block
[2] = 0;
4681 } else if (!ext4_has_inline_data(inode
)) {
4682 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4683 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4686 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4687 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4688 if (ei
->i_extra_isize
) {
4689 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4690 raw_inode
->i_version_hi
=
4691 cpu_to_le32(inode
->i_version
>> 32);
4692 raw_inode
->i_extra_isize
=
4693 cpu_to_le16(ei
->i_extra_isize
);
4697 BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
4698 EXT4_FEATURE_RO_COMPAT_PROJECT
) &&
4699 i_projid
!= EXT4_DEF_PROJID
);
4701 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4702 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4703 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4705 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4706 spin_unlock(&ei
->i_raw_lock
);
4707 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4708 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4711 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4712 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4715 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4716 if (set_large_file
) {
4717 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4718 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4721 ext4_update_dynamic_rev(sb
);
4722 ext4_set_feature_large_file(sb
);
4723 ext4_handle_sync(handle
);
4724 err
= ext4_handle_dirty_super(handle
, sb
);
4726 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4729 ext4_std_error(inode
->i_sb
, err
);
4734 * ext4_write_inode()
4736 * We are called from a few places:
4738 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4739 * Here, there will be no transaction running. We wait for any running
4740 * transaction to commit.
4742 * - Within flush work (sys_sync(), kupdate and such).
4743 * We wait on commit, if told to.
4745 * - Within iput_final() -> write_inode_now()
4746 * We wait on commit, if told to.
4748 * In all cases it is actually safe for us to return without doing anything,
4749 * because the inode has been copied into a raw inode buffer in
4750 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4753 * Note that we are absolutely dependent upon all inode dirtiers doing the
4754 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4755 * which we are interested.
4757 * It would be a bug for them to not do this. The code:
4759 * mark_inode_dirty(inode)
4761 * inode->i_size = expr;
4763 * is in error because write_inode() could occur while `stuff()' is running,
4764 * and the new i_size will be lost. Plus the inode will no longer be on the
4765 * superblock's dirty inode list.
4767 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4771 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4774 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4775 if (ext4_journal_current_handle()) {
4776 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4782 * No need to force transaction in WB_SYNC_NONE mode. Also
4783 * ext4_sync_fs() will force the commit after everything is
4786 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4789 err
= ext4_force_commit(inode
->i_sb
);
4791 struct ext4_iloc iloc
;
4793 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4797 * sync(2) will flush the whole buffer cache. No need to do
4798 * it here separately for each inode.
4800 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4801 sync_dirty_buffer(iloc
.bh
);
4802 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4803 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4804 "IO error syncing inode");
4813 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4814 * buffers that are attached to a page stradding i_size and are undergoing
4815 * commit. In that case we have to wait for commit to finish and try again.
4817 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4821 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4822 tid_t commit_tid
= 0;
4825 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4827 * All buffers in the last page remain valid? Then there's nothing to
4828 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4831 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4834 page
= find_lock_page(inode
->i_mapping
,
4835 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4838 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4839 PAGE_CACHE_SIZE
- offset
);
4841 page_cache_release(page
);
4845 read_lock(&journal
->j_state_lock
);
4846 if (journal
->j_committing_transaction
)
4847 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4848 read_unlock(&journal
->j_state_lock
);
4850 jbd2_log_wait_commit(journal
, commit_tid
);
4857 * Called from notify_change.
4859 * We want to trap VFS attempts to truncate the file as soon as
4860 * possible. In particular, we want to make sure that when the VFS
4861 * shrinks i_size, we put the inode on the orphan list and modify
4862 * i_disksize immediately, so that during the subsequent flushing of
4863 * dirty pages and freeing of disk blocks, we can guarantee that any
4864 * commit will leave the blocks being flushed in an unused state on
4865 * disk. (On recovery, the inode will get truncated and the blocks will
4866 * be freed, so we have a strong guarantee that no future commit will
4867 * leave these blocks visible to the user.)
4869 * Another thing we have to assure is that if we are in ordered mode
4870 * and inode is still attached to the committing transaction, we must
4871 * we start writeout of all the dirty pages which are being truncated.
4872 * This way we are sure that all the data written in the previous
4873 * transaction are already on disk (truncate waits for pages under
4876 * Called with inode->i_mutex down.
4878 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4880 struct inode
*inode
= d_inode(dentry
);
4883 const unsigned int ia_valid
= attr
->ia_valid
;
4885 error
= inode_change_ok(inode
, attr
);
4889 if (is_quota_modification(inode
, attr
)) {
4890 error
= dquot_initialize(inode
);
4894 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4895 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4898 /* (user+group)*(old+new) structure, inode write (sb,
4899 * inode block, ? - but truncate inode update has it) */
4900 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4901 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4902 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4903 if (IS_ERR(handle
)) {
4904 error
= PTR_ERR(handle
);
4907 error
= dquot_transfer(inode
, attr
);
4909 ext4_journal_stop(handle
);
4912 /* Update corresponding info in inode so that everything is in
4913 * one transaction */
4914 if (attr
->ia_valid
& ATTR_UID
)
4915 inode
->i_uid
= attr
->ia_uid
;
4916 if (attr
->ia_valid
& ATTR_GID
)
4917 inode
->i_gid
= attr
->ia_gid
;
4918 error
= ext4_mark_inode_dirty(handle
, inode
);
4919 ext4_journal_stop(handle
);
4922 if (attr
->ia_valid
& ATTR_SIZE
) {
4924 loff_t oldsize
= inode
->i_size
;
4925 int shrink
= (attr
->ia_size
<= inode
->i_size
);
4927 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4928 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4930 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4933 if (!S_ISREG(inode
->i_mode
))
4936 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4937 inode_inc_iversion(inode
);
4939 if (ext4_should_order_data(inode
) &&
4940 (attr
->ia_size
< inode
->i_size
)) {
4941 error
= ext4_begin_ordered_truncate(inode
,
4946 if (attr
->ia_size
!= inode
->i_size
) {
4947 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4948 if (IS_ERR(handle
)) {
4949 error
= PTR_ERR(handle
);
4952 if (ext4_handle_valid(handle
) && shrink
) {
4953 error
= ext4_orphan_add(handle
, inode
);
4957 * Update c/mtime on truncate up, ext4_truncate() will
4958 * update c/mtime in shrink case below
4961 inode
->i_mtime
= ext4_current_time(inode
);
4962 inode
->i_ctime
= inode
->i_mtime
;
4964 down_write(&EXT4_I(inode
)->i_data_sem
);
4965 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4966 rc
= ext4_mark_inode_dirty(handle
, inode
);
4970 * We have to update i_size under i_data_sem together
4971 * with i_disksize to avoid races with writeback code
4972 * running ext4_wb_update_i_disksize().
4975 i_size_write(inode
, attr
->ia_size
);
4976 up_write(&EXT4_I(inode
)->i_data_sem
);
4977 ext4_journal_stop(handle
);
4980 ext4_orphan_del(NULL
, inode
);
4985 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
4988 * Blocks are going to be removed from the inode. Wait
4989 * for dio in flight. Temporarily disable
4990 * dioread_nolock to prevent livelock.
4993 if (!ext4_should_journal_data(inode
)) {
4994 ext4_inode_block_unlocked_dio(inode
);
4995 inode_dio_wait(inode
);
4996 ext4_inode_resume_unlocked_dio(inode
);
4998 ext4_wait_for_tail_page_commit(inode
);
5000 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5002 * Truncate pagecache after we've waited for commit
5003 * in data=journal mode to make pages freeable.
5005 truncate_pagecache(inode
, inode
->i_size
);
5007 ext4_truncate(inode
);
5008 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5012 setattr_copy(inode
, attr
);
5013 mark_inode_dirty(inode
);
5017 * If the call to ext4_truncate failed to get a transaction handle at
5018 * all, we need to clean up the in-core orphan list manually.
5020 if (orphan
&& inode
->i_nlink
)
5021 ext4_orphan_del(NULL
, inode
);
5023 if (!rc
&& (ia_valid
& ATTR_MODE
))
5024 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5027 ext4_std_error(inode
->i_sb
, error
);
5033 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5036 struct inode
*inode
;
5037 unsigned long long delalloc_blocks
;
5039 inode
= d_inode(dentry
);
5040 generic_fillattr(inode
, stat
);
5043 * If there is inline data in the inode, the inode will normally not
5044 * have data blocks allocated (it may have an external xattr block).
5045 * Report at least one sector for such files, so tools like tar, rsync,
5046 * others doen't incorrectly think the file is completely sparse.
5048 if (unlikely(ext4_has_inline_data(inode
)))
5049 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5052 * We can't update i_blocks if the block allocation is delayed
5053 * otherwise in the case of system crash before the real block
5054 * allocation is done, we will have i_blocks inconsistent with
5055 * on-disk file blocks.
5056 * We always keep i_blocks updated together with real
5057 * allocation. But to not confuse with user, stat
5058 * will return the blocks that include the delayed allocation
5059 * blocks for this file.
5061 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5062 EXT4_I(inode
)->i_reserved_data_blocks
);
5063 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5067 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5070 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5071 return ext4_ind_trans_blocks(inode
, lblocks
);
5072 return ext4_ext_index_trans_blocks(inode
, pextents
);
5076 * Account for index blocks, block groups bitmaps and block group
5077 * descriptor blocks if modify datablocks and index blocks
5078 * worse case, the indexs blocks spread over different block groups
5080 * If datablocks are discontiguous, they are possible to spread over
5081 * different block groups too. If they are contiguous, with flexbg,
5082 * they could still across block group boundary.
5084 * Also account for superblock, inode, quota and xattr blocks
5086 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5089 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5095 * How many index blocks need to touch to map @lblocks logical blocks
5096 * to @pextents physical extents?
5098 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5103 * Now let's see how many group bitmaps and group descriptors need
5106 groups
= idxblocks
+ pextents
;
5108 if (groups
> ngroups
)
5110 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5111 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5113 /* bitmaps and block group descriptor blocks */
5114 ret
+= groups
+ gdpblocks
;
5116 /* Blocks for super block, inode, quota and xattr blocks */
5117 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5123 * Calculate the total number of credits to reserve to fit
5124 * the modification of a single pages into a single transaction,
5125 * which may include multiple chunks of block allocations.
5127 * This could be called via ext4_write_begin()
5129 * We need to consider the worse case, when
5130 * one new block per extent.
5132 int ext4_writepage_trans_blocks(struct inode
*inode
)
5134 int bpp
= ext4_journal_blocks_per_page(inode
);
5137 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5139 /* Account for data blocks for journalled mode */
5140 if (ext4_should_journal_data(inode
))
5146 * Calculate the journal credits for a chunk of data modification.
5148 * This is called from DIO, fallocate or whoever calling
5149 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5151 * journal buffers for data blocks are not included here, as DIO
5152 * and fallocate do no need to journal data buffers.
5154 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5156 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5160 * The caller must have previously called ext4_reserve_inode_write().
5161 * Give this, we know that the caller already has write access to iloc->bh.
5163 int ext4_mark_iloc_dirty(handle_t
*handle
,
5164 struct inode
*inode
, struct ext4_iloc
*iloc
)
5168 if (IS_I_VERSION(inode
))
5169 inode_inc_iversion(inode
);
5171 /* the do_update_inode consumes one bh->b_count */
5174 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5175 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5181 * On success, We end up with an outstanding reference count against
5182 * iloc->bh. This _must_ be cleaned up later.
5186 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5187 struct ext4_iloc
*iloc
)
5191 err
= ext4_get_inode_loc(inode
, iloc
);
5193 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5194 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5200 ext4_std_error(inode
->i_sb
, err
);
5205 * Expand an inode by new_extra_isize bytes.
5206 * Returns 0 on success or negative error number on failure.
5208 static int ext4_expand_extra_isize(struct inode
*inode
,
5209 unsigned int new_extra_isize
,
5210 struct ext4_iloc iloc
,
5213 struct ext4_inode
*raw_inode
;
5214 struct ext4_xattr_ibody_header
*header
;
5216 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5219 raw_inode
= ext4_raw_inode(&iloc
);
5221 header
= IHDR(inode
, raw_inode
);
5223 /* No extended attributes present */
5224 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5225 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5226 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5228 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5232 /* try to expand with EAs present */
5233 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5238 * What we do here is to mark the in-core inode as clean with respect to inode
5239 * dirtiness (it may still be data-dirty).
5240 * This means that the in-core inode may be reaped by prune_icache
5241 * without having to perform any I/O. This is a very good thing,
5242 * because *any* task may call prune_icache - even ones which
5243 * have a transaction open against a different journal.
5245 * Is this cheating? Not really. Sure, we haven't written the
5246 * inode out, but prune_icache isn't a user-visible syncing function.
5247 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5248 * we start and wait on commits.
5250 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5252 struct ext4_iloc iloc
;
5253 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5254 static unsigned int mnt_count
;
5258 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5259 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5260 if (ext4_handle_valid(handle
) &&
5261 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5262 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5264 * We need extra buffer credits since we may write into EA block
5265 * with this same handle. If journal_extend fails, then it will
5266 * only result in a minor loss of functionality for that inode.
5267 * If this is felt to be critical, then e2fsck should be run to
5268 * force a large enough s_min_extra_isize.
5270 if ((jbd2_journal_extend(handle
,
5271 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5272 ret
= ext4_expand_extra_isize(inode
,
5273 sbi
->s_want_extra_isize
,
5276 ext4_set_inode_state(inode
,
5277 EXT4_STATE_NO_EXPAND
);
5279 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5280 ext4_warning(inode
->i_sb
,
5281 "Unable to expand inode %lu. Delete"
5282 " some EAs or run e2fsck.",
5285 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5291 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5296 * ext4_dirty_inode() is called from __mark_inode_dirty()
5298 * We're really interested in the case where a file is being extended.
5299 * i_size has been changed by generic_commit_write() and we thus need
5300 * to include the updated inode in the current transaction.
5302 * Also, dquot_alloc_block() will always dirty the inode when blocks
5303 * are allocated to the file.
5305 * If the inode is marked synchronous, we don't honour that here - doing
5306 * so would cause a commit on atime updates, which we don't bother doing.
5307 * We handle synchronous inodes at the highest possible level.
5309 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5310 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5311 * to copy into the on-disk inode structure are the timestamp files.
5313 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5317 if (flags
== I_DIRTY_TIME
)
5319 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5323 ext4_mark_inode_dirty(handle
, inode
);
5325 ext4_journal_stop(handle
);
5332 * Bind an inode's backing buffer_head into this transaction, to prevent
5333 * it from being flushed to disk early. Unlike
5334 * ext4_reserve_inode_write, this leaves behind no bh reference and
5335 * returns no iloc structure, so the caller needs to repeat the iloc
5336 * lookup to mark the inode dirty later.
5338 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5340 struct ext4_iloc iloc
;
5344 err
= ext4_get_inode_loc(inode
, &iloc
);
5346 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5347 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5349 err
= ext4_handle_dirty_metadata(handle
,
5355 ext4_std_error(inode
->i_sb
, err
);
5360 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5367 * We have to be very careful here: changing a data block's
5368 * journaling status dynamically is dangerous. If we write a
5369 * data block to the journal, change the status and then delete
5370 * that block, we risk forgetting to revoke the old log record
5371 * from the journal and so a subsequent replay can corrupt data.
5372 * So, first we make sure that the journal is empty and that
5373 * nobody is changing anything.
5376 journal
= EXT4_JOURNAL(inode
);
5379 if (is_journal_aborted(journal
))
5381 /* We have to allocate physical blocks for delalloc blocks
5382 * before flushing journal. otherwise delalloc blocks can not
5383 * be allocated any more. even more truncate on delalloc blocks
5384 * could trigger BUG by flushing delalloc blocks in journal.
5385 * There is no delalloc block in non-journal data mode.
5387 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5388 err
= ext4_alloc_da_blocks(inode
);
5393 /* Wait for all existing dio workers */
5394 ext4_inode_block_unlocked_dio(inode
);
5395 inode_dio_wait(inode
);
5397 jbd2_journal_lock_updates(journal
);
5400 * OK, there are no updates running now, and all cached data is
5401 * synced to disk. We are now in a completely consistent state
5402 * which doesn't have anything in the journal, and we know that
5403 * no filesystem updates are running, so it is safe to modify
5404 * the inode's in-core data-journaling state flag now.
5408 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5410 err
= jbd2_journal_flush(journal
);
5412 jbd2_journal_unlock_updates(journal
);
5413 ext4_inode_resume_unlocked_dio(inode
);
5416 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5418 ext4_set_aops(inode
);
5420 jbd2_journal_unlock_updates(journal
);
5421 ext4_inode_resume_unlocked_dio(inode
);
5423 /* Finally we can mark the inode as dirty. */
5425 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5427 return PTR_ERR(handle
);
5429 err
= ext4_mark_inode_dirty(handle
, inode
);
5430 ext4_handle_sync(handle
);
5431 ext4_journal_stop(handle
);
5432 ext4_std_error(inode
->i_sb
, err
);
5437 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5439 return !buffer_mapped(bh
);
5442 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5444 struct page
*page
= vmf
->page
;
5448 struct file
*file
= vma
->vm_file
;
5449 struct inode
*inode
= file_inode(file
);
5450 struct address_space
*mapping
= inode
->i_mapping
;
5452 get_block_t
*get_block
;
5455 sb_start_pagefault(inode
->i_sb
);
5456 file_update_time(vma
->vm_file
);
5458 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5459 /* Delalloc case is easy... */
5460 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5461 !ext4_should_journal_data(inode
) &&
5462 !ext4_nonda_switch(inode
->i_sb
)) {
5464 ret
= block_page_mkwrite(vma
, vmf
,
5465 ext4_da_get_block_prep
);
5466 } while (ret
== -ENOSPC
&&
5467 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5472 size
= i_size_read(inode
);
5473 /* Page got truncated from under us? */
5474 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5476 ret
= VM_FAULT_NOPAGE
;
5480 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5481 len
= size
& ~PAGE_CACHE_MASK
;
5483 len
= PAGE_CACHE_SIZE
;
5485 * Return if we have all the buffers mapped. This avoids the need to do
5486 * journal_start/journal_stop which can block and take a long time
5488 if (page_has_buffers(page
)) {
5489 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5491 ext4_bh_unmapped
)) {
5492 /* Wait so that we don't change page under IO */
5493 wait_for_stable_page(page
);
5494 ret
= VM_FAULT_LOCKED
;
5499 /* OK, we need to fill the hole... */
5500 if (ext4_should_dioread_nolock(inode
))
5501 get_block
= ext4_get_block_write
;
5503 get_block
= ext4_get_block
;
5505 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5506 ext4_writepage_trans_blocks(inode
));
5507 if (IS_ERR(handle
)) {
5508 ret
= VM_FAULT_SIGBUS
;
5511 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5512 if (!ret
&& ext4_should_journal_data(inode
)) {
5513 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5514 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5516 ret
= VM_FAULT_SIGBUS
;
5517 ext4_journal_stop(handle
);
5520 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5522 ext4_journal_stop(handle
);
5523 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5526 ret
= block_page_mkwrite_return(ret
);
5528 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5529 sb_end_pagefault(inode
->i_sb
);
5533 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5535 struct inode
*inode
= file_inode(vma
->vm_file
);
5538 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5539 err
= filemap_fault(vma
, vmf
);
5540 up_read(&EXT4_I(inode
)->i_mmap_sem
);