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
,
76 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
77 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
83 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
84 struct ext4_inode_info
*ei
)
86 __u32 provided
, calculated
;
88 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
89 cpu_to_le32(EXT4_OS_LINUX
) ||
90 !ext4_has_metadata_csum(inode
->i_sb
))
93 provided
= le16_to_cpu(raw
->i_checksum_lo
);
94 calculated
= ext4_inode_csum(inode
, raw
, ei
);
95 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
96 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
97 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
101 return provided
== calculated
;
104 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
105 struct ext4_inode_info
*ei
)
109 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
110 cpu_to_le32(EXT4_OS_LINUX
) ||
111 !ext4_has_metadata_csum(inode
->i_sb
))
114 csum
= ext4_inode_csum(inode
, raw
, ei
);
115 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
116 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
117 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
118 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
124 trace_ext4_begin_ordered_truncate(inode
, new_size
);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode
)->jinode
)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
134 EXT4_I(inode
)->jinode
,
138 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
139 unsigned int length
);
140 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
141 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
142 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
146 * Test whether an inode is a fast symlink.
148 int ext4_inode_is_fast_symlink(struct inode
*inode
)
150 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
151 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
153 if (ext4_has_inline_data(inode
))
156 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
160 * Restart the transaction associated with *handle. This does a commit,
161 * so before we call here everything must be consistently dirtied against
164 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
170 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
171 * moment, get_block can be called only for blocks inside i_size since
172 * page cache has been already dropped and writes are blocked by
173 * i_mutex. So we can safely drop the i_data_sem here.
175 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
176 jbd_debug(2, "restarting handle %p\n", handle
);
177 up_write(&EXT4_I(inode
)->i_data_sem
);
178 ret
= ext4_journal_restart(handle
, nblocks
);
179 down_write(&EXT4_I(inode
)->i_data_sem
);
180 ext4_discard_preallocations(inode
);
186 * Called at the last iput() if i_nlink is zero.
188 void ext4_evict_inode(struct inode
*inode
)
193 trace_ext4_evict_inode(inode
);
195 if (inode
->i_nlink
) {
197 * When journalling data dirty buffers are tracked only in the
198 * journal. So although mm thinks everything is clean and
199 * ready for reaping the inode might still have some pages to
200 * write in the running transaction or waiting to be
201 * checkpointed. Thus calling jbd2_journal_invalidatepage()
202 * (via truncate_inode_pages()) to discard these buffers can
203 * cause data loss. Also even if we did not discard these
204 * buffers, we would have no way to find them after the inode
205 * is reaped and thus user could see stale data if he tries to
206 * read them before the transaction is checkpointed. So be
207 * careful and force everything to disk here... We use
208 * ei->i_datasync_tid to store the newest transaction
209 * containing inode's data.
211 * Note that directories do not have this problem because they
212 * don't use page cache.
214 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
215 ext4_should_journal_data(inode
) &&
216 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
217 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
218 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
220 jbd2_complete_transaction(journal
, commit_tid
);
221 filemap_write_and_wait(&inode
->i_data
);
223 truncate_inode_pages_final(&inode
->i_data
);
228 if (is_bad_inode(inode
))
230 dquot_initialize(inode
);
232 if (ext4_should_order_data(inode
))
233 ext4_begin_ordered_truncate(inode
, 0);
234 truncate_inode_pages_final(&inode
->i_data
);
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode
->i_sb
);
241 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
242 ext4_blocks_for_truncate(inode
)+3);
243 if (IS_ERR(handle
)) {
244 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
246 * If we're going to skip the normal cleanup, we still need to
247 * make sure that the in-core orphan linked list is properly
250 ext4_orphan_del(NULL
, inode
);
251 sb_end_intwrite(inode
->i_sb
);
256 ext4_handle_sync(handle
);
258 err
= ext4_mark_inode_dirty(handle
, inode
);
260 ext4_warning(inode
->i_sb
,
261 "couldn't mark inode dirty (err %d)", err
);
264 if (inode
->i_blocks
) {
265 err
= ext4_truncate(inode
);
267 ext4_error(inode
->i_sb
,
268 "couldn't truncate inode %lu (err %d)",
275 * ext4_ext_truncate() doesn't reserve any slop when it
276 * restarts journal transactions; therefore there may not be
277 * enough credits left in the handle to remove the inode from
278 * the orphan list and set the dtime field.
280 if (!ext4_handle_has_enough_credits(handle
, 3)) {
281 err
= ext4_journal_extend(handle
, 3);
283 err
= ext4_journal_restart(handle
, 3);
285 ext4_warning(inode
->i_sb
,
286 "couldn't extend journal (err %d)", err
);
288 ext4_journal_stop(handle
);
289 ext4_orphan_del(NULL
, inode
);
290 sb_end_intwrite(inode
->i_sb
);
296 * Kill off the orphan record which ext4_truncate created.
297 * AKPM: I think this can be inside the above `if'.
298 * Note that ext4_orphan_del() has to be able to cope with the
299 * deletion of a non-existent orphan - this is because we don't
300 * know if ext4_truncate() actually created an orphan record.
301 * (Well, we could do this if we need to, but heck - it works)
303 ext4_orphan_del(handle
, inode
);
304 EXT4_I(inode
)->i_dtime
= get_seconds();
307 * One subtle ordering requirement: if anything has gone wrong
308 * (transaction abort, IO errors, whatever), then we can still
309 * do these next steps (the fs will already have been marked as
310 * having errors), but we can't free the inode if the mark_dirty
313 if (ext4_mark_inode_dirty(handle
, inode
))
314 /* If that failed, just do the required in-core inode clear. */
315 ext4_clear_inode(inode
);
317 ext4_free_inode(handle
, inode
);
318 ext4_journal_stop(handle
);
319 sb_end_intwrite(inode
->i_sb
);
322 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
326 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
328 return &EXT4_I(inode
)->i_reserved_quota
;
333 * Called with i_data_sem down, which is important since we can call
334 * ext4_discard_preallocations() from here.
336 void ext4_da_update_reserve_space(struct inode
*inode
,
337 int used
, int quota_claim
)
339 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
340 struct ext4_inode_info
*ei
= EXT4_I(inode
);
342 spin_lock(&ei
->i_block_reservation_lock
);
343 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
344 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
345 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
346 "with only %d reserved data blocks",
347 __func__
, inode
->i_ino
, used
,
348 ei
->i_reserved_data_blocks
);
350 used
= ei
->i_reserved_data_blocks
;
353 /* Update per-inode reservations */
354 ei
->i_reserved_data_blocks
-= used
;
355 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
357 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
359 /* Update quota subsystem for data blocks */
361 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
364 * We did fallocate with an offset that is already delayed
365 * allocated. So on delayed allocated writeback we should
366 * not re-claim the quota for fallocated blocks.
368 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
372 * If we have done all the pending block allocations and if
373 * there aren't any writers on the inode, we can discard the
374 * inode's preallocations.
376 if ((ei
->i_reserved_data_blocks
== 0) &&
377 (atomic_read(&inode
->i_writecount
) == 0))
378 ext4_discard_preallocations(inode
);
381 static int __check_block_validity(struct inode
*inode
, const char *func
,
383 struct ext4_map_blocks
*map
)
385 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
387 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
388 "lblock %lu mapped to illegal pblock "
389 "(length %d)", (unsigned long) map
->m_lblk
,
391 return -EFSCORRUPTED
;
396 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
401 if (ext4_encrypted_inode(inode
))
402 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
404 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
411 #define check_block_validity(inode, map) \
412 __check_block_validity((inode), __func__, __LINE__, (map))
414 #ifdef ES_AGGRESSIVE_TEST
415 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
417 struct ext4_map_blocks
*es_map
,
418 struct ext4_map_blocks
*map
,
425 * There is a race window that the result is not the same.
426 * e.g. xfstests #223 when dioread_nolock enables. The reason
427 * is that we lookup a block mapping in extent status tree with
428 * out taking i_data_sem. So at the time the unwritten extent
429 * could be converted.
431 down_read(&EXT4_I(inode
)->i_data_sem
);
432 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
433 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
434 EXT4_GET_BLOCKS_KEEP_SIZE
);
436 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
437 EXT4_GET_BLOCKS_KEEP_SIZE
);
439 up_read((&EXT4_I(inode
)->i_data_sem
));
442 * We don't check m_len because extent will be collpased in status
443 * tree. So the m_len might not equal.
445 if (es_map
->m_lblk
!= map
->m_lblk
||
446 es_map
->m_flags
!= map
->m_flags
||
447 es_map
->m_pblk
!= map
->m_pblk
) {
448 printk("ES cache assertion failed for inode: %lu "
449 "es_cached ex [%d/%d/%llu/%x] != "
450 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
451 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
452 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
453 map
->m_len
, map
->m_pblk
, map
->m_flags
,
457 #endif /* ES_AGGRESSIVE_TEST */
460 * The ext4_map_blocks() function tries to look up the requested blocks,
461 * and returns if the blocks are already mapped.
463 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
464 * and store the allocated blocks in the result buffer head and mark it
467 * If file type is extents based, it will call ext4_ext_map_blocks(),
468 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
471 * On success, it returns the number of blocks being mapped or allocated. if
472 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
473 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
475 * It returns 0 if plain look up failed (blocks have not been allocated), in
476 * that case, @map is returned as unmapped but we still do fill map->m_len to
477 * indicate the length of a hole starting at map->m_lblk.
479 * It returns the error in case of allocation failure.
481 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
482 struct ext4_map_blocks
*map
, int flags
)
484 struct extent_status es
;
487 #ifdef ES_AGGRESSIVE_TEST
488 struct ext4_map_blocks orig_map
;
490 memcpy(&orig_map
, map
, sizeof(*map
));
494 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
495 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
496 (unsigned long) map
->m_lblk
);
499 * ext4_map_blocks returns an int, and m_len is an unsigned int
501 if (unlikely(map
->m_len
> INT_MAX
))
502 map
->m_len
= INT_MAX
;
504 /* We can handle the block number less than EXT_MAX_BLOCKS */
505 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
506 return -EFSCORRUPTED
;
508 /* Lookup extent status tree firstly */
509 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
510 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
511 map
->m_pblk
= ext4_es_pblock(&es
) +
512 map
->m_lblk
- es
.es_lblk
;
513 map
->m_flags
|= ext4_es_is_written(&es
) ?
514 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
515 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
516 if (retval
> map
->m_len
)
519 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
521 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
522 if (retval
> map
->m_len
)
529 #ifdef ES_AGGRESSIVE_TEST
530 ext4_map_blocks_es_recheck(handle
, inode
, map
,
537 * Try to see if we can get the block without requesting a new
540 down_read(&EXT4_I(inode
)->i_data_sem
);
541 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
542 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
543 EXT4_GET_BLOCKS_KEEP_SIZE
);
545 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
546 EXT4_GET_BLOCKS_KEEP_SIZE
);
551 if (unlikely(retval
!= map
->m_len
)) {
552 ext4_warning(inode
->i_sb
,
553 "ES len assertion failed for inode "
554 "%lu: retval %d != map->m_len %d",
555 inode
->i_ino
, retval
, map
->m_len
);
559 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
560 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
561 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
562 !(status
& EXTENT_STATUS_WRITTEN
) &&
563 ext4_find_delalloc_range(inode
, map
->m_lblk
,
564 map
->m_lblk
+ map
->m_len
- 1))
565 status
|= EXTENT_STATUS_DELAYED
;
566 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
567 map
->m_len
, map
->m_pblk
, status
);
571 up_read((&EXT4_I(inode
)->i_data_sem
));
574 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
575 ret
= check_block_validity(inode
, map
);
580 /* If it is only a block(s) look up */
581 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
585 * Returns if the blocks have already allocated
587 * Note that if blocks have been preallocated
588 * ext4_ext_get_block() returns the create = 0
589 * with buffer head unmapped.
591 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
593 * If we need to convert extent to unwritten
594 * we continue and do the actual work in
595 * ext4_ext_map_blocks()
597 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
601 * Here we clear m_flags because after allocating an new extent,
602 * it will be set again.
604 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
607 * New blocks allocate and/or writing to unwritten extent
608 * will possibly result in updating i_data, so we take
609 * the write lock of i_data_sem, and call get_block()
610 * with create == 1 flag.
612 down_write(&EXT4_I(inode
)->i_data_sem
);
615 * We need to check for EXT4 here because migrate
616 * could have changed the inode type in between
618 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
619 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
621 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
623 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
625 * We allocated new blocks which will result in
626 * i_data's format changing. Force the migrate
627 * to fail by clearing migrate flags
629 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
633 * Update reserved blocks/metadata blocks after successful
634 * block allocation which had been deferred till now. We don't
635 * support fallocate for non extent files. So we can update
636 * reserve space here.
639 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
640 ext4_da_update_reserve_space(inode
, retval
, 1);
646 if (unlikely(retval
!= map
->m_len
)) {
647 ext4_warning(inode
->i_sb
,
648 "ES len assertion failed for inode "
649 "%lu: retval %d != map->m_len %d",
650 inode
->i_ino
, retval
, map
->m_len
);
655 * We have to zeroout blocks before inserting them into extent
656 * status tree. Otherwise someone could look them up there and
657 * use them before they are really zeroed. We also have to
658 * unmap metadata before zeroing as otherwise writeback can
659 * overwrite zeros with stale data from block device.
661 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
662 map
->m_flags
& EXT4_MAP_MAPPED
&&
663 map
->m_flags
& EXT4_MAP_NEW
) {
664 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
666 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
667 map
->m_pblk
, map
->m_len
);
675 * If the extent has been zeroed out, we don't need to update
676 * extent status tree.
678 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
679 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
680 if (ext4_es_is_written(&es
))
683 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
684 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
685 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
686 !(status
& EXTENT_STATUS_WRITTEN
) &&
687 ext4_find_delalloc_range(inode
, map
->m_lblk
,
688 map
->m_lblk
+ map
->m_len
- 1))
689 status
|= EXTENT_STATUS_DELAYED
;
690 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
691 map
->m_pblk
, status
);
699 up_write((&EXT4_I(inode
)->i_data_sem
));
700 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
701 ret
= check_block_validity(inode
, map
);
706 * Inodes with freshly allocated blocks where contents will be
707 * visible after transaction commit must be on transaction's
710 if (map
->m_flags
& EXT4_MAP_NEW
&&
711 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
712 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
713 !IS_NOQUOTA(inode
) &&
714 ext4_should_order_data(inode
)) {
715 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
716 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
718 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
727 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
728 * we have to be careful as someone else may be manipulating b_state as well.
730 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
732 unsigned long old_state
;
733 unsigned long new_state
;
735 flags
&= EXT4_MAP_FLAGS
;
737 /* Dummy buffer_head? Set non-atomically. */
739 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
743 * Someone else may be modifying b_state. Be careful! This is ugly but
744 * once we get rid of using bh as a container for mapping information
745 * to pass to / from get_block functions, this can go away.
748 old_state
= READ_ONCE(bh
->b_state
);
749 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
751 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
754 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
755 struct buffer_head
*bh
, int flags
)
757 struct ext4_map_blocks map
;
760 if (ext4_has_inline_data(inode
))
764 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
766 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
769 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
770 ext4_update_bh_state(bh
, map
.m_flags
);
771 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
773 } else if (ret
== 0) {
774 /* hole case, need to fill in bh->b_size */
775 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
780 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
781 struct buffer_head
*bh
, int create
)
783 return _ext4_get_block(inode
, iblock
, bh
,
784 create
? EXT4_GET_BLOCKS_CREATE
: 0);
788 * Get block function used when preparing for buffered write if we require
789 * creating an unwritten extent if blocks haven't been allocated. The extent
790 * will be converted to written after the IO is complete.
792 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
793 struct buffer_head
*bh_result
, int create
)
795 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
796 inode
->i_ino
, create
);
797 return _ext4_get_block(inode
, iblock
, bh_result
,
798 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
801 /* Maximum number of blocks we map for direct IO at once. */
802 #define DIO_MAX_BLOCKS 4096
805 * Get blocks function for the cases that need to start a transaction -
806 * generally difference cases of direct IO and DAX IO. It also handles retries
809 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
810 struct buffer_head
*bh_result
, int flags
)
817 /* Trim mapping request to maximum we can map at once for DIO */
818 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
819 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
820 dio_credits
= ext4_chunk_trans_blocks(inode
,
821 bh_result
->b_size
>> inode
->i_blkbits
);
823 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
825 return PTR_ERR(handle
);
827 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
828 ext4_journal_stop(handle
);
830 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
835 /* Get block function for DIO reads and writes to inodes without extents */
836 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
837 struct buffer_head
*bh
, int create
)
839 /* We don't expect handle for direct IO */
840 WARN_ON_ONCE(ext4_journal_current_handle());
843 return _ext4_get_block(inode
, iblock
, bh
, 0);
844 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
848 * Get block function for AIO DIO writes when we create unwritten extent if
849 * blocks are not allocated yet. The extent will be converted to written
850 * after IO is complete.
852 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
853 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
857 /* We don't expect handle for direct IO */
858 WARN_ON_ONCE(ext4_journal_current_handle());
860 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
861 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
864 * When doing DIO using unwritten extents, we need io_end to convert
865 * unwritten extents to written on IO completion. We allocate io_end
866 * once we spot unwritten extent and store it in b_private. Generic
867 * DIO code keeps b_private set and furthermore passes the value to
868 * our completion callback in 'private' argument.
870 if (!ret
&& buffer_unwritten(bh_result
)) {
871 if (!bh_result
->b_private
) {
872 ext4_io_end_t
*io_end
;
874 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
877 bh_result
->b_private
= io_end
;
878 ext4_set_io_unwritten_flag(inode
, io_end
);
880 set_buffer_defer_completion(bh_result
);
887 * Get block function for non-AIO DIO writes when we create unwritten extent if
888 * blocks are not allocated yet. The extent will be converted to written
889 * after IO is complete from ext4_ext_direct_IO() function.
891 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
892 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
896 /* We don't expect handle for direct IO */
897 WARN_ON_ONCE(ext4_journal_current_handle());
899 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
900 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
903 * Mark inode as having pending DIO writes to unwritten extents.
904 * ext4_ext_direct_IO() checks this flag and converts extents to
907 if (!ret
&& buffer_unwritten(bh_result
))
908 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
913 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
914 struct buffer_head
*bh_result
, int create
)
918 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
919 inode
->i_ino
, create
);
920 /* We don't expect handle for direct IO */
921 WARN_ON_ONCE(ext4_journal_current_handle());
923 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
925 * Blocks should have been preallocated! ext4_file_write_iter() checks
928 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
935 * `handle' can be NULL if create is zero
937 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
938 ext4_lblk_t block
, int map_flags
)
940 struct ext4_map_blocks map
;
941 struct buffer_head
*bh
;
942 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
945 J_ASSERT(handle
!= NULL
|| create
== 0);
949 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
952 return create
? ERR_PTR(-ENOSPC
) : NULL
;
956 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
958 return ERR_PTR(-ENOMEM
);
959 if (map
.m_flags
& EXT4_MAP_NEW
) {
960 J_ASSERT(create
!= 0);
961 J_ASSERT(handle
!= NULL
);
964 * Now that we do not always journal data, we should
965 * keep in mind whether this should always journal the
966 * new buffer as metadata. For now, regular file
967 * writes use ext4_get_block instead, so it's not a
971 BUFFER_TRACE(bh
, "call get_create_access");
972 err
= ext4_journal_get_create_access(handle
, bh
);
977 if (!buffer_uptodate(bh
)) {
978 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
979 set_buffer_uptodate(bh
);
982 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
983 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
987 BUFFER_TRACE(bh
, "not a new buffer");
994 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
995 ext4_lblk_t block
, int map_flags
)
997 struct buffer_head
*bh
;
999 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1002 if (!bh
|| buffer_uptodate(bh
))
1004 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1006 if (buffer_uptodate(bh
))
1009 return ERR_PTR(-EIO
);
1012 int ext4_walk_page_buffers(handle_t
*handle
,
1013 struct buffer_head
*head
,
1017 int (*fn
)(handle_t
*handle
,
1018 struct buffer_head
*bh
))
1020 struct buffer_head
*bh
;
1021 unsigned block_start
, block_end
;
1022 unsigned blocksize
= head
->b_size
;
1024 struct buffer_head
*next
;
1026 for (bh
= head
, block_start
= 0;
1027 ret
== 0 && (bh
!= head
|| !block_start
);
1028 block_start
= block_end
, bh
= next
) {
1029 next
= bh
->b_this_page
;
1030 block_end
= block_start
+ blocksize
;
1031 if (block_end
<= from
|| block_start
>= to
) {
1032 if (partial
&& !buffer_uptodate(bh
))
1036 err
= (*fn
)(handle
, bh
);
1044 * To preserve ordering, it is essential that the hole instantiation and
1045 * the data write be encapsulated in a single transaction. We cannot
1046 * close off a transaction and start a new one between the ext4_get_block()
1047 * and the commit_write(). So doing the jbd2_journal_start at the start of
1048 * prepare_write() is the right place.
1050 * Also, this function can nest inside ext4_writepage(). In that case, we
1051 * *know* that ext4_writepage() has generated enough buffer credits to do the
1052 * whole page. So we won't block on the journal in that case, which is good,
1053 * because the caller may be PF_MEMALLOC.
1055 * By accident, ext4 can be reentered when a transaction is open via
1056 * quota file writes. If we were to commit the transaction while thus
1057 * reentered, there can be a deadlock - we would be holding a quota
1058 * lock, and the commit would never complete if another thread had a
1059 * transaction open and was blocking on the quota lock - a ranking
1062 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1063 * will _not_ run commit under these circumstances because handle->h_ref
1064 * is elevated. We'll still have enough credits for the tiny quotafile
1067 int do_journal_get_write_access(handle_t
*handle
,
1068 struct buffer_head
*bh
)
1070 int dirty
= buffer_dirty(bh
);
1073 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1076 * __block_write_begin() could have dirtied some buffers. Clean
1077 * the dirty bit as jbd2_journal_get_write_access() could complain
1078 * otherwise about fs integrity issues. Setting of the dirty bit
1079 * by __block_write_begin() isn't a real problem here as we clear
1080 * the bit before releasing a page lock and thus writeback cannot
1081 * ever write the buffer.
1084 clear_buffer_dirty(bh
);
1085 BUFFER_TRACE(bh
, "get write access");
1086 ret
= ext4_journal_get_write_access(handle
, bh
);
1088 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1092 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1093 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1094 get_block_t
*get_block
)
1096 unsigned from
= pos
& (PAGE_SIZE
- 1);
1097 unsigned to
= from
+ len
;
1098 struct inode
*inode
= page
->mapping
->host
;
1099 unsigned block_start
, block_end
;
1102 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1104 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1105 bool decrypt
= false;
1107 BUG_ON(!PageLocked(page
));
1108 BUG_ON(from
> PAGE_SIZE
);
1109 BUG_ON(to
> PAGE_SIZE
);
1112 if (!page_has_buffers(page
))
1113 create_empty_buffers(page
, blocksize
, 0);
1114 head
= page_buffers(page
);
1115 bbits
= ilog2(blocksize
);
1116 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1118 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1119 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1120 block_end
= block_start
+ blocksize
;
1121 if (block_end
<= from
|| block_start
>= to
) {
1122 if (PageUptodate(page
)) {
1123 if (!buffer_uptodate(bh
))
1124 set_buffer_uptodate(bh
);
1129 clear_buffer_new(bh
);
1130 if (!buffer_mapped(bh
)) {
1131 WARN_ON(bh
->b_size
!= blocksize
);
1132 err
= get_block(inode
, block
, bh
, 1);
1135 if (buffer_new(bh
)) {
1136 clean_bdev_bh_alias(bh
);
1137 if (PageUptodate(page
)) {
1138 clear_buffer_new(bh
);
1139 set_buffer_uptodate(bh
);
1140 mark_buffer_dirty(bh
);
1143 if (block_end
> to
|| block_start
< from
)
1144 zero_user_segments(page
, to
, block_end
,
1149 if (PageUptodate(page
)) {
1150 if (!buffer_uptodate(bh
))
1151 set_buffer_uptodate(bh
);
1154 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1155 !buffer_unwritten(bh
) &&
1156 (block_start
< from
|| block_end
> to
)) {
1157 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1159 decrypt
= ext4_encrypted_inode(inode
) &&
1160 S_ISREG(inode
->i_mode
);
1164 * If we issued read requests, let them complete.
1166 while (wait_bh
> wait
) {
1167 wait_on_buffer(*--wait_bh
);
1168 if (!buffer_uptodate(*wait_bh
))
1172 page_zero_new_buffers(page
, from
, to
);
1174 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1175 PAGE_SIZE
, 0, page
->index
);
1180 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1181 loff_t pos
, unsigned len
, unsigned flags
,
1182 struct page
**pagep
, void **fsdata
)
1184 struct inode
*inode
= mapping
->host
;
1185 int ret
, needed_blocks
;
1192 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1194 * Reserve one block more for addition to orphan list in case
1195 * we allocate blocks but write fails for some reason
1197 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1198 index
= pos
>> PAGE_SHIFT
;
1199 from
= pos
& (PAGE_SIZE
- 1);
1202 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1203 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1212 * grab_cache_page_write_begin() can take a long time if the
1213 * system is thrashing due to memory pressure, or if the page
1214 * is being written back. So grab it first before we start
1215 * the transaction handle. This also allows us to allocate
1216 * the page (if needed) without using GFP_NOFS.
1219 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1225 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1226 if (IS_ERR(handle
)) {
1228 return PTR_ERR(handle
);
1232 if (page
->mapping
!= mapping
) {
1233 /* The page got truncated from under us */
1236 ext4_journal_stop(handle
);
1239 /* In case writeback began while the page was unlocked */
1240 wait_for_stable_page(page
);
1242 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1243 if (ext4_should_dioread_nolock(inode
))
1244 ret
= ext4_block_write_begin(page
, pos
, len
,
1245 ext4_get_block_unwritten
);
1247 ret
= ext4_block_write_begin(page
, pos
, len
,
1250 if (ext4_should_dioread_nolock(inode
))
1251 ret
= __block_write_begin(page
, pos
, len
,
1252 ext4_get_block_unwritten
);
1254 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1256 if (!ret
&& ext4_should_journal_data(inode
)) {
1257 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1259 do_journal_get_write_access
);
1265 * __block_write_begin may have instantiated a few blocks
1266 * outside i_size. Trim these off again. Don't need
1267 * i_size_read because we hold i_mutex.
1269 * Add inode to orphan list in case we crash before
1272 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1273 ext4_orphan_add(handle
, inode
);
1275 ext4_journal_stop(handle
);
1276 if (pos
+ len
> inode
->i_size
) {
1277 ext4_truncate_failed_write(inode
);
1279 * If truncate failed early the inode might
1280 * still be on the orphan list; we need to
1281 * make sure the inode is removed from the
1282 * orphan list in that case.
1285 ext4_orphan_del(NULL
, inode
);
1288 if (ret
== -ENOSPC
&&
1289 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1298 /* For write_end() in data=journal mode */
1299 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1302 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1304 set_buffer_uptodate(bh
);
1305 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1306 clear_buffer_meta(bh
);
1307 clear_buffer_prio(bh
);
1312 * We need to pick up the new inode size which generic_commit_write gave us
1313 * `file' can be NULL - eg, when called from page_symlink().
1315 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1316 * buffers are managed internally.
1318 static int ext4_write_end(struct file
*file
,
1319 struct address_space
*mapping
,
1320 loff_t pos
, unsigned len
, unsigned copied
,
1321 struct page
*page
, void *fsdata
)
1323 handle_t
*handle
= ext4_journal_current_handle();
1324 struct inode
*inode
= mapping
->host
;
1325 loff_t old_size
= inode
->i_size
;
1327 int i_size_changed
= 0;
1329 trace_ext4_write_end(inode
, pos
, len
, copied
);
1330 if (ext4_has_inline_data(inode
)) {
1331 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1340 copied
= block_write_end(file
, mapping
, pos
,
1341 len
, copied
, page
, fsdata
);
1343 * it's important to update i_size while still holding page lock:
1344 * page writeout could otherwise come in and zero beyond i_size.
1346 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1351 pagecache_isize_extended(inode
, old_size
, pos
);
1353 * Don't mark the inode dirty under page lock. First, it unnecessarily
1354 * makes the holding time of page lock longer. Second, it forces lock
1355 * ordering of page lock and transaction start for journaling
1359 ext4_mark_inode_dirty(handle
, inode
);
1361 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1362 /* if we have allocated more blocks and copied
1363 * less. We will have blocks allocated outside
1364 * inode->i_size. So truncate them
1366 ext4_orphan_add(handle
, inode
);
1368 ret2
= ext4_journal_stop(handle
);
1372 if (pos
+ len
> inode
->i_size
) {
1373 ext4_truncate_failed_write(inode
);
1375 * If truncate failed early the inode might still be
1376 * on the orphan list; we need to make sure the inode
1377 * is removed from the orphan list in that case.
1380 ext4_orphan_del(NULL
, inode
);
1383 return ret
? ret
: copied
;
1387 * This is a private version of page_zero_new_buffers() which doesn't
1388 * set the buffer to be dirty, since in data=journalled mode we need
1389 * to call ext4_handle_dirty_metadata() instead.
1391 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1393 unsigned from
, unsigned to
)
1395 unsigned int block_start
= 0, block_end
;
1396 struct buffer_head
*head
, *bh
;
1398 bh
= head
= page_buffers(page
);
1400 block_end
= block_start
+ bh
->b_size
;
1401 if (buffer_new(bh
)) {
1402 if (block_end
> from
&& block_start
< to
) {
1403 if (!PageUptodate(page
)) {
1404 unsigned start
, size
;
1406 start
= max(from
, block_start
);
1407 size
= min(to
, block_end
) - start
;
1409 zero_user(page
, start
, size
);
1410 write_end_fn(handle
, bh
);
1412 clear_buffer_new(bh
);
1415 block_start
= block_end
;
1416 bh
= bh
->b_this_page
;
1417 } while (bh
!= head
);
1420 static int ext4_journalled_write_end(struct file
*file
,
1421 struct address_space
*mapping
,
1422 loff_t pos
, unsigned len
, unsigned copied
,
1423 struct page
*page
, void *fsdata
)
1425 handle_t
*handle
= ext4_journal_current_handle();
1426 struct inode
*inode
= mapping
->host
;
1427 loff_t old_size
= inode
->i_size
;
1431 int size_changed
= 0;
1433 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1434 from
= pos
& (PAGE_SIZE
- 1);
1437 BUG_ON(!ext4_handle_valid(handle
));
1439 if (ext4_has_inline_data(inode
)) {
1440 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1448 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1450 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1452 if (unlikely(copied
< len
))
1453 ext4_journalled_zero_new_buffers(handle
, page
,
1455 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1456 from
+ copied
, &partial
,
1459 SetPageUptodate(page
);
1461 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1462 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1463 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1468 pagecache_isize_extended(inode
, old_size
, pos
);
1471 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1476 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1477 /* if we have allocated more blocks and copied
1478 * less. We will have blocks allocated outside
1479 * inode->i_size. So truncate them
1481 ext4_orphan_add(handle
, inode
);
1484 ret2
= ext4_journal_stop(handle
);
1487 if (pos
+ len
> inode
->i_size
) {
1488 ext4_truncate_failed_write(inode
);
1490 * If truncate failed early the inode might still be
1491 * on the orphan list; we need to make sure the inode
1492 * is removed from the orphan list in that case.
1495 ext4_orphan_del(NULL
, inode
);
1498 return ret
? ret
: copied
;
1502 * Reserve space for a single cluster
1504 static int ext4_da_reserve_space(struct inode
*inode
)
1506 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1507 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1511 * We will charge metadata quota at writeout time; this saves
1512 * us from metadata over-estimation, though we may go over by
1513 * a small amount in the end. Here we just reserve for data.
1515 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1519 spin_lock(&ei
->i_block_reservation_lock
);
1520 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1521 spin_unlock(&ei
->i_block_reservation_lock
);
1522 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1525 ei
->i_reserved_data_blocks
++;
1526 trace_ext4_da_reserve_space(inode
);
1527 spin_unlock(&ei
->i_block_reservation_lock
);
1529 return 0; /* success */
1532 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1534 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1535 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1538 return; /* Nothing to release, exit */
1540 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1542 trace_ext4_da_release_space(inode
, to_free
);
1543 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1545 * if there aren't enough reserved blocks, then the
1546 * counter is messed up somewhere. Since this
1547 * function is called from invalidate page, it's
1548 * harmless to return without any action.
1550 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1551 "ino %lu, to_free %d with only %d reserved "
1552 "data blocks", inode
->i_ino
, to_free
,
1553 ei
->i_reserved_data_blocks
);
1555 to_free
= ei
->i_reserved_data_blocks
;
1557 ei
->i_reserved_data_blocks
-= to_free
;
1559 /* update fs dirty data blocks counter */
1560 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1562 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1564 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1567 static void ext4_da_page_release_reservation(struct page
*page
,
1568 unsigned int offset
,
1569 unsigned int length
)
1571 int to_release
= 0, contiguous_blks
= 0;
1572 struct buffer_head
*head
, *bh
;
1573 unsigned int curr_off
= 0;
1574 struct inode
*inode
= page
->mapping
->host
;
1575 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1576 unsigned int stop
= offset
+ length
;
1580 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1582 head
= page_buffers(page
);
1585 unsigned int next_off
= curr_off
+ bh
->b_size
;
1587 if (next_off
> stop
)
1590 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1593 clear_buffer_delay(bh
);
1594 } else if (contiguous_blks
) {
1595 lblk
= page
->index
<<
1596 (PAGE_SHIFT
- inode
->i_blkbits
);
1597 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1599 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1600 contiguous_blks
= 0;
1602 curr_off
= next_off
;
1603 } while ((bh
= bh
->b_this_page
) != head
);
1605 if (contiguous_blks
) {
1606 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1607 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1608 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1611 /* If we have released all the blocks belonging to a cluster, then we
1612 * need to release the reserved space for that cluster. */
1613 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1614 while (num_clusters
> 0) {
1615 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1616 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1617 if (sbi
->s_cluster_ratio
== 1 ||
1618 !ext4_find_delalloc_cluster(inode
, lblk
))
1619 ext4_da_release_space(inode
, 1);
1626 * Delayed allocation stuff
1629 struct mpage_da_data
{
1630 struct inode
*inode
;
1631 struct writeback_control
*wbc
;
1633 pgoff_t first_page
; /* The first page to write */
1634 pgoff_t next_page
; /* Current page to examine */
1635 pgoff_t last_page
; /* Last page to examine */
1637 * Extent to map - this can be after first_page because that can be
1638 * fully mapped. We somewhat abuse m_flags to store whether the extent
1639 * is delalloc or unwritten.
1641 struct ext4_map_blocks map
;
1642 struct ext4_io_submit io_submit
; /* IO submission data */
1645 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1650 struct pagevec pvec
;
1651 struct inode
*inode
= mpd
->inode
;
1652 struct address_space
*mapping
= inode
->i_mapping
;
1654 /* This is necessary when next_page == 0. */
1655 if (mpd
->first_page
>= mpd
->next_page
)
1658 index
= mpd
->first_page
;
1659 end
= mpd
->next_page
- 1;
1661 ext4_lblk_t start
, last
;
1662 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1663 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1664 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1667 pagevec_init(&pvec
, 0);
1668 while (index
<= end
) {
1669 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1672 for (i
= 0; i
< nr_pages
; i
++) {
1673 struct page
*page
= pvec
.pages
[i
];
1674 if (page
->index
> end
)
1676 BUG_ON(!PageLocked(page
));
1677 BUG_ON(PageWriteback(page
));
1679 if (page_mapped(page
))
1680 clear_page_dirty_for_io(page
);
1681 block_invalidatepage(page
, 0, PAGE_SIZE
);
1682 ClearPageUptodate(page
);
1686 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1687 pagevec_release(&pvec
);
1691 static void ext4_print_free_blocks(struct inode
*inode
)
1693 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1694 struct super_block
*sb
= inode
->i_sb
;
1695 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1697 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1698 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1699 ext4_count_free_clusters(sb
)));
1700 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1701 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1702 (long long) EXT4_C2B(EXT4_SB(sb
),
1703 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1704 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1705 (long long) EXT4_C2B(EXT4_SB(sb
),
1706 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1707 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1708 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1709 ei
->i_reserved_data_blocks
);
1713 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1715 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1719 * This function is grabs code from the very beginning of
1720 * ext4_map_blocks, but assumes that the caller is from delayed write
1721 * time. This function looks up the requested blocks and sets the
1722 * buffer delay bit under the protection of i_data_sem.
1724 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1725 struct ext4_map_blocks
*map
,
1726 struct buffer_head
*bh
)
1728 struct extent_status es
;
1730 sector_t invalid_block
= ~((sector_t
) 0xffff);
1731 #ifdef ES_AGGRESSIVE_TEST
1732 struct ext4_map_blocks orig_map
;
1734 memcpy(&orig_map
, map
, sizeof(*map
));
1737 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1741 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1742 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1743 (unsigned long) map
->m_lblk
);
1745 /* Lookup extent status tree firstly */
1746 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1747 if (ext4_es_is_hole(&es
)) {
1749 down_read(&EXT4_I(inode
)->i_data_sem
);
1754 * Delayed extent could be allocated by fallocate.
1755 * So we need to check it.
1757 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1758 map_bh(bh
, inode
->i_sb
, invalid_block
);
1760 set_buffer_delay(bh
);
1764 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1765 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1766 if (retval
> map
->m_len
)
1767 retval
= map
->m_len
;
1768 map
->m_len
= retval
;
1769 if (ext4_es_is_written(&es
))
1770 map
->m_flags
|= EXT4_MAP_MAPPED
;
1771 else if (ext4_es_is_unwritten(&es
))
1772 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1776 #ifdef ES_AGGRESSIVE_TEST
1777 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1783 * Try to see if we can get the block without requesting a new
1784 * file system block.
1786 down_read(&EXT4_I(inode
)->i_data_sem
);
1787 if (ext4_has_inline_data(inode
))
1789 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1790 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1792 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1798 * XXX: __block_prepare_write() unmaps passed block,
1802 * If the block was allocated from previously allocated cluster,
1803 * then we don't need to reserve it again. However we still need
1804 * to reserve metadata for every block we're going to write.
1806 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1807 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1808 ret
= ext4_da_reserve_space(inode
);
1810 /* not enough space to reserve */
1816 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1817 ~0, EXTENT_STATUS_DELAYED
);
1823 map_bh(bh
, inode
->i_sb
, invalid_block
);
1825 set_buffer_delay(bh
);
1826 } else if (retval
> 0) {
1828 unsigned int status
;
1830 if (unlikely(retval
!= map
->m_len
)) {
1831 ext4_warning(inode
->i_sb
,
1832 "ES len assertion failed for inode "
1833 "%lu: retval %d != map->m_len %d",
1834 inode
->i_ino
, retval
, map
->m_len
);
1838 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1839 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1840 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1841 map
->m_pblk
, status
);
1847 up_read((&EXT4_I(inode
)->i_data_sem
));
1853 * This is a special get_block_t callback which is used by
1854 * ext4_da_write_begin(). It will either return mapped block or
1855 * reserve space for a single block.
1857 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1858 * We also have b_blocknr = -1 and b_bdev initialized properly
1860 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1861 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1862 * initialized properly.
1864 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1865 struct buffer_head
*bh
, int create
)
1867 struct ext4_map_blocks map
;
1870 BUG_ON(create
== 0);
1871 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1873 map
.m_lblk
= iblock
;
1877 * first, we need to know whether the block is allocated already
1878 * preallocated blocks are unmapped but should treated
1879 * the same as allocated blocks.
1881 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1885 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1886 ext4_update_bh_state(bh
, map
.m_flags
);
1888 if (buffer_unwritten(bh
)) {
1889 /* A delayed write to unwritten bh should be marked
1890 * new and mapped. Mapped ensures that we don't do
1891 * get_block multiple times when we write to the same
1892 * offset and new ensures that we do proper zero out
1893 * for partial write.
1896 set_buffer_mapped(bh
);
1901 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1907 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1913 static int __ext4_journalled_writepage(struct page
*page
,
1916 struct address_space
*mapping
= page
->mapping
;
1917 struct inode
*inode
= mapping
->host
;
1918 struct buffer_head
*page_bufs
= NULL
;
1919 handle_t
*handle
= NULL
;
1920 int ret
= 0, err
= 0;
1921 int inline_data
= ext4_has_inline_data(inode
);
1922 struct buffer_head
*inode_bh
= NULL
;
1924 ClearPageChecked(page
);
1927 BUG_ON(page
->index
!= 0);
1928 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1929 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1930 if (inode_bh
== NULL
)
1933 page_bufs
= page_buffers(page
);
1938 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1942 * We need to release the page lock before we start the
1943 * journal, so grab a reference so the page won't disappear
1944 * out from under us.
1949 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1950 ext4_writepage_trans_blocks(inode
));
1951 if (IS_ERR(handle
)) {
1952 ret
= PTR_ERR(handle
);
1954 goto out_no_pagelock
;
1956 BUG_ON(!ext4_handle_valid(handle
));
1960 if (page
->mapping
!= mapping
) {
1961 /* The page got truncated from under us */
1962 ext4_journal_stop(handle
);
1968 BUFFER_TRACE(inode_bh
, "get write access");
1969 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1971 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1974 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1975 do_journal_get_write_access
);
1977 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1982 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1983 err
= ext4_journal_stop(handle
);
1987 if (!ext4_has_inline_data(inode
))
1988 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1990 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1999 * Note that we don't need to start a transaction unless we're journaling data
2000 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2001 * need to file the inode to the transaction's list in ordered mode because if
2002 * we are writing back data added by write(), the inode is already there and if
2003 * we are writing back data modified via mmap(), no one guarantees in which
2004 * transaction the data will hit the disk. In case we are journaling data, we
2005 * cannot start transaction directly because transaction start ranks above page
2006 * lock so we have to do some magic.
2008 * This function can get called via...
2009 * - ext4_writepages after taking page lock (have journal handle)
2010 * - journal_submit_inode_data_buffers (no journal handle)
2011 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2012 * - grab_page_cache when doing write_begin (have journal handle)
2014 * We don't do any block allocation in this function. If we have page with
2015 * multiple blocks we need to write those buffer_heads that are mapped. This
2016 * is important for mmaped based write. So if we do with blocksize 1K
2017 * truncate(f, 1024);
2018 * a = mmap(f, 0, 4096);
2020 * truncate(f, 4096);
2021 * we have in the page first buffer_head mapped via page_mkwrite call back
2022 * but other buffer_heads would be unmapped but dirty (dirty done via the
2023 * do_wp_page). So writepage should write the first block. If we modify
2024 * the mmap area beyond 1024 we will again get a page_fault and the
2025 * page_mkwrite callback will do the block allocation and mark the
2026 * buffer_heads mapped.
2028 * We redirty the page if we have any buffer_heads that is either delay or
2029 * unwritten in the page.
2031 * We can get recursively called as show below.
2033 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2036 * But since we don't do any block allocation we should not deadlock.
2037 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2039 static int ext4_writepage(struct page
*page
,
2040 struct writeback_control
*wbc
)
2045 struct buffer_head
*page_bufs
= NULL
;
2046 struct inode
*inode
= page
->mapping
->host
;
2047 struct ext4_io_submit io_submit
;
2048 bool keep_towrite
= false;
2050 trace_ext4_writepage(page
);
2051 size
= i_size_read(inode
);
2052 if (page
->index
== size
>> PAGE_SHIFT
)
2053 len
= size
& ~PAGE_MASK
;
2057 page_bufs
= page_buffers(page
);
2059 * We cannot do block allocation or other extent handling in this
2060 * function. If there are buffers needing that, we have to redirty
2061 * the page. But we may reach here when we do a journal commit via
2062 * journal_submit_inode_data_buffers() and in that case we must write
2063 * allocated buffers to achieve data=ordered mode guarantees.
2065 * Also, if there is only one buffer per page (the fs block
2066 * size == the page size), if one buffer needs block
2067 * allocation or needs to modify the extent tree to clear the
2068 * unwritten flag, we know that the page can't be written at
2069 * all, so we might as well refuse the write immediately.
2070 * Unfortunately if the block size != page size, we can't as
2071 * easily detect this case using ext4_walk_page_buffers(), but
2072 * for the extremely common case, this is an optimization that
2073 * skips a useless round trip through ext4_bio_write_page().
2075 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2076 ext4_bh_delay_or_unwritten
)) {
2077 redirty_page_for_writepage(wbc
, page
);
2078 if ((current
->flags
& PF_MEMALLOC
) ||
2079 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2081 * For memory cleaning there's no point in writing only
2082 * some buffers. So just bail out. Warn if we came here
2083 * from direct reclaim.
2085 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2090 keep_towrite
= true;
2093 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2095 * It's mmapped pagecache. Add buffers and journal it. There
2096 * doesn't seem much point in redirtying the page here.
2098 return __ext4_journalled_writepage(page
, len
);
2100 ext4_io_submit_init(&io_submit
, wbc
);
2101 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2102 if (!io_submit
.io_end
) {
2103 redirty_page_for_writepage(wbc
, page
);
2107 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2108 ext4_io_submit(&io_submit
);
2109 /* Drop io_end reference we got from init */
2110 ext4_put_io_end_defer(io_submit
.io_end
);
2114 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2117 loff_t size
= i_size_read(mpd
->inode
);
2120 BUG_ON(page
->index
!= mpd
->first_page
);
2121 if (page
->index
== size
>> PAGE_SHIFT
)
2122 len
= size
& ~PAGE_MASK
;
2125 clear_page_dirty_for_io(page
);
2126 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2128 mpd
->wbc
->nr_to_write
--;
2134 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2137 * mballoc gives us at most this number of blocks...
2138 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2139 * The rest of mballoc seems to handle chunks up to full group size.
2141 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2144 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2146 * @mpd - extent of blocks
2147 * @lblk - logical number of the block in the file
2148 * @bh - buffer head we want to add to the extent
2150 * The function is used to collect contig. blocks in the same state. If the
2151 * buffer doesn't require mapping for writeback and we haven't started the
2152 * extent of buffers to map yet, the function returns 'true' immediately - the
2153 * caller can write the buffer right away. Otherwise the function returns true
2154 * if the block has been added to the extent, false if the block couldn't be
2157 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2158 struct buffer_head
*bh
)
2160 struct ext4_map_blocks
*map
= &mpd
->map
;
2162 /* Buffer that doesn't need mapping for writeback? */
2163 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2164 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2165 /* So far no extent to map => we write the buffer right away */
2166 if (map
->m_len
== 0)
2171 /* First block in the extent? */
2172 if (map
->m_len
== 0) {
2175 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2179 /* Don't go larger than mballoc is willing to allocate */
2180 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2183 /* Can we merge the block to our big extent? */
2184 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2185 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2193 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2195 * @mpd - extent of blocks for mapping
2196 * @head - the first buffer in the page
2197 * @bh - buffer we should start processing from
2198 * @lblk - logical number of the block in the file corresponding to @bh
2200 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2201 * the page for IO if all buffers in this page were mapped and there's no
2202 * accumulated extent of buffers to map or add buffers in the page to the
2203 * extent of buffers to map. The function returns 1 if the caller can continue
2204 * by processing the next page, 0 if it should stop adding buffers to the
2205 * extent to map because we cannot extend it anymore. It can also return value
2206 * < 0 in case of error during IO submission.
2208 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2209 struct buffer_head
*head
,
2210 struct buffer_head
*bh
,
2213 struct inode
*inode
= mpd
->inode
;
2215 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2216 >> inode
->i_blkbits
;
2219 BUG_ON(buffer_locked(bh
));
2221 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2222 /* Found extent to map? */
2225 /* Everything mapped so far and we hit EOF */
2228 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2229 /* So far everything mapped? Submit the page for IO. */
2230 if (mpd
->map
.m_len
== 0) {
2231 err
= mpage_submit_page(mpd
, head
->b_page
);
2235 return lblk
< blocks
;
2239 * mpage_map_buffers - update buffers corresponding to changed extent and
2240 * submit fully mapped pages for IO
2242 * @mpd - description of extent to map, on return next extent to map
2244 * Scan buffers corresponding to changed extent (we expect corresponding pages
2245 * to be already locked) and update buffer state according to new extent state.
2246 * We map delalloc buffers to their physical location, clear unwritten bits,
2247 * and mark buffers as uninit when we perform writes to unwritten extents
2248 * and do extent conversion after IO is finished. If the last page is not fully
2249 * mapped, we update @map to the next extent in the last page that needs
2250 * mapping. Otherwise we submit the page for IO.
2252 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2254 struct pagevec pvec
;
2256 struct inode
*inode
= mpd
->inode
;
2257 struct buffer_head
*head
, *bh
;
2258 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2264 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2265 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2266 lblk
= start
<< bpp_bits
;
2267 pblock
= mpd
->map
.m_pblk
;
2269 pagevec_init(&pvec
, 0);
2270 while (start
<= end
) {
2271 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2275 for (i
= 0; i
< nr_pages
; i
++) {
2276 struct page
*page
= pvec
.pages
[i
];
2278 if (page
->index
> end
)
2280 /* Up to 'end' pages must be contiguous */
2281 BUG_ON(page
->index
!= start
);
2282 bh
= head
= page_buffers(page
);
2284 if (lblk
< mpd
->map
.m_lblk
)
2286 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2288 * Buffer after end of mapped extent.
2289 * Find next buffer in the page to map.
2292 mpd
->map
.m_flags
= 0;
2294 * FIXME: If dioread_nolock supports
2295 * blocksize < pagesize, we need to make
2296 * sure we add size mapped so far to
2297 * io_end->size as the following call
2298 * can submit the page for IO.
2300 err
= mpage_process_page_bufs(mpd
, head
,
2302 pagevec_release(&pvec
);
2307 if (buffer_delay(bh
)) {
2308 clear_buffer_delay(bh
);
2309 bh
->b_blocknr
= pblock
++;
2311 clear_buffer_unwritten(bh
);
2312 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2315 * FIXME: This is going to break if dioread_nolock
2316 * supports blocksize < pagesize as we will try to
2317 * convert potentially unmapped parts of inode.
2319 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2320 /* Page fully mapped - let IO run! */
2321 err
= mpage_submit_page(mpd
, page
);
2323 pagevec_release(&pvec
);
2328 pagevec_release(&pvec
);
2330 /* Extent fully mapped and matches with page boundary. We are done. */
2332 mpd
->map
.m_flags
= 0;
2336 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2338 struct inode
*inode
= mpd
->inode
;
2339 struct ext4_map_blocks
*map
= &mpd
->map
;
2340 int get_blocks_flags
;
2341 int err
, dioread_nolock
;
2343 trace_ext4_da_write_pages_extent(inode
, map
);
2345 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2346 * to convert an unwritten extent to be initialized (in the case
2347 * where we have written into one or more preallocated blocks). It is
2348 * possible that we're going to need more metadata blocks than
2349 * previously reserved. However we must not fail because we're in
2350 * writeback and there is nothing we can do about it so it might result
2351 * in data loss. So use reserved blocks to allocate metadata if
2354 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2355 * the blocks in question are delalloc blocks. This indicates
2356 * that the blocks and quotas has already been checked when
2357 * the data was copied into the page cache.
2359 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2360 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2361 EXT4_GET_BLOCKS_IO_SUBMIT
;
2362 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2364 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2365 if (map
->m_flags
& (1 << BH_Delay
))
2366 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2368 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2371 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2372 if (!mpd
->io_submit
.io_end
->handle
&&
2373 ext4_handle_valid(handle
)) {
2374 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2375 handle
->h_rsv_handle
= NULL
;
2377 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2380 BUG_ON(map
->m_len
== 0);
2381 if (map
->m_flags
& EXT4_MAP_NEW
) {
2382 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
2389 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2390 * mpd->len and submit pages underlying it for IO
2392 * @handle - handle for journal operations
2393 * @mpd - extent to map
2394 * @give_up_on_write - we set this to true iff there is a fatal error and there
2395 * is no hope of writing the data. The caller should discard
2396 * dirty pages to avoid infinite loops.
2398 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2399 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2400 * them to initialized or split the described range from larger unwritten
2401 * extent. Note that we need not map all the described range since allocation
2402 * can return less blocks or the range is covered by more unwritten extents. We
2403 * cannot map more because we are limited by reserved transaction credits. On
2404 * the other hand we always make sure that the last touched page is fully
2405 * mapped so that it can be written out (and thus forward progress is
2406 * guaranteed). After mapping we submit all mapped pages for IO.
2408 static int mpage_map_and_submit_extent(handle_t
*handle
,
2409 struct mpage_da_data
*mpd
,
2410 bool *give_up_on_write
)
2412 struct inode
*inode
= mpd
->inode
;
2413 struct ext4_map_blocks
*map
= &mpd
->map
;
2418 mpd
->io_submit
.io_end
->offset
=
2419 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2421 err
= mpage_map_one_extent(handle
, mpd
);
2423 struct super_block
*sb
= inode
->i_sb
;
2425 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2426 goto invalidate_dirty_pages
;
2428 * Let the uper layers retry transient errors.
2429 * In the case of ENOSPC, if ext4_count_free_blocks()
2430 * is non-zero, a commit should free up blocks.
2432 if ((err
== -ENOMEM
) ||
2433 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2435 goto update_disksize
;
2438 ext4_msg(sb
, KERN_CRIT
,
2439 "Delayed block allocation failed for "
2440 "inode %lu at logical offset %llu with"
2441 " max blocks %u with error %d",
2443 (unsigned long long)map
->m_lblk
,
2444 (unsigned)map
->m_len
, -err
);
2445 ext4_msg(sb
, KERN_CRIT
,
2446 "This should not happen!! Data will "
2449 ext4_print_free_blocks(inode
);
2450 invalidate_dirty_pages
:
2451 *give_up_on_write
= true;
2456 * Update buffer state, submit mapped pages, and get us new
2459 err
= mpage_map_and_submit_buffers(mpd
);
2461 goto update_disksize
;
2462 } while (map
->m_len
);
2466 * Update on-disk size after IO is submitted. Races with
2467 * truncate are avoided by checking i_size under i_data_sem.
2469 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2470 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2474 down_write(&EXT4_I(inode
)->i_data_sem
);
2475 i_size
= i_size_read(inode
);
2476 if (disksize
> i_size
)
2478 if (disksize
> EXT4_I(inode
)->i_disksize
)
2479 EXT4_I(inode
)->i_disksize
= disksize
;
2480 err2
= ext4_mark_inode_dirty(handle
, inode
);
2481 up_write(&EXT4_I(inode
)->i_data_sem
);
2483 ext4_error(inode
->i_sb
,
2484 "Failed to mark inode %lu dirty",
2493 * Calculate the total number of credits to reserve for one writepages
2494 * iteration. This is called from ext4_writepages(). We map an extent of
2495 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2496 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2497 * bpp - 1 blocks in bpp different extents.
2499 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2501 int bpp
= ext4_journal_blocks_per_page(inode
);
2503 return ext4_meta_trans_blocks(inode
,
2504 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2508 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2509 * and underlying extent to map
2511 * @mpd - where to look for pages
2513 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2514 * IO immediately. When we find a page which isn't mapped we start accumulating
2515 * extent of buffers underlying these pages that needs mapping (formed by
2516 * either delayed or unwritten buffers). We also lock the pages containing
2517 * these buffers. The extent found is returned in @mpd structure (starting at
2518 * mpd->lblk with length mpd->len blocks).
2520 * Note that this function can attach bios to one io_end structure which are
2521 * neither logically nor physically contiguous. Although it may seem as an
2522 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2523 * case as we need to track IO to all buffers underlying a page in one io_end.
2525 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2527 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2528 struct pagevec pvec
;
2529 unsigned int nr_pages
;
2530 long left
= mpd
->wbc
->nr_to_write
;
2531 pgoff_t index
= mpd
->first_page
;
2532 pgoff_t end
= mpd
->last_page
;
2535 int blkbits
= mpd
->inode
->i_blkbits
;
2537 struct buffer_head
*head
;
2539 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2540 tag
= PAGECACHE_TAG_TOWRITE
;
2542 tag
= PAGECACHE_TAG_DIRTY
;
2544 pagevec_init(&pvec
, 0);
2546 mpd
->next_page
= index
;
2547 while (index
<= end
) {
2548 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2549 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2553 for (i
= 0; i
< nr_pages
; i
++) {
2554 struct page
*page
= pvec
.pages
[i
];
2557 * At this point, the page may be truncated or
2558 * invalidated (changing page->mapping to NULL), or
2559 * even swizzled back from swapper_space to tmpfs file
2560 * mapping. However, page->index will not change
2561 * because we have a reference on the page.
2563 if (page
->index
> end
)
2567 * Accumulated enough dirty pages? This doesn't apply
2568 * to WB_SYNC_ALL mode. For integrity sync we have to
2569 * keep going because someone may be concurrently
2570 * dirtying pages, and we might have synced a lot of
2571 * newly appeared dirty pages, but have not synced all
2572 * of the old dirty pages.
2574 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2577 /* If we can't merge this page, we are done. */
2578 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2583 * If the page is no longer dirty, or its mapping no
2584 * longer corresponds to inode we are writing (which
2585 * means it has been truncated or invalidated), or the
2586 * page is already under writeback and we are not doing
2587 * a data integrity writeback, skip the page
2589 if (!PageDirty(page
) ||
2590 (PageWriteback(page
) &&
2591 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2592 unlikely(page
->mapping
!= mapping
)) {
2597 wait_on_page_writeback(page
);
2598 BUG_ON(PageWriteback(page
));
2600 if (mpd
->map
.m_len
== 0)
2601 mpd
->first_page
= page
->index
;
2602 mpd
->next_page
= page
->index
+ 1;
2603 /* Add all dirty buffers to mpd */
2604 lblk
= ((ext4_lblk_t
)page
->index
) <<
2605 (PAGE_SHIFT
- blkbits
);
2606 head
= page_buffers(page
);
2607 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2613 pagevec_release(&pvec
);
2618 pagevec_release(&pvec
);
2622 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2625 struct address_space
*mapping
= data
;
2626 int ret
= ext4_writepage(page
, wbc
);
2627 mapping_set_error(mapping
, ret
);
2631 static int ext4_writepages(struct address_space
*mapping
,
2632 struct writeback_control
*wbc
)
2634 pgoff_t writeback_index
= 0;
2635 long nr_to_write
= wbc
->nr_to_write
;
2636 int range_whole
= 0;
2638 handle_t
*handle
= NULL
;
2639 struct mpage_da_data mpd
;
2640 struct inode
*inode
= mapping
->host
;
2641 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2642 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2644 struct blk_plug plug
;
2645 bool give_up_on_write
= false;
2647 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2648 trace_ext4_writepages(inode
, wbc
);
2650 if (dax_mapping(mapping
)) {
2651 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2653 goto out_writepages
;
2657 * No pages to write? This is mainly a kludge to avoid starting
2658 * a transaction for special inodes like journal inode on last iput()
2659 * because that could violate lock ordering on umount
2661 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2662 goto out_writepages
;
2664 if (ext4_should_journal_data(inode
)) {
2665 struct blk_plug plug
;
2667 blk_start_plug(&plug
);
2668 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2669 blk_finish_plug(&plug
);
2670 goto out_writepages
;
2674 * If the filesystem has aborted, it is read-only, so return
2675 * right away instead of dumping stack traces later on that
2676 * will obscure the real source of the problem. We test
2677 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2678 * the latter could be true if the filesystem is mounted
2679 * read-only, and in that case, ext4_writepages should
2680 * *never* be called, so if that ever happens, we would want
2683 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2685 goto out_writepages
;
2688 if (ext4_should_dioread_nolock(inode
)) {
2690 * We may need to convert up to one extent per block in
2691 * the page and we may dirty the inode.
2693 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2697 * If we have inline data and arrive here, it means that
2698 * we will soon create the block for the 1st page, so
2699 * we'd better clear the inline data here.
2701 if (ext4_has_inline_data(inode
)) {
2702 /* Just inode will be modified... */
2703 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2704 if (IS_ERR(handle
)) {
2705 ret
= PTR_ERR(handle
);
2706 goto out_writepages
;
2708 BUG_ON(ext4_test_inode_state(inode
,
2709 EXT4_STATE_MAY_INLINE_DATA
));
2710 ext4_destroy_inline_data(handle
, inode
);
2711 ext4_journal_stop(handle
);
2714 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2717 if (wbc
->range_cyclic
) {
2718 writeback_index
= mapping
->writeback_index
;
2719 if (writeback_index
)
2721 mpd
.first_page
= writeback_index
;
2724 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2725 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2730 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2732 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2733 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2735 blk_start_plug(&plug
);
2736 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2737 /* For each extent of pages we use new io_end */
2738 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2739 if (!mpd
.io_submit
.io_end
) {
2745 * We have two constraints: We find one extent to map and we
2746 * must always write out whole page (makes a difference when
2747 * blocksize < pagesize) so that we don't block on IO when we
2748 * try to write out the rest of the page. Journalled mode is
2749 * not supported by delalloc.
2751 BUG_ON(ext4_should_journal_data(inode
));
2752 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2754 /* start a new transaction */
2755 handle
= ext4_journal_start_with_reserve(inode
,
2756 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2757 if (IS_ERR(handle
)) {
2758 ret
= PTR_ERR(handle
);
2759 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2760 "%ld pages, ino %lu; err %d", __func__
,
2761 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2762 /* Release allocated io_end */
2763 ext4_put_io_end(mpd
.io_submit
.io_end
);
2767 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2768 ret
= mpage_prepare_extent_to_map(&mpd
);
2771 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2775 * We scanned the whole range (or exhausted
2776 * nr_to_write), submitted what was mapped and
2777 * didn't find anything needing mapping. We are
2784 * Caution: If the handle is synchronous,
2785 * ext4_journal_stop() can wait for transaction commit
2786 * to finish which may depend on writeback of pages to
2787 * complete or on page lock to be released. In that
2788 * case, we have to wait until after after we have
2789 * submitted all the IO, released page locks we hold,
2790 * and dropped io_end reference (for extent conversion
2791 * to be able to complete) before stopping the handle.
2793 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2794 ext4_journal_stop(handle
);
2797 /* Submit prepared bio */
2798 ext4_io_submit(&mpd
.io_submit
);
2799 /* Unlock pages we didn't use */
2800 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2802 * Drop our io_end reference we got from init. We have
2803 * to be careful and use deferred io_end finishing if
2804 * we are still holding the transaction as we can
2805 * release the last reference to io_end which may end
2806 * up doing unwritten extent conversion.
2809 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2810 ext4_journal_stop(handle
);
2812 ext4_put_io_end(mpd
.io_submit
.io_end
);
2814 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2816 * Commit the transaction which would
2817 * free blocks released in the transaction
2820 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2824 /* Fatal error - ENOMEM, EIO... */
2828 blk_finish_plug(&plug
);
2829 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2831 mpd
.last_page
= writeback_index
- 1;
2837 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2839 * Set the writeback_index so that range_cyclic
2840 * mode will write it back later
2842 mapping
->writeback_index
= mpd
.first_page
;
2845 trace_ext4_writepages_result(inode
, wbc
, ret
,
2846 nr_to_write
- wbc
->nr_to_write
);
2847 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2851 static int ext4_nonda_switch(struct super_block
*sb
)
2853 s64 free_clusters
, dirty_clusters
;
2854 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2857 * switch to non delalloc mode if we are running low
2858 * on free block. The free block accounting via percpu
2859 * counters can get slightly wrong with percpu_counter_batch getting
2860 * accumulated on each CPU without updating global counters
2861 * Delalloc need an accurate free block accounting. So switch
2862 * to non delalloc when we are near to error range.
2865 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2867 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2869 * Start pushing delalloc when 1/2 of free blocks are dirty.
2871 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2872 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2874 if (2 * free_clusters
< 3 * dirty_clusters
||
2875 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2877 * free block count is less than 150% of dirty blocks
2878 * or free blocks is less than watermark
2885 /* We always reserve for an inode update; the superblock could be there too */
2886 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2888 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2891 if (pos
+ len
<= 0x7fffffffULL
)
2894 /* We might need to update the superblock to set LARGE_FILE */
2898 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2899 loff_t pos
, unsigned len
, unsigned flags
,
2900 struct page
**pagep
, void **fsdata
)
2902 int ret
, retries
= 0;
2905 struct inode
*inode
= mapping
->host
;
2908 index
= pos
>> PAGE_SHIFT
;
2910 if (ext4_nonda_switch(inode
->i_sb
) ||
2911 S_ISLNK(inode
->i_mode
)) {
2912 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2913 return ext4_write_begin(file
, mapping
, pos
,
2914 len
, flags
, pagep
, fsdata
);
2916 *fsdata
= (void *)0;
2917 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2919 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2920 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2930 * grab_cache_page_write_begin() can take a long time if the
2931 * system is thrashing due to memory pressure, or if the page
2932 * is being written back. So grab it first before we start
2933 * the transaction handle. This also allows us to allocate
2934 * the page (if needed) without using GFP_NOFS.
2937 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2943 * With delayed allocation, we don't log the i_disksize update
2944 * if there is delayed block allocation. But we still need
2945 * to journalling the i_disksize update if writes to the end
2946 * of file which has an already mapped buffer.
2949 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2950 ext4_da_write_credits(inode
, pos
, len
));
2951 if (IS_ERR(handle
)) {
2953 return PTR_ERR(handle
);
2957 if (page
->mapping
!= mapping
) {
2958 /* The page got truncated from under us */
2961 ext4_journal_stop(handle
);
2964 /* In case writeback began while the page was unlocked */
2965 wait_for_stable_page(page
);
2967 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2968 ret
= ext4_block_write_begin(page
, pos
, len
,
2969 ext4_da_get_block_prep
);
2971 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2975 ext4_journal_stop(handle
);
2977 * block_write_begin may have instantiated a few blocks
2978 * outside i_size. Trim these off again. Don't need
2979 * i_size_read because we hold i_mutex.
2981 if (pos
+ len
> inode
->i_size
)
2982 ext4_truncate_failed_write(inode
);
2984 if (ret
== -ENOSPC
&&
2985 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2997 * Check if we should update i_disksize
2998 * when write to the end of file but not require block allocation
3000 static int ext4_da_should_update_i_disksize(struct page
*page
,
3001 unsigned long offset
)
3003 struct buffer_head
*bh
;
3004 struct inode
*inode
= page
->mapping
->host
;
3008 bh
= page_buffers(page
);
3009 idx
= offset
>> inode
->i_blkbits
;
3011 for (i
= 0; i
< idx
; i
++)
3012 bh
= bh
->b_this_page
;
3014 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3019 static int ext4_da_write_end(struct file
*file
,
3020 struct address_space
*mapping
,
3021 loff_t pos
, unsigned len
, unsigned copied
,
3022 struct page
*page
, void *fsdata
)
3024 struct inode
*inode
= mapping
->host
;
3026 handle_t
*handle
= ext4_journal_current_handle();
3028 unsigned long start
, end
;
3029 int write_mode
= (int)(unsigned long)fsdata
;
3031 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3032 return ext4_write_end(file
, mapping
, pos
,
3033 len
, copied
, page
, fsdata
);
3035 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3036 start
= pos
& (PAGE_SIZE
- 1);
3037 end
= start
+ copied
- 1;
3040 * generic_write_end() will run mark_inode_dirty() if i_size
3041 * changes. So let's piggyback the i_disksize mark_inode_dirty
3044 new_i_size
= pos
+ copied
;
3045 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3046 if (ext4_has_inline_data(inode
) ||
3047 ext4_da_should_update_i_disksize(page
, end
)) {
3048 ext4_update_i_disksize(inode
, new_i_size
);
3049 /* We need to mark inode dirty even if
3050 * new_i_size is less that inode->i_size
3051 * bu greater than i_disksize.(hint delalloc)
3053 ext4_mark_inode_dirty(handle
, inode
);
3057 if (write_mode
!= CONVERT_INLINE_DATA
&&
3058 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3059 ext4_has_inline_data(inode
))
3060 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3063 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3069 ret2
= ext4_journal_stop(handle
);
3073 return ret
? ret
: copied
;
3076 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3077 unsigned int length
)
3080 * Drop reserved blocks
3082 BUG_ON(!PageLocked(page
));
3083 if (!page_has_buffers(page
))
3086 ext4_da_page_release_reservation(page
, offset
, length
);
3089 ext4_invalidatepage(page
, offset
, length
);
3095 * Force all delayed allocation blocks to be allocated for a given inode.
3097 int ext4_alloc_da_blocks(struct inode
*inode
)
3099 trace_ext4_alloc_da_blocks(inode
);
3101 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3105 * We do something simple for now. The filemap_flush() will
3106 * also start triggering a write of the data blocks, which is
3107 * not strictly speaking necessary (and for users of
3108 * laptop_mode, not even desirable). However, to do otherwise
3109 * would require replicating code paths in:
3111 * ext4_writepages() ->
3112 * write_cache_pages() ---> (via passed in callback function)
3113 * __mpage_da_writepage() -->
3114 * mpage_add_bh_to_extent()
3115 * mpage_da_map_blocks()
3117 * The problem is that write_cache_pages(), located in
3118 * mm/page-writeback.c, marks pages clean in preparation for
3119 * doing I/O, which is not desirable if we're not planning on
3122 * We could call write_cache_pages(), and then redirty all of
3123 * the pages by calling redirty_page_for_writepage() but that
3124 * would be ugly in the extreme. So instead we would need to
3125 * replicate parts of the code in the above functions,
3126 * simplifying them because we wouldn't actually intend to
3127 * write out the pages, but rather only collect contiguous
3128 * logical block extents, call the multi-block allocator, and
3129 * then update the buffer heads with the block allocations.
3131 * For now, though, we'll cheat by calling filemap_flush(),
3132 * which will map the blocks, and start the I/O, but not
3133 * actually wait for the I/O to complete.
3135 return filemap_flush(inode
->i_mapping
);
3139 * bmap() is special. It gets used by applications such as lilo and by
3140 * the swapper to find the on-disk block of a specific piece of data.
3142 * Naturally, this is dangerous if the block concerned is still in the
3143 * journal. If somebody makes a swapfile on an ext4 data-journaling
3144 * filesystem and enables swap, then they may get a nasty shock when the
3145 * data getting swapped to that swapfile suddenly gets overwritten by
3146 * the original zero's written out previously to the journal and
3147 * awaiting writeback in the kernel's buffer cache.
3149 * So, if we see any bmap calls here on a modified, data-journaled file,
3150 * take extra steps to flush any blocks which might be in the cache.
3152 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3154 struct inode
*inode
= mapping
->host
;
3159 * We can get here for an inline file via the FIBMAP ioctl
3161 if (ext4_has_inline_data(inode
))
3164 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3165 test_opt(inode
->i_sb
, DELALLOC
)) {
3167 * With delalloc we want to sync the file
3168 * so that we can make sure we allocate
3171 filemap_write_and_wait(mapping
);
3174 if (EXT4_JOURNAL(inode
) &&
3175 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3177 * This is a REALLY heavyweight approach, but the use of
3178 * bmap on dirty files is expected to be extremely rare:
3179 * only if we run lilo or swapon on a freshly made file
3180 * do we expect this to happen.
3182 * (bmap requires CAP_SYS_RAWIO so this does not
3183 * represent an unprivileged user DOS attack --- we'd be
3184 * in trouble if mortal users could trigger this path at
3187 * NB. EXT4_STATE_JDATA is not set on files other than
3188 * regular files. If somebody wants to bmap a directory
3189 * or symlink and gets confused because the buffer
3190 * hasn't yet been flushed to disk, they deserve
3191 * everything they get.
3194 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3195 journal
= EXT4_JOURNAL(inode
);
3196 jbd2_journal_lock_updates(journal
);
3197 err
= jbd2_journal_flush(journal
);
3198 jbd2_journal_unlock_updates(journal
);
3204 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3207 static int ext4_readpage(struct file
*file
, struct page
*page
)
3210 struct inode
*inode
= page
->mapping
->host
;
3212 trace_ext4_readpage(page
);
3214 if (ext4_has_inline_data(inode
))
3215 ret
= ext4_readpage_inline(inode
, page
);
3218 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3224 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3225 struct list_head
*pages
, unsigned nr_pages
)
3227 struct inode
*inode
= mapping
->host
;
3229 /* If the file has inline data, no need to do readpages. */
3230 if (ext4_has_inline_data(inode
))
3233 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3236 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3237 unsigned int length
)
3239 trace_ext4_invalidatepage(page
, offset
, length
);
3241 /* No journalling happens on data buffers when this function is used */
3242 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3244 block_invalidatepage(page
, offset
, length
);
3247 static int __ext4_journalled_invalidatepage(struct page
*page
,
3248 unsigned int offset
,
3249 unsigned int length
)
3251 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3253 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3256 * If it's a full truncate we just forget about the pending dirtying
3258 if (offset
== 0 && length
== PAGE_SIZE
)
3259 ClearPageChecked(page
);
3261 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3264 /* Wrapper for aops... */
3265 static void ext4_journalled_invalidatepage(struct page
*page
,
3266 unsigned int offset
,
3267 unsigned int length
)
3269 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3272 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3274 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3276 trace_ext4_releasepage(page
);
3278 /* Page has dirty journalled data -> cannot release */
3279 if (PageChecked(page
))
3282 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3284 return try_to_free_buffers(page
);
3287 #ifdef CONFIG_FS_DAX
3288 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3289 unsigned flags
, struct iomap
*iomap
)
3291 unsigned int blkbits
= inode
->i_blkbits
;
3292 unsigned long first_block
= offset
>> blkbits
;
3293 unsigned long last_block
= (offset
+ length
- 1) >> blkbits
;
3294 struct ext4_map_blocks map
;
3297 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3300 map
.m_lblk
= first_block
;
3301 map
.m_len
= last_block
- first_block
+ 1;
3303 if (!(flags
& IOMAP_WRITE
)) {
3304 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3310 /* Trim mapping request to maximum we can map at once for DIO */
3311 if (map
.m_len
> DIO_MAX_BLOCKS
)
3312 map
.m_len
= DIO_MAX_BLOCKS
;
3313 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3316 * Either we allocate blocks and then we don't get unwritten
3317 * extent so we have reserved enough credits, or the blocks
3318 * are already allocated and unwritten and in that case
3319 * extent conversion fits in the credits as well.
3321 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3324 return PTR_ERR(handle
);
3326 ret
= ext4_map_blocks(handle
, inode
, &map
,
3327 EXT4_GET_BLOCKS_CREATE_ZERO
);
3329 ext4_journal_stop(handle
);
3330 if (ret
== -ENOSPC
&&
3331 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3337 * If we added blocks beyond i_size, we need to make sure they
3338 * will get truncated if we crash before updating i_size in
3339 * ext4_iomap_end(). For faults we don't need to do that (and
3340 * even cannot because for orphan list operations inode_lock is
3341 * required) - if we happen to instantiate block beyond i_size,
3342 * it is because we race with truncate which has already added
3343 * the inode to the orphan list.
3345 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3346 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3349 err
= ext4_orphan_add(handle
, inode
);
3351 ext4_journal_stop(handle
);
3355 ext4_journal_stop(handle
);
3359 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3360 iomap
->offset
= first_block
<< blkbits
;
3363 iomap
->type
= IOMAP_HOLE
;
3364 iomap
->blkno
= IOMAP_NULL_BLOCK
;
3365 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3367 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3368 iomap
->type
= IOMAP_MAPPED
;
3369 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3370 iomap
->type
= IOMAP_UNWRITTEN
;
3375 iomap
->blkno
= (sector_t
)map
.m_pblk
<< (blkbits
- 9);
3376 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3379 if (map
.m_flags
& EXT4_MAP_NEW
)
3380 iomap
->flags
|= IOMAP_F_NEW
;
3384 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3385 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3389 int blkbits
= inode
->i_blkbits
;
3390 bool truncate
= false;
3392 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3395 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3396 if (IS_ERR(handle
)) {
3397 ret
= PTR_ERR(handle
);
3400 if (ext4_update_inode_size(inode
, offset
+ written
))
3401 ext4_mark_inode_dirty(handle
, inode
);
3403 * We may need to truncate allocated but not written blocks beyond EOF.
3405 if (iomap
->offset
+ iomap
->length
>
3406 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3407 ext4_lblk_t written_blk
, end_blk
;
3409 written_blk
= (offset
+ written
) >> blkbits
;
3410 end_blk
= (offset
+ length
) >> blkbits
;
3411 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3415 * Remove inode from orphan list if we were extending a inode and
3416 * everything went fine.
3418 if (!truncate
&& inode
->i_nlink
&&
3419 !list_empty(&EXT4_I(inode
)->i_orphan
))
3420 ext4_orphan_del(handle
, inode
);
3421 ext4_journal_stop(handle
);
3423 ext4_truncate_failed_write(inode
);
3426 * If truncate failed early the inode might still be on the
3427 * orphan list; we need to make sure the inode is removed from
3428 * the orphan list in that case.
3431 ext4_orphan_del(NULL
, inode
);
3436 struct iomap_ops ext4_iomap_ops
= {
3437 .iomap_begin
= ext4_iomap_begin
,
3438 .iomap_end
= ext4_iomap_end
,
3443 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3444 ssize_t size
, void *private)
3446 ext4_io_end_t
*io_end
= private;
3448 /* if not async direct IO just return */
3452 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3453 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3454 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3457 * Error during AIO DIO. We cannot convert unwritten extents as the
3458 * data was not written. Just clear the unwritten flag and drop io_end.
3461 ext4_clear_io_unwritten_flag(io_end
);
3464 io_end
->offset
= offset
;
3465 io_end
->size
= size
;
3466 ext4_put_io_end(io_end
);
3472 * Handling of direct IO writes.
3474 * For ext4 extent files, ext4 will do direct-io write even to holes,
3475 * preallocated extents, and those write extend the file, no need to
3476 * fall back to buffered IO.
3478 * For holes, we fallocate those blocks, mark them as unwritten
3479 * If those blocks were preallocated, we mark sure they are split, but
3480 * still keep the range to write as unwritten.
3482 * The unwritten extents will be converted to written when DIO is completed.
3483 * For async direct IO, since the IO may still pending when return, we
3484 * set up an end_io call back function, which will do the conversion
3485 * when async direct IO completed.
3487 * If the O_DIRECT write will extend the file then add this inode to the
3488 * orphan list. So recovery will truncate it back to the original size
3489 * if the machine crashes during the write.
3492 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3494 struct file
*file
= iocb
->ki_filp
;
3495 struct inode
*inode
= file
->f_mapping
->host
;
3496 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3498 loff_t offset
= iocb
->ki_pos
;
3499 size_t count
= iov_iter_count(iter
);
3501 get_block_t
*get_block_func
= NULL
;
3503 loff_t final_size
= offset
+ count
;
3507 if (final_size
> inode
->i_size
) {
3508 /* Credits for sb + inode write */
3509 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3510 if (IS_ERR(handle
)) {
3511 ret
= PTR_ERR(handle
);
3514 ret
= ext4_orphan_add(handle
, inode
);
3516 ext4_journal_stop(handle
);
3520 ei
->i_disksize
= inode
->i_size
;
3521 ext4_journal_stop(handle
);
3524 BUG_ON(iocb
->private == NULL
);
3527 * Make all waiters for direct IO properly wait also for extent
3528 * conversion. This also disallows race between truncate() and
3529 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3531 inode_dio_begin(inode
);
3533 /* If we do a overwrite dio, i_mutex locking can be released */
3534 overwrite
= *((int *)iocb
->private);
3537 inode_unlock(inode
);
3540 * For extent mapped files we could direct write to holes and fallocate.
3542 * Allocated blocks to fill the hole are marked as unwritten to prevent
3543 * parallel buffered read to expose the stale data before DIO complete
3546 * As to previously fallocated extents, ext4 get_block will just simply
3547 * mark the buffer mapped but still keep the extents unwritten.
3549 * For non AIO case, we will convert those unwritten extents to written
3550 * after return back from blockdev_direct_IO. That way we save us from
3551 * allocating io_end structure and also the overhead of offloading
3552 * the extent convertion to a workqueue.
3554 * For async DIO, the conversion needs to be deferred when the
3555 * IO is completed. The ext4 end_io callback function will be
3556 * called to take care of the conversion work. Here for async
3557 * case, we allocate an io_end structure to hook to the iocb.
3559 iocb
->private = NULL
;
3561 get_block_func
= ext4_dio_get_block_overwrite
;
3562 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3563 round_down(offset
, 1 << inode
->i_blkbits
) >= inode
->i_size
) {
3564 get_block_func
= ext4_dio_get_block
;
3565 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3566 } else if (is_sync_kiocb(iocb
)) {
3567 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3568 dio_flags
= DIO_LOCKING
;
3570 get_block_func
= ext4_dio_get_block_unwritten_async
;
3571 dio_flags
= DIO_LOCKING
;
3573 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3574 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3576 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3577 get_block_func
, ext4_end_io_dio
, NULL
,
3580 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3581 EXT4_STATE_DIO_UNWRITTEN
)) {
3584 * for non AIO case, since the IO is already
3585 * completed, we could do the conversion right here
3587 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3591 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3594 inode_dio_end(inode
);
3595 /* take i_mutex locking again if we do a ovewrite dio */
3599 if (ret
< 0 && final_size
> inode
->i_size
)
3600 ext4_truncate_failed_write(inode
);
3602 /* Handle extending of i_size after direct IO write */
3606 /* Credits for sb + inode write */
3607 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3608 if (IS_ERR(handle
)) {
3609 /* This is really bad luck. We've written the data
3610 * but cannot extend i_size. Bail out and pretend
3611 * the write failed... */
3612 ret
= PTR_ERR(handle
);
3614 ext4_orphan_del(NULL
, inode
);
3619 ext4_orphan_del(handle
, inode
);
3621 loff_t end
= offset
+ ret
;
3622 if (end
> inode
->i_size
) {
3623 ei
->i_disksize
= end
;
3624 i_size_write(inode
, end
);
3626 * We're going to return a positive `ret'
3627 * here due to non-zero-length I/O, so there's
3628 * no way of reporting error returns from
3629 * ext4_mark_inode_dirty() to userspace. So
3632 ext4_mark_inode_dirty(handle
, inode
);
3635 err
= ext4_journal_stop(handle
);
3643 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3645 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3646 struct inode
*inode
= mapping
->host
;
3647 size_t count
= iov_iter_count(iter
);
3651 * Shared inode_lock is enough for us - it protects against concurrent
3652 * writes & truncates and since we take care of writing back page cache,
3653 * we are protected against page writeback as well.
3655 inode_lock_shared(inode
);
3656 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3657 iocb
->ki_pos
+ count
);
3660 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3661 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3663 inode_unlock_shared(inode
);
3667 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3669 struct file
*file
= iocb
->ki_filp
;
3670 struct inode
*inode
= file
->f_mapping
->host
;
3671 size_t count
= iov_iter_count(iter
);
3672 loff_t offset
= iocb
->ki_pos
;
3675 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3676 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3681 * If we are doing data journalling we don't support O_DIRECT
3683 if (ext4_should_journal_data(inode
))
3686 /* Let buffer I/O handle the inline data case. */
3687 if (ext4_has_inline_data(inode
))
3690 /* DAX uses iomap path now */
3691 if (WARN_ON_ONCE(IS_DAX(inode
)))
3694 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3695 if (iov_iter_rw(iter
) == READ
)
3696 ret
= ext4_direct_IO_read(iocb
, iter
);
3698 ret
= ext4_direct_IO_write(iocb
, iter
);
3699 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3704 * Pages can be marked dirty completely asynchronously from ext4's journalling
3705 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3706 * much here because ->set_page_dirty is called under VFS locks. The page is
3707 * not necessarily locked.
3709 * We cannot just dirty the page and leave attached buffers clean, because the
3710 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3711 * or jbddirty because all the journalling code will explode.
3713 * So what we do is to mark the page "pending dirty" and next time writepage
3714 * is called, propagate that into the buffers appropriately.
3716 static int ext4_journalled_set_page_dirty(struct page
*page
)
3718 SetPageChecked(page
);
3719 return __set_page_dirty_nobuffers(page
);
3722 static int ext4_set_page_dirty(struct page
*page
)
3724 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3725 WARN_ON_ONCE(!page_has_buffers(page
));
3726 return __set_page_dirty_buffers(page
);
3729 static const struct address_space_operations ext4_aops
= {
3730 .readpage
= ext4_readpage
,
3731 .readpages
= ext4_readpages
,
3732 .writepage
= ext4_writepage
,
3733 .writepages
= ext4_writepages
,
3734 .write_begin
= ext4_write_begin
,
3735 .write_end
= ext4_write_end
,
3736 .set_page_dirty
= ext4_set_page_dirty
,
3738 .invalidatepage
= ext4_invalidatepage
,
3739 .releasepage
= ext4_releasepage
,
3740 .direct_IO
= ext4_direct_IO
,
3741 .migratepage
= buffer_migrate_page
,
3742 .is_partially_uptodate
= block_is_partially_uptodate
,
3743 .error_remove_page
= generic_error_remove_page
,
3746 static const struct address_space_operations ext4_journalled_aops
= {
3747 .readpage
= ext4_readpage
,
3748 .readpages
= ext4_readpages
,
3749 .writepage
= ext4_writepage
,
3750 .writepages
= ext4_writepages
,
3751 .write_begin
= ext4_write_begin
,
3752 .write_end
= ext4_journalled_write_end
,
3753 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3755 .invalidatepage
= ext4_journalled_invalidatepage
,
3756 .releasepage
= ext4_releasepage
,
3757 .direct_IO
= ext4_direct_IO
,
3758 .is_partially_uptodate
= block_is_partially_uptodate
,
3759 .error_remove_page
= generic_error_remove_page
,
3762 static const struct address_space_operations ext4_da_aops
= {
3763 .readpage
= ext4_readpage
,
3764 .readpages
= ext4_readpages
,
3765 .writepage
= ext4_writepage
,
3766 .writepages
= ext4_writepages
,
3767 .write_begin
= ext4_da_write_begin
,
3768 .write_end
= ext4_da_write_end
,
3769 .set_page_dirty
= ext4_set_page_dirty
,
3771 .invalidatepage
= ext4_da_invalidatepage
,
3772 .releasepage
= ext4_releasepage
,
3773 .direct_IO
= ext4_direct_IO
,
3774 .migratepage
= buffer_migrate_page
,
3775 .is_partially_uptodate
= block_is_partially_uptodate
,
3776 .error_remove_page
= generic_error_remove_page
,
3779 void ext4_set_aops(struct inode
*inode
)
3781 switch (ext4_inode_journal_mode(inode
)) {
3782 case EXT4_INODE_ORDERED_DATA_MODE
:
3783 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3785 case EXT4_INODE_JOURNAL_DATA_MODE
:
3786 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3791 if (test_opt(inode
->i_sb
, DELALLOC
))
3792 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3794 inode
->i_mapping
->a_ops
= &ext4_aops
;
3797 static int __ext4_block_zero_page_range(handle_t
*handle
,
3798 struct address_space
*mapping
, loff_t from
, loff_t length
)
3800 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3801 unsigned offset
= from
& (PAGE_SIZE
-1);
3802 unsigned blocksize
, pos
;
3804 struct inode
*inode
= mapping
->host
;
3805 struct buffer_head
*bh
;
3809 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3810 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3814 blocksize
= inode
->i_sb
->s_blocksize
;
3816 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3818 if (!page_has_buffers(page
))
3819 create_empty_buffers(page
, blocksize
, 0);
3821 /* Find the buffer that contains "offset" */
3822 bh
= page_buffers(page
);
3824 while (offset
>= pos
) {
3825 bh
= bh
->b_this_page
;
3829 if (buffer_freed(bh
)) {
3830 BUFFER_TRACE(bh
, "freed: skip");
3833 if (!buffer_mapped(bh
)) {
3834 BUFFER_TRACE(bh
, "unmapped");
3835 ext4_get_block(inode
, iblock
, bh
, 0);
3836 /* unmapped? It's a hole - nothing to do */
3837 if (!buffer_mapped(bh
)) {
3838 BUFFER_TRACE(bh
, "still unmapped");
3843 /* Ok, it's mapped. Make sure it's up-to-date */
3844 if (PageUptodate(page
))
3845 set_buffer_uptodate(bh
);
3847 if (!buffer_uptodate(bh
)) {
3849 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3851 /* Uhhuh. Read error. Complain and punt. */
3852 if (!buffer_uptodate(bh
))
3854 if (S_ISREG(inode
->i_mode
) &&
3855 ext4_encrypted_inode(inode
)) {
3856 /* We expect the key to be set. */
3857 BUG_ON(!fscrypt_has_encryption_key(inode
));
3858 BUG_ON(blocksize
!= PAGE_SIZE
);
3859 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
3860 page
, PAGE_SIZE
, 0, page
->index
));
3863 if (ext4_should_journal_data(inode
)) {
3864 BUFFER_TRACE(bh
, "get write access");
3865 err
= ext4_journal_get_write_access(handle
, bh
);
3869 zero_user(page
, offset
, length
);
3870 BUFFER_TRACE(bh
, "zeroed end of block");
3872 if (ext4_should_journal_data(inode
)) {
3873 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3876 mark_buffer_dirty(bh
);
3877 if (ext4_should_order_data(inode
))
3878 err
= ext4_jbd2_inode_add_write(handle
, inode
);
3888 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3889 * starting from file offset 'from'. The range to be zero'd must
3890 * be contained with in one block. If the specified range exceeds
3891 * the end of the block it will be shortened to end of the block
3892 * that cooresponds to 'from'
3894 static int ext4_block_zero_page_range(handle_t
*handle
,
3895 struct address_space
*mapping
, loff_t from
, loff_t length
)
3897 struct inode
*inode
= mapping
->host
;
3898 unsigned offset
= from
& (PAGE_SIZE
-1);
3899 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3900 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3903 * correct length if it does not fall between
3904 * 'from' and the end of the block
3906 if (length
> max
|| length
< 0)
3909 if (IS_DAX(inode
)) {
3910 return iomap_zero_range(inode
, from
, length
, NULL
,
3913 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3917 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3918 * up to the end of the block which corresponds to `from'.
3919 * This required during truncate. We need to physically zero the tail end
3920 * of that block so it doesn't yield old data if the file is later grown.
3922 static int ext4_block_truncate_page(handle_t
*handle
,
3923 struct address_space
*mapping
, loff_t from
)
3925 unsigned offset
= from
& (PAGE_SIZE
-1);
3928 struct inode
*inode
= mapping
->host
;
3930 /* If we are processing an encrypted inode during orphan list handling */
3931 if (ext4_encrypted_inode(inode
) && !fscrypt_has_encryption_key(inode
))
3934 blocksize
= inode
->i_sb
->s_blocksize
;
3935 length
= blocksize
- (offset
& (blocksize
- 1));
3937 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3940 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3941 loff_t lstart
, loff_t length
)
3943 struct super_block
*sb
= inode
->i_sb
;
3944 struct address_space
*mapping
= inode
->i_mapping
;
3945 unsigned partial_start
, partial_end
;
3946 ext4_fsblk_t start
, end
;
3947 loff_t byte_end
= (lstart
+ length
- 1);
3950 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3951 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3953 start
= lstart
>> sb
->s_blocksize_bits
;
3954 end
= byte_end
>> sb
->s_blocksize_bits
;
3956 /* Handle partial zero within the single block */
3958 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3959 err
= ext4_block_zero_page_range(handle
, mapping
,
3963 /* Handle partial zero out on the start of the range */
3964 if (partial_start
) {
3965 err
= ext4_block_zero_page_range(handle
, mapping
,
3966 lstart
, sb
->s_blocksize
);
3970 /* Handle partial zero out on the end of the range */
3971 if (partial_end
!= sb
->s_blocksize
- 1)
3972 err
= ext4_block_zero_page_range(handle
, mapping
,
3973 byte_end
- partial_end
,
3978 int ext4_can_truncate(struct inode
*inode
)
3980 if (S_ISREG(inode
->i_mode
))
3982 if (S_ISDIR(inode
->i_mode
))
3984 if (S_ISLNK(inode
->i_mode
))
3985 return !ext4_inode_is_fast_symlink(inode
);
3990 * We have to make sure i_disksize gets properly updated before we truncate
3991 * page cache due to hole punching or zero range. Otherwise i_disksize update
3992 * can get lost as it may have been postponed to submission of writeback but
3993 * that will never happen after we truncate page cache.
3995 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3999 loff_t size
= i_size_read(inode
);
4001 WARN_ON(!inode_is_locked(inode
));
4002 if (offset
> size
|| offset
+ len
< size
)
4005 if (EXT4_I(inode
)->i_disksize
>= size
)
4008 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4010 return PTR_ERR(handle
);
4011 ext4_update_i_disksize(inode
, size
);
4012 ext4_mark_inode_dirty(handle
, inode
);
4013 ext4_journal_stop(handle
);
4019 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4020 * associated with the given offset and length
4022 * @inode: File inode
4023 * @offset: The offset where the hole will begin
4024 * @len: The length of the hole
4026 * Returns: 0 on success or negative on failure
4029 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4031 struct super_block
*sb
= inode
->i_sb
;
4032 ext4_lblk_t first_block
, stop_block
;
4033 struct address_space
*mapping
= inode
->i_mapping
;
4034 loff_t first_block_offset
, last_block_offset
;
4036 unsigned int credits
;
4039 if (!S_ISREG(inode
->i_mode
))
4042 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4045 * Write out all dirty pages to avoid race conditions
4046 * Then release them.
4048 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4049 ret
= filemap_write_and_wait_range(mapping
, offset
,
4050 offset
+ length
- 1);
4057 /* No need to punch hole beyond i_size */
4058 if (offset
>= inode
->i_size
)
4062 * If the hole extends beyond i_size, set the hole
4063 * to end after the page that contains i_size
4065 if (offset
+ length
> inode
->i_size
) {
4066 length
= inode
->i_size
+
4067 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4071 if (offset
& (sb
->s_blocksize
- 1) ||
4072 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4074 * Attach jinode to inode for jbd2 if we do any zeroing of
4077 ret
= ext4_inode_attach_jinode(inode
);
4083 /* Wait all existing dio workers, newcomers will block on i_mutex */
4084 ext4_inode_block_unlocked_dio(inode
);
4085 inode_dio_wait(inode
);
4088 * Prevent page faults from reinstantiating pages we have released from
4091 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4092 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4093 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4095 /* Now release the pages and zero block aligned part of pages*/
4096 if (last_block_offset
> first_block_offset
) {
4097 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4100 truncate_pagecache_range(inode
, first_block_offset
,
4104 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4105 credits
= ext4_writepage_trans_blocks(inode
);
4107 credits
= ext4_blocks_for_truncate(inode
);
4108 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4109 if (IS_ERR(handle
)) {
4110 ret
= PTR_ERR(handle
);
4111 ext4_std_error(sb
, ret
);
4115 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4120 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4121 EXT4_BLOCK_SIZE_BITS(sb
);
4122 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4124 /* If there are no blocks to remove, return now */
4125 if (first_block
>= stop_block
)
4128 down_write(&EXT4_I(inode
)->i_data_sem
);
4129 ext4_discard_preallocations(inode
);
4131 ret
= ext4_es_remove_extent(inode
, first_block
,
4132 stop_block
- first_block
);
4134 up_write(&EXT4_I(inode
)->i_data_sem
);
4138 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4139 ret
= ext4_ext_remove_space(inode
, first_block
,
4142 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4145 up_write(&EXT4_I(inode
)->i_data_sem
);
4147 ext4_handle_sync(handle
);
4149 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4150 ext4_mark_inode_dirty(handle
, inode
);
4152 ext4_journal_stop(handle
);
4154 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4155 ext4_inode_resume_unlocked_dio(inode
);
4157 inode_unlock(inode
);
4161 int ext4_inode_attach_jinode(struct inode
*inode
)
4163 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4164 struct jbd2_inode
*jinode
;
4166 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4169 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4170 spin_lock(&inode
->i_lock
);
4173 spin_unlock(&inode
->i_lock
);
4176 ei
->jinode
= jinode
;
4177 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4180 spin_unlock(&inode
->i_lock
);
4181 if (unlikely(jinode
!= NULL
))
4182 jbd2_free_inode(jinode
);
4189 * We block out ext4_get_block() block instantiations across the entire
4190 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4191 * simultaneously on behalf of the same inode.
4193 * As we work through the truncate and commit bits of it to the journal there
4194 * is one core, guiding principle: the file's tree must always be consistent on
4195 * disk. We must be able to restart the truncate after a crash.
4197 * The file's tree may be transiently inconsistent in memory (although it
4198 * probably isn't), but whenever we close off and commit a journal transaction,
4199 * the contents of (the filesystem + the journal) must be consistent and
4200 * restartable. It's pretty simple, really: bottom up, right to left (although
4201 * left-to-right works OK too).
4203 * Note that at recovery time, journal replay occurs *before* the restart of
4204 * truncate against the orphan inode list.
4206 * The committed inode has the new, desired i_size (which is the same as
4207 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4208 * that this inode's truncate did not complete and it will again call
4209 * ext4_truncate() to have another go. So there will be instantiated blocks
4210 * to the right of the truncation point in a crashed ext4 filesystem. But
4211 * that's fine - as long as they are linked from the inode, the post-crash
4212 * ext4_truncate() run will find them and release them.
4214 int ext4_truncate(struct inode
*inode
)
4216 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4217 unsigned int credits
;
4220 struct address_space
*mapping
= inode
->i_mapping
;
4223 * There is a possibility that we're either freeing the inode
4224 * or it's a completely new inode. In those cases we might not
4225 * have i_mutex locked because it's not necessary.
4227 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4228 WARN_ON(!inode_is_locked(inode
));
4229 trace_ext4_truncate_enter(inode
);
4231 if (!ext4_can_truncate(inode
))
4234 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4236 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4237 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4239 if (ext4_has_inline_data(inode
)) {
4242 ext4_inline_data_truncate(inode
, &has_inline
);
4247 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4248 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4249 if (ext4_inode_attach_jinode(inode
) < 0)
4253 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4254 credits
= ext4_writepage_trans_blocks(inode
);
4256 credits
= ext4_blocks_for_truncate(inode
);
4258 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4260 return PTR_ERR(handle
);
4262 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4263 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4266 * We add the inode to the orphan list, so that if this
4267 * truncate spans multiple transactions, and we crash, we will
4268 * resume the truncate when the filesystem recovers. It also
4269 * marks the inode dirty, to catch the new size.
4271 * Implication: the file must always be in a sane, consistent
4272 * truncatable state while each transaction commits.
4274 err
= ext4_orphan_add(handle
, inode
);
4278 down_write(&EXT4_I(inode
)->i_data_sem
);
4280 ext4_discard_preallocations(inode
);
4282 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4283 err
= ext4_ext_truncate(handle
, inode
);
4285 ext4_ind_truncate(handle
, inode
);
4287 up_write(&ei
->i_data_sem
);
4292 ext4_handle_sync(handle
);
4296 * If this was a simple ftruncate() and the file will remain alive,
4297 * then we need to clear up the orphan record which we created above.
4298 * However, if this was a real unlink then we were called by
4299 * ext4_evict_inode(), and we allow that function to clean up the
4300 * orphan info for us.
4303 ext4_orphan_del(handle
, inode
);
4305 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4306 ext4_mark_inode_dirty(handle
, inode
);
4307 ext4_journal_stop(handle
);
4309 trace_ext4_truncate_exit(inode
);
4314 * ext4_get_inode_loc returns with an extra refcount against the inode's
4315 * underlying buffer_head on success. If 'in_mem' is true, we have all
4316 * data in memory that is needed to recreate the on-disk version of this
4319 static int __ext4_get_inode_loc(struct inode
*inode
,
4320 struct ext4_iloc
*iloc
, int in_mem
)
4322 struct ext4_group_desc
*gdp
;
4323 struct buffer_head
*bh
;
4324 struct super_block
*sb
= inode
->i_sb
;
4326 int inodes_per_block
, inode_offset
;
4329 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4330 return -EFSCORRUPTED
;
4332 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4333 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4338 * Figure out the offset within the block group inode table
4340 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4341 inode_offset
= ((inode
->i_ino
- 1) %
4342 EXT4_INODES_PER_GROUP(sb
));
4343 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4344 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4346 bh
= sb_getblk(sb
, block
);
4349 if (!buffer_uptodate(bh
)) {
4353 * If the buffer has the write error flag, we have failed
4354 * to write out another inode in the same block. In this
4355 * case, we don't have to read the block because we may
4356 * read the old inode data successfully.
4358 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4359 set_buffer_uptodate(bh
);
4361 if (buffer_uptodate(bh
)) {
4362 /* someone brought it uptodate while we waited */
4368 * If we have all information of the inode in memory and this
4369 * is the only valid inode in the block, we need not read the
4373 struct buffer_head
*bitmap_bh
;
4376 start
= inode_offset
& ~(inodes_per_block
- 1);
4378 /* Is the inode bitmap in cache? */
4379 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4380 if (unlikely(!bitmap_bh
))
4384 * If the inode bitmap isn't in cache then the
4385 * optimisation may end up performing two reads instead
4386 * of one, so skip it.
4388 if (!buffer_uptodate(bitmap_bh
)) {
4392 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4393 if (i
== inode_offset
)
4395 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4399 if (i
== start
+ inodes_per_block
) {
4400 /* all other inodes are free, so skip I/O */
4401 memset(bh
->b_data
, 0, bh
->b_size
);
4402 set_buffer_uptodate(bh
);
4410 * If we need to do any I/O, try to pre-readahead extra
4411 * blocks from the inode table.
4413 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4414 ext4_fsblk_t b
, end
, table
;
4416 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4418 table
= ext4_inode_table(sb
, gdp
);
4419 /* s_inode_readahead_blks is always a power of 2 */
4420 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4424 num
= EXT4_INODES_PER_GROUP(sb
);
4425 if (ext4_has_group_desc_csum(sb
))
4426 num
-= ext4_itable_unused_count(sb
, gdp
);
4427 table
+= num
/ inodes_per_block
;
4431 sb_breadahead(sb
, b
++);
4435 * There are other valid inodes in the buffer, this inode
4436 * has in-inode xattrs, or we don't have this inode in memory.
4437 * Read the block from disk.
4439 trace_ext4_load_inode(inode
);
4441 bh
->b_end_io
= end_buffer_read_sync
;
4442 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4444 if (!buffer_uptodate(bh
)) {
4445 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4446 "unable to read itable block");
4456 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4458 /* We have all inode data except xattrs in memory here. */
4459 return __ext4_get_inode_loc(inode
, iloc
,
4460 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4463 void ext4_set_inode_flags(struct inode
*inode
)
4465 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4466 unsigned int new_fl
= 0;
4468 if (flags
& EXT4_SYNC_FL
)
4470 if (flags
& EXT4_APPEND_FL
)
4472 if (flags
& EXT4_IMMUTABLE_FL
)
4473 new_fl
|= S_IMMUTABLE
;
4474 if (flags
& EXT4_NOATIME_FL
)
4475 new_fl
|= S_NOATIME
;
4476 if (flags
& EXT4_DIRSYNC_FL
)
4477 new_fl
|= S_DIRSYNC
;
4478 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
) &&
4479 !ext4_should_journal_data(inode
) && !ext4_has_inline_data(inode
) &&
4480 !ext4_encrypted_inode(inode
))
4482 inode_set_flags(inode
, new_fl
,
4483 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4486 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4487 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4489 unsigned int vfs_fl
;
4490 unsigned long old_fl
, new_fl
;
4493 vfs_fl
= ei
->vfs_inode
.i_flags
;
4494 old_fl
= ei
->i_flags
;
4495 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4496 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4498 if (vfs_fl
& S_SYNC
)
4499 new_fl
|= EXT4_SYNC_FL
;
4500 if (vfs_fl
& S_APPEND
)
4501 new_fl
|= EXT4_APPEND_FL
;
4502 if (vfs_fl
& S_IMMUTABLE
)
4503 new_fl
|= EXT4_IMMUTABLE_FL
;
4504 if (vfs_fl
& S_NOATIME
)
4505 new_fl
|= EXT4_NOATIME_FL
;
4506 if (vfs_fl
& S_DIRSYNC
)
4507 new_fl
|= EXT4_DIRSYNC_FL
;
4508 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4511 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4512 struct ext4_inode_info
*ei
)
4515 struct inode
*inode
= &(ei
->vfs_inode
);
4516 struct super_block
*sb
= inode
->i_sb
;
4518 if (ext4_has_feature_huge_file(sb
)) {
4519 /* we are using combined 48 bit field */
4520 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4521 le32_to_cpu(raw_inode
->i_blocks_lo
);
4522 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4523 /* i_blocks represent file system block size */
4524 return i_blocks
<< (inode
->i_blkbits
- 9);
4529 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4533 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4534 struct ext4_inode
*raw_inode
,
4535 struct ext4_inode_info
*ei
)
4537 __le32
*magic
= (void *)raw_inode
+
4538 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4539 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4540 EXT4_INODE_SIZE(inode
->i_sb
) &&
4541 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4542 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4543 ext4_find_inline_data_nolock(inode
);
4545 EXT4_I(inode
)->i_inline_off
= 0;
4548 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4550 if (!ext4_has_feature_project(inode
->i_sb
))
4552 *projid
= EXT4_I(inode
)->i_projid
;
4556 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4558 struct ext4_iloc iloc
;
4559 struct ext4_inode
*raw_inode
;
4560 struct ext4_inode_info
*ei
;
4561 struct inode
*inode
;
4562 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4570 inode
= iget_locked(sb
, ino
);
4572 return ERR_PTR(-ENOMEM
);
4573 if (!(inode
->i_state
& I_NEW
))
4579 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4582 raw_inode
= ext4_raw_inode(&iloc
);
4584 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4585 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4586 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4587 EXT4_INODE_SIZE(inode
->i_sb
) ||
4588 (ei
->i_extra_isize
& 3)) {
4589 EXT4_ERROR_INODE(inode
,
4590 "bad extra_isize %u (inode size %u)",
4592 EXT4_INODE_SIZE(inode
->i_sb
));
4593 ret
= -EFSCORRUPTED
;
4597 ei
->i_extra_isize
= 0;
4599 /* Precompute checksum seed for inode metadata */
4600 if (ext4_has_metadata_csum(sb
)) {
4601 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4603 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4604 __le32 gen
= raw_inode
->i_generation
;
4605 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4607 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4611 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4612 EXT4_ERROR_INODE(inode
, "checksum invalid");
4617 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4618 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4619 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4620 if (ext4_has_feature_project(sb
) &&
4621 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4622 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4623 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4625 i_projid
= EXT4_DEF_PROJID
;
4627 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4628 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4629 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4631 i_uid_write(inode
, i_uid
);
4632 i_gid_write(inode
, i_gid
);
4633 ei
->i_projid
= make_kprojid(sb
->s_user_ns
, i_projid
);
4634 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4636 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4637 ei
->i_inline_off
= 0;
4638 ei
->i_dir_start_lookup
= 0;
4639 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4640 /* We now have enough fields to check if the inode was active or not.
4641 * This is needed because nfsd might try to access dead inodes
4642 * the test is that same one that e2fsck uses
4643 * NeilBrown 1999oct15
4645 if (inode
->i_nlink
== 0) {
4646 if ((inode
->i_mode
== 0 ||
4647 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4648 ino
!= EXT4_BOOT_LOADER_INO
) {
4649 /* this inode is deleted */
4653 /* The only unlinked inodes we let through here have
4654 * valid i_mode and are being read by the orphan
4655 * recovery code: that's fine, we're about to complete
4656 * the process of deleting those.
4657 * OR it is the EXT4_BOOT_LOADER_INO which is
4658 * not initialized on a new filesystem. */
4660 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4661 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4662 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4663 if (ext4_has_feature_64bit(sb
))
4665 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4666 inode
->i_size
= ext4_isize(raw_inode
);
4667 if ((size
= i_size_read(inode
)) < 0) {
4668 EXT4_ERROR_INODE(inode
, "bad i_size value: %lld", size
);
4669 ret
= -EFSCORRUPTED
;
4672 ei
->i_disksize
= inode
->i_size
;
4674 ei
->i_reserved_quota
= 0;
4676 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4677 ei
->i_block_group
= iloc
.block_group
;
4678 ei
->i_last_alloc_group
= ~0;
4680 * NOTE! The in-memory inode i_data array is in little-endian order
4681 * even on big-endian machines: we do NOT byteswap the block numbers!
4683 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4684 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4685 INIT_LIST_HEAD(&ei
->i_orphan
);
4688 * Set transaction id's of transactions that have to be committed
4689 * to finish f[data]sync. We set them to currently running transaction
4690 * as we cannot be sure that the inode or some of its metadata isn't
4691 * part of the transaction - the inode could have been reclaimed and
4692 * now it is reread from disk.
4695 transaction_t
*transaction
;
4698 read_lock(&journal
->j_state_lock
);
4699 if (journal
->j_running_transaction
)
4700 transaction
= journal
->j_running_transaction
;
4702 transaction
= journal
->j_committing_transaction
;
4704 tid
= transaction
->t_tid
;
4706 tid
= journal
->j_commit_sequence
;
4707 read_unlock(&journal
->j_state_lock
);
4708 ei
->i_sync_tid
= tid
;
4709 ei
->i_datasync_tid
= tid
;
4712 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4713 if (ei
->i_extra_isize
== 0) {
4714 /* The extra space is currently unused. Use it. */
4715 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4716 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4717 EXT4_GOOD_OLD_INODE_SIZE
;
4719 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4723 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4724 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4725 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4726 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4728 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4729 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4730 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4731 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4733 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4738 if (ei
->i_file_acl
&&
4739 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4740 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4742 ret
= -EFSCORRUPTED
;
4744 } else if (!ext4_has_inline_data(inode
)) {
4745 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4746 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4747 (S_ISLNK(inode
->i_mode
) &&
4748 !ext4_inode_is_fast_symlink(inode
))))
4749 /* Validate extent which is part of inode */
4750 ret
= ext4_ext_check_inode(inode
);
4751 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4752 (S_ISLNK(inode
->i_mode
) &&
4753 !ext4_inode_is_fast_symlink(inode
))) {
4754 /* Validate block references which are part of inode */
4755 ret
= ext4_ind_check_inode(inode
);
4761 if (S_ISREG(inode
->i_mode
)) {
4762 inode
->i_op
= &ext4_file_inode_operations
;
4763 inode
->i_fop
= &ext4_file_operations
;
4764 ext4_set_aops(inode
);
4765 } else if (S_ISDIR(inode
->i_mode
)) {
4766 inode
->i_op
= &ext4_dir_inode_operations
;
4767 inode
->i_fop
= &ext4_dir_operations
;
4768 } else if (S_ISLNK(inode
->i_mode
)) {
4769 if (ext4_encrypted_inode(inode
)) {
4770 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4771 ext4_set_aops(inode
);
4772 } else if (ext4_inode_is_fast_symlink(inode
)) {
4773 inode
->i_link
= (char *)ei
->i_data
;
4774 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4775 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4776 sizeof(ei
->i_data
) - 1);
4778 inode
->i_op
= &ext4_symlink_inode_operations
;
4779 ext4_set_aops(inode
);
4781 inode_nohighmem(inode
);
4782 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4783 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4784 inode
->i_op
= &ext4_special_inode_operations
;
4785 if (raw_inode
->i_block
[0])
4786 init_special_inode(inode
, inode
->i_mode
,
4787 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4789 init_special_inode(inode
, inode
->i_mode
,
4790 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4791 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4792 make_bad_inode(inode
);
4794 ret
= -EFSCORRUPTED
;
4795 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4799 ext4_set_inode_flags(inode
);
4800 unlock_new_inode(inode
);
4806 return ERR_PTR(ret
);
4809 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4811 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4812 return ERR_PTR(-EFSCORRUPTED
);
4813 return ext4_iget(sb
, ino
);
4816 static int ext4_inode_blocks_set(handle_t
*handle
,
4817 struct ext4_inode
*raw_inode
,
4818 struct ext4_inode_info
*ei
)
4820 struct inode
*inode
= &(ei
->vfs_inode
);
4821 u64 i_blocks
= inode
->i_blocks
;
4822 struct super_block
*sb
= inode
->i_sb
;
4824 if (i_blocks
<= ~0U) {
4826 * i_blocks can be represented in a 32 bit variable
4827 * as multiple of 512 bytes
4829 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4830 raw_inode
->i_blocks_high
= 0;
4831 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4834 if (!ext4_has_feature_huge_file(sb
))
4837 if (i_blocks
<= 0xffffffffffffULL
) {
4839 * i_blocks can be represented in a 48 bit variable
4840 * as multiple of 512 bytes
4842 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4843 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4844 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4846 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4847 /* i_block is stored in file system block size */
4848 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4849 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4850 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4855 struct other_inode
{
4856 unsigned long orig_ino
;
4857 struct ext4_inode
*raw_inode
;
4860 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4863 struct other_inode
*oi
= (struct other_inode
*) data
;
4865 if ((inode
->i_ino
!= ino
) ||
4866 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4867 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4868 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4870 spin_lock(&inode
->i_lock
);
4871 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4872 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4873 (inode
->i_state
& I_DIRTY_TIME
)) {
4874 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4876 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4877 spin_unlock(&inode
->i_lock
);
4879 spin_lock(&ei
->i_raw_lock
);
4880 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4881 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4882 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4883 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4884 spin_unlock(&ei
->i_raw_lock
);
4885 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4888 spin_unlock(&inode
->i_lock
);
4893 * Opportunistically update the other time fields for other inodes in
4894 * the same inode table block.
4896 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4897 unsigned long orig_ino
, char *buf
)
4899 struct other_inode oi
;
4901 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4902 int inode_size
= EXT4_INODE_SIZE(sb
);
4904 oi
.orig_ino
= orig_ino
;
4906 * Calculate the first inode in the inode table block. Inode
4907 * numbers are one-based. That is, the first inode in a block
4908 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4910 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4911 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4912 if (ino
== orig_ino
)
4914 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4915 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4920 * Post the struct inode info into an on-disk inode location in the
4921 * buffer-cache. This gobbles the caller's reference to the
4922 * buffer_head in the inode location struct.
4924 * The caller must have write access to iloc->bh.
4926 static int ext4_do_update_inode(handle_t
*handle
,
4927 struct inode
*inode
,
4928 struct ext4_iloc
*iloc
)
4930 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4931 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4932 struct buffer_head
*bh
= iloc
->bh
;
4933 struct super_block
*sb
= inode
->i_sb
;
4934 int err
= 0, rc
, block
;
4935 int need_datasync
= 0, set_large_file
= 0;
4940 spin_lock(&ei
->i_raw_lock
);
4942 /* For fields not tracked in the in-memory inode,
4943 * initialise them to zero for new inodes. */
4944 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4945 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4947 ext4_get_inode_flags(ei
);
4948 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4949 i_uid
= i_uid_read(inode
);
4950 i_gid
= i_gid_read(inode
);
4951 i_projid
= from_kprojid(sb
->s_user_ns
, ei
->i_projid
);
4952 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4953 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4954 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4956 * Fix up interoperability with old kernels. Otherwise, old inodes get
4957 * re-used with the upper 16 bits of the uid/gid intact
4959 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4960 raw_inode
->i_uid_high
= 0;
4961 raw_inode
->i_gid_high
= 0;
4963 raw_inode
->i_uid_high
=
4964 cpu_to_le16(high_16_bits(i_uid
));
4965 raw_inode
->i_gid_high
=
4966 cpu_to_le16(high_16_bits(i_gid
));
4969 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4970 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4971 raw_inode
->i_uid_high
= 0;
4972 raw_inode
->i_gid_high
= 0;
4974 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4976 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4977 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4978 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4979 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4981 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4983 spin_unlock(&ei
->i_raw_lock
);
4986 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4987 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4988 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4989 raw_inode
->i_file_acl_high
=
4990 cpu_to_le16(ei
->i_file_acl
>> 32);
4991 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4992 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4993 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4996 if (ei
->i_disksize
> 0x7fffffffULL
) {
4997 if (!ext4_has_feature_large_file(sb
) ||
4998 EXT4_SB(sb
)->s_es
->s_rev_level
==
4999 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5002 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5003 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5004 if (old_valid_dev(inode
->i_rdev
)) {
5005 raw_inode
->i_block
[0] =
5006 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5007 raw_inode
->i_block
[1] = 0;
5009 raw_inode
->i_block
[0] = 0;
5010 raw_inode
->i_block
[1] =
5011 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5012 raw_inode
->i_block
[2] = 0;
5014 } else if (!ext4_has_inline_data(inode
)) {
5015 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5016 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5019 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5020 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5021 if (ei
->i_extra_isize
) {
5022 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5023 raw_inode
->i_version_hi
=
5024 cpu_to_le32(inode
->i_version
>> 32);
5025 raw_inode
->i_extra_isize
=
5026 cpu_to_le16(ei
->i_extra_isize
);
5030 if (i_projid
!= (projid_t
)-1) {
5031 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5032 i_projid
!= EXT4_DEF_PROJID
);
5034 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5035 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5036 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5039 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5040 spin_unlock(&ei
->i_raw_lock
);
5041 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
5042 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5045 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5046 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5049 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5050 if (set_large_file
) {
5051 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5052 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5055 ext4_update_dynamic_rev(sb
);
5056 ext4_set_feature_large_file(sb
);
5057 ext4_handle_sync(handle
);
5058 err
= ext4_handle_dirty_super(handle
, sb
);
5060 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5063 ext4_std_error(inode
->i_sb
, err
);
5068 * ext4_write_inode()
5070 * We are called from a few places:
5072 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5073 * Here, there will be no transaction running. We wait for any running
5074 * transaction to commit.
5076 * - Within flush work (sys_sync(), kupdate and such).
5077 * We wait on commit, if told to.
5079 * - Within iput_final() -> write_inode_now()
5080 * We wait on commit, if told to.
5082 * In all cases it is actually safe for us to return without doing anything,
5083 * because the inode has been copied into a raw inode buffer in
5084 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5087 * Note that we are absolutely dependent upon all inode dirtiers doing the
5088 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5089 * which we are interested.
5091 * It would be a bug for them to not do this. The code:
5093 * mark_inode_dirty(inode)
5095 * inode->i_size = expr;
5097 * is in error because write_inode() could occur while `stuff()' is running,
5098 * and the new i_size will be lost. Plus the inode will no longer be on the
5099 * superblock's dirty inode list.
5101 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5105 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
5108 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5109 if (ext4_journal_current_handle()) {
5110 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5116 * No need to force transaction in WB_SYNC_NONE mode. Also
5117 * ext4_sync_fs() will force the commit after everything is
5120 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5123 err
= ext4_force_commit(inode
->i_sb
);
5125 struct ext4_iloc iloc
;
5127 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5131 * sync(2) will flush the whole buffer cache. No need to do
5132 * it here separately for each inode.
5134 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5135 sync_dirty_buffer(iloc
.bh
);
5136 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5137 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5138 "IO error syncing inode");
5147 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5148 * buffers that are attached to a page stradding i_size and are undergoing
5149 * commit. In that case we have to wait for commit to finish and try again.
5151 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5155 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5156 tid_t commit_tid
= 0;
5159 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5161 * All buffers in the last page remain valid? Then there's nothing to
5162 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5165 if (offset
> PAGE_SIZE
- (1 << inode
->i_blkbits
))
5168 page
= find_lock_page(inode
->i_mapping
,
5169 inode
->i_size
>> PAGE_SHIFT
);
5172 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5173 PAGE_SIZE
- offset
);
5179 read_lock(&journal
->j_state_lock
);
5180 if (journal
->j_committing_transaction
)
5181 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5182 read_unlock(&journal
->j_state_lock
);
5184 jbd2_log_wait_commit(journal
, commit_tid
);
5191 * Called from notify_change.
5193 * We want to trap VFS attempts to truncate the file as soon as
5194 * possible. In particular, we want to make sure that when the VFS
5195 * shrinks i_size, we put the inode on the orphan list and modify
5196 * i_disksize immediately, so that during the subsequent flushing of
5197 * dirty pages and freeing of disk blocks, we can guarantee that any
5198 * commit will leave the blocks being flushed in an unused state on
5199 * disk. (On recovery, the inode will get truncated and the blocks will
5200 * be freed, so we have a strong guarantee that no future commit will
5201 * leave these blocks visible to the user.)
5203 * Another thing we have to assure is that if we are in ordered mode
5204 * and inode is still attached to the committing transaction, we must
5205 * we start writeout of all the dirty pages which are being truncated.
5206 * This way we are sure that all the data written in the previous
5207 * transaction are already on disk (truncate waits for pages under
5210 * Called with inode->i_mutex down.
5212 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5214 struct inode
*inode
= d_inode(dentry
);
5217 const unsigned int ia_valid
= attr
->ia_valid
;
5219 error
= setattr_prepare(dentry
, attr
);
5223 if (is_quota_modification(inode
, attr
)) {
5224 error
= dquot_initialize(inode
);
5228 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5229 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5232 /* (user+group)*(old+new) structure, inode write (sb,
5233 * inode block, ? - but truncate inode update has it) */
5234 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5235 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5236 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5237 if (IS_ERR(handle
)) {
5238 error
= PTR_ERR(handle
);
5241 error
= dquot_transfer(inode
, attr
);
5243 ext4_journal_stop(handle
);
5246 /* Update corresponding info in inode so that everything is in
5247 * one transaction */
5248 if (attr
->ia_valid
& ATTR_UID
)
5249 inode
->i_uid
= attr
->ia_uid
;
5250 if (attr
->ia_valid
& ATTR_GID
)
5251 inode
->i_gid
= attr
->ia_gid
;
5252 error
= ext4_mark_inode_dirty(handle
, inode
);
5253 ext4_journal_stop(handle
);
5256 if (attr
->ia_valid
& ATTR_SIZE
) {
5258 loff_t oldsize
= inode
->i_size
;
5259 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5261 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5262 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5264 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5267 if (!S_ISREG(inode
->i_mode
))
5270 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5271 inode_inc_iversion(inode
);
5273 if (ext4_should_order_data(inode
) &&
5274 (attr
->ia_size
< inode
->i_size
)) {
5275 error
= ext4_begin_ordered_truncate(inode
,
5280 if (attr
->ia_size
!= inode
->i_size
) {
5281 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5282 if (IS_ERR(handle
)) {
5283 error
= PTR_ERR(handle
);
5286 if (ext4_handle_valid(handle
) && shrink
) {
5287 error
= ext4_orphan_add(handle
, inode
);
5291 * Update c/mtime on truncate up, ext4_truncate() will
5292 * update c/mtime in shrink case below
5295 inode
->i_mtime
= current_time(inode
);
5296 inode
->i_ctime
= inode
->i_mtime
;
5298 down_write(&EXT4_I(inode
)->i_data_sem
);
5299 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5300 rc
= ext4_mark_inode_dirty(handle
, inode
);
5304 * We have to update i_size under i_data_sem together
5305 * with i_disksize to avoid races with writeback code
5306 * running ext4_wb_update_i_disksize().
5309 i_size_write(inode
, attr
->ia_size
);
5310 up_write(&EXT4_I(inode
)->i_data_sem
);
5311 ext4_journal_stop(handle
);
5314 ext4_orphan_del(NULL
, inode
);
5319 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5322 * Blocks are going to be removed from the inode. Wait
5323 * for dio in flight. Temporarily disable
5324 * dioread_nolock to prevent livelock.
5327 if (!ext4_should_journal_data(inode
)) {
5328 ext4_inode_block_unlocked_dio(inode
);
5329 inode_dio_wait(inode
);
5330 ext4_inode_resume_unlocked_dio(inode
);
5332 ext4_wait_for_tail_page_commit(inode
);
5334 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5336 * Truncate pagecache after we've waited for commit
5337 * in data=journal mode to make pages freeable.
5339 truncate_pagecache(inode
, inode
->i_size
);
5341 rc
= ext4_truncate(inode
);
5345 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5349 setattr_copy(inode
, attr
);
5350 mark_inode_dirty(inode
);
5354 * If the call to ext4_truncate failed to get a transaction handle at
5355 * all, we need to clean up the in-core orphan list manually.
5357 if (orphan
&& inode
->i_nlink
)
5358 ext4_orphan_del(NULL
, inode
);
5360 if (!error
&& (ia_valid
& ATTR_MODE
))
5361 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5364 ext4_std_error(inode
->i_sb
, error
);
5370 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5373 struct inode
*inode
;
5374 unsigned long long delalloc_blocks
;
5376 inode
= d_inode(dentry
);
5377 generic_fillattr(inode
, stat
);
5380 * If there is inline data in the inode, the inode will normally not
5381 * have data blocks allocated (it may have an external xattr block).
5382 * Report at least one sector for such files, so tools like tar, rsync,
5383 * others doen't incorrectly think the file is completely sparse.
5385 if (unlikely(ext4_has_inline_data(inode
)))
5386 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5389 * We can't update i_blocks if the block allocation is delayed
5390 * otherwise in the case of system crash before the real block
5391 * allocation is done, we will have i_blocks inconsistent with
5392 * on-disk file blocks.
5393 * We always keep i_blocks updated together with real
5394 * allocation. But to not confuse with user, stat
5395 * will return the blocks that include the delayed allocation
5396 * blocks for this file.
5398 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5399 EXT4_I(inode
)->i_reserved_data_blocks
);
5400 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5404 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5407 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5408 return ext4_ind_trans_blocks(inode
, lblocks
);
5409 return ext4_ext_index_trans_blocks(inode
, pextents
);
5413 * Account for index blocks, block groups bitmaps and block group
5414 * descriptor blocks if modify datablocks and index blocks
5415 * worse case, the indexs blocks spread over different block groups
5417 * If datablocks are discontiguous, they are possible to spread over
5418 * different block groups too. If they are contiguous, with flexbg,
5419 * they could still across block group boundary.
5421 * Also account for superblock, inode, quota and xattr blocks
5423 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5426 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5432 * How many index blocks need to touch to map @lblocks logical blocks
5433 * to @pextents physical extents?
5435 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5440 * Now let's see how many group bitmaps and group descriptors need
5443 groups
= idxblocks
+ pextents
;
5445 if (groups
> ngroups
)
5447 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5448 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5450 /* bitmaps and block group descriptor blocks */
5451 ret
+= groups
+ gdpblocks
;
5453 /* Blocks for super block, inode, quota and xattr blocks */
5454 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5460 * Calculate the total number of credits to reserve to fit
5461 * the modification of a single pages into a single transaction,
5462 * which may include multiple chunks of block allocations.
5464 * This could be called via ext4_write_begin()
5466 * We need to consider the worse case, when
5467 * one new block per extent.
5469 int ext4_writepage_trans_blocks(struct inode
*inode
)
5471 int bpp
= ext4_journal_blocks_per_page(inode
);
5474 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5476 /* Account for data blocks for journalled mode */
5477 if (ext4_should_journal_data(inode
))
5483 * Calculate the journal credits for a chunk of data modification.
5485 * This is called from DIO, fallocate or whoever calling
5486 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5488 * journal buffers for data blocks are not included here, as DIO
5489 * and fallocate do no need to journal data buffers.
5491 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5493 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5497 * The caller must have previously called ext4_reserve_inode_write().
5498 * Give this, we know that the caller already has write access to iloc->bh.
5500 int ext4_mark_iloc_dirty(handle_t
*handle
,
5501 struct inode
*inode
, struct ext4_iloc
*iloc
)
5505 if (IS_I_VERSION(inode
))
5506 inode_inc_iversion(inode
);
5508 /* the do_update_inode consumes one bh->b_count */
5511 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5512 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5518 * On success, We end up with an outstanding reference count against
5519 * iloc->bh. This _must_ be cleaned up later.
5523 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5524 struct ext4_iloc
*iloc
)
5528 err
= ext4_get_inode_loc(inode
, iloc
);
5530 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5531 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5537 ext4_std_error(inode
->i_sb
, err
);
5542 * Expand an inode by new_extra_isize bytes.
5543 * Returns 0 on success or negative error number on failure.
5545 static int ext4_expand_extra_isize(struct inode
*inode
,
5546 unsigned int new_extra_isize
,
5547 struct ext4_iloc iloc
,
5550 struct ext4_inode
*raw_inode
;
5551 struct ext4_xattr_ibody_header
*header
;
5553 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5556 raw_inode
= ext4_raw_inode(&iloc
);
5558 header
= IHDR(inode
, raw_inode
);
5560 /* No extended attributes present */
5561 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5562 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5563 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5565 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5569 /* try to expand with EAs present */
5570 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5575 * What we do here is to mark the in-core inode as clean with respect to inode
5576 * dirtiness (it may still be data-dirty).
5577 * This means that the in-core inode may be reaped by prune_icache
5578 * without having to perform any I/O. This is a very good thing,
5579 * because *any* task may call prune_icache - even ones which
5580 * have a transaction open against a different journal.
5582 * Is this cheating? Not really. Sure, we haven't written the
5583 * inode out, but prune_icache isn't a user-visible syncing function.
5584 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5585 * we start and wait on commits.
5587 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5589 struct ext4_iloc iloc
;
5590 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5591 static unsigned int mnt_count
;
5595 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5596 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5599 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5600 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5602 * In nojournal mode, we can immediately attempt to expand
5603 * the inode. When journaled, we first need to obtain extra
5604 * buffer credits since we may write into the EA block
5605 * with this same handle. If journal_extend fails, then it will
5606 * only result in a minor loss of functionality for that inode.
5607 * If this is felt to be critical, then e2fsck should be run to
5608 * force a large enough s_min_extra_isize.
5610 if (!ext4_handle_valid(handle
) ||
5611 jbd2_journal_extend(handle
,
5612 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) == 0) {
5613 ret
= ext4_expand_extra_isize(inode
,
5614 sbi
->s_want_extra_isize
,
5618 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5619 ext4_warning(inode
->i_sb
,
5620 "Unable to expand inode %lu. Delete"
5621 " some EAs or run e2fsck.",
5624 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5629 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5633 * ext4_dirty_inode() is called from __mark_inode_dirty()
5635 * We're really interested in the case where a file is being extended.
5636 * i_size has been changed by generic_commit_write() and we thus need
5637 * to include the updated inode in the current transaction.
5639 * Also, dquot_alloc_block() will always dirty the inode when blocks
5640 * are allocated to the file.
5642 * If the inode is marked synchronous, we don't honour that here - doing
5643 * so would cause a commit on atime updates, which we don't bother doing.
5644 * We handle synchronous inodes at the highest possible level.
5646 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5647 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5648 * to copy into the on-disk inode structure are the timestamp files.
5650 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5654 if (flags
== I_DIRTY_TIME
)
5656 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5660 ext4_mark_inode_dirty(handle
, inode
);
5662 ext4_journal_stop(handle
);
5669 * Bind an inode's backing buffer_head into this transaction, to prevent
5670 * it from being flushed to disk early. Unlike
5671 * ext4_reserve_inode_write, this leaves behind no bh reference and
5672 * returns no iloc structure, so the caller needs to repeat the iloc
5673 * lookup to mark the inode dirty later.
5675 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5677 struct ext4_iloc iloc
;
5681 err
= ext4_get_inode_loc(inode
, &iloc
);
5683 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5684 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5686 err
= ext4_handle_dirty_metadata(handle
,
5692 ext4_std_error(inode
->i_sb
, err
);
5697 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5702 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5705 * We have to be very careful here: changing a data block's
5706 * journaling status dynamically is dangerous. If we write a
5707 * data block to the journal, change the status and then delete
5708 * that block, we risk forgetting to revoke the old log record
5709 * from the journal and so a subsequent replay can corrupt data.
5710 * So, first we make sure that the journal is empty and that
5711 * nobody is changing anything.
5714 journal
= EXT4_JOURNAL(inode
);
5717 if (is_journal_aborted(journal
))
5720 /* Wait for all existing dio workers */
5721 ext4_inode_block_unlocked_dio(inode
);
5722 inode_dio_wait(inode
);
5725 * Before flushing the journal and switching inode's aops, we have
5726 * to flush all dirty data the inode has. There can be outstanding
5727 * delayed allocations, there can be unwritten extents created by
5728 * fallocate or buffered writes in dioread_nolock mode covered by
5729 * dirty data which can be converted only after flushing the dirty
5730 * data (and journalled aops don't know how to handle these cases).
5733 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5734 err
= filemap_write_and_wait(inode
->i_mapping
);
5736 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5737 ext4_inode_resume_unlocked_dio(inode
);
5742 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5743 jbd2_journal_lock_updates(journal
);
5746 * OK, there are no updates running now, and all cached data is
5747 * synced to disk. We are now in a completely consistent state
5748 * which doesn't have anything in the journal, and we know that
5749 * no filesystem updates are running, so it is safe to modify
5750 * the inode's in-core data-journaling state flag now.
5754 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5756 err
= jbd2_journal_flush(journal
);
5758 jbd2_journal_unlock_updates(journal
);
5759 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5760 ext4_inode_resume_unlocked_dio(inode
);
5763 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5765 ext4_set_aops(inode
);
5767 * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated.
5768 * E.g. S_DAX may get cleared / set.
5770 ext4_set_inode_flags(inode
);
5772 jbd2_journal_unlock_updates(journal
);
5773 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5776 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5777 ext4_inode_resume_unlocked_dio(inode
);
5779 /* Finally we can mark the inode as dirty. */
5781 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5783 return PTR_ERR(handle
);
5785 err
= ext4_mark_inode_dirty(handle
, inode
);
5786 ext4_handle_sync(handle
);
5787 ext4_journal_stop(handle
);
5788 ext4_std_error(inode
->i_sb
, err
);
5793 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5795 return !buffer_mapped(bh
);
5798 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5800 struct page
*page
= vmf
->page
;
5804 struct file
*file
= vma
->vm_file
;
5805 struct inode
*inode
= file_inode(file
);
5806 struct address_space
*mapping
= inode
->i_mapping
;
5808 get_block_t
*get_block
;
5811 sb_start_pagefault(inode
->i_sb
);
5812 file_update_time(vma
->vm_file
);
5814 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5816 ret
= ext4_convert_inline_data(inode
);
5820 /* Delalloc case is easy... */
5821 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5822 !ext4_should_journal_data(inode
) &&
5823 !ext4_nonda_switch(inode
->i_sb
)) {
5825 ret
= block_page_mkwrite(vma
, vmf
,
5826 ext4_da_get_block_prep
);
5827 } while (ret
== -ENOSPC
&&
5828 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5833 size
= i_size_read(inode
);
5834 /* Page got truncated from under us? */
5835 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5837 ret
= VM_FAULT_NOPAGE
;
5841 if (page
->index
== size
>> PAGE_SHIFT
)
5842 len
= size
& ~PAGE_MASK
;
5846 * Return if we have all the buffers mapped. This avoids the need to do
5847 * journal_start/journal_stop which can block and take a long time
5849 if (page_has_buffers(page
)) {
5850 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5852 ext4_bh_unmapped
)) {
5853 /* Wait so that we don't change page under IO */
5854 wait_for_stable_page(page
);
5855 ret
= VM_FAULT_LOCKED
;
5860 /* OK, we need to fill the hole... */
5861 if (ext4_should_dioread_nolock(inode
))
5862 get_block
= ext4_get_block_unwritten
;
5864 get_block
= ext4_get_block
;
5866 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5867 ext4_writepage_trans_blocks(inode
));
5868 if (IS_ERR(handle
)) {
5869 ret
= VM_FAULT_SIGBUS
;
5872 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5873 if (!ret
&& ext4_should_journal_data(inode
)) {
5874 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5875 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
5877 ret
= VM_FAULT_SIGBUS
;
5878 ext4_journal_stop(handle
);
5881 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5883 ext4_journal_stop(handle
);
5884 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5887 ret
= block_page_mkwrite_return(ret
);
5889 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5890 sb_end_pagefault(inode
->i_sb
);
5894 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5896 struct inode
*inode
= file_inode(vma
->vm_file
);
5899 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5900 err
= filemap_fault(vma
, vmf
);
5901 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5907 * Find the first extent at or after @lblk in an inode that is not a hole.
5908 * Search for @map_len blocks at most. The extent is returned in @result.
5910 * The function returns 1 if we found an extent. The function returns 0 in
5911 * case there is no extent at or after @lblk and in that case also sets
5912 * @result->es_len to 0. In case of error, the error code is returned.
5914 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
5915 unsigned int map_len
, struct extent_status
*result
)
5917 struct ext4_map_blocks map
;
5918 struct extent_status es
= {};
5922 map
.m_len
= map_len
;
5925 * For non-extent based files this loop may iterate several times since
5926 * we do not determine full hole size.
5928 while (map
.m_len
> 0) {
5929 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
5932 /* There's extent covering m_lblk? Just return it. */
5936 ext4_es_store_pblock(result
, map
.m_pblk
);
5937 result
->es_lblk
= map
.m_lblk
;
5938 result
->es_len
= map
.m_len
;
5939 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
5940 status
= EXTENT_STATUS_UNWRITTEN
;
5942 status
= EXTENT_STATUS_WRITTEN
;
5943 ext4_es_store_status(result
, status
);
5946 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
5947 map
.m_lblk
+ map
.m_len
- 1,
5949 /* Is delalloc data before next block in extent tree? */
5950 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
5951 ext4_lblk_t offset
= 0;
5953 if (es
.es_lblk
< lblk
)
5954 offset
= lblk
- es
.es_lblk
;
5955 result
->es_lblk
= es
.es_lblk
+ offset
;
5956 ext4_es_store_pblock(result
,
5957 ext4_es_pblock(&es
) + offset
);
5958 result
->es_len
= es
.es_len
- offset
;
5959 ext4_es_store_status(result
, ext4_es_status(&es
));
5963 /* There's a hole at m_lblk, advance us after it */
5964 map
.m_lblk
+= map
.m_len
;
5965 map_len
-= map
.m_len
;
5966 map
.m_len
= map_len
;