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>
40 #include <linux/iomap.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
52 struct ext4_inode_info
*ei
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
57 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
58 unsigned int csum_size
= sizeof(dummy_csum
);
60 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
61 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
63 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
64 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
66 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
67 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
68 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
69 EXT4_GOOD_OLD_INODE_SIZE
,
70 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
71 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
72 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
75 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
76 EXT4_INODE_SIZE(inode
->i_sb
) -
84 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
85 struct ext4_inode_info
*ei
)
87 __u32 provided
, calculated
;
89 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
90 cpu_to_le32(EXT4_OS_LINUX
) ||
91 !ext4_has_metadata_csum(inode
->i_sb
))
94 provided
= le16_to_cpu(raw
->i_checksum_lo
);
95 calculated
= ext4_inode_csum(inode
, raw
, ei
);
96 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
97 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
98 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
100 calculated
&= 0xFFFF;
102 return provided
== calculated
;
105 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
106 struct ext4_inode_info
*ei
)
110 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
111 cpu_to_le32(EXT4_OS_LINUX
) ||
112 !ext4_has_metadata_csum(inode
->i_sb
))
115 csum
= ext4_inode_csum(inode
, raw
, ei
);
116 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
117 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
118 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
119 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
125 trace_ext4_begin_ordered_truncate(inode
, new_size
);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode
)->jinode
)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
135 EXT4_I(inode
)->jinode
,
139 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
140 unsigned int length
);
141 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
142 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
143 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
147 * Test whether an inode is a fast symlink.
149 int ext4_inode_is_fast_symlink(struct inode
*inode
)
151 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
152 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
154 if (ext4_has_inline_data(inode
))
157 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
161 * Restart the transaction associated with *handle. This does a commit,
162 * so before we call here everything must be consistently dirtied against
165 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
171 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
172 * moment, get_block can be called only for blocks inside i_size since
173 * page cache has been already dropped and writes are blocked by
174 * i_mutex. So we can safely drop the i_data_sem here.
176 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
177 jbd_debug(2, "restarting handle %p\n", handle
);
178 up_write(&EXT4_I(inode
)->i_data_sem
);
179 ret
= ext4_journal_restart(handle
, nblocks
);
180 down_write(&EXT4_I(inode
)->i_data_sem
);
181 ext4_discard_preallocations(inode
);
187 * Called at the last iput() if i_nlink is zero.
189 void ext4_evict_inode(struct inode
*inode
)
194 trace_ext4_evict_inode(inode
);
196 if (inode
->i_nlink
) {
198 * When journalling data dirty buffers are tracked only in the
199 * journal. So although mm thinks everything is clean and
200 * ready for reaping the inode might still have some pages to
201 * write in the running transaction or waiting to be
202 * checkpointed. Thus calling jbd2_journal_invalidatepage()
203 * (via truncate_inode_pages()) to discard these buffers can
204 * cause data loss. Also even if we did not discard these
205 * buffers, we would have no way to find them after the inode
206 * is reaped and thus user could see stale data if he tries to
207 * read them before the transaction is checkpointed. So be
208 * careful and force everything to disk here... We use
209 * ei->i_datasync_tid to store the newest transaction
210 * containing inode's data.
212 * Note that directories do not have this problem because they
213 * don't use page cache.
215 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
216 ext4_should_journal_data(inode
) &&
217 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
218 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
219 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
221 jbd2_complete_transaction(journal
, commit_tid
);
222 filemap_write_and_wait(&inode
->i_data
);
224 truncate_inode_pages_final(&inode
->i_data
);
229 if (is_bad_inode(inode
))
231 dquot_initialize(inode
);
233 if (ext4_should_order_data(inode
))
234 ext4_begin_ordered_truncate(inode
, 0);
235 truncate_inode_pages_final(&inode
->i_data
);
238 * Protect us against freezing - iput() caller didn't have to have any
239 * protection against it
241 sb_start_intwrite(inode
->i_sb
);
242 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
243 ext4_blocks_for_truncate(inode
)+3);
244 if (IS_ERR(handle
)) {
245 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
247 * If we're going to skip the normal cleanup, we still need to
248 * make sure that the in-core orphan linked list is properly
251 ext4_orphan_del(NULL
, inode
);
252 sb_end_intwrite(inode
->i_sb
);
257 ext4_handle_sync(handle
);
259 err
= ext4_mark_inode_dirty(handle
, inode
);
261 ext4_warning(inode
->i_sb
,
262 "couldn't mark inode dirty (err %d)", err
);
265 if (inode
->i_blocks
) {
266 err
= ext4_truncate(inode
);
268 ext4_error(inode
->i_sb
,
269 "couldn't truncate inode %lu (err %d)",
276 * ext4_ext_truncate() doesn't reserve any slop when it
277 * restarts journal transactions; therefore there may not be
278 * enough credits left in the handle to remove the inode from
279 * the orphan list and set the dtime field.
281 if (!ext4_handle_has_enough_credits(handle
, 3)) {
282 err
= ext4_journal_extend(handle
, 3);
284 err
= ext4_journal_restart(handle
, 3);
286 ext4_warning(inode
->i_sb
,
287 "couldn't extend journal (err %d)", err
);
289 ext4_journal_stop(handle
);
290 ext4_orphan_del(NULL
, inode
);
291 sb_end_intwrite(inode
->i_sb
);
297 * Kill off the orphan record which ext4_truncate created.
298 * AKPM: I think this can be inside the above `if'.
299 * Note that ext4_orphan_del() has to be able to cope with the
300 * deletion of a non-existent orphan - this is because we don't
301 * know if ext4_truncate() actually created an orphan record.
302 * (Well, we could do this if we need to, but heck - it works)
304 ext4_orphan_del(handle
, inode
);
305 EXT4_I(inode
)->i_dtime
= get_seconds();
308 * One subtle ordering requirement: if anything has gone wrong
309 * (transaction abort, IO errors, whatever), then we can still
310 * do these next steps (the fs will already have been marked as
311 * having errors), but we can't free the inode if the mark_dirty
314 if (ext4_mark_inode_dirty(handle
, inode
))
315 /* If that failed, just do the required in-core inode clear. */
316 ext4_clear_inode(inode
);
318 ext4_free_inode(handle
, inode
);
319 ext4_journal_stop(handle
);
320 sb_end_intwrite(inode
->i_sb
);
323 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
327 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
329 return &EXT4_I(inode
)->i_reserved_quota
;
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
337 void ext4_da_update_reserve_space(struct inode
*inode
,
338 int used
, int quota_claim
)
340 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
341 struct ext4_inode_info
*ei
= EXT4_I(inode
);
343 spin_lock(&ei
->i_block_reservation_lock
);
344 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
345 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
346 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__
, inode
->i_ino
, used
,
349 ei
->i_reserved_data_blocks
);
351 used
= ei
->i_reserved_data_blocks
;
354 /* Update per-inode reservations */
355 ei
->i_reserved_data_blocks
-= used
;
356 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
358 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
360 /* Update quota subsystem for data blocks */
362 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
365 * We did fallocate with an offset that is already delayed
366 * allocated. So on delayed allocated writeback we should
367 * not re-claim the quota for fallocated blocks.
369 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
373 * If we have done all the pending block allocations and if
374 * there aren't any writers on the inode, we can discard the
375 * inode's preallocations.
377 if ((ei
->i_reserved_data_blocks
== 0) &&
378 (atomic_read(&inode
->i_writecount
) == 0))
379 ext4_discard_preallocations(inode
);
382 static int __check_block_validity(struct inode
*inode
, const char *func
,
384 struct ext4_map_blocks
*map
)
386 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
388 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
389 "lblock %lu mapped to illegal pblock "
390 "(length %d)", (unsigned long) map
->m_lblk
,
392 return -EFSCORRUPTED
;
397 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
402 if (ext4_encrypted_inode(inode
))
403 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
405 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
412 #define check_block_validity(inode, map) \
413 __check_block_validity((inode), __func__, __LINE__, (map))
415 #ifdef ES_AGGRESSIVE_TEST
416 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
418 struct ext4_map_blocks
*es_map
,
419 struct ext4_map_blocks
*map
,
426 * There is a race window that the result is not the same.
427 * e.g. xfstests #223 when dioread_nolock enables. The reason
428 * is that we lookup a block mapping in extent status tree with
429 * out taking i_data_sem. So at the time the unwritten extent
430 * could be converted.
432 down_read(&EXT4_I(inode
)->i_data_sem
);
433 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
434 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
435 EXT4_GET_BLOCKS_KEEP_SIZE
);
437 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
438 EXT4_GET_BLOCKS_KEEP_SIZE
);
440 up_read((&EXT4_I(inode
)->i_data_sem
));
443 * We don't check m_len because extent will be collpased in status
444 * tree. So the m_len might not equal.
446 if (es_map
->m_lblk
!= map
->m_lblk
||
447 es_map
->m_flags
!= map
->m_flags
||
448 es_map
->m_pblk
!= map
->m_pblk
) {
449 printk("ES cache assertion failed for inode: %lu "
450 "es_cached ex [%d/%d/%llu/%x] != "
451 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
452 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
453 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
454 map
->m_len
, map
->m_pblk
, map
->m_flags
,
458 #endif /* ES_AGGRESSIVE_TEST */
461 * The ext4_map_blocks() function tries to look up the requested blocks,
462 * and returns if the blocks are already mapped.
464 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
465 * and store the allocated blocks in the result buffer head and mark it
468 * If file type is extents based, it will call ext4_ext_map_blocks(),
469 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
472 * On success, it returns the number of blocks being mapped or allocated. if
473 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
474 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
476 * It returns 0 if plain look up failed (blocks have not been allocated), in
477 * that case, @map is returned as unmapped but we still do fill map->m_len to
478 * indicate the length of a hole starting at map->m_lblk.
480 * It returns the error in case of allocation failure.
482 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
483 struct ext4_map_blocks
*map
, int flags
)
485 struct extent_status es
;
488 #ifdef ES_AGGRESSIVE_TEST
489 struct ext4_map_blocks orig_map
;
491 memcpy(&orig_map
, map
, sizeof(*map
));
495 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
496 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
497 (unsigned long) map
->m_lblk
);
500 * ext4_map_blocks returns an int, and m_len is an unsigned int
502 if (unlikely(map
->m_len
> INT_MAX
))
503 map
->m_len
= INT_MAX
;
505 /* We can handle the block number less than EXT_MAX_BLOCKS */
506 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
507 return -EFSCORRUPTED
;
509 /* Lookup extent status tree firstly */
510 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
511 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
512 map
->m_pblk
= ext4_es_pblock(&es
) +
513 map
->m_lblk
- es
.es_lblk
;
514 map
->m_flags
|= ext4_es_is_written(&es
) ?
515 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
516 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
517 if (retval
> map
->m_len
)
520 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
522 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
523 if (retval
> map
->m_len
)
530 #ifdef ES_AGGRESSIVE_TEST
531 ext4_map_blocks_es_recheck(handle
, inode
, map
,
538 * Try to see if we can get the block without requesting a new
541 down_read(&EXT4_I(inode
)->i_data_sem
);
542 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
543 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
544 EXT4_GET_BLOCKS_KEEP_SIZE
);
546 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
547 EXT4_GET_BLOCKS_KEEP_SIZE
);
552 if (unlikely(retval
!= map
->m_len
)) {
553 ext4_warning(inode
->i_sb
,
554 "ES len assertion failed for inode "
555 "%lu: retval %d != map->m_len %d",
556 inode
->i_ino
, retval
, map
->m_len
);
560 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
561 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
562 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
563 !(status
& EXTENT_STATUS_WRITTEN
) &&
564 ext4_find_delalloc_range(inode
, map
->m_lblk
,
565 map
->m_lblk
+ map
->m_len
- 1))
566 status
|= EXTENT_STATUS_DELAYED
;
567 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
568 map
->m_len
, map
->m_pblk
, status
);
572 up_read((&EXT4_I(inode
)->i_data_sem
));
575 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
576 ret
= check_block_validity(inode
, map
);
581 /* If it is only a block(s) look up */
582 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
586 * Returns if the blocks have already allocated
588 * Note that if blocks have been preallocated
589 * ext4_ext_get_block() returns the create = 0
590 * with buffer head unmapped.
592 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
594 * If we need to convert extent to unwritten
595 * we continue and do the actual work in
596 * ext4_ext_map_blocks()
598 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
602 * Here we clear m_flags because after allocating an new extent,
603 * it will be set again.
605 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
608 * New blocks allocate and/or writing to unwritten extent
609 * will possibly result in updating i_data, so we take
610 * the write lock of i_data_sem, and call get_block()
611 * with create == 1 flag.
613 down_write(&EXT4_I(inode
)->i_data_sem
);
616 * We need to check for EXT4 here because migrate
617 * could have changed the inode type in between
619 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
620 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
622 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
624 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
626 * We allocated new blocks which will result in
627 * i_data's format changing. Force the migrate
628 * to fail by clearing migrate flags
630 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
634 * Update reserved blocks/metadata blocks after successful
635 * block allocation which had been deferred till now. We don't
636 * support fallocate for non extent files. So we can update
637 * reserve space here.
640 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
641 ext4_da_update_reserve_space(inode
, retval
, 1);
647 if (unlikely(retval
!= map
->m_len
)) {
648 ext4_warning(inode
->i_sb
,
649 "ES len assertion failed for inode "
650 "%lu: retval %d != map->m_len %d",
651 inode
->i_ino
, retval
, map
->m_len
);
656 * We have to zeroout blocks before inserting them into extent
657 * status tree. Otherwise someone could look them up there and
658 * use them before they are really zeroed. We also have to
659 * unmap metadata before zeroing as otherwise writeback can
660 * overwrite zeros with stale data from block device.
662 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
663 map
->m_flags
& EXT4_MAP_MAPPED
&&
664 map
->m_flags
& EXT4_MAP_NEW
) {
667 for (i
= 0; i
< map
->m_len
; i
++) {
668 unmap_underlying_metadata(inode
->i_sb
->s_bdev
,
671 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
672 map
->m_pblk
, map
->m_len
);
680 * If the extent has been zeroed out, we don't need to update
681 * extent status tree.
683 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
684 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
685 if (ext4_es_is_written(&es
))
688 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
689 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
690 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
691 !(status
& EXTENT_STATUS_WRITTEN
) &&
692 ext4_find_delalloc_range(inode
, map
->m_lblk
,
693 map
->m_lblk
+ map
->m_len
- 1))
694 status
|= EXTENT_STATUS_DELAYED
;
695 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
696 map
->m_pblk
, status
);
704 up_write((&EXT4_I(inode
)->i_data_sem
));
705 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
706 ret
= check_block_validity(inode
, map
);
711 * Inodes with freshly allocated blocks where contents will be
712 * visible after transaction commit must be on transaction's
715 if (map
->m_flags
& EXT4_MAP_NEW
&&
716 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
717 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
718 !IS_NOQUOTA(inode
) &&
719 ext4_should_order_data(inode
)) {
720 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
721 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
723 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
732 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
733 * we have to be careful as someone else may be manipulating b_state as well.
735 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
737 unsigned long old_state
;
738 unsigned long new_state
;
740 flags
&= EXT4_MAP_FLAGS
;
742 /* Dummy buffer_head? Set non-atomically. */
744 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
748 * Someone else may be modifying b_state. Be careful! This is ugly but
749 * once we get rid of using bh as a container for mapping information
750 * to pass to / from get_block functions, this can go away.
753 old_state
= READ_ONCE(bh
->b_state
);
754 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
756 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
759 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
760 struct buffer_head
*bh
, int flags
)
762 struct ext4_map_blocks map
;
765 if (ext4_has_inline_data(inode
))
769 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
771 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
774 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
775 ext4_update_bh_state(bh
, map
.m_flags
);
776 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
778 } else if (ret
== 0) {
779 /* hole case, need to fill in bh->b_size */
780 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
785 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
786 struct buffer_head
*bh
, int create
)
788 return _ext4_get_block(inode
, iblock
, bh
,
789 create
? EXT4_GET_BLOCKS_CREATE
: 0);
793 * Get block function used when preparing for buffered write if we require
794 * creating an unwritten extent if blocks haven't been allocated. The extent
795 * will be converted to written after the IO is complete.
797 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
798 struct buffer_head
*bh_result
, int create
)
800 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
801 inode
->i_ino
, create
);
802 return _ext4_get_block(inode
, iblock
, bh_result
,
803 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
806 /* Maximum number of blocks we map for direct IO at once. */
807 #define DIO_MAX_BLOCKS 4096
810 * Get blocks function for the cases that need to start a transaction -
811 * generally difference cases of direct IO and DAX IO. It also handles retries
814 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
815 struct buffer_head
*bh_result
, int flags
)
822 /* Trim mapping request to maximum we can map at once for DIO */
823 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
824 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
825 dio_credits
= ext4_chunk_trans_blocks(inode
,
826 bh_result
->b_size
>> inode
->i_blkbits
);
828 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
830 return PTR_ERR(handle
);
832 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
833 ext4_journal_stop(handle
);
835 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
840 /* Get block function for DIO reads and writes to inodes without extents */
841 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
842 struct buffer_head
*bh
, int create
)
844 /* We don't expect handle for direct IO */
845 WARN_ON_ONCE(ext4_journal_current_handle());
848 return _ext4_get_block(inode
, iblock
, bh
, 0);
849 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
853 * Get block function for AIO DIO writes when we create unwritten extent if
854 * blocks are not allocated yet. The extent will be converted to written
855 * after IO is complete.
857 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
858 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
862 /* We don't expect handle for direct IO */
863 WARN_ON_ONCE(ext4_journal_current_handle());
865 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
866 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
869 * When doing DIO using unwritten extents, we need io_end to convert
870 * unwritten extents to written on IO completion. We allocate io_end
871 * once we spot unwritten extent and store it in b_private. Generic
872 * DIO code keeps b_private set and furthermore passes the value to
873 * our completion callback in 'private' argument.
875 if (!ret
&& buffer_unwritten(bh_result
)) {
876 if (!bh_result
->b_private
) {
877 ext4_io_end_t
*io_end
;
879 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
882 bh_result
->b_private
= io_end
;
883 ext4_set_io_unwritten_flag(inode
, io_end
);
885 set_buffer_defer_completion(bh_result
);
892 * Get block function for non-AIO DIO writes when we create unwritten extent if
893 * blocks are not allocated yet. The extent will be converted to written
894 * after IO is complete from ext4_ext_direct_IO() function.
896 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
897 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
901 /* We don't expect handle for direct IO */
902 WARN_ON_ONCE(ext4_journal_current_handle());
904 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
905 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
908 * Mark inode as having pending DIO writes to unwritten extents.
909 * ext4_ext_direct_IO() checks this flag and converts extents to
912 if (!ret
&& buffer_unwritten(bh_result
))
913 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
918 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
919 struct buffer_head
*bh_result
, int create
)
923 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
924 inode
->i_ino
, create
);
925 /* We don't expect handle for direct IO */
926 WARN_ON_ONCE(ext4_journal_current_handle());
928 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
930 * Blocks should have been preallocated! ext4_file_write_iter() checks
933 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
940 * `handle' can be NULL if create is zero
942 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
943 ext4_lblk_t block
, int map_flags
)
945 struct ext4_map_blocks map
;
946 struct buffer_head
*bh
;
947 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
950 J_ASSERT(handle
!= NULL
|| create
== 0);
954 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
957 return create
? ERR_PTR(-ENOSPC
) : NULL
;
961 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
963 return ERR_PTR(-ENOMEM
);
964 if (map
.m_flags
& EXT4_MAP_NEW
) {
965 J_ASSERT(create
!= 0);
966 J_ASSERT(handle
!= NULL
);
969 * Now that we do not always journal data, we should
970 * keep in mind whether this should always journal the
971 * new buffer as metadata. For now, regular file
972 * writes use ext4_get_block instead, so it's not a
976 BUFFER_TRACE(bh
, "call get_create_access");
977 err
= ext4_journal_get_create_access(handle
, bh
);
982 if (!buffer_uptodate(bh
)) {
983 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
984 set_buffer_uptodate(bh
);
987 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
988 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
992 BUFFER_TRACE(bh
, "not a new buffer");
999 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1000 ext4_lblk_t block
, int map_flags
)
1002 struct buffer_head
*bh
;
1004 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1007 if (!bh
|| buffer_uptodate(bh
))
1009 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1011 if (buffer_uptodate(bh
))
1014 return ERR_PTR(-EIO
);
1017 int ext4_walk_page_buffers(handle_t
*handle
,
1018 struct buffer_head
*head
,
1022 int (*fn
)(handle_t
*handle
,
1023 struct buffer_head
*bh
))
1025 struct buffer_head
*bh
;
1026 unsigned block_start
, block_end
;
1027 unsigned blocksize
= head
->b_size
;
1029 struct buffer_head
*next
;
1031 for (bh
= head
, block_start
= 0;
1032 ret
== 0 && (bh
!= head
|| !block_start
);
1033 block_start
= block_end
, bh
= next
) {
1034 next
= bh
->b_this_page
;
1035 block_end
= block_start
+ blocksize
;
1036 if (block_end
<= from
|| block_start
>= to
) {
1037 if (partial
&& !buffer_uptodate(bh
))
1041 err
= (*fn
)(handle
, bh
);
1049 * To preserve ordering, it is essential that the hole instantiation and
1050 * the data write be encapsulated in a single transaction. We cannot
1051 * close off a transaction and start a new one between the ext4_get_block()
1052 * and the commit_write(). So doing the jbd2_journal_start at the start of
1053 * prepare_write() is the right place.
1055 * Also, this function can nest inside ext4_writepage(). In that case, we
1056 * *know* that ext4_writepage() has generated enough buffer credits to do the
1057 * whole page. So we won't block on the journal in that case, which is good,
1058 * because the caller may be PF_MEMALLOC.
1060 * By accident, ext4 can be reentered when a transaction is open via
1061 * quota file writes. If we were to commit the transaction while thus
1062 * reentered, there can be a deadlock - we would be holding a quota
1063 * lock, and the commit would never complete if another thread had a
1064 * transaction open and was blocking on the quota lock - a ranking
1067 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1068 * will _not_ run commit under these circumstances because handle->h_ref
1069 * is elevated. We'll still have enough credits for the tiny quotafile
1072 int do_journal_get_write_access(handle_t
*handle
,
1073 struct buffer_head
*bh
)
1075 int dirty
= buffer_dirty(bh
);
1078 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1081 * __block_write_begin() could have dirtied some buffers. Clean
1082 * the dirty bit as jbd2_journal_get_write_access() could complain
1083 * otherwise about fs integrity issues. Setting of the dirty bit
1084 * by __block_write_begin() isn't a real problem here as we clear
1085 * the bit before releasing a page lock and thus writeback cannot
1086 * ever write the buffer.
1089 clear_buffer_dirty(bh
);
1090 BUFFER_TRACE(bh
, "get write access");
1091 ret
= ext4_journal_get_write_access(handle
, bh
);
1093 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1097 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1098 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1099 get_block_t
*get_block
)
1101 unsigned from
= pos
& (PAGE_SIZE
- 1);
1102 unsigned to
= from
+ len
;
1103 struct inode
*inode
= page
->mapping
->host
;
1104 unsigned block_start
, block_end
;
1107 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1109 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1110 bool decrypt
= false;
1112 BUG_ON(!PageLocked(page
));
1113 BUG_ON(from
> PAGE_SIZE
);
1114 BUG_ON(to
> PAGE_SIZE
);
1117 if (!page_has_buffers(page
))
1118 create_empty_buffers(page
, blocksize
, 0);
1119 head
= page_buffers(page
);
1120 bbits
= ilog2(blocksize
);
1121 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1123 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1124 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1125 block_end
= block_start
+ blocksize
;
1126 if (block_end
<= from
|| block_start
>= to
) {
1127 if (PageUptodate(page
)) {
1128 if (!buffer_uptodate(bh
))
1129 set_buffer_uptodate(bh
);
1134 clear_buffer_new(bh
);
1135 if (!buffer_mapped(bh
)) {
1136 WARN_ON(bh
->b_size
!= blocksize
);
1137 err
= get_block(inode
, block
, bh
, 1);
1140 if (buffer_new(bh
)) {
1141 unmap_underlying_metadata(bh
->b_bdev
,
1143 if (PageUptodate(page
)) {
1144 clear_buffer_new(bh
);
1145 set_buffer_uptodate(bh
);
1146 mark_buffer_dirty(bh
);
1149 if (block_end
> to
|| block_start
< from
)
1150 zero_user_segments(page
, to
, block_end
,
1155 if (PageUptodate(page
)) {
1156 if (!buffer_uptodate(bh
))
1157 set_buffer_uptodate(bh
);
1160 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1161 !buffer_unwritten(bh
) &&
1162 (block_start
< from
|| block_end
> to
)) {
1163 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1165 decrypt
= ext4_encrypted_inode(inode
) &&
1166 S_ISREG(inode
->i_mode
);
1170 * If we issued read requests, let them complete.
1172 while (wait_bh
> wait
) {
1173 wait_on_buffer(*--wait_bh
);
1174 if (!buffer_uptodate(*wait_bh
))
1178 page_zero_new_buffers(page
, from
, to
);
1180 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1181 PAGE_SIZE
, 0, page
->index
);
1186 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1187 loff_t pos
, unsigned len
, unsigned flags
,
1188 struct page
**pagep
, void **fsdata
)
1190 struct inode
*inode
= mapping
->host
;
1191 int ret
, needed_blocks
;
1198 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1200 * Reserve one block more for addition to orphan list in case
1201 * we allocate blocks but write fails for some reason
1203 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1204 index
= pos
>> PAGE_SHIFT
;
1205 from
= pos
& (PAGE_SIZE
- 1);
1208 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1209 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1218 * grab_cache_page_write_begin() can take a long time if the
1219 * system is thrashing due to memory pressure, or if the page
1220 * is being written back. So grab it first before we start
1221 * the transaction handle. This also allows us to allocate
1222 * the page (if needed) without using GFP_NOFS.
1225 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1231 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1232 if (IS_ERR(handle
)) {
1234 return PTR_ERR(handle
);
1238 if (page
->mapping
!= mapping
) {
1239 /* The page got truncated from under us */
1242 ext4_journal_stop(handle
);
1245 /* In case writeback began while the page was unlocked */
1246 wait_for_stable_page(page
);
1248 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1249 if (ext4_should_dioread_nolock(inode
))
1250 ret
= ext4_block_write_begin(page
, pos
, len
,
1251 ext4_get_block_unwritten
);
1253 ret
= ext4_block_write_begin(page
, pos
, len
,
1256 if (ext4_should_dioread_nolock(inode
))
1257 ret
= __block_write_begin(page
, pos
, len
,
1258 ext4_get_block_unwritten
);
1260 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1262 if (!ret
&& ext4_should_journal_data(inode
)) {
1263 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1265 do_journal_get_write_access
);
1271 * __block_write_begin may have instantiated a few blocks
1272 * outside i_size. Trim these off again. Don't need
1273 * i_size_read because we hold i_mutex.
1275 * Add inode to orphan list in case we crash before
1278 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1279 ext4_orphan_add(handle
, inode
);
1281 ext4_journal_stop(handle
);
1282 if (pos
+ len
> inode
->i_size
) {
1283 ext4_truncate_failed_write(inode
);
1285 * If truncate failed early the inode might
1286 * still be on the orphan list; we need to
1287 * make sure the inode is removed from the
1288 * orphan list in that case.
1291 ext4_orphan_del(NULL
, inode
);
1294 if (ret
== -ENOSPC
&&
1295 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1304 /* For write_end() in data=journal mode */
1305 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1308 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1310 set_buffer_uptodate(bh
);
1311 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1312 clear_buffer_meta(bh
);
1313 clear_buffer_prio(bh
);
1318 * We need to pick up the new inode size which generic_commit_write gave us
1319 * `file' can be NULL - eg, when called from page_symlink().
1321 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1322 * buffers are managed internally.
1324 static int ext4_write_end(struct file
*file
,
1325 struct address_space
*mapping
,
1326 loff_t pos
, unsigned len
, unsigned copied
,
1327 struct page
*page
, void *fsdata
)
1329 handle_t
*handle
= ext4_journal_current_handle();
1330 struct inode
*inode
= mapping
->host
;
1331 loff_t old_size
= inode
->i_size
;
1333 int i_size_changed
= 0;
1335 trace_ext4_write_end(inode
, pos
, len
, copied
);
1336 if (ext4_has_inline_data(inode
)) {
1337 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1343 copied
= block_write_end(file
, mapping
, pos
,
1344 len
, copied
, page
, fsdata
);
1346 * it's important to update i_size while still holding page lock:
1347 * page writeout could otherwise come in and zero beyond i_size.
1349 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1354 pagecache_isize_extended(inode
, old_size
, pos
);
1356 * Don't mark the inode dirty under page lock. First, it unnecessarily
1357 * makes the holding time of page lock longer. Second, it forces lock
1358 * ordering of page lock and transaction start for journaling
1362 ext4_mark_inode_dirty(handle
, inode
);
1364 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1365 /* if we have allocated more blocks and copied
1366 * less. We will have blocks allocated outside
1367 * inode->i_size. So truncate them
1369 ext4_orphan_add(handle
, inode
);
1371 ret2
= ext4_journal_stop(handle
);
1375 if (pos
+ len
> inode
->i_size
) {
1376 ext4_truncate_failed_write(inode
);
1378 * If truncate failed early the inode might still be
1379 * on the orphan list; we need to make sure the inode
1380 * is removed from the orphan list in that case.
1383 ext4_orphan_del(NULL
, inode
);
1386 return ret
? ret
: copied
;
1390 * This is a private version of page_zero_new_buffers() which doesn't
1391 * set the buffer to be dirty, since in data=journalled mode we need
1392 * to call ext4_handle_dirty_metadata() instead.
1394 static void zero_new_buffers(struct page
*page
, unsigned from
, unsigned to
)
1396 unsigned int block_start
= 0, block_end
;
1397 struct buffer_head
*head
, *bh
;
1399 bh
= head
= page_buffers(page
);
1401 block_end
= block_start
+ bh
->b_size
;
1402 if (buffer_new(bh
)) {
1403 if (block_end
> from
&& block_start
< to
) {
1404 if (!PageUptodate(page
)) {
1405 unsigned start
, size
;
1407 start
= max(from
, block_start
);
1408 size
= min(to
, block_end
) - start
;
1410 zero_user(page
, start
, size
);
1411 set_buffer_uptodate(bh
);
1413 clear_buffer_new(bh
);
1416 block_start
= block_end
;
1417 bh
= bh
->b_this_page
;
1418 } while (bh
!= head
);
1421 static int ext4_journalled_write_end(struct file
*file
,
1422 struct address_space
*mapping
,
1423 loff_t pos
, unsigned len
, unsigned copied
,
1424 struct page
*page
, void *fsdata
)
1426 handle_t
*handle
= ext4_journal_current_handle();
1427 struct inode
*inode
= mapping
->host
;
1428 loff_t old_size
= inode
->i_size
;
1432 int size_changed
= 0;
1434 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1435 from
= pos
& (PAGE_SIZE
- 1);
1438 BUG_ON(!ext4_handle_valid(handle
));
1440 if (ext4_has_inline_data(inode
))
1441 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1445 if (!PageUptodate(page
))
1447 zero_new_buffers(page
, from
+copied
, to
);
1450 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1451 to
, &partial
, write_end_fn
);
1453 SetPageUptodate(page
);
1455 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1456 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1457 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1462 pagecache_isize_extended(inode
, old_size
, pos
);
1465 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1470 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1471 /* if we have allocated more blocks and copied
1472 * less. We will have blocks allocated outside
1473 * inode->i_size. So truncate them
1475 ext4_orphan_add(handle
, inode
);
1477 ret2
= ext4_journal_stop(handle
);
1480 if (pos
+ len
> inode
->i_size
) {
1481 ext4_truncate_failed_write(inode
);
1483 * If truncate failed early the inode might still be
1484 * on the orphan list; we need to make sure the inode
1485 * is removed from the orphan list in that case.
1488 ext4_orphan_del(NULL
, inode
);
1491 return ret
? ret
: copied
;
1495 * Reserve space for a single cluster
1497 static int ext4_da_reserve_space(struct inode
*inode
)
1499 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1500 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1504 * We will charge metadata quota at writeout time; this saves
1505 * us from metadata over-estimation, though we may go over by
1506 * a small amount in the end. Here we just reserve for data.
1508 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1512 spin_lock(&ei
->i_block_reservation_lock
);
1513 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1514 spin_unlock(&ei
->i_block_reservation_lock
);
1515 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1518 ei
->i_reserved_data_blocks
++;
1519 trace_ext4_da_reserve_space(inode
);
1520 spin_unlock(&ei
->i_block_reservation_lock
);
1522 return 0; /* success */
1525 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1527 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1528 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1531 return; /* Nothing to release, exit */
1533 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1535 trace_ext4_da_release_space(inode
, to_free
);
1536 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1538 * if there aren't enough reserved blocks, then the
1539 * counter is messed up somewhere. Since this
1540 * function is called from invalidate page, it's
1541 * harmless to return without any action.
1543 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1544 "ino %lu, to_free %d with only %d reserved "
1545 "data blocks", inode
->i_ino
, to_free
,
1546 ei
->i_reserved_data_blocks
);
1548 to_free
= ei
->i_reserved_data_blocks
;
1550 ei
->i_reserved_data_blocks
-= to_free
;
1552 /* update fs dirty data blocks counter */
1553 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1555 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1557 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1560 static void ext4_da_page_release_reservation(struct page
*page
,
1561 unsigned int offset
,
1562 unsigned int length
)
1564 int to_release
= 0, contiguous_blks
= 0;
1565 struct buffer_head
*head
, *bh
;
1566 unsigned int curr_off
= 0;
1567 struct inode
*inode
= page
->mapping
->host
;
1568 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1569 unsigned int stop
= offset
+ length
;
1573 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1575 head
= page_buffers(page
);
1578 unsigned int next_off
= curr_off
+ bh
->b_size
;
1580 if (next_off
> stop
)
1583 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1586 clear_buffer_delay(bh
);
1587 } else if (contiguous_blks
) {
1588 lblk
= page
->index
<<
1589 (PAGE_SHIFT
- inode
->i_blkbits
);
1590 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1592 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1593 contiguous_blks
= 0;
1595 curr_off
= next_off
;
1596 } while ((bh
= bh
->b_this_page
) != head
);
1598 if (contiguous_blks
) {
1599 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1600 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1601 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1604 /* If we have released all the blocks belonging to a cluster, then we
1605 * need to release the reserved space for that cluster. */
1606 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1607 while (num_clusters
> 0) {
1608 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1609 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1610 if (sbi
->s_cluster_ratio
== 1 ||
1611 !ext4_find_delalloc_cluster(inode
, lblk
))
1612 ext4_da_release_space(inode
, 1);
1619 * Delayed allocation stuff
1622 struct mpage_da_data
{
1623 struct inode
*inode
;
1624 struct writeback_control
*wbc
;
1626 pgoff_t first_page
; /* The first page to write */
1627 pgoff_t next_page
; /* Current page to examine */
1628 pgoff_t last_page
; /* Last page to examine */
1630 * Extent to map - this can be after first_page because that can be
1631 * fully mapped. We somewhat abuse m_flags to store whether the extent
1632 * is delalloc or unwritten.
1634 struct ext4_map_blocks map
;
1635 struct ext4_io_submit io_submit
; /* IO submission data */
1638 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1643 struct pagevec pvec
;
1644 struct inode
*inode
= mpd
->inode
;
1645 struct address_space
*mapping
= inode
->i_mapping
;
1647 /* This is necessary when next_page == 0. */
1648 if (mpd
->first_page
>= mpd
->next_page
)
1651 index
= mpd
->first_page
;
1652 end
= mpd
->next_page
- 1;
1654 ext4_lblk_t start
, last
;
1655 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1656 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1657 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1660 pagevec_init(&pvec
, 0);
1661 while (index
<= end
) {
1662 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1665 for (i
= 0; i
< nr_pages
; i
++) {
1666 struct page
*page
= pvec
.pages
[i
];
1667 if (page
->index
> end
)
1669 BUG_ON(!PageLocked(page
));
1670 BUG_ON(PageWriteback(page
));
1672 if (page_mapped(page
))
1673 clear_page_dirty_for_io(page
);
1674 block_invalidatepage(page
, 0, PAGE_SIZE
);
1675 ClearPageUptodate(page
);
1679 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1680 pagevec_release(&pvec
);
1684 static void ext4_print_free_blocks(struct inode
*inode
)
1686 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1687 struct super_block
*sb
= inode
->i_sb
;
1688 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1690 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1691 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1692 ext4_count_free_clusters(sb
)));
1693 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1694 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1695 (long long) EXT4_C2B(EXT4_SB(sb
),
1696 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1697 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1698 (long long) EXT4_C2B(EXT4_SB(sb
),
1699 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1700 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1701 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1702 ei
->i_reserved_data_blocks
);
1706 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1708 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1712 * This function is grabs code from the very beginning of
1713 * ext4_map_blocks, but assumes that the caller is from delayed write
1714 * time. This function looks up the requested blocks and sets the
1715 * buffer delay bit under the protection of i_data_sem.
1717 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1718 struct ext4_map_blocks
*map
,
1719 struct buffer_head
*bh
)
1721 struct extent_status es
;
1723 sector_t invalid_block
= ~((sector_t
) 0xffff);
1724 #ifdef ES_AGGRESSIVE_TEST
1725 struct ext4_map_blocks orig_map
;
1727 memcpy(&orig_map
, map
, sizeof(*map
));
1730 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1734 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1735 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1736 (unsigned long) map
->m_lblk
);
1738 /* Lookup extent status tree firstly */
1739 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1740 if (ext4_es_is_hole(&es
)) {
1742 down_read(&EXT4_I(inode
)->i_data_sem
);
1747 * Delayed extent could be allocated by fallocate.
1748 * So we need to check it.
1750 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1751 map_bh(bh
, inode
->i_sb
, invalid_block
);
1753 set_buffer_delay(bh
);
1757 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1758 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1759 if (retval
> map
->m_len
)
1760 retval
= map
->m_len
;
1761 map
->m_len
= retval
;
1762 if (ext4_es_is_written(&es
))
1763 map
->m_flags
|= EXT4_MAP_MAPPED
;
1764 else if (ext4_es_is_unwritten(&es
))
1765 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1769 #ifdef ES_AGGRESSIVE_TEST
1770 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1776 * Try to see if we can get the block without requesting a new
1777 * file system block.
1779 down_read(&EXT4_I(inode
)->i_data_sem
);
1780 if (ext4_has_inline_data(inode
))
1782 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1783 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1785 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1791 * XXX: __block_prepare_write() unmaps passed block,
1795 * If the block was allocated from previously allocated cluster,
1796 * then we don't need to reserve it again. However we still need
1797 * to reserve metadata for every block we're going to write.
1799 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1800 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1801 ret
= ext4_da_reserve_space(inode
);
1803 /* not enough space to reserve */
1809 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1810 ~0, EXTENT_STATUS_DELAYED
);
1816 map_bh(bh
, inode
->i_sb
, invalid_block
);
1818 set_buffer_delay(bh
);
1819 } else if (retval
> 0) {
1821 unsigned int status
;
1823 if (unlikely(retval
!= map
->m_len
)) {
1824 ext4_warning(inode
->i_sb
,
1825 "ES len assertion failed for inode "
1826 "%lu: retval %d != map->m_len %d",
1827 inode
->i_ino
, retval
, map
->m_len
);
1831 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1832 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1833 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1834 map
->m_pblk
, status
);
1840 up_read((&EXT4_I(inode
)->i_data_sem
));
1846 * This is a special get_block_t callback which is used by
1847 * ext4_da_write_begin(). It will either return mapped block or
1848 * reserve space for a single block.
1850 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1851 * We also have b_blocknr = -1 and b_bdev initialized properly
1853 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1854 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1855 * initialized properly.
1857 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1858 struct buffer_head
*bh
, int create
)
1860 struct ext4_map_blocks map
;
1863 BUG_ON(create
== 0);
1864 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1866 map
.m_lblk
= iblock
;
1870 * first, we need to know whether the block is allocated already
1871 * preallocated blocks are unmapped but should treated
1872 * the same as allocated blocks.
1874 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1878 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1879 ext4_update_bh_state(bh
, map
.m_flags
);
1881 if (buffer_unwritten(bh
)) {
1882 /* A delayed write to unwritten bh should be marked
1883 * new and mapped. Mapped ensures that we don't do
1884 * get_block multiple times when we write to the same
1885 * offset and new ensures that we do proper zero out
1886 * for partial write.
1889 set_buffer_mapped(bh
);
1894 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1900 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1906 static int __ext4_journalled_writepage(struct page
*page
,
1909 struct address_space
*mapping
= page
->mapping
;
1910 struct inode
*inode
= mapping
->host
;
1911 struct buffer_head
*page_bufs
= NULL
;
1912 handle_t
*handle
= NULL
;
1913 int ret
= 0, err
= 0;
1914 int inline_data
= ext4_has_inline_data(inode
);
1915 struct buffer_head
*inode_bh
= NULL
;
1917 ClearPageChecked(page
);
1920 BUG_ON(page
->index
!= 0);
1921 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1922 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1923 if (inode_bh
== NULL
)
1926 page_bufs
= page_buffers(page
);
1931 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1935 * We need to release the page lock before we start the
1936 * journal, so grab a reference so the page won't disappear
1937 * out from under us.
1942 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1943 ext4_writepage_trans_blocks(inode
));
1944 if (IS_ERR(handle
)) {
1945 ret
= PTR_ERR(handle
);
1947 goto out_no_pagelock
;
1949 BUG_ON(!ext4_handle_valid(handle
));
1953 if (page
->mapping
!= mapping
) {
1954 /* The page got truncated from under us */
1955 ext4_journal_stop(handle
);
1961 BUFFER_TRACE(inode_bh
, "get write access");
1962 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1964 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1967 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1968 do_journal_get_write_access
);
1970 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1975 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1976 err
= ext4_journal_stop(handle
);
1980 if (!ext4_has_inline_data(inode
))
1981 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1983 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1992 * Note that we don't need to start a transaction unless we're journaling data
1993 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1994 * need to file the inode to the transaction's list in ordered mode because if
1995 * we are writing back data added by write(), the inode is already there and if
1996 * we are writing back data modified via mmap(), no one guarantees in which
1997 * transaction the data will hit the disk. In case we are journaling data, we
1998 * cannot start transaction directly because transaction start ranks above page
1999 * lock so we have to do some magic.
2001 * This function can get called via...
2002 * - ext4_writepages after taking page lock (have journal handle)
2003 * - journal_submit_inode_data_buffers (no journal handle)
2004 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2005 * - grab_page_cache when doing write_begin (have journal handle)
2007 * We don't do any block allocation in this function. If we have page with
2008 * multiple blocks we need to write those buffer_heads that are mapped. This
2009 * is important for mmaped based write. So if we do with blocksize 1K
2010 * truncate(f, 1024);
2011 * a = mmap(f, 0, 4096);
2013 * truncate(f, 4096);
2014 * we have in the page first buffer_head mapped via page_mkwrite call back
2015 * but other buffer_heads would be unmapped but dirty (dirty done via the
2016 * do_wp_page). So writepage should write the first block. If we modify
2017 * the mmap area beyond 1024 we will again get a page_fault and the
2018 * page_mkwrite callback will do the block allocation and mark the
2019 * buffer_heads mapped.
2021 * We redirty the page if we have any buffer_heads that is either delay or
2022 * unwritten in the page.
2024 * We can get recursively called as show below.
2026 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2029 * But since we don't do any block allocation we should not deadlock.
2030 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2032 static int ext4_writepage(struct page
*page
,
2033 struct writeback_control
*wbc
)
2038 struct buffer_head
*page_bufs
= NULL
;
2039 struct inode
*inode
= page
->mapping
->host
;
2040 struct ext4_io_submit io_submit
;
2041 bool keep_towrite
= false;
2043 trace_ext4_writepage(page
);
2044 size
= i_size_read(inode
);
2045 if (page
->index
== size
>> PAGE_SHIFT
)
2046 len
= size
& ~PAGE_MASK
;
2050 page_bufs
= page_buffers(page
);
2052 * We cannot do block allocation or other extent handling in this
2053 * function. If there are buffers needing that, we have to redirty
2054 * the page. But we may reach here when we do a journal commit via
2055 * journal_submit_inode_data_buffers() and in that case we must write
2056 * allocated buffers to achieve data=ordered mode guarantees.
2058 * Also, if there is only one buffer per page (the fs block
2059 * size == the page size), if one buffer needs block
2060 * allocation or needs to modify the extent tree to clear the
2061 * unwritten flag, we know that the page can't be written at
2062 * all, so we might as well refuse the write immediately.
2063 * Unfortunately if the block size != page size, we can't as
2064 * easily detect this case using ext4_walk_page_buffers(), but
2065 * for the extremely common case, this is an optimization that
2066 * skips a useless round trip through ext4_bio_write_page().
2068 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2069 ext4_bh_delay_or_unwritten
)) {
2070 redirty_page_for_writepage(wbc
, page
);
2071 if ((current
->flags
& PF_MEMALLOC
) ||
2072 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2074 * For memory cleaning there's no point in writing only
2075 * some buffers. So just bail out. Warn if we came here
2076 * from direct reclaim.
2078 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2083 keep_towrite
= true;
2086 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2088 * It's mmapped pagecache. Add buffers and journal it. There
2089 * doesn't seem much point in redirtying the page here.
2091 return __ext4_journalled_writepage(page
, len
);
2093 ext4_io_submit_init(&io_submit
, wbc
);
2094 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2095 if (!io_submit
.io_end
) {
2096 redirty_page_for_writepage(wbc
, page
);
2100 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2101 ext4_io_submit(&io_submit
);
2102 /* Drop io_end reference we got from init */
2103 ext4_put_io_end_defer(io_submit
.io_end
);
2107 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2110 loff_t size
= i_size_read(mpd
->inode
);
2113 BUG_ON(page
->index
!= mpd
->first_page
);
2114 if (page
->index
== size
>> PAGE_SHIFT
)
2115 len
= size
& ~PAGE_MASK
;
2118 clear_page_dirty_for_io(page
);
2119 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2121 mpd
->wbc
->nr_to_write
--;
2127 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2130 * mballoc gives us at most this number of blocks...
2131 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2132 * The rest of mballoc seems to handle chunks up to full group size.
2134 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2137 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2139 * @mpd - extent of blocks
2140 * @lblk - logical number of the block in the file
2141 * @bh - buffer head we want to add to the extent
2143 * The function is used to collect contig. blocks in the same state. If the
2144 * buffer doesn't require mapping for writeback and we haven't started the
2145 * extent of buffers to map yet, the function returns 'true' immediately - the
2146 * caller can write the buffer right away. Otherwise the function returns true
2147 * if the block has been added to the extent, false if the block couldn't be
2150 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2151 struct buffer_head
*bh
)
2153 struct ext4_map_blocks
*map
= &mpd
->map
;
2155 /* Buffer that doesn't need mapping for writeback? */
2156 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2157 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2158 /* So far no extent to map => we write the buffer right away */
2159 if (map
->m_len
== 0)
2164 /* First block in the extent? */
2165 if (map
->m_len
== 0) {
2168 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2172 /* Don't go larger than mballoc is willing to allocate */
2173 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2176 /* Can we merge the block to our big extent? */
2177 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2178 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2186 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2188 * @mpd - extent of blocks for mapping
2189 * @head - the first buffer in the page
2190 * @bh - buffer we should start processing from
2191 * @lblk - logical number of the block in the file corresponding to @bh
2193 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2194 * the page for IO if all buffers in this page were mapped and there's no
2195 * accumulated extent of buffers to map or add buffers in the page to the
2196 * extent of buffers to map. The function returns 1 if the caller can continue
2197 * by processing the next page, 0 if it should stop adding buffers to the
2198 * extent to map because we cannot extend it anymore. It can also return value
2199 * < 0 in case of error during IO submission.
2201 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2202 struct buffer_head
*head
,
2203 struct buffer_head
*bh
,
2206 struct inode
*inode
= mpd
->inode
;
2208 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2209 >> inode
->i_blkbits
;
2212 BUG_ON(buffer_locked(bh
));
2214 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2215 /* Found extent to map? */
2218 /* Everything mapped so far and we hit EOF */
2221 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2222 /* So far everything mapped? Submit the page for IO. */
2223 if (mpd
->map
.m_len
== 0) {
2224 err
= mpage_submit_page(mpd
, head
->b_page
);
2228 return lblk
< blocks
;
2232 * mpage_map_buffers - update buffers corresponding to changed extent and
2233 * submit fully mapped pages for IO
2235 * @mpd - description of extent to map, on return next extent to map
2237 * Scan buffers corresponding to changed extent (we expect corresponding pages
2238 * to be already locked) and update buffer state according to new extent state.
2239 * We map delalloc buffers to their physical location, clear unwritten bits,
2240 * and mark buffers as uninit when we perform writes to unwritten extents
2241 * and do extent conversion after IO is finished. If the last page is not fully
2242 * mapped, we update @map to the next extent in the last page that needs
2243 * mapping. Otherwise we submit the page for IO.
2245 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2247 struct pagevec pvec
;
2249 struct inode
*inode
= mpd
->inode
;
2250 struct buffer_head
*head
, *bh
;
2251 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2257 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2258 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2259 lblk
= start
<< bpp_bits
;
2260 pblock
= mpd
->map
.m_pblk
;
2262 pagevec_init(&pvec
, 0);
2263 while (start
<= end
) {
2264 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2268 for (i
= 0; i
< nr_pages
; i
++) {
2269 struct page
*page
= pvec
.pages
[i
];
2271 if (page
->index
> end
)
2273 /* Up to 'end' pages must be contiguous */
2274 BUG_ON(page
->index
!= start
);
2275 bh
= head
= page_buffers(page
);
2277 if (lblk
< mpd
->map
.m_lblk
)
2279 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2281 * Buffer after end of mapped extent.
2282 * Find next buffer in the page to map.
2285 mpd
->map
.m_flags
= 0;
2287 * FIXME: If dioread_nolock supports
2288 * blocksize < pagesize, we need to make
2289 * sure we add size mapped so far to
2290 * io_end->size as the following call
2291 * can submit the page for IO.
2293 err
= mpage_process_page_bufs(mpd
, head
,
2295 pagevec_release(&pvec
);
2300 if (buffer_delay(bh
)) {
2301 clear_buffer_delay(bh
);
2302 bh
->b_blocknr
= pblock
++;
2304 clear_buffer_unwritten(bh
);
2305 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2308 * FIXME: This is going to break if dioread_nolock
2309 * supports blocksize < pagesize as we will try to
2310 * convert potentially unmapped parts of inode.
2312 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2313 /* Page fully mapped - let IO run! */
2314 err
= mpage_submit_page(mpd
, page
);
2316 pagevec_release(&pvec
);
2321 pagevec_release(&pvec
);
2323 /* Extent fully mapped and matches with page boundary. We are done. */
2325 mpd
->map
.m_flags
= 0;
2329 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2331 struct inode
*inode
= mpd
->inode
;
2332 struct ext4_map_blocks
*map
= &mpd
->map
;
2333 int get_blocks_flags
;
2334 int err
, dioread_nolock
;
2336 trace_ext4_da_write_pages_extent(inode
, map
);
2338 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2339 * to convert an unwritten extent to be initialized (in the case
2340 * where we have written into one or more preallocated blocks). It is
2341 * possible that we're going to need more metadata blocks than
2342 * previously reserved. However we must not fail because we're in
2343 * writeback and there is nothing we can do about it so it might result
2344 * in data loss. So use reserved blocks to allocate metadata if
2347 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2348 * the blocks in question are delalloc blocks. This indicates
2349 * that the blocks and quotas has already been checked when
2350 * the data was copied into the page cache.
2352 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2353 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2354 EXT4_GET_BLOCKS_IO_SUBMIT
;
2355 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2357 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2358 if (map
->m_flags
& (1 << BH_Delay
))
2359 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2361 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2364 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2365 if (!mpd
->io_submit
.io_end
->handle
&&
2366 ext4_handle_valid(handle
)) {
2367 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2368 handle
->h_rsv_handle
= NULL
;
2370 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2373 BUG_ON(map
->m_len
== 0);
2374 if (map
->m_flags
& EXT4_MAP_NEW
) {
2375 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2378 for (i
= 0; i
< map
->m_len
; i
++)
2379 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2385 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2386 * mpd->len and submit pages underlying it for IO
2388 * @handle - handle for journal operations
2389 * @mpd - extent to map
2390 * @give_up_on_write - we set this to true iff there is a fatal error and there
2391 * is no hope of writing the data. The caller should discard
2392 * dirty pages to avoid infinite loops.
2394 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2395 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2396 * them to initialized or split the described range from larger unwritten
2397 * extent. Note that we need not map all the described range since allocation
2398 * can return less blocks or the range is covered by more unwritten extents. We
2399 * cannot map more because we are limited by reserved transaction credits. On
2400 * the other hand we always make sure that the last touched page is fully
2401 * mapped so that it can be written out (and thus forward progress is
2402 * guaranteed). After mapping we submit all mapped pages for IO.
2404 static int mpage_map_and_submit_extent(handle_t
*handle
,
2405 struct mpage_da_data
*mpd
,
2406 bool *give_up_on_write
)
2408 struct inode
*inode
= mpd
->inode
;
2409 struct ext4_map_blocks
*map
= &mpd
->map
;
2414 mpd
->io_submit
.io_end
->offset
=
2415 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2417 err
= mpage_map_one_extent(handle
, mpd
);
2419 struct super_block
*sb
= inode
->i_sb
;
2421 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2422 goto invalidate_dirty_pages
;
2424 * Let the uper layers retry transient errors.
2425 * In the case of ENOSPC, if ext4_count_free_blocks()
2426 * is non-zero, a commit should free up blocks.
2428 if ((err
== -ENOMEM
) ||
2429 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2431 goto update_disksize
;
2434 ext4_msg(sb
, KERN_CRIT
,
2435 "Delayed block allocation failed for "
2436 "inode %lu at logical offset %llu with"
2437 " max blocks %u with error %d",
2439 (unsigned long long)map
->m_lblk
,
2440 (unsigned)map
->m_len
, -err
);
2441 ext4_msg(sb
, KERN_CRIT
,
2442 "This should not happen!! Data will "
2445 ext4_print_free_blocks(inode
);
2446 invalidate_dirty_pages
:
2447 *give_up_on_write
= true;
2452 * Update buffer state, submit mapped pages, and get us new
2455 err
= mpage_map_and_submit_buffers(mpd
);
2457 goto update_disksize
;
2458 } while (map
->m_len
);
2462 * Update on-disk size after IO is submitted. Races with
2463 * truncate are avoided by checking i_size under i_data_sem.
2465 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2466 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2470 down_write(&EXT4_I(inode
)->i_data_sem
);
2471 i_size
= i_size_read(inode
);
2472 if (disksize
> i_size
)
2474 if (disksize
> EXT4_I(inode
)->i_disksize
)
2475 EXT4_I(inode
)->i_disksize
= disksize
;
2476 err2
= ext4_mark_inode_dirty(handle
, inode
);
2477 up_write(&EXT4_I(inode
)->i_data_sem
);
2479 ext4_error(inode
->i_sb
,
2480 "Failed to mark inode %lu dirty",
2489 * Calculate the total number of credits to reserve for one writepages
2490 * iteration. This is called from ext4_writepages(). We map an extent of
2491 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2492 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2493 * bpp - 1 blocks in bpp different extents.
2495 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2497 int bpp
= ext4_journal_blocks_per_page(inode
);
2499 return ext4_meta_trans_blocks(inode
,
2500 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2504 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2505 * and underlying extent to map
2507 * @mpd - where to look for pages
2509 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2510 * IO immediately. When we find a page which isn't mapped we start accumulating
2511 * extent of buffers underlying these pages that needs mapping (formed by
2512 * either delayed or unwritten buffers). We also lock the pages containing
2513 * these buffers. The extent found is returned in @mpd structure (starting at
2514 * mpd->lblk with length mpd->len blocks).
2516 * Note that this function can attach bios to one io_end structure which are
2517 * neither logically nor physically contiguous. Although it may seem as an
2518 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2519 * case as we need to track IO to all buffers underlying a page in one io_end.
2521 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2523 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2524 struct pagevec pvec
;
2525 unsigned int nr_pages
;
2526 long left
= mpd
->wbc
->nr_to_write
;
2527 pgoff_t index
= mpd
->first_page
;
2528 pgoff_t end
= mpd
->last_page
;
2531 int blkbits
= mpd
->inode
->i_blkbits
;
2533 struct buffer_head
*head
;
2535 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2536 tag
= PAGECACHE_TAG_TOWRITE
;
2538 tag
= PAGECACHE_TAG_DIRTY
;
2540 pagevec_init(&pvec
, 0);
2542 mpd
->next_page
= index
;
2543 while (index
<= end
) {
2544 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2545 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2549 for (i
= 0; i
< nr_pages
; i
++) {
2550 struct page
*page
= pvec
.pages
[i
];
2553 * At this point, the page may be truncated or
2554 * invalidated (changing page->mapping to NULL), or
2555 * even swizzled back from swapper_space to tmpfs file
2556 * mapping. However, page->index will not change
2557 * because we have a reference on the page.
2559 if (page
->index
> end
)
2563 * Accumulated enough dirty pages? This doesn't apply
2564 * to WB_SYNC_ALL mode. For integrity sync we have to
2565 * keep going because someone may be concurrently
2566 * dirtying pages, and we might have synced a lot of
2567 * newly appeared dirty pages, but have not synced all
2568 * of the old dirty pages.
2570 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2573 /* If we can't merge this page, we are done. */
2574 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2579 * If the page is no longer dirty, or its mapping no
2580 * longer corresponds to inode we are writing (which
2581 * means it has been truncated or invalidated), or the
2582 * page is already under writeback and we are not doing
2583 * a data integrity writeback, skip the page
2585 if (!PageDirty(page
) ||
2586 (PageWriteback(page
) &&
2587 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2588 unlikely(page
->mapping
!= mapping
)) {
2593 wait_on_page_writeback(page
);
2594 BUG_ON(PageWriteback(page
));
2596 if (mpd
->map
.m_len
== 0)
2597 mpd
->first_page
= page
->index
;
2598 mpd
->next_page
= page
->index
+ 1;
2599 /* Add all dirty buffers to mpd */
2600 lblk
= ((ext4_lblk_t
)page
->index
) <<
2601 (PAGE_SHIFT
- blkbits
);
2602 head
= page_buffers(page
);
2603 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2609 pagevec_release(&pvec
);
2614 pagevec_release(&pvec
);
2618 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2621 struct address_space
*mapping
= data
;
2622 int ret
= ext4_writepage(page
, wbc
);
2623 mapping_set_error(mapping
, ret
);
2627 static int ext4_writepages(struct address_space
*mapping
,
2628 struct writeback_control
*wbc
)
2630 pgoff_t writeback_index
= 0;
2631 long nr_to_write
= wbc
->nr_to_write
;
2632 int range_whole
= 0;
2634 handle_t
*handle
= NULL
;
2635 struct mpage_da_data mpd
;
2636 struct inode
*inode
= mapping
->host
;
2637 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2638 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2640 struct blk_plug plug
;
2641 bool give_up_on_write
= false;
2643 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2644 trace_ext4_writepages(inode
, wbc
);
2646 if (dax_mapping(mapping
)) {
2647 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2649 goto out_writepages
;
2653 * No pages to write? This is mainly a kludge to avoid starting
2654 * a transaction for special inodes like journal inode on last iput()
2655 * because that could violate lock ordering on umount
2657 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2658 goto out_writepages
;
2660 if (ext4_should_journal_data(inode
)) {
2661 struct blk_plug plug
;
2663 blk_start_plug(&plug
);
2664 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2665 blk_finish_plug(&plug
);
2666 goto out_writepages
;
2670 * If the filesystem has aborted, it is read-only, so return
2671 * right away instead of dumping stack traces later on that
2672 * will obscure the real source of the problem. We test
2673 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2674 * the latter could be true if the filesystem is mounted
2675 * read-only, and in that case, ext4_writepages should
2676 * *never* be called, so if that ever happens, we would want
2679 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2681 goto out_writepages
;
2684 if (ext4_should_dioread_nolock(inode
)) {
2686 * We may need to convert up to one extent per block in
2687 * the page and we may dirty the inode.
2689 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2693 * If we have inline data and arrive here, it means that
2694 * we will soon create the block for the 1st page, so
2695 * we'd better clear the inline data here.
2697 if (ext4_has_inline_data(inode
)) {
2698 /* Just inode will be modified... */
2699 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2700 if (IS_ERR(handle
)) {
2701 ret
= PTR_ERR(handle
);
2702 goto out_writepages
;
2704 BUG_ON(ext4_test_inode_state(inode
,
2705 EXT4_STATE_MAY_INLINE_DATA
));
2706 ext4_destroy_inline_data(handle
, inode
);
2707 ext4_journal_stop(handle
);
2710 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2713 if (wbc
->range_cyclic
) {
2714 writeback_index
= mapping
->writeback_index
;
2715 if (writeback_index
)
2717 mpd
.first_page
= writeback_index
;
2720 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2721 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2726 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2728 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2729 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2731 blk_start_plug(&plug
);
2732 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2733 /* For each extent of pages we use new io_end */
2734 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2735 if (!mpd
.io_submit
.io_end
) {
2741 * We have two constraints: We find one extent to map and we
2742 * must always write out whole page (makes a difference when
2743 * blocksize < pagesize) so that we don't block on IO when we
2744 * try to write out the rest of the page. Journalled mode is
2745 * not supported by delalloc.
2747 BUG_ON(ext4_should_journal_data(inode
));
2748 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2750 /* start a new transaction */
2751 handle
= ext4_journal_start_with_reserve(inode
,
2752 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2753 if (IS_ERR(handle
)) {
2754 ret
= PTR_ERR(handle
);
2755 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2756 "%ld pages, ino %lu; err %d", __func__
,
2757 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2758 /* Release allocated io_end */
2759 ext4_put_io_end(mpd
.io_submit
.io_end
);
2763 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2764 ret
= mpage_prepare_extent_to_map(&mpd
);
2767 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2771 * We scanned the whole range (or exhausted
2772 * nr_to_write), submitted what was mapped and
2773 * didn't find anything needing mapping. We are
2780 * Caution: If the handle is synchronous,
2781 * ext4_journal_stop() can wait for transaction commit
2782 * to finish which may depend on writeback of pages to
2783 * complete or on page lock to be released. In that
2784 * case, we have to wait until after after we have
2785 * submitted all the IO, released page locks we hold,
2786 * and dropped io_end reference (for extent conversion
2787 * to be able to complete) before stopping the handle.
2789 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2790 ext4_journal_stop(handle
);
2793 /* Submit prepared bio */
2794 ext4_io_submit(&mpd
.io_submit
);
2795 /* Unlock pages we didn't use */
2796 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2798 * Drop our io_end reference we got from init. We have
2799 * to be careful and use deferred io_end finishing if
2800 * we are still holding the transaction as we can
2801 * release the last reference to io_end which may end
2802 * up doing unwritten extent conversion.
2805 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2806 ext4_journal_stop(handle
);
2808 ext4_put_io_end(mpd
.io_submit
.io_end
);
2810 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2812 * Commit the transaction which would
2813 * free blocks released in the transaction
2816 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2820 /* Fatal error - ENOMEM, EIO... */
2824 blk_finish_plug(&plug
);
2825 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2827 mpd
.last_page
= writeback_index
- 1;
2833 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2835 * Set the writeback_index so that range_cyclic
2836 * mode will write it back later
2838 mapping
->writeback_index
= mpd
.first_page
;
2841 trace_ext4_writepages_result(inode
, wbc
, ret
,
2842 nr_to_write
- wbc
->nr_to_write
);
2843 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2847 static int ext4_nonda_switch(struct super_block
*sb
)
2849 s64 free_clusters
, dirty_clusters
;
2850 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2853 * switch to non delalloc mode if we are running low
2854 * on free block. The free block accounting via percpu
2855 * counters can get slightly wrong with percpu_counter_batch getting
2856 * accumulated on each CPU without updating global counters
2857 * Delalloc need an accurate free block accounting. So switch
2858 * to non delalloc when we are near to error range.
2861 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2863 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2865 * Start pushing delalloc when 1/2 of free blocks are dirty.
2867 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2868 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2870 if (2 * free_clusters
< 3 * dirty_clusters
||
2871 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2873 * free block count is less than 150% of dirty blocks
2874 * or free blocks is less than watermark
2881 /* We always reserve for an inode update; the superblock could be there too */
2882 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2884 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2887 if (pos
+ len
<= 0x7fffffffULL
)
2890 /* We might need to update the superblock to set LARGE_FILE */
2894 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2895 loff_t pos
, unsigned len
, unsigned flags
,
2896 struct page
**pagep
, void **fsdata
)
2898 int ret
, retries
= 0;
2901 struct inode
*inode
= mapping
->host
;
2904 index
= pos
>> PAGE_SHIFT
;
2906 if (ext4_nonda_switch(inode
->i_sb
)) {
2907 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2908 return ext4_write_begin(file
, mapping
, pos
,
2909 len
, flags
, pagep
, fsdata
);
2911 *fsdata
= (void *)0;
2912 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2914 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2915 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2925 * grab_cache_page_write_begin() can take a long time if the
2926 * system is thrashing due to memory pressure, or if the page
2927 * is being written back. So grab it first before we start
2928 * the transaction handle. This also allows us to allocate
2929 * the page (if needed) without using GFP_NOFS.
2932 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2938 * With delayed allocation, we don't log the i_disksize update
2939 * if there is delayed block allocation. But we still need
2940 * to journalling the i_disksize update if writes to the end
2941 * of file which has an already mapped buffer.
2944 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2945 ext4_da_write_credits(inode
, pos
, len
));
2946 if (IS_ERR(handle
)) {
2948 return PTR_ERR(handle
);
2952 if (page
->mapping
!= mapping
) {
2953 /* The page got truncated from under us */
2956 ext4_journal_stop(handle
);
2959 /* In case writeback began while the page was unlocked */
2960 wait_for_stable_page(page
);
2962 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2963 ret
= ext4_block_write_begin(page
, pos
, len
,
2964 ext4_da_get_block_prep
);
2966 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2970 ext4_journal_stop(handle
);
2972 * block_write_begin may have instantiated a few blocks
2973 * outside i_size. Trim these off again. Don't need
2974 * i_size_read because we hold i_mutex.
2976 if (pos
+ len
> inode
->i_size
)
2977 ext4_truncate_failed_write(inode
);
2979 if (ret
== -ENOSPC
&&
2980 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2992 * Check if we should update i_disksize
2993 * when write to the end of file but not require block allocation
2995 static int ext4_da_should_update_i_disksize(struct page
*page
,
2996 unsigned long offset
)
2998 struct buffer_head
*bh
;
2999 struct inode
*inode
= page
->mapping
->host
;
3003 bh
= page_buffers(page
);
3004 idx
= offset
>> inode
->i_blkbits
;
3006 for (i
= 0; i
< idx
; i
++)
3007 bh
= bh
->b_this_page
;
3009 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3014 static int ext4_da_write_end(struct file
*file
,
3015 struct address_space
*mapping
,
3016 loff_t pos
, unsigned len
, unsigned copied
,
3017 struct page
*page
, void *fsdata
)
3019 struct inode
*inode
= mapping
->host
;
3021 handle_t
*handle
= ext4_journal_current_handle();
3023 unsigned long start
, end
;
3024 int write_mode
= (int)(unsigned long)fsdata
;
3026 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3027 return ext4_write_end(file
, mapping
, pos
,
3028 len
, copied
, page
, fsdata
);
3030 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3031 start
= pos
& (PAGE_SIZE
- 1);
3032 end
= start
+ copied
- 1;
3035 * generic_write_end() will run mark_inode_dirty() if i_size
3036 * changes. So let's piggyback the i_disksize mark_inode_dirty
3039 new_i_size
= pos
+ copied
;
3040 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3041 if (ext4_has_inline_data(inode
) ||
3042 ext4_da_should_update_i_disksize(page
, end
)) {
3043 ext4_update_i_disksize(inode
, new_i_size
);
3044 /* We need to mark inode dirty even if
3045 * new_i_size is less that inode->i_size
3046 * bu greater than i_disksize.(hint delalloc)
3048 ext4_mark_inode_dirty(handle
, inode
);
3052 if (write_mode
!= CONVERT_INLINE_DATA
&&
3053 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3054 ext4_has_inline_data(inode
))
3055 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3058 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3064 ret2
= ext4_journal_stop(handle
);
3068 return ret
? ret
: copied
;
3071 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3072 unsigned int length
)
3075 * Drop reserved blocks
3077 BUG_ON(!PageLocked(page
));
3078 if (!page_has_buffers(page
))
3081 ext4_da_page_release_reservation(page
, offset
, length
);
3084 ext4_invalidatepage(page
, offset
, length
);
3090 * Force all delayed allocation blocks to be allocated for a given inode.
3092 int ext4_alloc_da_blocks(struct inode
*inode
)
3094 trace_ext4_alloc_da_blocks(inode
);
3096 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3100 * We do something simple for now. The filemap_flush() will
3101 * also start triggering a write of the data blocks, which is
3102 * not strictly speaking necessary (and for users of
3103 * laptop_mode, not even desirable). However, to do otherwise
3104 * would require replicating code paths in:
3106 * ext4_writepages() ->
3107 * write_cache_pages() ---> (via passed in callback function)
3108 * __mpage_da_writepage() -->
3109 * mpage_add_bh_to_extent()
3110 * mpage_da_map_blocks()
3112 * The problem is that write_cache_pages(), located in
3113 * mm/page-writeback.c, marks pages clean in preparation for
3114 * doing I/O, which is not desirable if we're not planning on
3117 * We could call write_cache_pages(), and then redirty all of
3118 * the pages by calling redirty_page_for_writepage() but that
3119 * would be ugly in the extreme. So instead we would need to
3120 * replicate parts of the code in the above functions,
3121 * simplifying them because we wouldn't actually intend to
3122 * write out the pages, but rather only collect contiguous
3123 * logical block extents, call the multi-block allocator, and
3124 * then update the buffer heads with the block allocations.
3126 * For now, though, we'll cheat by calling filemap_flush(),
3127 * which will map the blocks, and start the I/O, but not
3128 * actually wait for the I/O to complete.
3130 return filemap_flush(inode
->i_mapping
);
3134 * bmap() is special. It gets used by applications such as lilo and by
3135 * the swapper to find the on-disk block of a specific piece of data.
3137 * Naturally, this is dangerous if the block concerned is still in the
3138 * journal. If somebody makes a swapfile on an ext4 data-journaling
3139 * filesystem and enables swap, then they may get a nasty shock when the
3140 * data getting swapped to that swapfile suddenly gets overwritten by
3141 * the original zero's written out previously to the journal and
3142 * awaiting writeback in the kernel's buffer cache.
3144 * So, if we see any bmap calls here on a modified, data-journaled file,
3145 * take extra steps to flush any blocks which might be in the cache.
3147 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3149 struct inode
*inode
= mapping
->host
;
3154 * We can get here for an inline file via the FIBMAP ioctl
3156 if (ext4_has_inline_data(inode
))
3159 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3160 test_opt(inode
->i_sb
, DELALLOC
)) {
3162 * With delalloc we want to sync the file
3163 * so that we can make sure we allocate
3166 filemap_write_and_wait(mapping
);
3169 if (EXT4_JOURNAL(inode
) &&
3170 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3172 * This is a REALLY heavyweight approach, but the use of
3173 * bmap on dirty files is expected to be extremely rare:
3174 * only if we run lilo or swapon on a freshly made file
3175 * do we expect this to happen.
3177 * (bmap requires CAP_SYS_RAWIO so this does not
3178 * represent an unprivileged user DOS attack --- we'd be
3179 * in trouble if mortal users could trigger this path at
3182 * NB. EXT4_STATE_JDATA is not set on files other than
3183 * regular files. If somebody wants to bmap a directory
3184 * or symlink and gets confused because the buffer
3185 * hasn't yet been flushed to disk, they deserve
3186 * everything they get.
3189 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3190 journal
= EXT4_JOURNAL(inode
);
3191 jbd2_journal_lock_updates(journal
);
3192 err
= jbd2_journal_flush(journal
);
3193 jbd2_journal_unlock_updates(journal
);
3199 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3202 static int ext4_readpage(struct file
*file
, struct page
*page
)
3205 struct inode
*inode
= page
->mapping
->host
;
3207 trace_ext4_readpage(page
);
3209 if (ext4_has_inline_data(inode
))
3210 ret
= ext4_readpage_inline(inode
, page
);
3213 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3219 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3220 struct list_head
*pages
, unsigned nr_pages
)
3222 struct inode
*inode
= mapping
->host
;
3224 /* If the file has inline data, no need to do readpages. */
3225 if (ext4_has_inline_data(inode
))
3228 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3231 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3232 unsigned int length
)
3234 trace_ext4_invalidatepage(page
, offset
, length
);
3236 /* No journalling happens on data buffers when this function is used */
3237 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3239 block_invalidatepage(page
, offset
, length
);
3242 static int __ext4_journalled_invalidatepage(struct page
*page
,
3243 unsigned int offset
,
3244 unsigned int length
)
3246 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3248 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3251 * If it's a full truncate we just forget about the pending dirtying
3253 if (offset
== 0 && length
== PAGE_SIZE
)
3254 ClearPageChecked(page
);
3256 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3259 /* Wrapper for aops... */
3260 static void ext4_journalled_invalidatepage(struct page
*page
,
3261 unsigned int offset
,
3262 unsigned int length
)
3264 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3267 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3269 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3271 trace_ext4_releasepage(page
);
3273 /* Page has dirty journalled data -> cannot release */
3274 if (PageChecked(page
))
3277 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3279 return try_to_free_buffers(page
);
3282 #ifdef CONFIG_FS_DAX
3284 * Get block function for DAX IO and mmap faults. It takes care of converting
3285 * unwritten extents to written ones and initializes new / converted blocks
3288 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3289 struct buffer_head
*bh_result
, int create
)
3293 ext4_debug("inode %lu, create flag %d\n", inode
->i_ino
, create
);
3295 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
3297 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3298 EXT4_GET_BLOCKS_PRE_IO
|
3299 EXT4_GET_BLOCKS_CREATE_ZERO
);
3303 if (buffer_unwritten(bh_result
)) {
3305 * We are protected by i_mmap_sem or i_mutex so we know block
3306 * cannot go away from under us even though we dropped
3307 * i_data_sem. Convert extent to written and write zeros there.
3309 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
3310 EXT4_GET_BLOCKS_CONVERT
|
3311 EXT4_GET_BLOCKS_CREATE_ZERO
);
3316 * At least for now we have to clear BH_New so that DAX code
3317 * doesn't attempt to zero blocks again in a racy way.
3319 clear_buffer_new(bh_result
);
3323 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3324 unsigned flags
, struct iomap
*iomap
)
3326 unsigned int blkbits
= inode
->i_blkbits
;
3327 unsigned long first_block
= offset
>> blkbits
;
3328 unsigned long last_block
= (offset
+ length
- 1) >> blkbits
;
3329 struct ext4_map_blocks map
;
3332 if (flags
& IOMAP_WRITE
)
3335 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3338 map
.m_lblk
= first_block
;
3339 map
.m_len
= last_block
- first_block
+ 1;
3341 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3346 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3347 iomap
->offset
= first_block
<< blkbits
;
3350 iomap
->type
= IOMAP_HOLE
;
3351 iomap
->blkno
= IOMAP_NULL_BLOCK
;
3352 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3354 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3355 iomap
->type
= IOMAP_MAPPED
;
3356 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3357 iomap
->type
= IOMAP_UNWRITTEN
;
3362 iomap
->blkno
= (sector_t
)map
.m_pblk
<< (blkbits
- 9);
3363 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3366 if (map
.m_flags
& EXT4_MAP_NEW
)
3367 iomap
->flags
|= IOMAP_F_NEW
;
3371 struct iomap_ops ext4_iomap_ops
= {
3372 .iomap_begin
= ext4_iomap_begin
,
3376 /* Just define empty function, it will never get called. */
3377 int ext4_dax_get_block(struct inode
*inode
, sector_t iblock
,
3378 struct buffer_head
*bh_result
, int create
)
3385 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3386 ssize_t size
, void *private)
3388 ext4_io_end_t
*io_end
= private;
3390 /* if not async direct IO just return */
3394 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3395 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3396 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3399 * Error during AIO DIO. We cannot convert unwritten extents as the
3400 * data was not written. Just clear the unwritten flag and drop io_end.
3403 ext4_clear_io_unwritten_flag(io_end
);
3406 io_end
->offset
= offset
;
3407 io_end
->size
= size
;
3408 ext4_put_io_end(io_end
);
3414 * Handling of direct IO writes.
3416 * For ext4 extent files, ext4 will do direct-io write even to holes,
3417 * preallocated extents, and those write extend the file, no need to
3418 * fall back to buffered IO.
3420 * For holes, we fallocate those blocks, mark them as unwritten
3421 * If those blocks were preallocated, we mark sure they are split, but
3422 * still keep the range to write as unwritten.
3424 * The unwritten extents will be converted to written when DIO is completed.
3425 * For async direct IO, since the IO may still pending when return, we
3426 * set up an end_io call back function, which will do the conversion
3427 * when async direct IO completed.
3429 * If the O_DIRECT write will extend the file then add this inode to the
3430 * orphan list. So recovery will truncate it back to the original size
3431 * if the machine crashes during the write.
3434 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3436 struct file
*file
= iocb
->ki_filp
;
3437 struct inode
*inode
= file
->f_mapping
->host
;
3438 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3440 loff_t offset
= iocb
->ki_pos
;
3441 size_t count
= iov_iter_count(iter
);
3443 get_block_t
*get_block_func
= NULL
;
3445 loff_t final_size
= offset
+ count
;
3449 if (final_size
> inode
->i_size
) {
3450 /* Credits for sb + inode write */
3451 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3452 if (IS_ERR(handle
)) {
3453 ret
= PTR_ERR(handle
);
3456 ret
= ext4_orphan_add(handle
, inode
);
3458 ext4_journal_stop(handle
);
3462 ei
->i_disksize
= inode
->i_size
;
3463 ext4_journal_stop(handle
);
3466 BUG_ON(iocb
->private == NULL
);
3469 * Make all waiters for direct IO properly wait also for extent
3470 * conversion. This also disallows race between truncate() and
3471 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3473 inode_dio_begin(inode
);
3475 /* If we do a overwrite dio, i_mutex locking can be released */
3476 overwrite
= *((int *)iocb
->private);
3479 inode_unlock(inode
);
3482 * For extent mapped files we could direct write to holes and fallocate.
3484 * Allocated blocks to fill the hole are marked as unwritten to prevent
3485 * parallel buffered read to expose the stale data before DIO complete
3488 * As to previously fallocated extents, ext4 get_block will just simply
3489 * mark the buffer mapped but still keep the extents unwritten.
3491 * For non AIO case, we will convert those unwritten extents to written
3492 * after return back from blockdev_direct_IO. That way we save us from
3493 * allocating io_end structure and also the overhead of offloading
3494 * the extent convertion to a workqueue.
3496 * For async DIO, the conversion needs to be deferred when the
3497 * IO is completed. The ext4 end_io callback function will be
3498 * called to take care of the conversion work. Here for async
3499 * case, we allocate an io_end structure to hook to the iocb.
3501 iocb
->private = NULL
;
3503 get_block_func
= ext4_dio_get_block_overwrite
;
3504 else if (IS_DAX(inode
)) {
3506 * We can avoid zeroing for aligned DAX writes beyond EOF. Other
3507 * writes need zeroing either because they can race with page
3508 * faults or because they use partial blocks.
3510 if (round_down(offset
, 1<<inode
->i_blkbits
) >= inode
->i_size
&&
3511 ext4_aligned_io(inode
, offset
, count
))
3512 get_block_func
= ext4_dio_get_block
;
3514 get_block_func
= ext4_dax_get_block
;
3515 dio_flags
= DIO_LOCKING
;
3516 } else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3517 round_down(offset
, 1 << inode
->i_blkbits
) >= inode
->i_size
) {
3518 get_block_func
= ext4_dio_get_block
;
3519 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3520 } else if (is_sync_kiocb(iocb
)) {
3521 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3522 dio_flags
= DIO_LOCKING
;
3524 get_block_func
= ext4_dio_get_block_unwritten_async
;
3525 dio_flags
= DIO_LOCKING
;
3527 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3528 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3530 if (IS_DAX(inode
)) {
3531 ret
= dax_do_io(iocb
, inode
, iter
, get_block_func
,
3532 ext4_end_io_dio
, dio_flags
);
3534 ret
= __blockdev_direct_IO(iocb
, inode
,
3535 inode
->i_sb
->s_bdev
, iter
,
3537 ext4_end_io_dio
, NULL
, dio_flags
);
3539 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3540 EXT4_STATE_DIO_UNWRITTEN
)) {
3543 * for non AIO case, since the IO is already
3544 * completed, we could do the conversion right here
3546 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3550 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3553 inode_dio_end(inode
);
3554 /* take i_mutex locking again if we do a ovewrite dio */
3558 if (ret
< 0 && final_size
> inode
->i_size
)
3559 ext4_truncate_failed_write(inode
);
3561 /* Handle extending of i_size after direct IO write */
3565 /* Credits for sb + inode write */
3566 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3567 if (IS_ERR(handle
)) {
3568 /* This is really bad luck. We've written the data
3569 * but cannot extend i_size. Bail out and pretend
3570 * the write failed... */
3571 ret
= PTR_ERR(handle
);
3573 ext4_orphan_del(NULL
, inode
);
3578 ext4_orphan_del(handle
, inode
);
3580 loff_t end
= offset
+ ret
;
3581 if (end
> inode
->i_size
) {
3582 ei
->i_disksize
= end
;
3583 i_size_write(inode
, end
);
3585 * We're going to return a positive `ret'
3586 * here due to non-zero-length I/O, so there's
3587 * no way of reporting error returns from
3588 * ext4_mark_inode_dirty() to userspace. So
3591 ext4_mark_inode_dirty(handle
, inode
);
3594 err
= ext4_journal_stop(handle
);
3602 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3604 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3605 struct inode
*inode
= mapping
->host
;
3609 * Shared inode_lock is enough for us - it protects against concurrent
3610 * writes & truncates and since we take care of writing back page cache,
3611 * we are protected against page writeback as well.
3613 inode_lock_shared(inode
);
3614 if (IS_DAX(inode
)) {
3615 ret
= dax_do_io(iocb
, inode
, iter
, ext4_dio_get_block
, NULL
, 0);
3617 size_t count
= iov_iter_count(iter
);
3619 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3620 iocb
->ki_pos
+ count
);
3623 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3624 iter
, ext4_dio_get_block
,
3628 inode_unlock_shared(inode
);
3632 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3634 struct file
*file
= iocb
->ki_filp
;
3635 struct inode
*inode
= file
->f_mapping
->host
;
3636 size_t count
= iov_iter_count(iter
);
3637 loff_t offset
= iocb
->ki_pos
;
3640 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3641 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3646 * If we are doing data journalling we don't support O_DIRECT
3648 if (ext4_should_journal_data(inode
))
3651 /* Let buffer I/O handle the inline data case. */
3652 if (ext4_has_inline_data(inode
))
3655 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3656 if (iov_iter_rw(iter
) == READ
)
3657 ret
= ext4_direct_IO_read(iocb
, iter
);
3659 ret
= ext4_direct_IO_write(iocb
, iter
);
3660 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3665 * Pages can be marked dirty completely asynchronously from ext4's journalling
3666 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3667 * much here because ->set_page_dirty is called under VFS locks. The page is
3668 * not necessarily locked.
3670 * We cannot just dirty the page and leave attached buffers clean, because the
3671 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3672 * or jbddirty because all the journalling code will explode.
3674 * So what we do is to mark the page "pending dirty" and next time writepage
3675 * is called, propagate that into the buffers appropriately.
3677 static int ext4_journalled_set_page_dirty(struct page
*page
)
3679 SetPageChecked(page
);
3680 return __set_page_dirty_nobuffers(page
);
3683 static const struct address_space_operations ext4_aops
= {
3684 .readpage
= ext4_readpage
,
3685 .readpages
= ext4_readpages
,
3686 .writepage
= ext4_writepage
,
3687 .writepages
= ext4_writepages
,
3688 .write_begin
= ext4_write_begin
,
3689 .write_end
= ext4_write_end
,
3691 .invalidatepage
= ext4_invalidatepage
,
3692 .releasepage
= ext4_releasepage
,
3693 .direct_IO
= ext4_direct_IO
,
3694 .migratepage
= buffer_migrate_page
,
3695 .is_partially_uptodate
= block_is_partially_uptodate
,
3696 .error_remove_page
= generic_error_remove_page
,
3699 static const struct address_space_operations ext4_journalled_aops
= {
3700 .readpage
= ext4_readpage
,
3701 .readpages
= ext4_readpages
,
3702 .writepage
= ext4_writepage
,
3703 .writepages
= ext4_writepages
,
3704 .write_begin
= ext4_write_begin
,
3705 .write_end
= ext4_journalled_write_end
,
3706 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3708 .invalidatepage
= ext4_journalled_invalidatepage
,
3709 .releasepage
= ext4_releasepage
,
3710 .direct_IO
= ext4_direct_IO
,
3711 .is_partially_uptodate
= block_is_partially_uptodate
,
3712 .error_remove_page
= generic_error_remove_page
,
3715 static const struct address_space_operations ext4_da_aops
= {
3716 .readpage
= ext4_readpage
,
3717 .readpages
= ext4_readpages
,
3718 .writepage
= ext4_writepage
,
3719 .writepages
= ext4_writepages
,
3720 .write_begin
= ext4_da_write_begin
,
3721 .write_end
= ext4_da_write_end
,
3723 .invalidatepage
= ext4_da_invalidatepage
,
3724 .releasepage
= ext4_releasepage
,
3725 .direct_IO
= ext4_direct_IO
,
3726 .migratepage
= buffer_migrate_page
,
3727 .is_partially_uptodate
= block_is_partially_uptodate
,
3728 .error_remove_page
= generic_error_remove_page
,
3731 void ext4_set_aops(struct inode
*inode
)
3733 switch (ext4_inode_journal_mode(inode
)) {
3734 case EXT4_INODE_ORDERED_DATA_MODE
:
3735 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3737 case EXT4_INODE_JOURNAL_DATA_MODE
:
3738 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3743 if (test_opt(inode
->i_sb
, DELALLOC
))
3744 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3746 inode
->i_mapping
->a_ops
= &ext4_aops
;
3749 static int __ext4_block_zero_page_range(handle_t
*handle
,
3750 struct address_space
*mapping
, loff_t from
, loff_t length
)
3752 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3753 unsigned offset
= from
& (PAGE_SIZE
-1);
3754 unsigned blocksize
, pos
;
3756 struct inode
*inode
= mapping
->host
;
3757 struct buffer_head
*bh
;
3761 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3762 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3766 blocksize
= inode
->i_sb
->s_blocksize
;
3768 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3770 if (!page_has_buffers(page
))
3771 create_empty_buffers(page
, blocksize
, 0);
3773 /* Find the buffer that contains "offset" */
3774 bh
= page_buffers(page
);
3776 while (offset
>= pos
) {
3777 bh
= bh
->b_this_page
;
3781 if (buffer_freed(bh
)) {
3782 BUFFER_TRACE(bh
, "freed: skip");
3785 if (!buffer_mapped(bh
)) {
3786 BUFFER_TRACE(bh
, "unmapped");
3787 ext4_get_block(inode
, iblock
, bh
, 0);
3788 /* unmapped? It's a hole - nothing to do */
3789 if (!buffer_mapped(bh
)) {
3790 BUFFER_TRACE(bh
, "still unmapped");
3795 /* Ok, it's mapped. Make sure it's up-to-date */
3796 if (PageUptodate(page
))
3797 set_buffer_uptodate(bh
);
3799 if (!buffer_uptodate(bh
)) {
3801 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3803 /* Uhhuh. Read error. Complain and punt. */
3804 if (!buffer_uptodate(bh
))
3806 if (S_ISREG(inode
->i_mode
) &&
3807 ext4_encrypted_inode(inode
)) {
3808 /* We expect the key to be set. */
3809 BUG_ON(!fscrypt_has_encryption_key(inode
));
3810 BUG_ON(blocksize
!= PAGE_SIZE
);
3811 BUG_ON(!PageLocked(page
));
3812 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
3813 page
, PAGE_SIZE
, 0, page
->index
));
3816 if (ext4_should_journal_data(inode
)) {
3817 BUFFER_TRACE(bh
, "get write access");
3818 err
= ext4_journal_get_write_access(handle
, bh
);
3822 zero_user(page
, offset
, length
);
3823 BUFFER_TRACE(bh
, "zeroed end of block");
3825 if (ext4_should_journal_data(inode
)) {
3826 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3829 mark_buffer_dirty(bh
);
3830 if (ext4_should_order_data(inode
))
3831 err
= ext4_jbd2_inode_add_write(handle
, inode
);
3841 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3842 * starting from file offset 'from'. The range to be zero'd must
3843 * be contained with in one block. If the specified range exceeds
3844 * the end of the block it will be shortened to end of the block
3845 * that cooresponds to 'from'
3847 static int ext4_block_zero_page_range(handle_t
*handle
,
3848 struct address_space
*mapping
, loff_t from
, loff_t length
)
3850 struct inode
*inode
= mapping
->host
;
3851 unsigned offset
= from
& (PAGE_SIZE
-1);
3852 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3853 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3856 * correct length if it does not fall between
3857 * 'from' and the end of the block
3859 if (length
> max
|| length
< 0)
3863 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3864 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3868 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3869 * up to the end of the block which corresponds to `from'.
3870 * This required during truncate. We need to physically zero the tail end
3871 * of that block so it doesn't yield old data if the file is later grown.
3873 static int ext4_block_truncate_page(handle_t
*handle
,
3874 struct address_space
*mapping
, loff_t from
)
3876 unsigned offset
= from
& (PAGE_SIZE
-1);
3879 struct inode
*inode
= mapping
->host
;
3881 blocksize
= inode
->i_sb
->s_blocksize
;
3882 length
= blocksize
- (offset
& (blocksize
- 1));
3884 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3887 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3888 loff_t lstart
, loff_t length
)
3890 struct super_block
*sb
= inode
->i_sb
;
3891 struct address_space
*mapping
= inode
->i_mapping
;
3892 unsigned partial_start
, partial_end
;
3893 ext4_fsblk_t start
, end
;
3894 loff_t byte_end
= (lstart
+ length
- 1);
3897 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3898 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3900 start
= lstart
>> sb
->s_blocksize_bits
;
3901 end
= byte_end
>> sb
->s_blocksize_bits
;
3903 /* Handle partial zero within the single block */
3905 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3906 err
= ext4_block_zero_page_range(handle
, mapping
,
3910 /* Handle partial zero out on the start of the range */
3911 if (partial_start
) {
3912 err
= ext4_block_zero_page_range(handle
, mapping
,
3913 lstart
, sb
->s_blocksize
);
3917 /* Handle partial zero out on the end of the range */
3918 if (partial_end
!= sb
->s_blocksize
- 1)
3919 err
= ext4_block_zero_page_range(handle
, mapping
,
3920 byte_end
- partial_end
,
3925 int ext4_can_truncate(struct inode
*inode
)
3927 if (S_ISREG(inode
->i_mode
))
3929 if (S_ISDIR(inode
->i_mode
))
3931 if (S_ISLNK(inode
->i_mode
))
3932 return !ext4_inode_is_fast_symlink(inode
);
3937 * We have to make sure i_disksize gets properly updated before we truncate
3938 * page cache due to hole punching or zero range. Otherwise i_disksize update
3939 * can get lost as it may have been postponed to submission of writeback but
3940 * that will never happen after we truncate page cache.
3942 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3946 loff_t size
= i_size_read(inode
);
3948 WARN_ON(!inode_is_locked(inode
));
3949 if (offset
> size
|| offset
+ len
< size
)
3952 if (EXT4_I(inode
)->i_disksize
>= size
)
3955 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3957 return PTR_ERR(handle
);
3958 ext4_update_i_disksize(inode
, size
);
3959 ext4_mark_inode_dirty(handle
, inode
);
3960 ext4_journal_stop(handle
);
3966 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3967 * associated with the given offset and length
3969 * @inode: File inode
3970 * @offset: The offset where the hole will begin
3971 * @len: The length of the hole
3973 * Returns: 0 on success or negative on failure
3976 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3978 struct super_block
*sb
= inode
->i_sb
;
3979 ext4_lblk_t first_block
, stop_block
;
3980 struct address_space
*mapping
= inode
->i_mapping
;
3981 loff_t first_block_offset
, last_block_offset
;
3983 unsigned int credits
;
3986 if (!S_ISREG(inode
->i_mode
))
3989 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3992 * Write out all dirty pages to avoid race conditions
3993 * Then release them.
3995 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3996 ret
= filemap_write_and_wait_range(mapping
, offset
,
3997 offset
+ length
- 1);
4004 /* No need to punch hole beyond i_size */
4005 if (offset
>= inode
->i_size
)
4009 * If the hole extends beyond i_size, set the hole
4010 * to end after the page that contains i_size
4012 if (offset
+ length
> inode
->i_size
) {
4013 length
= inode
->i_size
+
4014 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4018 if (offset
& (sb
->s_blocksize
- 1) ||
4019 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4021 * Attach jinode to inode for jbd2 if we do any zeroing of
4024 ret
= ext4_inode_attach_jinode(inode
);
4030 /* Wait all existing dio workers, newcomers will block on i_mutex */
4031 ext4_inode_block_unlocked_dio(inode
);
4032 inode_dio_wait(inode
);
4035 * Prevent page faults from reinstantiating pages we have released from
4038 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4039 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4040 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4042 /* Now release the pages and zero block aligned part of pages*/
4043 if (last_block_offset
> first_block_offset
) {
4044 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4047 truncate_pagecache_range(inode
, first_block_offset
,
4051 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4052 credits
= ext4_writepage_trans_blocks(inode
);
4054 credits
= ext4_blocks_for_truncate(inode
);
4055 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4056 if (IS_ERR(handle
)) {
4057 ret
= PTR_ERR(handle
);
4058 ext4_std_error(sb
, ret
);
4062 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4067 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4068 EXT4_BLOCK_SIZE_BITS(sb
);
4069 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4071 /* If there are no blocks to remove, return now */
4072 if (first_block
>= stop_block
)
4075 down_write(&EXT4_I(inode
)->i_data_sem
);
4076 ext4_discard_preallocations(inode
);
4078 ret
= ext4_es_remove_extent(inode
, first_block
,
4079 stop_block
- first_block
);
4081 up_write(&EXT4_I(inode
)->i_data_sem
);
4085 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4086 ret
= ext4_ext_remove_space(inode
, first_block
,
4089 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4092 up_write(&EXT4_I(inode
)->i_data_sem
);
4094 ext4_handle_sync(handle
);
4096 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4097 ext4_mark_inode_dirty(handle
, inode
);
4099 ext4_journal_stop(handle
);
4101 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4102 ext4_inode_resume_unlocked_dio(inode
);
4104 inode_unlock(inode
);
4108 int ext4_inode_attach_jinode(struct inode
*inode
)
4110 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4111 struct jbd2_inode
*jinode
;
4113 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4116 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4117 spin_lock(&inode
->i_lock
);
4120 spin_unlock(&inode
->i_lock
);
4123 ei
->jinode
= jinode
;
4124 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4127 spin_unlock(&inode
->i_lock
);
4128 if (unlikely(jinode
!= NULL
))
4129 jbd2_free_inode(jinode
);
4136 * We block out ext4_get_block() block instantiations across the entire
4137 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4138 * simultaneously on behalf of the same inode.
4140 * As we work through the truncate and commit bits of it to the journal there
4141 * is one core, guiding principle: the file's tree must always be consistent on
4142 * disk. We must be able to restart the truncate after a crash.
4144 * The file's tree may be transiently inconsistent in memory (although it
4145 * probably isn't), but whenever we close off and commit a journal transaction,
4146 * the contents of (the filesystem + the journal) must be consistent and
4147 * restartable. It's pretty simple, really: bottom up, right to left (although
4148 * left-to-right works OK too).
4150 * Note that at recovery time, journal replay occurs *before* the restart of
4151 * truncate against the orphan inode list.
4153 * The committed inode has the new, desired i_size (which is the same as
4154 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4155 * that this inode's truncate did not complete and it will again call
4156 * ext4_truncate() to have another go. So there will be instantiated blocks
4157 * to the right of the truncation point in a crashed ext4 filesystem. But
4158 * that's fine - as long as they are linked from the inode, the post-crash
4159 * ext4_truncate() run will find them and release them.
4161 int ext4_truncate(struct inode
*inode
)
4163 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4164 unsigned int credits
;
4167 struct address_space
*mapping
= inode
->i_mapping
;
4170 * There is a possibility that we're either freeing the inode
4171 * or it's a completely new inode. In those cases we might not
4172 * have i_mutex locked because it's not necessary.
4174 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4175 WARN_ON(!inode_is_locked(inode
));
4176 trace_ext4_truncate_enter(inode
);
4178 if (!ext4_can_truncate(inode
))
4181 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4183 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4184 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4186 if (ext4_has_inline_data(inode
)) {
4189 ext4_inline_data_truncate(inode
, &has_inline
);
4194 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4195 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4196 if (ext4_inode_attach_jinode(inode
) < 0)
4200 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4201 credits
= ext4_writepage_trans_blocks(inode
);
4203 credits
= ext4_blocks_for_truncate(inode
);
4205 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4207 return PTR_ERR(handle
);
4209 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4210 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4213 * We add the inode to the orphan list, so that if this
4214 * truncate spans multiple transactions, and we crash, we will
4215 * resume the truncate when the filesystem recovers. It also
4216 * marks the inode dirty, to catch the new size.
4218 * Implication: the file must always be in a sane, consistent
4219 * truncatable state while each transaction commits.
4221 err
= ext4_orphan_add(handle
, inode
);
4225 down_write(&EXT4_I(inode
)->i_data_sem
);
4227 ext4_discard_preallocations(inode
);
4229 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4230 err
= ext4_ext_truncate(handle
, inode
);
4232 ext4_ind_truncate(handle
, inode
);
4234 up_write(&ei
->i_data_sem
);
4239 ext4_handle_sync(handle
);
4243 * If this was a simple ftruncate() and the file will remain alive,
4244 * then we need to clear up the orphan record which we created above.
4245 * However, if this was a real unlink then we were called by
4246 * ext4_evict_inode(), and we allow that function to clean up the
4247 * orphan info for us.
4250 ext4_orphan_del(handle
, inode
);
4252 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4253 ext4_mark_inode_dirty(handle
, inode
);
4254 ext4_journal_stop(handle
);
4256 trace_ext4_truncate_exit(inode
);
4261 * ext4_get_inode_loc returns with an extra refcount against the inode's
4262 * underlying buffer_head on success. If 'in_mem' is true, we have all
4263 * data in memory that is needed to recreate the on-disk version of this
4266 static int __ext4_get_inode_loc(struct inode
*inode
,
4267 struct ext4_iloc
*iloc
, int in_mem
)
4269 struct ext4_group_desc
*gdp
;
4270 struct buffer_head
*bh
;
4271 struct super_block
*sb
= inode
->i_sb
;
4273 int inodes_per_block
, inode_offset
;
4276 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4277 return -EFSCORRUPTED
;
4279 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4280 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4285 * Figure out the offset within the block group inode table
4287 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4288 inode_offset
= ((inode
->i_ino
- 1) %
4289 EXT4_INODES_PER_GROUP(sb
));
4290 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4291 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4293 bh
= sb_getblk(sb
, block
);
4296 if (!buffer_uptodate(bh
)) {
4300 * If the buffer has the write error flag, we have failed
4301 * to write out another inode in the same block. In this
4302 * case, we don't have to read the block because we may
4303 * read the old inode data successfully.
4305 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4306 set_buffer_uptodate(bh
);
4308 if (buffer_uptodate(bh
)) {
4309 /* someone brought it uptodate while we waited */
4315 * If we have all information of the inode in memory and this
4316 * is the only valid inode in the block, we need not read the
4320 struct buffer_head
*bitmap_bh
;
4323 start
= inode_offset
& ~(inodes_per_block
- 1);
4325 /* Is the inode bitmap in cache? */
4326 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4327 if (unlikely(!bitmap_bh
))
4331 * If the inode bitmap isn't in cache then the
4332 * optimisation may end up performing two reads instead
4333 * of one, so skip it.
4335 if (!buffer_uptodate(bitmap_bh
)) {
4339 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4340 if (i
== inode_offset
)
4342 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4346 if (i
== start
+ inodes_per_block
) {
4347 /* all other inodes are free, so skip I/O */
4348 memset(bh
->b_data
, 0, bh
->b_size
);
4349 set_buffer_uptodate(bh
);
4357 * If we need to do any I/O, try to pre-readahead extra
4358 * blocks from the inode table.
4360 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4361 ext4_fsblk_t b
, end
, table
;
4363 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4365 table
= ext4_inode_table(sb
, gdp
);
4366 /* s_inode_readahead_blks is always a power of 2 */
4367 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4371 num
= EXT4_INODES_PER_GROUP(sb
);
4372 if (ext4_has_group_desc_csum(sb
))
4373 num
-= ext4_itable_unused_count(sb
, gdp
);
4374 table
+= num
/ inodes_per_block
;
4378 sb_breadahead(sb
, b
++);
4382 * There are other valid inodes in the buffer, this inode
4383 * has in-inode xattrs, or we don't have this inode in memory.
4384 * Read the block from disk.
4386 trace_ext4_load_inode(inode
);
4388 bh
->b_end_io
= end_buffer_read_sync
;
4389 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4391 if (!buffer_uptodate(bh
)) {
4392 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4393 "unable to read itable block");
4403 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4405 /* We have all inode data except xattrs in memory here. */
4406 return __ext4_get_inode_loc(inode
, iloc
,
4407 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4410 void ext4_set_inode_flags(struct inode
*inode
)
4412 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4413 unsigned int new_fl
= 0;
4415 if (flags
& EXT4_SYNC_FL
)
4417 if (flags
& EXT4_APPEND_FL
)
4419 if (flags
& EXT4_IMMUTABLE_FL
)
4420 new_fl
|= S_IMMUTABLE
;
4421 if (flags
& EXT4_NOATIME_FL
)
4422 new_fl
|= S_NOATIME
;
4423 if (flags
& EXT4_DIRSYNC_FL
)
4424 new_fl
|= S_DIRSYNC
;
4425 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
) &&
4426 !ext4_should_journal_data(inode
) && !ext4_has_inline_data(inode
) &&
4427 !ext4_encrypted_inode(inode
))
4429 inode_set_flags(inode
, new_fl
,
4430 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4433 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4434 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4436 unsigned int vfs_fl
;
4437 unsigned long old_fl
, new_fl
;
4440 vfs_fl
= ei
->vfs_inode
.i_flags
;
4441 old_fl
= ei
->i_flags
;
4442 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4443 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4445 if (vfs_fl
& S_SYNC
)
4446 new_fl
|= EXT4_SYNC_FL
;
4447 if (vfs_fl
& S_APPEND
)
4448 new_fl
|= EXT4_APPEND_FL
;
4449 if (vfs_fl
& S_IMMUTABLE
)
4450 new_fl
|= EXT4_IMMUTABLE_FL
;
4451 if (vfs_fl
& S_NOATIME
)
4452 new_fl
|= EXT4_NOATIME_FL
;
4453 if (vfs_fl
& S_DIRSYNC
)
4454 new_fl
|= EXT4_DIRSYNC_FL
;
4455 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4458 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4459 struct ext4_inode_info
*ei
)
4462 struct inode
*inode
= &(ei
->vfs_inode
);
4463 struct super_block
*sb
= inode
->i_sb
;
4465 if (ext4_has_feature_huge_file(sb
)) {
4466 /* we are using combined 48 bit field */
4467 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4468 le32_to_cpu(raw_inode
->i_blocks_lo
);
4469 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4470 /* i_blocks represent file system block size */
4471 return i_blocks
<< (inode
->i_blkbits
- 9);
4476 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4480 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4481 struct ext4_inode
*raw_inode
,
4482 struct ext4_inode_info
*ei
)
4484 __le32
*magic
= (void *)raw_inode
+
4485 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4486 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4487 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4488 ext4_find_inline_data_nolock(inode
);
4490 EXT4_I(inode
)->i_inline_off
= 0;
4493 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4495 if (!ext4_has_feature_project(inode
->i_sb
))
4497 *projid
= EXT4_I(inode
)->i_projid
;
4501 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4503 struct ext4_iloc iloc
;
4504 struct ext4_inode
*raw_inode
;
4505 struct ext4_inode_info
*ei
;
4506 struct inode
*inode
;
4507 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4514 inode
= iget_locked(sb
, ino
);
4516 return ERR_PTR(-ENOMEM
);
4517 if (!(inode
->i_state
& I_NEW
))
4523 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4526 raw_inode
= ext4_raw_inode(&iloc
);
4528 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4529 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4530 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4531 EXT4_INODE_SIZE(inode
->i_sb
)) {
4532 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4533 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4534 EXT4_INODE_SIZE(inode
->i_sb
));
4535 ret
= -EFSCORRUPTED
;
4539 ei
->i_extra_isize
= 0;
4541 /* Precompute checksum seed for inode metadata */
4542 if (ext4_has_metadata_csum(sb
)) {
4543 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4545 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4546 __le32 gen
= raw_inode
->i_generation
;
4547 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4549 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4553 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4554 EXT4_ERROR_INODE(inode
, "checksum invalid");
4559 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4560 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4561 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4562 if (ext4_has_feature_project(sb
) &&
4563 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4564 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4565 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4567 i_projid
= EXT4_DEF_PROJID
;
4569 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4570 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4571 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4573 i_uid_write(inode
, i_uid
);
4574 i_gid_write(inode
, i_gid
);
4575 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4576 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4578 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4579 ei
->i_inline_off
= 0;
4580 ei
->i_dir_start_lookup
= 0;
4581 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4582 /* We now have enough fields to check if the inode was active or not.
4583 * This is needed because nfsd might try to access dead inodes
4584 * the test is that same one that e2fsck uses
4585 * NeilBrown 1999oct15
4587 if (inode
->i_nlink
== 0) {
4588 if ((inode
->i_mode
== 0 ||
4589 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4590 ino
!= EXT4_BOOT_LOADER_INO
) {
4591 /* this inode is deleted */
4595 /* The only unlinked inodes we let through here have
4596 * valid i_mode and are being read by the orphan
4597 * recovery code: that's fine, we're about to complete
4598 * the process of deleting those.
4599 * OR it is the EXT4_BOOT_LOADER_INO which is
4600 * not initialized on a new filesystem. */
4602 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4603 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4604 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4605 if (ext4_has_feature_64bit(sb
))
4607 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4608 inode
->i_size
= ext4_isize(raw_inode
);
4609 ei
->i_disksize
= inode
->i_size
;
4611 ei
->i_reserved_quota
= 0;
4613 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4614 ei
->i_block_group
= iloc
.block_group
;
4615 ei
->i_last_alloc_group
= ~0;
4617 * NOTE! The in-memory inode i_data array is in little-endian order
4618 * even on big-endian machines: we do NOT byteswap the block numbers!
4620 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4621 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4622 INIT_LIST_HEAD(&ei
->i_orphan
);
4625 * Set transaction id's of transactions that have to be committed
4626 * to finish f[data]sync. We set them to currently running transaction
4627 * as we cannot be sure that the inode or some of its metadata isn't
4628 * part of the transaction - the inode could have been reclaimed and
4629 * now it is reread from disk.
4632 transaction_t
*transaction
;
4635 read_lock(&journal
->j_state_lock
);
4636 if (journal
->j_running_transaction
)
4637 transaction
= journal
->j_running_transaction
;
4639 transaction
= journal
->j_committing_transaction
;
4641 tid
= transaction
->t_tid
;
4643 tid
= journal
->j_commit_sequence
;
4644 read_unlock(&journal
->j_state_lock
);
4645 ei
->i_sync_tid
= tid
;
4646 ei
->i_datasync_tid
= tid
;
4649 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4650 if (ei
->i_extra_isize
== 0) {
4651 /* The extra space is currently unused. Use it. */
4652 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4653 EXT4_GOOD_OLD_INODE_SIZE
;
4655 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4659 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4660 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4661 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4662 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4664 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4665 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4666 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4667 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4669 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4674 if (ei
->i_file_acl
&&
4675 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4676 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4678 ret
= -EFSCORRUPTED
;
4680 } else if (!ext4_has_inline_data(inode
)) {
4681 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4682 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4683 (S_ISLNK(inode
->i_mode
) &&
4684 !ext4_inode_is_fast_symlink(inode
))))
4685 /* Validate extent which is part of inode */
4686 ret
= ext4_ext_check_inode(inode
);
4687 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4688 (S_ISLNK(inode
->i_mode
) &&
4689 !ext4_inode_is_fast_symlink(inode
))) {
4690 /* Validate block references which are part of inode */
4691 ret
= ext4_ind_check_inode(inode
);
4697 if (S_ISREG(inode
->i_mode
)) {
4698 inode
->i_op
= &ext4_file_inode_operations
;
4699 inode
->i_fop
= &ext4_file_operations
;
4700 ext4_set_aops(inode
);
4701 } else if (S_ISDIR(inode
->i_mode
)) {
4702 inode
->i_op
= &ext4_dir_inode_operations
;
4703 inode
->i_fop
= &ext4_dir_operations
;
4704 } else if (S_ISLNK(inode
->i_mode
)) {
4705 if (ext4_encrypted_inode(inode
)) {
4706 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4707 ext4_set_aops(inode
);
4708 } else if (ext4_inode_is_fast_symlink(inode
)) {
4709 inode
->i_link
= (char *)ei
->i_data
;
4710 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4711 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4712 sizeof(ei
->i_data
) - 1);
4714 inode
->i_op
= &ext4_symlink_inode_operations
;
4715 ext4_set_aops(inode
);
4717 inode_nohighmem(inode
);
4718 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4719 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4720 inode
->i_op
= &ext4_special_inode_operations
;
4721 if (raw_inode
->i_block
[0])
4722 init_special_inode(inode
, inode
->i_mode
,
4723 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4725 init_special_inode(inode
, inode
->i_mode
,
4726 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4727 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4728 make_bad_inode(inode
);
4730 ret
= -EFSCORRUPTED
;
4731 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4735 ext4_set_inode_flags(inode
);
4736 unlock_new_inode(inode
);
4742 return ERR_PTR(ret
);
4745 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4747 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4748 return ERR_PTR(-EFSCORRUPTED
);
4749 return ext4_iget(sb
, ino
);
4752 static int ext4_inode_blocks_set(handle_t
*handle
,
4753 struct ext4_inode
*raw_inode
,
4754 struct ext4_inode_info
*ei
)
4756 struct inode
*inode
= &(ei
->vfs_inode
);
4757 u64 i_blocks
= inode
->i_blocks
;
4758 struct super_block
*sb
= inode
->i_sb
;
4760 if (i_blocks
<= ~0U) {
4762 * i_blocks can be represented in a 32 bit variable
4763 * as multiple of 512 bytes
4765 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4766 raw_inode
->i_blocks_high
= 0;
4767 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4770 if (!ext4_has_feature_huge_file(sb
))
4773 if (i_blocks
<= 0xffffffffffffULL
) {
4775 * i_blocks can be represented in a 48 bit variable
4776 * as multiple of 512 bytes
4778 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4779 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4780 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4782 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4783 /* i_block is stored in file system block size */
4784 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4785 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4786 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4791 struct other_inode
{
4792 unsigned long orig_ino
;
4793 struct ext4_inode
*raw_inode
;
4796 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4799 struct other_inode
*oi
= (struct other_inode
*) data
;
4801 if ((inode
->i_ino
!= ino
) ||
4802 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4803 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4804 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4806 spin_lock(&inode
->i_lock
);
4807 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4808 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4809 (inode
->i_state
& I_DIRTY_TIME
)) {
4810 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4812 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4813 spin_unlock(&inode
->i_lock
);
4815 spin_lock(&ei
->i_raw_lock
);
4816 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4817 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4818 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4819 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4820 spin_unlock(&ei
->i_raw_lock
);
4821 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4824 spin_unlock(&inode
->i_lock
);
4829 * Opportunistically update the other time fields for other inodes in
4830 * the same inode table block.
4832 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4833 unsigned long orig_ino
, char *buf
)
4835 struct other_inode oi
;
4837 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4838 int inode_size
= EXT4_INODE_SIZE(sb
);
4840 oi
.orig_ino
= orig_ino
;
4842 * Calculate the first inode in the inode table block. Inode
4843 * numbers are one-based. That is, the first inode in a block
4844 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4846 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4847 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4848 if (ino
== orig_ino
)
4850 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4851 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4856 * Post the struct inode info into an on-disk inode location in the
4857 * buffer-cache. This gobbles the caller's reference to the
4858 * buffer_head in the inode location struct.
4860 * The caller must have write access to iloc->bh.
4862 static int ext4_do_update_inode(handle_t
*handle
,
4863 struct inode
*inode
,
4864 struct ext4_iloc
*iloc
)
4866 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4867 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4868 struct buffer_head
*bh
= iloc
->bh
;
4869 struct super_block
*sb
= inode
->i_sb
;
4870 int err
= 0, rc
, block
;
4871 int need_datasync
= 0, set_large_file
= 0;
4876 spin_lock(&ei
->i_raw_lock
);
4878 /* For fields not tracked in the in-memory inode,
4879 * initialise them to zero for new inodes. */
4880 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4881 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4883 ext4_get_inode_flags(ei
);
4884 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4885 i_uid
= i_uid_read(inode
);
4886 i_gid
= i_gid_read(inode
);
4887 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4888 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4889 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4890 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4892 * Fix up interoperability with old kernels. Otherwise, old inodes get
4893 * re-used with the upper 16 bits of the uid/gid intact
4895 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4896 raw_inode
->i_uid_high
= 0;
4897 raw_inode
->i_gid_high
= 0;
4899 raw_inode
->i_uid_high
=
4900 cpu_to_le16(high_16_bits(i_uid
));
4901 raw_inode
->i_gid_high
=
4902 cpu_to_le16(high_16_bits(i_gid
));
4905 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4906 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4907 raw_inode
->i_uid_high
= 0;
4908 raw_inode
->i_gid_high
= 0;
4910 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4912 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4913 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4914 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4915 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4917 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4919 spin_unlock(&ei
->i_raw_lock
);
4922 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4923 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4924 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4925 raw_inode
->i_file_acl_high
=
4926 cpu_to_le16(ei
->i_file_acl
>> 32);
4927 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4928 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4929 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4932 if (ei
->i_disksize
> 0x7fffffffULL
) {
4933 if (!ext4_has_feature_large_file(sb
) ||
4934 EXT4_SB(sb
)->s_es
->s_rev_level
==
4935 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4938 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4939 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4940 if (old_valid_dev(inode
->i_rdev
)) {
4941 raw_inode
->i_block
[0] =
4942 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4943 raw_inode
->i_block
[1] = 0;
4945 raw_inode
->i_block
[0] = 0;
4946 raw_inode
->i_block
[1] =
4947 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4948 raw_inode
->i_block
[2] = 0;
4950 } else if (!ext4_has_inline_data(inode
)) {
4951 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4952 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4955 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4956 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4957 if (ei
->i_extra_isize
) {
4958 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4959 raw_inode
->i_version_hi
=
4960 cpu_to_le32(inode
->i_version
>> 32);
4961 raw_inode
->i_extra_isize
=
4962 cpu_to_le16(ei
->i_extra_isize
);
4966 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
4967 i_projid
!= EXT4_DEF_PROJID
);
4969 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4970 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4971 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4973 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4974 spin_unlock(&ei
->i_raw_lock
);
4975 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4976 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4979 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4980 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4983 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4984 if (set_large_file
) {
4985 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4986 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4989 ext4_update_dynamic_rev(sb
);
4990 ext4_set_feature_large_file(sb
);
4991 ext4_handle_sync(handle
);
4992 err
= ext4_handle_dirty_super(handle
, sb
);
4994 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4997 ext4_std_error(inode
->i_sb
, err
);
5002 * ext4_write_inode()
5004 * We are called from a few places:
5006 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5007 * Here, there will be no transaction running. We wait for any running
5008 * transaction to commit.
5010 * - Within flush work (sys_sync(), kupdate and such).
5011 * We wait on commit, if told to.
5013 * - Within iput_final() -> write_inode_now()
5014 * We wait on commit, if told to.
5016 * In all cases it is actually safe for us to return without doing anything,
5017 * because the inode has been copied into a raw inode buffer in
5018 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5021 * Note that we are absolutely dependent upon all inode dirtiers doing the
5022 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5023 * which we are interested.
5025 * It would be a bug for them to not do this. The code:
5027 * mark_inode_dirty(inode)
5029 * inode->i_size = expr;
5031 * is in error because write_inode() could occur while `stuff()' is running,
5032 * and the new i_size will be lost. Plus the inode will no longer be on the
5033 * superblock's dirty inode list.
5035 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5039 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
5042 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5043 if (ext4_journal_current_handle()) {
5044 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5050 * No need to force transaction in WB_SYNC_NONE mode. Also
5051 * ext4_sync_fs() will force the commit after everything is
5054 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5057 err
= ext4_force_commit(inode
->i_sb
);
5059 struct ext4_iloc iloc
;
5061 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5065 * sync(2) will flush the whole buffer cache. No need to do
5066 * it here separately for each inode.
5068 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5069 sync_dirty_buffer(iloc
.bh
);
5070 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5071 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5072 "IO error syncing inode");
5081 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5082 * buffers that are attached to a page stradding i_size and are undergoing
5083 * commit. In that case we have to wait for commit to finish and try again.
5085 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5089 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5090 tid_t commit_tid
= 0;
5093 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5095 * All buffers in the last page remain valid? Then there's nothing to
5096 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5099 if (offset
> PAGE_SIZE
- (1 << inode
->i_blkbits
))
5102 page
= find_lock_page(inode
->i_mapping
,
5103 inode
->i_size
>> PAGE_SHIFT
);
5106 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5107 PAGE_SIZE
- offset
);
5113 read_lock(&journal
->j_state_lock
);
5114 if (journal
->j_committing_transaction
)
5115 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5116 read_unlock(&journal
->j_state_lock
);
5118 jbd2_log_wait_commit(journal
, commit_tid
);
5125 * Called from notify_change.
5127 * We want to trap VFS attempts to truncate the file as soon as
5128 * possible. In particular, we want to make sure that when the VFS
5129 * shrinks i_size, we put the inode on the orphan list and modify
5130 * i_disksize immediately, so that during the subsequent flushing of
5131 * dirty pages and freeing of disk blocks, we can guarantee that any
5132 * commit will leave the blocks being flushed in an unused state on
5133 * disk. (On recovery, the inode will get truncated and the blocks will
5134 * be freed, so we have a strong guarantee that no future commit will
5135 * leave these blocks visible to the user.)
5137 * Another thing we have to assure is that if we are in ordered mode
5138 * and inode is still attached to the committing transaction, we must
5139 * we start writeout of all the dirty pages which are being truncated.
5140 * This way we are sure that all the data written in the previous
5141 * transaction are already on disk (truncate waits for pages under
5144 * Called with inode->i_mutex down.
5146 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5148 struct inode
*inode
= d_inode(dentry
);
5151 const unsigned int ia_valid
= attr
->ia_valid
;
5153 error
= setattr_prepare(dentry
, attr
);
5157 if (is_quota_modification(inode
, attr
)) {
5158 error
= dquot_initialize(inode
);
5162 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5163 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5166 /* (user+group)*(old+new) structure, inode write (sb,
5167 * inode block, ? - but truncate inode update has it) */
5168 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5169 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5170 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5171 if (IS_ERR(handle
)) {
5172 error
= PTR_ERR(handle
);
5175 error
= dquot_transfer(inode
, attr
);
5177 ext4_journal_stop(handle
);
5180 /* Update corresponding info in inode so that everything is in
5181 * one transaction */
5182 if (attr
->ia_valid
& ATTR_UID
)
5183 inode
->i_uid
= attr
->ia_uid
;
5184 if (attr
->ia_valid
& ATTR_GID
)
5185 inode
->i_gid
= attr
->ia_gid
;
5186 error
= ext4_mark_inode_dirty(handle
, inode
);
5187 ext4_journal_stop(handle
);
5190 if (attr
->ia_valid
& ATTR_SIZE
) {
5192 loff_t oldsize
= inode
->i_size
;
5193 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5195 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5196 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5198 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5201 if (!S_ISREG(inode
->i_mode
))
5204 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5205 inode_inc_iversion(inode
);
5207 if (ext4_should_order_data(inode
) &&
5208 (attr
->ia_size
< inode
->i_size
)) {
5209 error
= ext4_begin_ordered_truncate(inode
,
5214 if (attr
->ia_size
!= inode
->i_size
) {
5215 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5216 if (IS_ERR(handle
)) {
5217 error
= PTR_ERR(handle
);
5220 if (ext4_handle_valid(handle
) && shrink
) {
5221 error
= ext4_orphan_add(handle
, inode
);
5225 * Update c/mtime on truncate up, ext4_truncate() will
5226 * update c/mtime in shrink case below
5229 inode
->i_mtime
= current_time(inode
);
5230 inode
->i_ctime
= inode
->i_mtime
;
5232 down_write(&EXT4_I(inode
)->i_data_sem
);
5233 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5234 rc
= ext4_mark_inode_dirty(handle
, inode
);
5238 * We have to update i_size under i_data_sem together
5239 * with i_disksize to avoid races with writeback code
5240 * running ext4_wb_update_i_disksize().
5243 i_size_write(inode
, attr
->ia_size
);
5244 up_write(&EXT4_I(inode
)->i_data_sem
);
5245 ext4_journal_stop(handle
);
5248 ext4_orphan_del(NULL
, inode
);
5253 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5256 * Blocks are going to be removed from the inode. Wait
5257 * for dio in flight. Temporarily disable
5258 * dioread_nolock to prevent livelock.
5261 if (!ext4_should_journal_data(inode
)) {
5262 ext4_inode_block_unlocked_dio(inode
);
5263 inode_dio_wait(inode
);
5264 ext4_inode_resume_unlocked_dio(inode
);
5266 ext4_wait_for_tail_page_commit(inode
);
5268 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5270 * Truncate pagecache after we've waited for commit
5271 * in data=journal mode to make pages freeable.
5273 truncate_pagecache(inode
, inode
->i_size
);
5275 rc
= ext4_truncate(inode
);
5279 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5283 setattr_copy(inode
, attr
);
5284 mark_inode_dirty(inode
);
5288 * If the call to ext4_truncate failed to get a transaction handle at
5289 * all, we need to clean up the in-core orphan list manually.
5291 if (orphan
&& inode
->i_nlink
)
5292 ext4_orphan_del(NULL
, inode
);
5294 if (!error
&& (ia_valid
& ATTR_MODE
))
5295 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5298 ext4_std_error(inode
->i_sb
, error
);
5304 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5307 struct inode
*inode
;
5308 unsigned long long delalloc_blocks
;
5310 inode
= d_inode(dentry
);
5311 generic_fillattr(inode
, stat
);
5314 * If there is inline data in the inode, the inode will normally not
5315 * have data blocks allocated (it may have an external xattr block).
5316 * Report at least one sector for such files, so tools like tar, rsync,
5317 * others doen't incorrectly think the file is completely sparse.
5319 if (unlikely(ext4_has_inline_data(inode
)))
5320 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5323 * We can't update i_blocks if the block allocation is delayed
5324 * otherwise in the case of system crash before the real block
5325 * allocation is done, we will have i_blocks inconsistent with
5326 * on-disk file blocks.
5327 * We always keep i_blocks updated together with real
5328 * allocation. But to not confuse with user, stat
5329 * will return the blocks that include the delayed allocation
5330 * blocks for this file.
5332 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5333 EXT4_I(inode
)->i_reserved_data_blocks
);
5334 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5338 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5341 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5342 return ext4_ind_trans_blocks(inode
, lblocks
);
5343 return ext4_ext_index_trans_blocks(inode
, pextents
);
5347 * Account for index blocks, block groups bitmaps and block group
5348 * descriptor blocks if modify datablocks and index blocks
5349 * worse case, the indexs blocks spread over different block groups
5351 * If datablocks are discontiguous, they are possible to spread over
5352 * different block groups too. If they are contiguous, with flexbg,
5353 * they could still across block group boundary.
5355 * Also account for superblock, inode, quota and xattr blocks
5357 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5360 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5366 * How many index blocks need to touch to map @lblocks logical blocks
5367 * to @pextents physical extents?
5369 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5374 * Now let's see how many group bitmaps and group descriptors need
5377 groups
= idxblocks
+ pextents
;
5379 if (groups
> ngroups
)
5381 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5382 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5384 /* bitmaps and block group descriptor blocks */
5385 ret
+= groups
+ gdpblocks
;
5387 /* Blocks for super block, inode, quota and xattr blocks */
5388 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5394 * Calculate the total number of credits to reserve to fit
5395 * the modification of a single pages into a single transaction,
5396 * which may include multiple chunks of block allocations.
5398 * This could be called via ext4_write_begin()
5400 * We need to consider the worse case, when
5401 * one new block per extent.
5403 int ext4_writepage_trans_blocks(struct inode
*inode
)
5405 int bpp
= ext4_journal_blocks_per_page(inode
);
5408 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5410 /* Account for data blocks for journalled mode */
5411 if (ext4_should_journal_data(inode
))
5417 * Calculate the journal credits for a chunk of data modification.
5419 * This is called from DIO, fallocate or whoever calling
5420 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5422 * journal buffers for data blocks are not included here, as DIO
5423 * and fallocate do no need to journal data buffers.
5425 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5427 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5431 * The caller must have previously called ext4_reserve_inode_write().
5432 * Give this, we know that the caller already has write access to iloc->bh.
5434 int ext4_mark_iloc_dirty(handle_t
*handle
,
5435 struct inode
*inode
, struct ext4_iloc
*iloc
)
5439 if (IS_I_VERSION(inode
))
5440 inode_inc_iversion(inode
);
5442 /* the do_update_inode consumes one bh->b_count */
5445 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5446 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5452 * On success, We end up with an outstanding reference count against
5453 * iloc->bh. This _must_ be cleaned up later.
5457 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5458 struct ext4_iloc
*iloc
)
5462 err
= ext4_get_inode_loc(inode
, iloc
);
5464 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5465 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5471 ext4_std_error(inode
->i_sb
, err
);
5476 * Expand an inode by new_extra_isize bytes.
5477 * Returns 0 on success or negative error number on failure.
5479 static int ext4_expand_extra_isize(struct inode
*inode
,
5480 unsigned int new_extra_isize
,
5481 struct ext4_iloc iloc
,
5484 struct ext4_inode
*raw_inode
;
5485 struct ext4_xattr_ibody_header
*header
;
5487 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5490 raw_inode
= ext4_raw_inode(&iloc
);
5492 header
= IHDR(inode
, raw_inode
);
5494 /* No extended attributes present */
5495 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5496 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5497 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5499 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5503 /* try to expand with EAs present */
5504 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5509 * What we do here is to mark the in-core inode as clean with respect to inode
5510 * dirtiness (it may still be data-dirty).
5511 * This means that the in-core inode may be reaped by prune_icache
5512 * without having to perform any I/O. This is a very good thing,
5513 * because *any* task may call prune_icache - even ones which
5514 * have a transaction open against a different journal.
5516 * Is this cheating? Not really. Sure, we haven't written the
5517 * inode out, but prune_icache isn't a user-visible syncing function.
5518 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5519 * we start and wait on commits.
5521 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5523 struct ext4_iloc iloc
;
5524 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5525 static unsigned int mnt_count
;
5529 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5530 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5533 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5534 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5536 * In nojournal mode, we can immediately attempt to expand
5537 * the inode. When journaled, we first need to obtain extra
5538 * buffer credits since we may write into the EA block
5539 * with this same handle. If journal_extend fails, then it will
5540 * only result in a minor loss of functionality for that inode.
5541 * If this is felt to be critical, then e2fsck should be run to
5542 * force a large enough s_min_extra_isize.
5544 if (!ext4_handle_valid(handle
) ||
5545 jbd2_journal_extend(handle
,
5546 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) == 0) {
5547 ret
= ext4_expand_extra_isize(inode
,
5548 sbi
->s_want_extra_isize
,
5552 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5553 ext4_warning(inode
->i_sb
,
5554 "Unable to expand inode %lu. Delete"
5555 " some EAs or run e2fsck.",
5558 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5563 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5567 * ext4_dirty_inode() is called from __mark_inode_dirty()
5569 * We're really interested in the case where a file is being extended.
5570 * i_size has been changed by generic_commit_write() and we thus need
5571 * to include the updated inode in the current transaction.
5573 * Also, dquot_alloc_block() will always dirty the inode when blocks
5574 * are allocated to the file.
5576 * If the inode is marked synchronous, we don't honour that here - doing
5577 * so would cause a commit on atime updates, which we don't bother doing.
5578 * We handle synchronous inodes at the highest possible level.
5580 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5581 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5582 * to copy into the on-disk inode structure are the timestamp files.
5584 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5588 if (flags
== I_DIRTY_TIME
)
5590 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5594 ext4_mark_inode_dirty(handle
, inode
);
5596 ext4_journal_stop(handle
);
5603 * Bind an inode's backing buffer_head into this transaction, to prevent
5604 * it from being flushed to disk early. Unlike
5605 * ext4_reserve_inode_write, this leaves behind no bh reference and
5606 * returns no iloc structure, so the caller needs to repeat the iloc
5607 * lookup to mark the inode dirty later.
5609 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5611 struct ext4_iloc iloc
;
5615 err
= ext4_get_inode_loc(inode
, &iloc
);
5617 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5618 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5620 err
= ext4_handle_dirty_metadata(handle
,
5626 ext4_std_error(inode
->i_sb
, err
);
5631 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5636 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5639 * We have to be very careful here: changing a data block's
5640 * journaling status dynamically is dangerous. If we write a
5641 * data block to the journal, change the status and then delete
5642 * that block, we risk forgetting to revoke the old log record
5643 * from the journal and so a subsequent replay can corrupt data.
5644 * So, first we make sure that the journal is empty and that
5645 * nobody is changing anything.
5648 journal
= EXT4_JOURNAL(inode
);
5651 if (is_journal_aborted(journal
))
5654 /* Wait for all existing dio workers */
5655 ext4_inode_block_unlocked_dio(inode
);
5656 inode_dio_wait(inode
);
5659 * Before flushing the journal and switching inode's aops, we have
5660 * to flush all dirty data the inode has. There can be outstanding
5661 * delayed allocations, there can be unwritten extents created by
5662 * fallocate or buffered writes in dioread_nolock mode covered by
5663 * dirty data which can be converted only after flushing the dirty
5664 * data (and journalled aops don't know how to handle these cases).
5667 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5668 err
= filemap_write_and_wait(inode
->i_mapping
);
5670 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5671 ext4_inode_resume_unlocked_dio(inode
);
5676 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5677 jbd2_journal_lock_updates(journal
);
5680 * OK, there are no updates running now, and all cached data is
5681 * synced to disk. We are now in a completely consistent state
5682 * which doesn't have anything in the journal, and we know that
5683 * no filesystem updates are running, so it is safe to modify
5684 * the inode's in-core data-journaling state flag now.
5688 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5690 err
= jbd2_journal_flush(journal
);
5692 jbd2_journal_unlock_updates(journal
);
5693 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5694 ext4_inode_resume_unlocked_dio(inode
);
5697 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5699 ext4_set_aops(inode
);
5701 * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated.
5702 * E.g. S_DAX may get cleared / set.
5704 ext4_set_inode_flags(inode
);
5706 jbd2_journal_unlock_updates(journal
);
5707 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5710 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5711 ext4_inode_resume_unlocked_dio(inode
);
5713 /* Finally we can mark the inode as dirty. */
5715 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5717 return PTR_ERR(handle
);
5719 err
= ext4_mark_inode_dirty(handle
, inode
);
5720 ext4_handle_sync(handle
);
5721 ext4_journal_stop(handle
);
5722 ext4_std_error(inode
->i_sb
, err
);
5727 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5729 return !buffer_mapped(bh
);
5732 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5734 struct page
*page
= vmf
->page
;
5738 struct file
*file
= vma
->vm_file
;
5739 struct inode
*inode
= file_inode(file
);
5740 struct address_space
*mapping
= inode
->i_mapping
;
5742 get_block_t
*get_block
;
5745 sb_start_pagefault(inode
->i_sb
);
5746 file_update_time(vma
->vm_file
);
5748 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5749 /* Delalloc case is easy... */
5750 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5751 !ext4_should_journal_data(inode
) &&
5752 !ext4_nonda_switch(inode
->i_sb
)) {
5754 ret
= block_page_mkwrite(vma
, vmf
,
5755 ext4_da_get_block_prep
);
5756 } while (ret
== -ENOSPC
&&
5757 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5762 size
= i_size_read(inode
);
5763 /* Page got truncated from under us? */
5764 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5766 ret
= VM_FAULT_NOPAGE
;
5770 if (page
->index
== size
>> PAGE_SHIFT
)
5771 len
= size
& ~PAGE_MASK
;
5775 * Return if we have all the buffers mapped. This avoids the need to do
5776 * journal_start/journal_stop which can block and take a long time
5778 if (page_has_buffers(page
)) {
5779 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5781 ext4_bh_unmapped
)) {
5782 /* Wait so that we don't change page under IO */
5783 wait_for_stable_page(page
);
5784 ret
= VM_FAULT_LOCKED
;
5789 /* OK, we need to fill the hole... */
5790 if (ext4_should_dioread_nolock(inode
))
5791 get_block
= ext4_get_block_unwritten
;
5793 get_block
= ext4_get_block
;
5795 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5796 ext4_writepage_trans_blocks(inode
));
5797 if (IS_ERR(handle
)) {
5798 ret
= VM_FAULT_SIGBUS
;
5801 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5802 if (!ret
&& ext4_should_journal_data(inode
)) {
5803 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5804 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
5806 ret
= VM_FAULT_SIGBUS
;
5807 ext4_journal_stop(handle
);
5810 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5812 ext4_journal_stop(handle
);
5813 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5816 ret
= block_page_mkwrite_return(ret
);
5818 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5819 sb_end_pagefault(inode
->i_sb
);
5823 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5825 struct inode
*inode
= file_inode(vma
->vm_file
);
5828 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5829 err
= filemap_fault(vma
, vmf
);
5830 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5836 * Find the first extent at or after @lblk in an inode that is not a hole.
5837 * Search for @map_len blocks at most. The extent is returned in @result.
5839 * The function returns 1 if we found an extent. The function returns 0 in
5840 * case there is no extent at or after @lblk and in that case also sets
5841 * @result->es_len to 0. In case of error, the error code is returned.
5843 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
5844 unsigned int map_len
, struct extent_status
*result
)
5846 struct ext4_map_blocks map
;
5847 struct extent_status es
= {};
5851 map
.m_len
= map_len
;
5854 * For non-extent based files this loop may iterate several times since
5855 * we do not determine full hole size.
5857 while (map
.m_len
> 0) {
5858 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
5861 /* There's extent covering m_lblk? Just return it. */
5865 ext4_es_store_pblock(result
, map
.m_pblk
);
5866 result
->es_lblk
= map
.m_lblk
;
5867 result
->es_len
= map
.m_len
;
5868 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
5869 status
= EXTENT_STATUS_UNWRITTEN
;
5871 status
= EXTENT_STATUS_WRITTEN
;
5872 ext4_es_store_status(result
, status
);
5875 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
5876 map
.m_lblk
+ map
.m_len
- 1,
5878 /* Is delalloc data before next block in extent tree? */
5879 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
5880 ext4_lblk_t offset
= 0;
5882 if (es
.es_lblk
< lblk
)
5883 offset
= lblk
- es
.es_lblk
;
5884 result
->es_lblk
= es
.es_lblk
+ offset
;
5885 ext4_es_store_pblock(result
,
5886 ext4_es_pblock(&es
) + offset
);
5887 result
->es_len
= es
.es_len
- offset
;
5888 ext4_es_store_status(result
, ext4_es_status(&es
));
5892 /* There's a hole at m_lblk, advance us after it */
5893 map
.m_lblk
+= map
.m_len
;
5894 map_len
-= map
.m_len
;
5895 map
.m_len
= map_len
;