1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
53 struct ext4_inode_info
*ei
)
55 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
59 unsigned int csum_size
= sizeof(dummy_csum
);
61 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
62 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
64 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
65 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
67 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
68 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
69 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
70 EXT4_GOOD_OLD_INODE_SIZE
,
71 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
72 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
73 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
77 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
78 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
84 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
85 struct ext4_inode_info
*ei
)
87 __u32 provided
, calculated
;
89 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
90 cpu_to_le32(EXT4_OS_LINUX
) ||
91 !ext4_has_metadata_csum(inode
->i_sb
))
94 provided
= le16_to_cpu(raw
->i_checksum_lo
);
95 calculated
= ext4_inode_csum(inode
, raw
, ei
);
96 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
97 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
98 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
100 calculated
&= 0xFFFF;
102 return provided
== calculated
;
105 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
106 struct ext4_inode_info
*ei
)
110 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
111 cpu_to_le32(EXT4_OS_LINUX
) ||
112 !ext4_has_metadata_csum(inode
->i_sb
))
115 csum
= ext4_inode_csum(inode
, raw
, ei
);
116 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
117 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
118 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
119 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
125 trace_ext4_begin_ordered_truncate(inode
, new_size
);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode
)->jinode
)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
135 EXT4_I(inode
)->jinode
,
139 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
140 unsigned int length
);
141 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
142 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
143 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
147 * Test whether an inode is a fast symlink.
148 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 int ext4_inode_is_fast_symlink(struct inode
*inode
)
152 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
153 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
154 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
156 if (ext4_has_inline_data(inode
))
159 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
161 return S_ISLNK(inode
->i_mode
) && inode
->i_size
&&
162 (inode
->i_size
< EXT4_N_BLOCKS
* 4);
166 * Restart the transaction associated with *handle. This does a commit,
167 * so before we call here everything must be consistently dirtied against
170 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
176 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
177 * moment, get_block can be called only for blocks inside i_size since
178 * page cache has been already dropped and writes are blocked by
179 * i_mutex. So we can safely drop the i_data_sem here.
181 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
182 jbd_debug(2, "restarting handle %p\n", handle
);
183 up_write(&EXT4_I(inode
)->i_data_sem
);
184 ret
= ext4_journal_restart(handle
, nblocks
);
185 down_write(&EXT4_I(inode
)->i_data_sem
);
186 ext4_discard_preallocations(inode
);
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_evict_inode(struct inode
*inode
)
199 * Credits for final inode cleanup and freeing:
200 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
201 * (xattr block freeing), bitmap, group descriptor (inode freeing)
203 int extra_credits
= 6;
204 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
206 trace_ext4_evict_inode(inode
);
208 if (inode
->i_nlink
) {
210 * When journalling data dirty buffers are tracked only in the
211 * journal. So although mm thinks everything is clean and
212 * ready for reaping the inode might still have some pages to
213 * write in the running transaction or waiting to be
214 * checkpointed. Thus calling jbd2_journal_invalidatepage()
215 * (via truncate_inode_pages()) to discard these buffers can
216 * cause data loss. Also even if we did not discard these
217 * buffers, we would have no way to find them after the inode
218 * is reaped and thus user could see stale data if he tries to
219 * read them before the transaction is checkpointed. So be
220 * careful and force everything to disk here... We use
221 * ei->i_datasync_tid to store the newest transaction
222 * containing inode's data.
224 * Note that directories do not have this problem because they
225 * don't use page cache.
227 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
228 ext4_should_journal_data(inode
) &&
229 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
230 inode
->i_data
.nrpages
) {
231 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
232 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
234 jbd2_complete_transaction(journal
, commit_tid
);
235 filemap_write_and_wait(&inode
->i_data
);
237 truncate_inode_pages_final(&inode
->i_data
);
242 if (is_bad_inode(inode
))
244 dquot_initialize(inode
);
246 if (ext4_should_order_data(inode
))
247 ext4_begin_ordered_truncate(inode
, 0);
248 truncate_inode_pages_final(&inode
->i_data
);
251 * Protect us against freezing - iput() caller didn't have to have any
252 * protection against it
254 sb_start_intwrite(inode
->i_sb
);
256 if (!IS_NOQUOTA(inode
))
257 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
260 * Block bitmap, group descriptor, and inode are accounted in both
261 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
263 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
264 ext4_blocks_for_truncate(inode
) + extra_credits
- 3);
265 if (IS_ERR(handle
)) {
266 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
268 * If we're going to skip the normal cleanup, we still need to
269 * make sure that the in-core orphan linked list is properly
272 ext4_orphan_del(NULL
, inode
);
273 sb_end_intwrite(inode
->i_sb
);
278 ext4_handle_sync(handle
);
281 * Set inode->i_size to 0 before calling ext4_truncate(). We need
282 * special handling of symlinks here because i_size is used to
283 * determine whether ext4_inode_info->i_data contains symlink data or
284 * block mappings. Setting i_size to 0 will remove its fast symlink
285 * status. Erase i_data so that it becomes a valid empty block map.
287 if (ext4_inode_is_fast_symlink(inode
))
288 memset(EXT4_I(inode
)->i_data
, 0, sizeof(EXT4_I(inode
)->i_data
));
290 err
= ext4_mark_inode_dirty(handle
, inode
);
292 ext4_warning(inode
->i_sb
,
293 "couldn't mark inode dirty (err %d)", err
);
296 if (inode
->i_blocks
) {
297 err
= ext4_truncate(inode
);
299 ext4_error(inode
->i_sb
,
300 "couldn't truncate inode %lu (err %d)",
306 /* Remove xattr references. */
307 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
310 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
312 ext4_journal_stop(handle
);
313 ext4_orphan_del(NULL
, inode
);
314 sb_end_intwrite(inode
->i_sb
);
315 ext4_xattr_inode_array_free(ea_inode_array
);
320 * Kill off the orphan record which ext4_truncate created.
321 * AKPM: I think this can be inside the above `if'.
322 * Note that ext4_orphan_del() has to be able to cope with the
323 * deletion of a non-existent orphan - this is because we don't
324 * know if ext4_truncate() actually created an orphan record.
325 * (Well, we could do this if we need to, but heck - it works)
327 ext4_orphan_del(handle
, inode
);
328 EXT4_I(inode
)->i_dtime
= get_seconds();
331 * One subtle ordering requirement: if anything has gone wrong
332 * (transaction abort, IO errors, whatever), then we can still
333 * do these next steps (the fs will already have been marked as
334 * having errors), but we can't free the inode if the mark_dirty
337 if (ext4_mark_inode_dirty(handle
, inode
))
338 /* If that failed, just do the required in-core inode clear. */
339 ext4_clear_inode(inode
);
341 ext4_free_inode(handle
, inode
);
342 ext4_journal_stop(handle
);
343 sb_end_intwrite(inode
->i_sb
);
344 ext4_xattr_inode_array_free(ea_inode_array
);
347 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
351 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
353 return &EXT4_I(inode
)->i_reserved_quota
;
358 * Called with i_data_sem down, which is important since we can call
359 * ext4_discard_preallocations() from here.
361 void ext4_da_update_reserve_space(struct inode
*inode
,
362 int used
, int quota_claim
)
364 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
365 struct ext4_inode_info
*ei
= EXT4_I(inode
);
367 spin_lock(&ei
->i_block_reservation_lock
);
368 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
369 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
370 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
371 "with only %d reserved data blocks",
372 __func__
, inode
->i_ino
, used
,
373 ei
->i_reserved_data_blocks
);
375 used
= ei
->i_reserved_data_blocks
;
378 /* Update per-inode reservations */
379 ei
->i_reserved_data_blocks
-= used
;
380 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
382 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
384 /* Update quota subsystem for data blocks */
386 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
389 * We did fallocate with an offset that is already delayed
390 * allocated. So on delayed allocated writeback we should
391 * not re-claim the quota for fallocated blocks.
393 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
397 * If we have done all the pending block allocations and if
398 * there aren't any writers on the inode, we can discard the
399 * inode's preallocations.
401 if ((ei
->i_reserved_data_blocks
== 0) &&
402 (atomic_read(&inode
->i_writecount
) == 0))
403 ext4_discard_preallocations(inode
);
406 static int __check_block_validity(struct inode
*inode
, const char *func
,
408 struct ext4_map_blocks
*map
)
410 if (ext4_has_feature_journal(inode
->i_sb
) &&
412 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_journal_inum
)))
414 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
416 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
417 "lblock %lu mapped to illegal pblock %llu "
418 "(length %d)", (unsigned long) map
->m_lblk
,
419 map
->m_pblk
, map
->m_len
);
420 return -EFSCORRUPTED
;
425 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
430 if (ext4_encrypted_inode(inode
))
431 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
433 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
440 #define check_block_validity(inode, map) \
441 __check_block_validity((inode), __func__, __LINE__, (map))
443 #ifdef ES_AGGRESSIVE_TEST
444 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
446 struct ext4_map_blocks
*es_map
,
447 struct ext4_map_blocks
*map
,
454 * There is a race window that the result is not the same.
455 * e.g. xfstests #223 when dioread_nolock enables. The reason
456 * is that we lookup a block mapping in extent status tree with
457 * out taking i_data_sem. So at the time the unwritten extent
458 * could be converted.
460 down_read(&EXT4_I(inode
)->i_data_sem
);
461 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
462 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
463 EXT4_GET_BLOCKS_KEEP_SIZE
);
465 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
466 EXT4_GET_BLOCKS_KEEP_SIZE
);
468 up_read((&EXT4_I(inode
)->i_data_sem
));
471 * We don't check m_len because extent will be collpased in status
472 * tree. So the m_len might not equal.
474 if (es_map
->m_lblk
!= map
->m_lblk
||
475 es_map
->m_flags
!= map
->m_flags
||
476 es_map
->m_pblk
!= map
->m_pblk
) {
477 printk("ES cache assertion failed for inode: %lu "
478 "es_cached ex [%d/%d/%llu/%x] != "
479 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
480 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
481 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
482 map
->m_len
, map
->m_pblk
, map
->m_flags
,
486 #endif /* ES_AGGRESSIVE_TEST */
489 * The ext4_map_blocks() function tries to look up the requested blocks,
490 * and returns if the blocks are already mapped.
492 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
493 * and store the allocated blocks in the result buffer head and mark it
496 * If file type is extents based, it will call ext4_ext_map_blocks(),
497 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
500 * On success, it returns the number of blocks being mapped or allocated. if
501 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
502 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
504 * It returns 0 if plain look up failed (blocks have not been allocated), in
505 * that case, @map is returned as unmapped but we still do fill map->m_len to
506 * indicate the length of a hole starting at map->m_lblk.
508 * It returns the error in case of allocation failure.
510 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
511 struct ext4_map_blocks
*map
, int flags
)
513 struct extent_status es
;
516 #ifdef ES_AGGRESSIVE_TEST
517 struct ext4_map_blocks orig_map
;
519 memcpy(&orig_map
, map
, sizeof(*map
));
523 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
524 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
525 (unsigned long) map
->m_lblk
);
528 * ext4_map_blocks returns an int, and m_len is an unsigned int
530 if (unlikely(map
->m_len
> INT_MAX
))
531 map
->m_len
= INT_MAX
;
533 /* We can handle the block number less than EXT_MAX_BLOCKS */
534 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
535 return -EFSCORRUPTED
;
537 /* Lookup extent status tree firstly */
538 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
539 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
540 map
->m_pblk
= ext4_es_pblock(&es
) +
541 map
->m_lblk
- es
.es_lblk
;
542 map
->m_flags
|= ext4_es_is_written(&es
) ?
543 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
544 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
545 if (retval
> map
->m_len
)
548 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
550 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
551 if (retval
> map
->m_len
)
558 #ifdef ES_AGGRESSIVE_TEST
559 ext4_map_blocks_es_recheck(handle
, inode
, map
,
566 * Try to see if we can get the block without requesting a new
569 down_read(&EXT4_I(inode
)->i_data_sem
);
570 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
571 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
572 EXT4_GET_BLOCKS_KEEP_SIZE
);
574 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
575 EXT4_GET_BLOCKS_KEEP_SIZE
);
580 if (unlikely(retval
!= map
->m_len
)) {
581 ext4_warning(inode
->i_sb
,
582 "ES len assertion failed for inode "
583 "%lu: retval %d != map->m_len %d",
584 inode
->i_ino
, retval
, map
->m_len
);
588 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
589 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
590 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
591 !(status
& EXTENT_STATUS_WRITTEN
) &&
592 ext4_find_delalloc_range(inode
, map
->m_lblk
,
593 map
->m_lblk
+ map
->m_len
- 1))
594 status
|= EXTENT_STATUS_DELAYED
;
595 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
596 map
->m_len
, map
->m_pblk
, status
);
600 up_read((&EXT4_I(inode
)->i_data_sem
));
603 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
604 ret
= check_block_validity(inode
, map
);
609 /* If it is only a block(s) look up */
610 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
614 * Returns if the blocks have already allocated
616 * Note that if blocks have been preallocated
617 * ext4_ext_get_block() returns the create = 0
618 * with buffer head unmapped.
620 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
622 * If we need to convert extent to unwritten
623 * we continue and do the actual work in
624 * ext4_ext_map_blocks()
626 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
630 * Here we clear m_flags because after allocating an new extent,
631 * it will be set again.
633 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
636 * New blocks allocate and/or writing to unwritten extent
637 * will possibly result in updating i_data, so we take
638 * the write lock of i_data_sem, and call get_block()
639 * with create == 1 flag.
641 down_write(&EXT4_I(inode
)->i_data_sem
);
644 * We need to check for EXT4 here because migrate
645 * could have changed the inode type in between
647 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
648 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
650 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
652 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
654 * We allocated new blocks which will result in
655 * i_data's format changing. Force the migrate
656 * to fail by clearing migrate flags
658 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
662 * Update reserved blocks/metadata blocks after successful
663 * block allocation which had been deferred till now. We don't
664 * support fallocate for non extent files. So we can update
665 * reserve space here.
668 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
669 ext4_da_update_reserve_space(inode
, retval
, 1);
675 if (unlikely(retval
!= map
->m_len
)) {
676 ext4_warning(inode
->i_sb
,
677 "ES len assertion failed for inode "
678 "%lu: retval %d != map->m_len %d",
679 inode
->i_ino
, retval
, map
->m_len
);
684 * We have to zeroout blocks before inserting them into extent
685 * status tree. Otherwise someone could look them up there and
686 * use them before they are really zeroed. We also have to
687 * unmap metadata before zeroing as otherwise writeback can
688 * overwrite zeros with stale data from block device.
690 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
691 map
->m_flags
& EXT4_MAP_MAPPED
&&
692 map
->m_flags
& EXT4_MAP_NEW
) {
693 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
695 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
696 map
->m_pblk
, map
->m_len
);
704 * If the extent has been zeroed out, we don't need to update
705 * extent status tree.
707 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
708 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
709 if (ext4_es_is_written(&es
))
712 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
713 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
714 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
715 !(status
& EXTENT_STATUS_WRITTEN
) &&
716 ext4_find_delalloc_range(inode
, map
->m_lblk
,
717 map
->m_lblk
+ map
->m_len
- 1))
718 status
|= EXTENT_STATUS_DELAYED
;
719 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
720 map
->m_pblk
, status
);
728 up_write((&EXT4_I(inode
)->i_data_sem
));
729 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
730 ret
= check_block_validity(inode
, map
);
735 * Inodes with freshly allocated blocks where contents will be
736 * visible after transaction commit must be on transaction's
739 if (map
->m_flags
& EXT4_MAP_NEW
&&
740 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
741 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
742 !ext4_is_quota_file(inode
) &&
743 ext4_should_order_data(inode
)) {
745 (loff_t
)map
->m_lblk
<< inode
->i_blkbits
;
746 loff_t length
= (loff_t
)map
->m_len
<< inode
->i_blkbits
;
748 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
749 ret
= ext4_jbd2_inode_add_wait(handle
, inode
,
752 ret
= ext4_jbd2_inode_add_write(handle
, inode
,
762 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
763 * we have to be careful as someone else may be manipulating b_state as well.
765 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
767 unsigned long old_state
;
768 unsigned long new_state
;
770 flags
&= EXT4_MAP_FLAGS
;
772 /* Dummy buffer_head? Set non-atomically. */
774 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
778 * Someone else may be modifying b_state. Be careful! This is ugly but
779 * once we get rid of using bh as a container for mapping information
780 * to pass to / from get_block functions, this can go away.
783 old_state
= READ_ONCE(bh
->b_state
);
784 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
786 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
789 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
790 struct buffer_head
*bh
, int flags
)
792 struct ext4_map_blocks map
;
795 if (ext4_has_inline_data(inode
))
799 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
801 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
804 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
805 ext4_update_bh_state(bh
, map
.m_flags
);
806 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
808 } else if (ret
== 0) {
809 /* hole case, need to fill in bh->b_size */
810 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
815 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
816 struct buffer_head
*bh
, int create
)
818 return _ext4_get_block(inode
, iblock
, bh
,
819 create
? EXT4_GET_BLOCKS_CREATE
: 0);
823 * Get block function used when preparing for buffered write if we require
824 * creating an unwritten extent if blocks haven't been allocated. The extent
825 * will be converted to written after the IO is complete.
827 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
828 struct buffer_head
*bh_result
, int create
)
830 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
831 inode
->i_ino
, create
);
832 return _ext4_get_block(inode
, iblock
, bh_result
,
833 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
836 /* Maximum number of blocks we map for direct IO at once. */
837 #define DIO_MAX_BLOCKS 4096
840 * Get blocks function for the cases that need to start a transaction -
841 * generally difference cases of direct IO and DAX IO. It also handles retries
844 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
845 struct buffer_head
*bh_result
, int flags
)
852 /* Trim mapping request to maximum we can map at once for DIO */
853 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
854 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
855 dio_credits
= ext4_chunk_trans_blocks(inode
,
856 bh_result
->b_size
>> inode
->i_blkbits
);
858 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
860 return PTR_ERR(handle
);
862 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
863 ext4_journal_stop(handle
);
865 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
870 /* Get block function for DIO reads and writes to inodes without extents */
871 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
872 struct buffer_head
*bh
, int create
)
874 /* We don't expect handle for direct IO */
875 WARN_ON_ONCE(ext4_journal_current_handle());
878 return _ext4_get_block(inode
, iblock
, bh
, 0);
879 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
883 * Get block function for AIO DIO writes when we create unwritten extent if
884 * blocks are not allocated yet. The extent will be converted to written
885 * after IO is complete.
887 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
888 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
892 /* We don't expect handle for direct IO */
893 WARN_ON_ONCE(ext4_journal_current_handle());
895 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
896 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
899 * When doing DIO using unwritten extents, we need io_end to convert
900 * unwritten extents to written on IO completion. We allocate io_end
901 * once we spot unwritten extent and store it in b_private. Generic
902 * DIO code keeps b_private set and furthermore passes the value to
903 * our completion callback in 'private' argument.
905 if (!ret
&& buffer_unwritten(bh_result
)) {
906 if (!bh_result
->b_private
) {
907 ext4_io_end_t
*io_end
;
909 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
912 bh_result
->b_private
= io_end
;
913 ext4_set_io_unwritten_flag(inode
, io_end
);
915 set_buffer_defer_completion(bh_result
);
922 * Get block function for non-AIO DIO writes when we create unwritten extent if
923 * blocks are not allocated yet. The extent will be converted to written
924 * after IO is complete by ext4_direct_IO_write().
926 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
927 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
931 /* We don't expect handle for direct IO */
932 WARN_ON_ONCE(ext4_journal_current_handle());
934 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
935 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
938 * Mark inode as having pending DIO writes to unwritten extents.
939 * ext4_direct_IO_write() checks this flag and converts extents to
942 if (!ret
&& buffer_unwritten(bh_result
))
943 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
948 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
949 struct buffer_head
*bh_result
, int create
)
953 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
954 inode
->i_ino
, create
);
955 /* We don't expect handle for direct IO */
956 WARN_ON_ONCE(ext4_journal_current_handle());
958 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
960 * Blocks should have been preallocated! ext4_file_write_iter() checks
963 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
970 * `handle' can be NULL if create is zero
972 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
973 ext4_lblk_t block
, int map_flags
)
975 struct ext4_map_blocks map
;
976 struct buffer_head
*bh
;
977 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
980 J_ASSERT(handle
!= NULL
|| create
== 0);
984 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
987 return create
? ERR_PTR(-ENOSPC
) : NULL
;
991 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
993 return ERR_PTR(-ENOMEM
);
994 if (map
.m_flags
& EXT4_MAP_NEW
) {
995 J_ASSERT(create
!= 0);
996 J_ASSERT(handle
!= NULL
);
999 * Now that we do not always journal data, we should
1000 * keep in mind whether this should always journal the
1001 * new buffer as metadata. For now, regular file
1002 * writes use ext4_get_block instead, so it's not a
1006 BUFFER_TRACE(bh
, "call get_create_access");
1007 err
= ext4_journal_get_create_access(handle
, bh
);
1008 if (unlikely(err
)) {
1012 if (!buffer_uptodate(bh
)) {
1013 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1014 set_buffer_uptodate(bh
);
1017 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1018 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1022 BUFFER_TRACE(bh
, "not a new buffer");
1026 return ERR_PTR(err
);
1029 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1030 ext4_lblk_t block
, int map_flags
)
1032 struct buffer_head
*bh
;
1034 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1037 if (!bh
|| buffer_uptodate(bh
))
1039 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1041 if (buffer_uptodate(bh
))
1044 return ERR_PTR(-EIO
);
1047 /* Read a contiguous batch of blocks. */
1048 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
1049 bool wait
, struct buffer_head
**bhs
)
1053 for (i
= 0; i
< bh_count
; i
++) {
1054 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
1055 if (IS_ERR(bhs
[i
])) {
1056 err
= PTR_ERR(bhs
[i
]);
1062 for (i
= 0; i
< bh_count
; i
++)
1063 /* Note that NULL bhs[i] is valid because of holes. */
1064 if (bhs
[i
] && !buffer_uptodate(bhs
[i
]))
1065 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1,
1071 for (i
= 0; i
< bh_count
; i
++)
1073 wait_on_buffer(bhs
[i
]);
1075 for (i
= 0; i
< bh_count
; i
++) {
1076 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
1084 for (i
= 0; i
< bh_count
; i
++) {
1091 int ext4_walk_page_buffers(handle_t
*handle
,
1092 struct buffer_head
*head
,
1096 int (*fn
)(handle_t
*handle
,
1097 struct buffer_head
*bh
))
1099 struct buffer_head
*bh
;
1100 unsigned block_start
, block_end
;
1101 unsigned blocksize
= head
->b_size
;
1103 struct buffer_head
*next
;
1105 for (bh
= head
, block_start
= 0;
1106 ret
== 0 && (bh
!= head
|| !block_start
);
1107 block_start
= block_end
, bh
= next
) {
1108 next
= bh
->b_this_page
;
1109 block_end
= block_start
+ blocksize
;
1110 if (block_end
<= from
|| block_start
>= to
) {
1111 if (partial
&& !buffer_uptodate(bh
))
1115 err
= (*fn
)(handle
, bh
);
1123 * To preserve ordering, it is essential that the hole instantiation and
1124 * the data write be encapsulated in a single transaction. We cannot
1125 * close off a transaction and start a new one between the ext4_get_block()
1126 * and the commit_write(). So doing the jbd2_journal_start at the start of
1127 * prepare_write() is the right place.
1129 * Also, this function can nest inside ext4_writepage(). In that case, we
1130 * *know* that ext4_writepage() has generated enough buffer credits to do the
1131 * whole page. So we won't block on the journal in that case, which is good,
1132 * because the caller may be PF_MEMALLOC.
1134 * By accident, ext4 can be reentered when a transaction is open via
1135 * quota file writes. If we were to commit the transaction while thus
1136 * reentered, there can be a deadlock - we would be holding a quota
1137 * lock, and the commit would never complete if another thread had a
1138 * transaction open and was blocking on the quota lock - a ranking
1141 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1142 * will _not_ run commit under these circumstances because handle->h_ref
1143 * is elevated. We'll still have enough credits for the tiny quotafile
1146 int do_journal_get_write_access(handle_t
*handle
,
1147 struct buffer_head
*bh
)
1149 int dirty
= buffer_dirty(bh
);
1152 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1155 * __block_write_begin() could have dirtied some buffers. Clean
1156 * the dirty bit as jbd2_journal_get_write_access() could complain
1157 * otherwise about fs integrity issues. Setting of the dirty bit
1158 * by __block_write_begin() isn't a real problem here as we clear
1159 * the bit before releasing a page lock and thus writeback cannot
1160 * ever write the buffer.
1163 clear_buffer_dirty(bh
);
1164 BUFFER_TRACE(bh
, "get write access");
1165 ret
= ext4_journal_get_write_access(handle
, bh
);
1167 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1171 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1172 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1173 get_block_t
*get_block
)
1175 unsigned from
= pos
& (PAGE_SIZE
- 1);
1176 unsigned to
= from
+ len
;
1177 struct inode
*inode
= page
->mapping
->host
;
1178 unsigned block_start
, block_end
;
1181 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1183 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1184 bool decrypt
= false;
1186 BUG_ON(!PageLocked(page
));
1187 BUG_ON(from
> PAGE_SIZE
);
1188 BUG_ON(to
> PAGE_SIZE
);
1191 if (!page_has_buffers(page
))
1192 create_empty_buffers(page
, blocksize
, 0);
1193 head
= page_buffers(page
);
1194 bbits
= ilog2(blocksize
);
1195 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1197 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1198 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1199 block_end
= block_start
+ blocksize
;
1200 if (block_end
<= from
|| block_start
>= to
) {
1201 if (PageUptodate(page
)) {
1202 if (!buffer_uptodate(bh
))
1203 set_buffer_uptodate(bh
);
1208 clear_buffer_new(bh
);
1209 if (!buffer_mapped(bh
)) {
1210 WARN_ON(bh
->b_size
!= blocksize
);
1211 err
= get_block(inode
, block
, bh
, 1);
1214 if (buffer_new(bh
)) {
1215 clean_bdev_bh_alias(bh
);
1216 if (PageUptodate(page
)) {
1217 clear_buffer_new(bh
);
1218 set_buffer_uptodate(bh
);
1219 mark_buffer_dirty(bh
);
1222 if (block_end
> to
|| block_start
< from
)
1223 zero_user_segments(page
, to
, block_end
,
1228 if (PageUptodate(page
)) {
1229 if (!buffer_uptodate(bh
))
1230 set_buffer_uptodate(bh
);
1233 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1234 !buffer_unwritten(bh
) &&
1235 (block_start
< from
|| block_end
> to
)) {
1236 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1238 decrypt
= ext4_encrypted_inode(inode
) &&
1239 S_ISREG(inode
->i_mode
);
1243 * If we issued read requests, let them complete.
1245 while (wait_bh
> wait
) {
1246 wait_on_buffer(*--wait_bh
);
1247 if (!buffer_uptodate(*wait_bh
))
1251 page_zero_new_buffers(page
, from
, to
);
1253 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1254 PAGE_SIZE
, 0, page
->index
);
1259 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1260 loff_t pos
, unsigned len
, unsigned flags
,
1261 struct page
**pagep
, void **fsdata
)
1263 struct inode
*inode
= mapping
->host
;
1264 int ret
, needed_blocks
;
1271 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1274 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1276 * Reserve one block more for addition to orphan list in case
1277 * we allocate blocks but write fails for some reason
1279 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1280 index
= pos
>> PAGE_SHIFT
;
1281 from
= pos
& (PAGE_SIZE
- 1);
1284 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1285 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1294 * grab_cache_page_write_begin() can take a long time if the
1295 * system is thrashing due to memory pressure, or if the page
1296 * is being written back. So grab it first before we start
1297 * the transaction handle. This also allows us to allocate
1298 * the page (if needed) without using GFP_NOFS.
1301 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1307 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1308 if (IS_ERR(handle
)) {
1310 return PTR_ERR(handle
);
1314 if (page
->mapping
!= mapping
) {
1315 /* The page got truncated from under us */
1318 ext4_journal_stop(handle
);
1321 /* In case writeback began while the page was unlocked */
1322 wait_for_stable_page(page
);
1324 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1325 if (ext4_should_dioread_nolock(inode
))
1326 ret
= ext4_block_write_begin(page
, pos
, len
,
1327 ext4_get_block_unwritten
);
1329 ret
= ext4_block_write_begin(page
, pos
, len
,
1332 if (ext4_should_dioread_nolock(inode
))
1333 ret
= __block_write_begin(page
, pos
, len
,
1334 ext4_get_block_unwritten
);
1336 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1338 if (!ret
&& ext4_should_journal_data(inode
)) {
1339 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1341 do_journal_get_write_access
);
1347 * __block_write_begin may have instantiated a few blocks
1348 * outside i_size. Trim these off again. Don't need
1349 * i_size_read because we hold i_mutex.
1351 * Add inode to orphan list in case we crash before
1354 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1355 ext4_orphan_add(handle
, inode
);
1357 ext4_journal_stop(handle
);
1358 if (pos
+ len
> inode
->i_size
) {
1359 ext4_truncate_failed_write(inode
);
1361 * If truncate failed early the inode might
1362 * still be on the orphan list; we need to
1363 * make sure the inode is removed from the
1364 * orphan list in that case.
1367 ext4_orphan_del(NULL
, inode
);
1370 if (ret
== -ENOSPC
&&
1371 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1380 /* For write_end() in data=journal mode */
1381 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1384 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1386 set_buffer_uptodate(bh
);
1387 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1388 clear_buffer_meta(bh
);
1389 clear_buffer_prio(bh
);
1394 * We need to pick up the new inode size which generic_commit_write gave us
1395 * `file' can be NULL - eg, when called from page_symlink().
1397 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1398 * buffers are managed internally.
1400 static int ext4_write_end(struct file
*file
,
1401 struct address_space
*mapping
,
1402 loff_t pos
, unsigned len
, unsigned copied
,
1403 struct page
*page
, void *fsdata
)
1405 handle_t
*handle
= ext4_journal_current_handle();
1406 struct inode
*inode
= mapping
->host
;
1407 loff_t old_size
= inode
->i_size
;
1409 int i_size_changed
= 0;
1410 int inline_data
= ext4_has_inline_data(inode
);
1412 trace_ext4_write_end(inode
, pos
, len
, copied
);
1414 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1423 copied
= block_write_end(file
, mapping
, pos
,
1424 len
, copied
, page
, fsdata
);
1426 * it's important to update i_size while still holding page lock:
1427 * page writeout could otherwise come in and zero beyond i_size.
1429 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1434 pagecache_isize_extended(inode
, old_size
, pos
);
1436 * Don't mark the inode dirty under page lock. First, it unnecessarily
1437 * makes the holding time of page lock longer. Second, it forces lock
1438 * ordering of page lock and transaction start for journaling
1441 if (i_size_changed
|| inline_data
)
1442 ext4_mark_inode_dirty(handle
, inode
);
1444 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1445 /* if we have allocated more blocks and copied
1446 * less. We will have blocks allocated outside
1447 * inode->i_size. So truncate them
1449 ext4_orphan_add(handle
, inode
);
1451 ret2
= ext4_journal_stop(handle
);
1455 if (pos
+ len
> inode
->i_size
) {
1456 ext4_truncate_failed_write(inode
);
1458 * If truncate failed early the inode might still be
1459 * on the orphan list; we need to make sure the inode
1460 * is removed from the orphan list in that case.
1463 ext4_orphan_del(NULL
, inode
);
1466 return ret
? ret
: copied
;
1470 * This is a private version of page_zero_new_buffers() which doesn't
1471 * set the buffer to be dirty, since in data=journalled mode we need
1472 * to call ext4_handle_dirty_metadata() instead.
1474 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1476 unsigned from
, unsigned to
)
1478 unsigned int block_start
= 0, block_end
;
1479 struct buffer_head
*head
, *bh
;
1481 bh
= head
= page_buffers(page
);
1483 block_end
= block_start
+ bh
->b_size
;
1484 if (buffer_new(bh
)) {
1485 if (block_end
> from
&& block_start
< to
) {
1486 if (!PageUptodate(page
)) {
1487 unsigned start
, size
;
1489 start
= max(from
, block_start
);
1490 size
= min(to
, block_end
) - start
;
1492 zero_user(page
, start
, size
);
1493 write_end_fn(handle
, bh
);
1495 clear_buffer_new(bh
);
1498 block_start
= block_end
;
1499 bh
= bh
->b_this_page
;
1500 } while (bh
!= head
);
1503 static int ext4_journalled_write_end(struct file
*file
,
1504 struct address_space
*mapping
,
1505 loff_t pos
, unsigned len
, unsigned copied
,
1506 struct page
*page
, void *fsdata
)
1508 handle_t
*handle
= ext4_journal_current_handle();
1509 struct inode
*inode
= mapping
->host
;
1510 loff_t old_size
= inode
->i_size
;
1514 int size_changed
= 0;
1515 int inline_data
= ext4_has_inline_data(inode
);
1517 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1518 from
= pos
& (PAGE_SIZE
- 1);
1521 BUG_ON(!ext4_handle_valid(handle
));
1524 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1532 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1534 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1536 if (unlikely(copied
< len
))
1537 ext4_journalled_zero_new_buffers(handle
, page
,
1539 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1540 from
+ copied
, &partial
,
1543 SetPageUptodate(page
);
1545 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1546 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1547 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1552 pagecache_isize_extended(inode
, old_size
, pos
);
1554 if (size_changed
|| inline_data
) {
1555 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1560 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1561 /* if we have allocated more blocks and copied
1562 * less. We will have blocks allocated outside
1563 * inode->i_size. So truncate them
1565 ext4_orphan_add(handle
, inode
);
1568 ret2
= ext4_journal_stop(handle
);
1571 if (pos
+ len
> inode
->i_size
) {
1572 ext4_truncate_failed_write(inode
);
1574 * If truncate failed early the inode might still be
1575 * on the orphan list; we need to make sure the inode
1576 * is removed from the orphan list in that case.
1579 ext4_orphan_del(NULL
, inode
);
1582 return ret
? ret
: copied
;
1586 * Reserve space for a single cluster
1588 static int ext4_da_reserve_space(struct inode
*inode
)
1590 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1591 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1595 * We will charge metadata quota at writeout time; this saves
1596 * us from metadata over-estimation, though we may go over by
1597 * a small amount in the end. Here we just reserve for data.
1599 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1603 spin_lock(&ei
->i_block_reservation_lock
);
1604 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1605 spin_unlock(&ei
->i_block_reservation_lock
);
1606 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1609 ei
->i_reserved_data_blocks
++;
1610 trace_ext4_da_reserve_space(inode
);
1611 spin_unlock(&ei
->i_block_reservation_lock
);
1613 return 0; /* success */
1616 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1618 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1619 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1622 return; /* Nothing to release, exit */
1624 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1626 trace_ext4_da_release_space(inode
, to_free
);
1627 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1629 * if there aren't enough reserved blocks, then the
1630 * counter is messed up somewhere. Since this
1631 * function is called from invalidate page, it's
1632 * harmless to return without any action.
1634 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1635 "ino %lu, to_free %d with only %d reserved "
1636 "data blocks", inode
->i_ino
, to_free
,
1637 ei
->i_reserved_data_blocks
);
1639 to_free
= ei
->i_reserved_data_blocks
;
1641 ei
->i_reserved_data_blocks
-= to_free
;
1643 /* update fs dirty data blocks counter */
1644 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1646 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1648 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1651 static void ext4_da_page_release_reservation(struct page
*page
,
1652 unsigned int offset
,
1653 unsigned int length
)
1655 int to_release
= 0, contiguous_blks
= 0;
1656 struct buffer_head
*head
, *bh
;
1657 unsigned int curr_off
= 0;
1658 struct inode
*inode
= page
->mapping
->host
;
1659 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1660 unsigned int stop
= offset
+ length
;
1664 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1666 head
= page_buffers(page
);
1669 unsigned int next_off
= curr_off
+ bh
->b_size
;
1671 if (next_off
> stop
)
1674 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1677 clear_buffer_delay(bh
);
1678 } else if (contiguous_blks
) {
1679 lblk
= page
->index
<<
1680 (PAGE_SHIFT
- inode
->i_blkbits
);
1681 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1683 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1684 contiguous_blks
= 0;
1686 curr_off
= next_off
;
1687 } while ((bh
= bh
->b_this_page
) != head
);
1689 if (contiguous_blks
) {
1690 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1691 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1692 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1695 /* If we have released all the blocks belonging to a cluster, then we
1696 * need to release the reserved space for that cluster. */
1697 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1698 while (num_clusters
> 0) {
1699 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1700 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1701 if (sbi
->s_cluster_ratio
== 1 ||
1702 !ext4_find_delalloc_cluster(inode
, lblk
))
1703 ext4_da_release_space(inode
, 1);
1710 * Delayed allocation stuff
1713 struct mpage_da_data
{
1714 struct inode
*inode
;
1715 struct writeback_control
*wbc
;
1717 pgoff_t first_page
; /* The first page to write */
1718 pgoff_t next_page
; /* Current page to examine */
1719 pgoff_t last_page
; /* Last page to examine */
1721 * Extent to map - this can be after first_page because that can be
1722 * fully mapped. We somewhat abuse m_flags to store whether the extent
1723 * is delalloc or unwritten.
1725 struct ext4_map_blocks map
;
1726 struct ext4_io_submit io_submit
; /* IO submission data */
1727 unsigned int do_map
:1;
1730 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1735 struct pagevec pvec
;
1736 struct inode
*inode
= mpd
->inode
;
1737 struct address_space
*mapping
= inode
->i_mapping
;
1739 /* This is necessary when next_page == 0. */
1740 if (mpd
->first_page
>= mpd
->next_page
)
1743 index
= mpd
->first_page
;
1744 end
= mpd
->next_page
- 1;
1746 ext4_lblk_t start
, last
;
1747 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1748 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1749 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1752 pagevec_init(&pvec
);
1753 while (index
<= end
) {
1754 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1757 for (i
= 0; i
< nr_pages
; i
++) {
1758 struct page
*page
= pvec
.pages
[i
];
1760 BUG_ON(!PageLocked(page
));
1761 BUG_ON(PageWriteback(page
));
1763 if (page_mapped(page
))
1764 clear_page_dirty_for_io(page
);
1765 block_invalidatepage(page
, 0, PAGE_SIZE
);
1766 ClearPageUptodate(page
);
1770 pagevec_release(&pvec
);
1774 static void ext4_print_free_blocks(struct inode
*inode
)
1776 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1777 struct super_block
*sb
= inode
->i_sb
;
1778 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1780 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1781 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1782 ext4_count_free_clusters(sb
)));
1783 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1784 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1785 (long long) EXT4_C2B(EXT4_SB(sb
),
1786 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1787 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1788 (long long) EXT4_C2B(EXT4_SB(sb
),
1789 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1790 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1791 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1792 ei
->i_reserved_data_blocks
);
1796 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1798 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1802 * This function is grabs code from the very beginning of
1803 * ext4_map_blocks, but assumes that the caller is from delayed write
1804 * time. This function looks up the requested blocks and sets the
1805 * buffer delay bit under the protection of i_data_sem.
1807 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1808 struct ext4_map_blocks
*map
,
1809 struct buffer_head
*bh
)
1811 struct extent_status es
;
1813 sector_t invalid_block
= ~((sector_t
) 0xffff);
1814 #ifdef ES_AGGRESSIVE_TEST
1815 struct ext4_map_blocks orig_map
;
1817 memcpy(&orig_map
, map
, sizeof(*map
));
1820 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1824 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1825 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1826 (unsigned long) map
->m_lblk
);
1828 /* Lookup extent status tree firstly */
1829 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1830 if (ext4_es_is_hole(&es
)) {
1832 down_read(&EXT4_I(inode
)->i_data_sem
);
1837 * Delayed extent could be allocated by fallocate.
1838 * So we need to check it.
1840 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1841 map_bh(bh
, inode
->i_sb
, invalid_block
);
1843 set_buffer_delay(bh
);
1847 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1848 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1849 if (retval
> map
->m_len
)
1850 retval
= map
->m_len
;
1851 map
->m_len
= retval
;
1852 if (ext4_es_is_written(&es
))
1853 map
->m_flags
|= EXT4_MAP_MAPPED
;
1854 else if (ext4_es_is_unwritten(&es
))
1855 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1859 #ifdef ES_AGGRESSIVE_TEST
1860 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1866 * Try to see if we can get the block without requesting a new
1867 * file system block.
1869 down_read(&EXT4_I(inode
)->i_data_sem
);
1870 if (ext4_has_inline_data(inode
))
1872 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1873 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1875 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1881 * XXX: __block_prepare_write() unmaps passed block,
1885 * If the block was allocated from previously allocated cluster,
1886 * then we don't need to reserve it again. However we still need
1887 * to reserve metadata for every block we're going to write.
1889 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1890 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1891 ret
= ext4_da_reserve_space(inode
);
1893 /* not enough space to reserve */
1899 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1900 ~0, EXTENT_STATUS_DELAYED
);
1906 map_bh(bh
, inode
->i_sb
, invalid_block
);
1908 set_buffer_delay(bh
);
1909 } else if (retval
> 0) {
1911 unsigned int status
;
1913 if (unlikely(retval
!= map
->m_len
)) {
1914 ext4_warning(inode
->i_sb
,
1915 "ES len assertion failed for inode "
1916 "%lu: retval %d != map->m_len %d",
1917 inode
->i_ino
, retval
, map
->m_len
);
1921 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1922 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1923 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1924 map
->m_pblk
, status
);
1930 up_read((&EXT4_I(inode
)->i_data_sem
));
1936 * This is a special get_block_t callback which is used by
1937 * ext4_da_write_begin(). It will either return mapped block or
1938 * reserve space for a single block.
1940 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1941 * We also have b_blocknr = -1 and b_bdev initialized properly
1943 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1944 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1945 * initialized properly.
1947 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1948 struct buffer_head
*bh
, int create
)
1950 struct ext4_map_blocks map
;
1953 BUG_ON(create
== 0);
1954 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1956 map
.m_lblk
= iblock
;
1960 * first, we need to know whether the block is allocated already
1961 * preallocated blocks are unmapped but should treated
1962 * the same as allocated blocks.
1964 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1968 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1969 ext4_update_bh_state(bh
, map
.m_flags
);
1971 if (buffer_unwritten(bh
)) {
1972 /* A delayed write to unwritten bh should be marked
1973 * new and mapped. Mapped ensures that we don't do
1974 * get_block multiple times when we write to the same
1975 * offset and new ensures that we do proper zero out
1976 * for partial write.
1979 set_buffer_mapped(bh
);
1984 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1990 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1996 static int __ext4_journalled_writepage(struct page
*page
,
1999 struct address_space
*mapping
= page
->mapping
;
2000 struct inode
*inode
= mapping
->host
;
2001 struct buffer_head
*page_bufs
= NULL
;
2002 handle_t
*handle
= NULL
;
2003 int ret
= 0, err
= 0;
2004 int inline_data
= ext4_has_inline_data(inode
);
2005 struct buffer_head
*inode_bh
= NULL
;
2007 ClearPageChecked(page
);
2010 BUG_ON(page
->index
!= 0);
2011 BUG_ON(len
> ext4_get_max_inline_size(inode
));
2012 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
2013 if (inode_bh
== NULL
)
2016 page_bufs
= page_buffers(page
);
2021 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2025 * We need to release the page lock before we start the
2026 * journal, so grab a reference so the page won't disappear
2027 * out from under us.
2032 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2033 ext4_writepage_trans_blocks(inode
));
2034 if (IS_ERR(handle
)) {
2035 ret
= PTR_ERR(handle
);
2037 goto out_no_pagelock
;
2039 BUG_ON(!ext4_handle_valid(handle
));
2043 if (page
->mapping
!= mapping
) {
2044 /* The page got truncated from under us */
2045 ext4_journal_stop(handle
);
2051 ret
= ext4_mark_inode_dirty(handle
, inode
);
2053 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2054 do_journal_get_write_access
);
2056 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2061 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2062 err
= ext4_journal_stop(handle
);
2066 if (!ext4_has_inline_data(inode
))
2067 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
2069 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2078 * Note that we don't need to start a transaction unless we're journaling data
2079 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2080 * need to file the inode to the transaction's list in ordered mode because if
2081 * we are writing back data added by write(), the inode is already there and if
2082 * we are writing back data modified via mmap(), no one guarantees in which
2083 * transaction the data will hit the disk. In case we are journaling data, we
2084 * cannot start transaction directly because transaction start ranks above page
2085 * lock so we have to do some magic.
2087 * This function can get called via...
2088 * - ext4_writepages after taking page lock (have journal handle)
2089 * - journal_submit_inode_data_buffers (no journal handle)
2090 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2091 * - grab_page_cache when doing write_begin (have journal handle)
2093 * We don't do any block allocation in this function. If we have page with
2094 * multiple blocks we need to write those buffer_heads that are mapped. This
2095 * is important for mmaped based write. So if we do with blocksize 1K
2096 * truncate(f, 1024);
2097 * a = mmap(f, 0, 4096);
2099 * truncate(f, 4096);
2100 * we have in the page first buffer_head mapped via page_mkwrite call back
2101 * but other buffer_heads would be unmapped but dirty (dirty done via the
2102 * do_wp_page). So writepage should write the first block. If we modify
2103 * the mmap area beyond 1024 we will again get a page_fault and the
2104 * page_mkwrite callback will do the block allocation and mark the
2105 * buffer_heads mapped.
2107 * We redirty the page if we have any buffer_heads that is either delay or
2108 * unwritten in the page.
2110 * We can get recursively called as show below.
2112 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2115 * But since we don't do any block allocation we should not deadlock.
2116 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2118 static int ext4_writepage(struct page
*page
,
2119 struct writeback_control
*wbc
)
2124 struct buffer_head
*page_bufs
= NULL
;
2125 struct inode
*inode
= page
->mapping
->host
;
2126 struct ext4_io_submit io_submit
;
2127 bool keep_towrite
= false;
2129 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
2130 ext4_invalidatepage(page
, 0, PAGE_SIZE
);
2135 trace_ext4_writepage(page
);
2136 size
= i_size_read(inode
);
2137 if (page
->index
== size
>> PAGE_SHIFT
)
2138 len
= size
& ~PAGE_MASK
;
2142 page_bufs
= page_buffers(page
);
2144 * We cannot do block allocation or other extent handling in this
2145 * function. If there are buffers needing that, we have to redirty
2146 * the page. But we may reach here when we do a journal commit via
2147 * journal_submit_inode_data_buffers() and in that case we must write
2148 * allocated buffers to achieve data=ordered mode guarantees.
2150 * Also, if there is only one buffer per page (the fs block
2151 * size == the page size), if one buffer needs block
2152 * allocation or needs to modify the extent tree to clear the
2153 * unwritten flag, we know that the page can't be written at
2154 * all, so we might as well refuse the write immediately.
2155 * Unfortunately if the block size != page size, we can't as
2156 * easily detect this case using ext4_walk_page_buffers(), but
2157 * for the extremely common case, this is an optimization that
2158 * skips a useless round trip through ext4_bio_write_page().
2160 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2161 ext4_bh_delay_or_unwritten
)) {
2162 redirty_page_for_writepage(wbc
, page
);
2163 if ((current
->flags
& PF_MEMALLOC
) ||
2164 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2166 * For memory cleaning there's no point in writing only
2167 * some buffers. So just bail out. Warn if we came here
2168 * from direct reclaim.
2170 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2175 keep_towrite
= true;
2178 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2180 * It's mmapped pagecache. Add buffers and journal it. There
2181 * doesn't seem much point in redirtying the page here.
2183 return __ext4_journalled_writepage(page
, len
);
2185 ext4_io_submit_init(&io_submit
, wbc
);
2186 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2187 if (!io_submit
.io_end
) {
2188 redirty_page_for_writepage(wbc
, page
);
2192 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2193 ext4_io_submit(&io_submit
);
2194 /* Drop io_end reference we got from init */
2195 ext4_put_io_end_defer(io_submit
.io_end
);
2199 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2205 BUG_ON(page
->index
!= mpd
->first_page
);
2206 clear_page_dirty_for_io(page
);
2208 * We have to be very careful here! Nothing protects writeback path
2209 * against i_size changes and the page can be writeably mapped into
2210 * page tables. So an application can be growing i_size and writing
2211 * data through mmap while writeback runs. clear_page_dirty_for_io()
2212 * write-protects our page in page tables and the page cannot get
2213 * written to again until we release page lock. So only after
2214 * clear_page_dirty_for_io() we are safe to sample i_size for
2215 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2216 * on the barrier provided by TestClearPageDirty in
2217 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2218 * after page tables are updated.
2220 size
= i_size_read(mpd
->inode
);
2221 if (page
->index
== size
>> PAGE_SHIFT
)
2222 len
= size
& ~PAGE_MASK
;
2225 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2227 mpd
->wbc
->nr_to_write
--;
2233 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2236 * mballoc gives us at most this number of blocks...
2237 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2238 * The rest of mballoc seems to handle chunks up to full group size.
2240 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2243 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2245 * @mpd - extent of blocks
2246 * @lblk - logical number of the block in the file
2247 * @bh - buffer head we want to add to the extent
2249 * The function is used to collect contig. blocks in the same state. If the
2250 * buffer doesn't require mapping for writeback and we haven't started the
2251 * extent of buffers to map yet, the function returns 'true' immediately - the
2252 * caller can write the buffer right away. Otherwise the function returns true
2253 * if the block has been added to the extent, false if the block couldn't be
2256 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2257 struct buffer_head
*bh
)
2259 struct ext4_map_blocks
*map
= &mpd
->map
;
2261 /* Buffer that doesn't need mapping for writeback? */
2262 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2263 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2264 /* So far no extent to map => we write the buffer right away */
2265 if (map
->m_len
== 0)
2270 /* First block in the extent? */
2271 if (map
->m_len
== 0) {
2272 /* We cannot map unless handle is started... */
2277 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2281 /* Don't go larger than mballoc is willing to allocate */
2282 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2285 /* Can we merge the block to our big extent? */
2286 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2287 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2295 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2297 * @mpd - extent of blocks for mapping
2298 * @head - the first buffer in the page
2299 * @bh - buffer we should start processing from
2300 * @lblk - logical number of the block in the file corresponding to @bh
2302 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2303 * the page for IO if all buffers in this page were mapped and there's no
2304 * accumulated extent of buffers to map or add buffers in the page to the
2305 * extent of buffers to map. The function returns 1 if the caller can continue
2306 * by processing the next page, 0 if it should stop adding buffers to the
2307 * extent to map because we cannot extend it anymore. It can also return value
2308 * < 0 in case of error during IO submission.
2310 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2311 struct buffer_head
*head
,
2312 struct buffer_head
*bh
,
2315 struct inode
*inode
= mpd
->inode
;
2317 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2318 >> inode
->i_blkbits
;
2321 BUG_ON(buffer_locked(bh
));
2323 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2324 /* Found extent to map? */
2327 /* Buffer needs mapping and handle is not started? */
2330 /* Everything mapped so far and we hit EOF */
2333 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2334 /* So far everything mapped? Submit the page for IO. */
2335 if (mpd
->map
.m_len
== 0) {
2336 err
= mpage_submit_page(mpd
, head
->b_page
);
2340 return lblk
< blocks
;
2344 * mpage_map_buffers - update buffers corresponding to changed extent and
2345 * submit fully mapped pages for IO
2347 * @mpd - description of extent to map, on return next extent to map
2349 * Scan buffers corresponding to changed extent (we expect corresponding pages
2350 * to be already locked) and update buffer state according to new extent state.
2351 * We map delalloc buffers to their physical location, clear unwritten bits,
2352 * and mark buffers as uninit when we perform writes to unwritten extents
2353 * and do extent conversion after IO is finished. If the last page is not fully
2354 * mapped, we update @map to the next extent in the last page that needs
2355 * mapping. Otherwise we submit the page for IO.
2357 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2359 struct pagevec pvec
;
2361 struct inode
*inode
= mpd
->inode
;
2362 struct buffer_head
*head
, *bh
;
2363 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2369 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2370 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2371 lblk
= start
<< bpp_bits
;
2372 pblock
= mpd
->map
.m_pblk
;
2374 pagevec_init(&pvec
);
2375 while (start
<= end
) {
2376 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2380 for (i
= 0; i
< nr_pages
; i
++) {
2381 struct page
*page
= pvec
.pages
[i
];
2383 bh
= head
= page_buffers(page
);
2385 if (lblk
< mpd
->map
.m_lblk
)
2387 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2389 * Buffer after end of mapped extent.
2390 * Find next buffer in the page to map.
2393 mpd
->map
.m_flags
= 0;
2395 * FIXME: If dioread_nolock supports
2396 * blocksize < pagesize, we need to make
2397 * sure we add size mapped so far to
2398 * io_end->size as the following call
2399 * can submit the page for IO.
2401 err
= mpage_process_page_bufs(mpd
, head
,
2403 pagevec_release(&pvec
);
2408 if (buffer_delay(bh
)) {
2409 clear_buffer_delay(bh
);
2410 bh
->b_blocknr
= pblock
++;
2412 clear_buffer_unwritten(bh
);
2413 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2416 * FIXME: This is going to break if dioread_nolock
2417 * supports blocksize < pagesize as we will try to
2418 * convert potentially unmapped parts of inode.
2420 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2421 /* Page fully mapped - let IO run! */
2422 err
= mpage_submit_page(mpd
, page
);
2424 pagevec_release(&pvec
);
2428 pagevec_release(&pvec
);
2430 /* Extent fully mapped and matches with page boundary. We are done. */
2432 mpd
->map
.m_flags
= 0;
2436 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2438 struct inode
*inode
= mpd
->inode
;
2439 struct ext4_map_blocks
*map
= &mpd
->map
;
2440 int get_blocks_flags
;
2441 int err
, dioread_nolock
;
2443 trace_ext4_da_write_pages_extent(inode
, map
);
2445 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2446 * to convert an unwritten extent to be initialized (in the case
2447 * where we have written into one or more preallocated blocks). It is
2448 * possible that we're going to need more metadata blocks than
2449 * previously reserved. However we must not fail because we're in
2450 * writeback and there is nothing we can do about it so it might result
2451 * in data loss. So use reserved blocks to allocate metadata if
2454 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2455 * the blocks in question are delalloc blocks. This indicates
2456 * that the blocks and quotas has already been checked when
2457 * the data was copied into the page cache.
2459 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2460 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2461 EXT4_GET_BLOCKS_IO_SUBMIT
;
2462 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2464 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2465 if (map
->m_flags
& (1 << BH_Delay
))
2466 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2468 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2471 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2472 if (!mpd
->io_submit
.io_end
->handle
&&
2473 ext4_handle_valid(handle
)) {
2474 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2475 handle
->h_rsv_handle
= NULL
;
2477 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2480 BUG_ON(map
->m_len
== 0);
2481 if (map
->m_flags
& EXT4_MAP_NEW
) {
2482 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
2489 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2490 * mpd->len and submit pages underlying it for IO
2492 * @handle - handle for journal operations
2493 * @mpd - extent to map
2494 * @give_up_on_write - we set this to true iff there is a fatal error and there
2495 * is no hope of writing the data. The caller should discard
2496 * dirty pages to avoid infinite loops.
2498 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2499 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2500 * them to initialized or split the described range from larger unwritten
2501 * extent. Note that we need not map all the described range since allocation
2502 * can return less blocks or the range is covered by more unwritten extents. We
2503 * cannot map more because we are limited by reserved transaction credits. On
2504 * the other hand we always make sure that the last touched page is fully
2505 * mapped so that it can be written out (and thus forward progress is
2506 * guaranteed). After mapping we submit all mapped pages for IO.
2508 static int mpage_map_and_submit_extent(handle_t
*handle
,
2509 struct mpage_da_data
*mpd
,
2510 bool *give_up_on_write
)
2512 struct inode
*inode
= mpd
->inode
;
2513 struct ext4_map_blocks
*map
= &mpd
->map
;
2518 mpd
->io_submit
.io_end
->offset
=
2519 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2521 err
= mpage_map_one_extent(handle
, mpd
);
2523 struct super_block
*sb
= inode
->i_sb
;
2525 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2526 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2527 goto invalidate_dirty_pages
;
2529 * Let the uper layers retry transient errors.
2530 * In the case of ENOSPC, if ext4_count_free_blocks()
2531 * is non-zero, a commit should free up blocks.
2533 if ((err
== -ENOMEM
) ||
2534 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2536 goto update_disksize
;
2539 ext4_msg(sb
, KERN_CRIT
,
2540 "Delayed block allocation failed for "
2541 "inode %lu at logical offset %llu with"
2542 " max blocks %u with error %d",
2544 (unsigned long long)map
->m_lblk
,
2545 (unsigned)map
->m_len
, -err
);
2546 ext4_msg(sb
, KERN_CRIT
,
2547 "This should not happen!! Data will "
2550 ext4_print_free_blocks(inode
);
2551 invalidate_dirty_pages
:
2552 *give_up_on_write
= true;
2557 * Update buffer state, submit mapped pages, and get us new
2560 err
= mpage_map_and_submit_buffers(mpd
);
2562 goto update_disksize
;
2563 } while (map
->m_len
);
2567 * Update on-disk size after IO is submitted. Races with
2568 * truncate are avoided by checking i_size under i_data_sem.
2570 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2571 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2575 down_write(&EXT4_I(inode
)->i_data_sem
);
2576 i_size
= i_size_read(inode
);
2577 if (disksize
> i_size
)
2579 if (disksize
> EXT4_I(inode
)->i_disksize
)
2580 EXT4_I(inode
)->i_disksize
= disksize
;
2581 up_write(&EXT4_I(inode
)->i_data_sem
);
2582 err2
= ext4_mark_inode_dirty(handle
, inode
);
2584 ext4_error(inode
->i_sb
,
2585 "Failed to mark inode %lu dirty",
2594 * Calculate the total number of credits to reserve for one writepages
2595 * iteration. This is called from ext4_writepages(). We map an extent of
2596 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2597 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2598 * bpp - 1 blocks in bpp different extents.
2600 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2602 int bpp
= ext4_journal_blocks_per_page(inode
);
2604 return ext4_meta_trans_blocks(inode
,
2605 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2609 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2610 * and underlying extent to map
2612 * @mpd - where to look for pages
2614 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2615 * IO immediately. When we find a page which isn't mapped we start accumulating
2616 * extent of buffers underlying these pages that needs mapping (formed by
2617 * either delayed or unwritten buffers). We also lock the pages containing
2618 * these buffers. The extent found is returned in @mpd structure (starting at
2619 * mpd->lblk with length mpd->len blocks).
2621 * Note that this function can attach bios to one io_end structure which are
2622 * neither logically nor physically contiguous. Although it may seem as an
2623 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2624 * case as we need to track IO to all buffers underlying a page in one io_end.
2626 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2628 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2629 struct pagevec pvec
;
2630 unsigned int nr_pages
;
2631 long left
= mpd
->wbc
->nr_to_write
;
2632 pgoff_t index
= mpd
->first_page
;
2633 pgoff_t end
= mpd
->last_page
;
2636 int blkbits
= mpd
->inode
->i_blkbits
;
2638 struct buffer_head
*head
;
2640 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2641 tag
= PAGECACHE_TAG_TOWRITE
;
2643 tag
= PAGECACHE_TAG_DIRTY
;
2645 pagevec_init(&pvec
);
2647 mpd
->next_page
= index
;
2648 while (index
<= end
) {
2649 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
, end
,
2654 for (i
= 0; i
< nr_pages
; i
++) {
2655 struct page
*page
= pvec
.pages
[i
];
2658 * Accumulated enough dirty pages? This doesn't apply
2659 * to WB_SYNC_ALL mode. For integrity sync we have to
2660 * keep going because someone may be concurrently
2661 * dirtying pages, and we might have synced a lot of
2662 * newly appeared dirty pages, but have not synced all
2663 * of the old dirty pages.
2665 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2668 /* If we can't merge this page, we are done. */
2669 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2674 * If the page is no longer dirty, or its mapping no
2675 * longer corresponds to inode we are writing (which
2676 * means it has been truncated or invalidated), or the
2677 * page is already under writeback and we are not doing
2678 * a data integrity writeback, skip the page
2680 if (!PageDirty(page
) ||
2681 (PageWriteback(page
) &&
2682 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2683 unlikely(page
->mapping
!= mapping
)) {
2688 wait_on_page_writeback(page
);
2689 BUG_ON(PageWriteback(page
));
2691 if (mpd
->map
.m_len
== 0)
2692 mpd
->first_page
= page
->index
;
2693 mpd
->next_page
= page
->index
+ 1;
2694 /* Add all dirty buffers to mpd */
2695 lblk
= ((ext4_lblk_t
)page
->index
) <<
2696 (PAGE_SHIFT
- blkbits
);
2697 head
= page_buffers(page
);
2698 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2704 pagevec_release(&pvec
);
2709 pagevec_release(&pvec
);
2713 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2716 struct address_space
*mapping
= data
;
2717 int ret
= ext4_writepage(page
, wbc
);
2718 mapping_set_error(mapping
, ret
);
2722 static int ext4_writepages(struct address_space
*mapping
,
2723 struct writeback_control
*wbc
)
2725 pgoff_t writeback_index
= 0;
2726 long nr_to_write
= wbc
->nr_to_write
;
2727 int range_whole
= 0;
2729 handle_t
*handle
= NULL
;
2730 struct mpage_da_data mpd
;
2731 struct inode
*inode
= mapping
->host
;
2732 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2733 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2735 struct blk_plug plug
;
2736 bool give_up_on_write
= false;
2738 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2741 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2742 trace_ext4_writepages(inode
, wbc
);
2744 if (dax_mapping(mapping
)) {
2745 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2747 goto out_writepages
;
2751 * No pages to write? This is mainly a kludge to avoid starting
2752 * a transaction for special inodes like journal inode on last iput()
2753 * because that could violate lock ordering on umount
2755 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2756 goto out_writepages
;
2758 if (ext4_should_journal_data(inode
)) {
2759 struct blk_plug plug
;
2761 blk_start_plug(&plug
);
2762 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2763 blk_finish_plug(&plug
);
2764 goto out_writepages
;
2768 * If the filesystem has aborted, it is read-only, so return
2769 * right away instead of dumping stack traces later on that
2770 * will obscure the real source of the problem. We test
2771 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2772 * the latter could be true if the filesystem is mounted
2773 * read-only, and in that case, ext4_writepages should
2774 * *never* be called, so if that ever happens, we would want
2777 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2778 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2780 goto out_writepages
;
2783 if (ext4_should_dioread_nolock(inode
)) {
2785 * We may need to convert up to one extent per block in
2786 * the page and we may dirty the inode.
2788 rsv_blocks
= 1 + ext4_chunk_trans_blocks(inode
,
2789 PAGE_SIZE
>> inode
->i_blkbits
);
2793 * If we have inline data and arrive here, it means that
2794 * we will soon create the block for the 1st page, so
2795 * we'd better clear the inline data here.
2797 if (ext4_has_inline_data(inode
)) {
2798 /* Just inode will be modified... */
2799 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2800 if (IS_ERR(handle
)) {
2801 ret
= PTR_ERR(handle
);
2802 goto out_writepages
;
2804 BUG_ON(ext4_test_inode_state(inode
,
2805 EXT4_STATE_MAY_INLINE_DATA
));
2806 ext4_destroy_inline_data(handle
, inode
);
2807 ext4_journal_stop(handle
);
2810 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2813 if (wbc
->range_cyclic
) {
2814 writeback_index
= mapping
->writeback_index
;
2815 if (writeback_index
)
2817 mpd
.first_page
= writeback_index
;
2820 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2821 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2826 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2828 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2829 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2831 blk_start_plug(&plug
);
2834 * First writeback pages that don't need mapping - we can avoid
2835 * starting a transaction unnecessarily and also avoid being blocked
2836 * in the block layer on device congestion while having transaction
2840 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2841 if (!mpd
.io_submit
.io_end
) {
2845 ret
= mpage_prepare_extent_to_map(&mpd
);
2846 /* Submit prepared bio */
2847 ext4_io_submit(&mpd
.io_submit
);
2848 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2849 mpd
.io_submit
.io_end
= NULL
;
2850 /* Unlock pages we didn't use */
2851 mpage_release_unused_pages(&mpd
, false);
2855 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2856 /* For each extent of pages we use new io_end */
2857 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2858 if (!mpd
.io_submit
.io_end
) {
2864 * We have two constraints: We find one extent to map and we
2865 * must always write out whole page (makes a difference when
2866 * blocksize < pagesize) so that we don't block on IO when we
2867 * try to write out the rest of the page. Journalled mode is
2868 * not supported by delalloc.
2870 BUG_ON(ext4_should_journal_data(inode
));
2871 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2873 /* start a new transaction */
2874 handle
= ext4_journal_start_with_reserve(inode
,
2875 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2876 if (IS_ERR(handle
)) {
2877 ret
= PTR_ERR(handle
);
2878 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2879 "%ld pages, ino %lu; err %d", __func__
,
2880 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2881 /* Release allocated io_end */
2882 ext4_put_io_end(mpd
.io_submit
.io_end
);
2883 mpd
.io_submit
.io_end
= NULL
;
2888 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2889 ret
= mpage_prepare_extent_to_map(&mpd
);
2892 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2896 * We scanned the whole range (or exhausted
2897 * nr_to_write), submitted what was mapped and
2898 * didn't find anything needing mapping. We are
2905 * Caution: If the handle is synchronous,
2906 * ext4_journal_stop() can wait for transaction commit
2907 * to finish which may depend on writeback of pages to
2908 * complete or on page lock to be released. In that
2909 * case, we have to wait until after after we have
2910 * submitted all the IO, released page locks we hold,
2911 * and dropped io_end reference (for extent conversion
2912 * to be able to complete) before stopping the handle.
2914 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2915 ext4_journal_stop(handle
);
2919 /* Submit prepared bio */
2920 ext4_io_submit(&mpd
.io_submit
);
2921 /* Unlock pages we didn't use */
2922 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2924 * Drop our io_end reference we got from init. We have
2925 * to be careful and use deferred io_end finishing if
2926 * we are still holding the transaction as we can
2927 * release the last reference to io_end which may end
2928 * up doing unwritten extent conversion.
2931 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2932 ext4_journal_stop(handle
);
2934 ext4_put_io_end(mpd
.io_submit
.io_end
);
2935 mpd
.io_submit
.io_end
= NULL
;
2937 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2939 * Commit the transaction which would
2940 * free blocks released in the transaction
2943 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2947 /* Fatal error - ENOMEM, EIO... */
2952 blk_finish_plug(&plug
);
2953 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2955 mpd
.last_page
= writeback_index
- 1;
2961 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2963 * Set the writeback_index so that range_cyclic
2964 * mode will write it back later
2966 mapping
->writeback_index
= mpd
.first_page
;
2969 trace_ext4_writepages_result(inode
, wbc
, ret
,
2970 nr_to_write
- wbc
->nr_to_write
);
2971 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2975 static int ext4_nonda_switch(struct super_block
*sb
)
2977 s64 free_clusters
, dirty_clusters
;
2978 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2981 * switch to non delalloc mode if we are running low
2982 * on free block. The free block accounting via percpu
2983 * counters can get slightly wrong with percpu_counter_batch getting
2984 * accumulated on each CPU without updating global counters
2985 * Delalloc need an accurate free block accounting. So switch
2986 * to non delalloc when we are near to error range.
2989 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2991 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2993 * Start pushing delalloc when 1/2 of free blocks are dirty.
2995 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2996 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2998 if (2 * free_clusters
< 3 * dirty_clusters
||
2999 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
3001 * free block count is less than 150% of dirty blocks
3002 * or free blocks is less than watermark
3009 /* We always reserve for an inode update; the superblock could be there too */
3010 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
3012 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
3015 if (pos
+ len
<= 0x7fffffffULL
)
3018 /* We might need to update the superblock to set LARGE_FILE */
3022 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3023 loff_t pos
, unsigned len
, unsigned flags
,
3024 struct page
**pagep
, void **fsdata
)
3026 int ret
, retries
= 0;
3029 struct inode
*inode
= mapping
->host
;
3032 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
3035 index
= pos
>> PAGE_SHIFT
;
3037 if (ext4_nonda_switch(inode
->i_sb
) ||
3038 S_ISLNK(inode
->i_mode
)) {
3039 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3040 return ext4_write_begin(file
, mapping
, pos
,
3041 len
, flags
, pagep
, fsdata
);
3043 *fsdata
= (void *)0;
3044 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3046 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
3047 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
3057 * grab_cache_page_write_begin() can take a long time if the
3058 * system is thrashing due to memory pressure, or if the page
3059 * is being written back. So grab it first before we start
3060 * the transaction handle. This also allows us to allocate
3061 * the page (if needed) without using GFP_NOFS.
3064 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3070 * With delayed allocation, we don't log the i_disksize update
3071 * if there is delayed block allocation. But we still need
3072 * to journalling the i_disksize update if writes to the end
3073 * of file which has an already mapped buffer.
3076 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
3077 ext4_da_write_credits(inode
, pos
, len
));
3078 if (IS_ERR(handle
)) {
3080 return PTR_ERR(handle
);
3084 if (page
->mapping
!= mapping
) {
3085 /* The page got truncated from under us */
3088 ext4_journal_stop(handle
);
3091 /* In case writeback began while the page was unlocked */
3092 wait_for_stable_page(page
);
3094 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3095 ret
= ext4_block_write_begin(page
, pos
, len
,
3096 ext4_da_get_block_prep
);
3098 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3102 ext4_journal_stop(handle
);
3104 * block_write_begin may have instantiated a few blocks
3105 * outside i_size. Trim these off again. Don't need
3106 * i_size_read because we hold i_mutex.
3108 if (pos
+ len
> inode
->i_size
)
3109 ext4_truncate_failed_write(inode
);
3111 if (ret
== -ENOSPC
&&
3112 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3124 * Check if we should update i_disksize
3125 * when write to the end of file but not require block allocation
3127 static int ext4_da_should_update_i_disksize(struct page
*page
,
3128 unsigned long offset
)
3130 struct buffer_head
*bh
;
3131 struct inode
*inode
= page
->mapping
->host
;
3135 bh
= page_buffers(page
);
3136 idx
= offset
>> inode
->i_blkbits
;
3138 for (i
= 0; i
< idx
; i
++)
3139 bh
= bh
->b_this_page
;
3141 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3146 static int ext4_da_write_end(struct file
*file
,
3147 struct address_space
*mapping
,
3148 loff_t pos
, unsigned len
, unsigned copied
,
3149 struct page
*page
, void *fsdata
)
3151 struct inode
*inode
= mapping
->host
;
3153 handle_t
*handle
= ext4_journal_current_handle();
3155 unsigned long start
, end
;
3156 int write_mode
= (int)(unsigned long)fsdata
;
3158 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3159 return ext4_write_end(file
, mapping
, pos
,
3160 len
, copied
, page
, fsdata
);
3162 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3163 start
= pos
& (PAGE_SIZE
- 1);
3164 end
= start
+ copied
- 1;
3167 * generic_write_end() will run mark_inode_dirty() if i_size
3168 * changes. So let's piggyback the i_disksize mark_inode_dirty
3171 new_i_size
= pos
+ copied
;
3172 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3173 if (ext4_has_inline_data(inode
) ||
3174 ext4_da_should_update_i_disksize(page
, end
)) {
3175 ext4_update_i_disksize(inode
, new_i_size
);
3176 /* We need to mark inode dirty even if
3177 * new_i_size is less that inode->i_size
3178 * bu greater than i_disksize.(hint delalloc)
3180 ext4_mark_inode_dirty(handle
, inode
);
3184 if (write_mode
!= CONVERT_INLINE_DATA
&&
3185 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3186 ext4_has_inline_data(inode
))
3187 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3190 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3196 ret2
= ext4_journal_stop(handle
);
3200 return ret
? ret
: copied
;
3203 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3204 unsigned int length
)
3207 * Drop reserved blocks
3209 BUG_ON(!PageLocked(page
));
3210 if (!page_has_buffers(page
))
3213 ext4_da_page_release_reservation(page
, offset
, length
);
3216 ext4_invalidatepage(page
, offset
, length
);
3222 * Force all delayed allocation blocks to be allocated for a given inode.
3224 int ext4_alloc_da_blocks(struct inode
*inode
)
3226 trace_ext4_alloc_da_blocks(inode
);
3228 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3232 * We do something simple for now. The filemap_flush() will
3233 * also start triggering a write of the data blocks, which is
3234 * not strictly speaking necessary (and for users of
3235 * laptop_mode, not even desirable). However, to do otherwise
3236 * would require replicating code paths in:
3238 * ext4_writepages() ->
3239 * write_cache_pages() ---> (via passed in callback function)
3240 * __mpage_da_writepage() -->
3241 * mpage_add_bh_to_extent()
3242 * mpage_da_map_blocks()
3244 * The problem is that write_cache_pages(), located in
3245 * mm/page-writeback.c, marks pages clean in preparation for
3246 * doing I/O, which is not desirable if we're not planning on
3249 * We could call write_cache_pages(), and then redirty all of
3250 * the pages by calling redirty_page_for_writepage() but that
3251 * would be ugly in the extreme. So instead we would need to
3252 * replicate parts of the code in the above functions,
3253 * simplifying them because we wouldn't actually intend to
3254 * write out the pages, but rather only collect contiguous
3255 * logical block extents, call the multi-block allocator, and
3256 * then update the buffer heads with the block allocations.
3258 * For now, though, we'll cheat by calling filemap_flush(),
3259 * which will map the blocks, and start the I/O, but not
3260 * actually wait for the I/O to complete.
3262 return filemap_flush(inode
->i_mapping
);
3266 * bmap() is special. It gets used by applications such as lilo and by
3267 * the swapper to find the on-disk block of a specific piece of data.
3269 * Naturally, this is dangerous if the block concerned is still in the
3270 * journal. If somebody makes a swapfile on an ext4 data-journaling
3271 * filesystem and enables swap, then they may get a nasty shock when the
3272 * data getting swapped to that swapfile suddenly gets overwritten by
3273 * the original zero's written out previously to the journal and
3274 * awaiting writeback in the kernel's buffer cache.
3276 * So, if we see any bmap calls here on a modified, data-journaled file,
3277 * take extra steps to flush any blocks which might be in the cache.
3279 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3281 struct inode
*inode
= mapping
->host
;
3286 * We can get here for an inline file via the FIBMAP ioctl
3288 if (ext4_has_inline_data(inode
))
3291 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3292 test_opt(inode
->i_sb
, DELALLOC
)) {
3294 * With delalloc we want to sync the file
3295 * so that we can make sure we allocate
3298 filemap_write_and_wait(mapping
);
3301 if (EXT4_JOURNAL(inode
) &&
3302 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3304 * This is a REALLY heavyweight approach, but the use of
3305 * bmap on dirty files is expected to be extremely rare:
3306 * only if we run lilo or swapon on a freshly made file
3307 * do we expect this to happen.
3309 * (bmap requires CAP_SYS_RAWIO so this does not
3310 * represent an unprivileged user DOS attack --- we'd be
3311 * in trouble if mortal users could trigger this path at
3314 * NB. EXT4_STATE_JDATA is not set on files other than
3315 * regular files. If somebody wants to bmap a directory
3316 * or symlink and gets confused because the buffer
3317 * hasn't yet been flushed to disk, they deserve
3318 * everything they get.
3321 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3322 journal
= EXT4_JOURNAL(inode
);
3323 jbd2_journal_lock_updates(journal
);
3324 err
= jbd2_journal_flush(journal
);
3325 jbd2_journal_unlock_updates(journal
);
3331 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3334 static int ext4_readpage(struct file
*file
, struct page
*page
)
3337 struct inode
*inode
= page
->mapping
->host
;
3339 trace_ext4_readpage(page
);
3341 if (ext4_has_inline_data(inode
))
3342 ret
= ext4_readpage_inline(inode
, page
);
3345 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3351 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3352 struct list_head
*pages
, unsigned nr_pages
)
3354 struct inode
*inode
= mapping
->host
;
3356 /* If the file has inline data, no need to do readpages. */
3357 if (ext4_has_inline_data(inode
))
3360 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3363 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3364 unsigned int length
)
3366 trace_ext4_invalidatepage(page
, offset
, length
);
3368 /* No journalling happens on data buffers when this function is used */
3369 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3371 block_invalidatepage(page
, offset
, length
);
3374 static int __ext4_journalled_invalidatepage(struct page
*page
,
3375 unsigned int offset
,
3376 unsigned int length
)
3378 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3380 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3383 * If it's a full truncate we just forget about the pending dirtying
3385 if (offset
== 0 && length
== PAGE_SIZE
)
3386 ClearPageChecked(page
);
3388 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3391 /* Wrapper for aops... */
3392 static void ext4_journalled_invalidatepage(struct page
*page
,
3393 unsigned int offset
,
3394 unsigned int length
)
3396 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3399 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3401 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3403 trace_ext4_releasepage(page
);
3405 /* Page has dirty journalled data -> cannot release */
3406 if (PageChecked(page
))
3409 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3411 return try_to_free_buffers(page
);
3414 static bool ext4_inode_datasync_dirty(struct inode
*inode
)
3416 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
3419 return !jbd2_transaction_committed(journal
,
3420 EXT4_I(inode
)->i_datasync_tid
);
3421 /* Any metadata buffers to write? */
3422 if (!list_empty(&inode
->i_mapping
->private_list
))
3424 return inode
->i_state
& I_DIRTY_DATASYNC
;
3427 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3428 unsigned flags
, struct iomap
*iomap
)
3430 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3431 unsigned int blkbits
= inode
->i_blkbits
;
3432 unsigned long first_block
, last_block
;
3433 struct ext4_map_blocks map
;
3434 bool delalloc
= false;
3437 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3439 first_block
= offset
>> blkbits
;
3440 last_block
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3441 EXT4_MAX_LOGICAL_BLOCK
);
3443 if (flags
& IOMAP_REPORT
) {
3444 if (ext4_has_inline_data(inode
)) {
3445 ret
= ext4_inline_data_iomap(inode
, iomap
);
3446 if (ret
!= -EAGAIN
) {
3447 if (ret
== 0 && offset
>= iomap
->length
)
3453 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3457 map
.m_lblk
= first_block
;
3458 map
.m_len
= last_block
- first_block
+ 1;
3460 if (flags
& IOMAP_REPORT
) {
3461 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3466 ext4_lblk_t end
= map
.m_lblk
+ map
.m_len
- 1;
3467 struct extent_status es
;
3469 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
, end
, &es
);
3471 if (!es
.es_len
|| es
.es_lblk
> end
) {
3472 /* entire range is a hole */
3473 } else if (es
.es_lblk
> map
.m_lblk
) {
3474 /* range starts with a hole */
3475 map
.m_len
= es
.es_lblk
- map
.m_lblk
;
3477 ext4_lblk_t offs
= 0;
3479 if (es
.es_lblk
< map
.m_lblk
)
3480 offs
= map
.m_lblk
- es
.es_lblk
;
3481 map
.m_lblk
= es
.es_lblk
+ offs
;
3482 map
.m_len
= es
.es_len
- offs
;
3486 } else if (flags
& IOMAP_WRITE
) {
3491 /* Trim mapping request to maximum we can map at once for DIO */
3492 if (map
.m_len
> DIO_MAX_BLOCKS
)
3493 map
.m_len
= DIO_MAX_BLOCKS
;
3494 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3497 * Either we allocate blocks and then we don't get unwritten
3498 * extent so we have reserved enough credits, or the blocks
3499 * are already allocated and unwritten and in that case
3500 * extent conversion fits in the credits as well.
3502 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3505 return PTR_ERR(handle
);
3507 ret
= ext4_map_blocks(handle
, inode
, &map
,
3508 EXT4_GET_BLOCKS_CREATE_ZERO
);
3510 ext4_journal_stop(handle
);
3511 if (ret
== -ENOSPC
&&
3512 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3518 * If we added blocks beyond i_size, we need to make sure they
3519 * will get truncated if we crash before updating i_size in
3520 * ext4_iomap_end(). For faults we don't need to do that (and
3521 * even cannot because for orphan list operations inode_lock is
3522 * required) - if we happen to instantiate block beyond i_size,
3523 * it is because we race with truncate which has already added
3524 * the inode to the orphan list.
3526 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3527 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3530 err
= ext4_orphan_add(handle
, inode
);
3532 ext4_journal_stop(handle
);
3536 ext4_journal_stop(handle
);
3538 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3544 if (ext4_inode_datasync_dirty(inode
))
3545 iomap
->flags
|= IOMAP_F_DIRTY
;
3546 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3547 iomap
->dax_dev
= sbi
->s_daxdev
;
3548 iomap
->offset
= (u64
)first_block
<< blkbits
;
3549 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3552 iomap
->type
= delalloc
? IOMAP_DELALLOC
: IOMAP_HOLE
;
3553 iomap
->addr
= IOMAP_NULL_ADDR
;
3555 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3556 iomap
->type
= IOMAP_MAPPED
;
3557 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3558 iomap
->type
= IOMAP_UNWRITTEN
;
3563 iomap
->addr
= (u64
)map
.m_pblk
<< blkbits
;
3566 if (map
.m_flags
& EXT4_MAP_NEW
)
3567 iomap
->flags
|= IOMAP_F_NEW
;
3572 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3573 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3577 int blkbits
= inode
->i_blkbits
;
3578 bool truncate
= false;
3580 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3583 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3584 if (IS_ERR(handle
)) {
3585 ret
= PTR_ERR(handle
);
3588 if (ext4_update_inode_size(inode
, offset
+ written
))
3589 ext4_mark_inode_dirty(handle
, inode
);
3591 * We may need to truncate allocated but not written blocks beyond EOF.
3593 if (iomap
->offset
+ iomap
->length
>
3594 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3595 ext4_lblk_t written_blk
, end_blk
;
3597 written_blk
= (offset
+ written
) >> blkbits
;
3598 end_blk
= (offset
+ length
) >> blkbits
;
3599 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3603 * Remove inode from orphan list if we were extending a inode and
3604 * everything went fine.
3606 if (!truncate
&& inode
->i_nlink
&&
3607 !list_empty(&EXT4_I(inode
)->i_orphan
))
3608 ext4_orphan_del(handle
, inode
);
3609 ext4_journal_stop(handle
);
3611 ext4_truncate_failed_write(inode
);
3614 * If truncate failed early the inode might still be on the
3615 * orphan list; we need to make sure the inode is removed from
3616 * the orphan list in that case.
3619 ext4_orphan_del(NULL
, inode
);
3624 const struct iomap_ops ext4_iomap_ops
= {
3625 .iomap_begin
= ext4_iomap_begin
,
3626 .iomap_end
= ext4_iomap_end
,
3629 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3630 ssize_t size
, void *private)
3632 ext4_io_end_t
*io_end
= private;
3634 /* if not async direct IO just return */
3638 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3639 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3640 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3643 * Error during AIO DIO. We cannot convert unwritten extents as the
3644 * data was not written. Just clear the unwritten flag and drop io_end.
3647 ext4_clear_io_unwritten_flag(io_end
);
3650 io_end
->offset
= offset
;
3651 io_end
->size
= size
;
3652 ext4_put_io_end(io_end
);
3658 * Handling of direct IO writes.
3660 * For ext4 extent files, ext4 will do direct-io write even to holes,
3661 * preallocated extents, and those write extend the file, no need to
3662 * fall back to buffered IO.
3664 * For holes, we fallocate those blocks, mark them as unwritten
3665 * If those blocks were preallocated, we mark sure they are split, but
3666 * still keep the range to write as unwritten.
3668 * The unwritten extents will be converted to written when DIO is completed.
3669 * For async direct IO, since the IO may still pending when return, we
3670 * set up an end_io call back function, which will do the conversion
3671 * when async direct IO completed.
3673 * If the O_DIRECT write will extend the file then add this inode to the
3674 * orphan list. So recovery will truncate it back to the original size
3675 * if the machine crashes during the write.
3678 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3680 struct file
*file
= iocb
->ki_filp
;
3681 struct inode
*inode
= file
->f_mapping
->host
;
3683 loff_t offset
= iocb
->ki_pos
;
3684 size_t count
= iov_iter_count(iter
);
3686 get_block_t
*get_block_func
= NULL
;
3688 loff_t final_size
= offset
+ count
;
3692 if (final_size
> inode
->i_size
) {
3693 /* Credits for sb + inode write */
3694 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3695 if (IS_ERR(handle
)) {
3696 ret
= PTR_ERR(handle
);
3699 ret
= ext4_orphan_add(handle
, inode
);
3701 ext4_journal_stop(handle
);
3705 ext4_update_i_disksize(inode
, inode
->i_size
);
3706 ext4_journal_stop(handle
);
3709 BUG_ON(iocb
->private == NULL
);
3712 * Make all waiters for direct IO properly wait also for extent
3713 * conversion. This also disallows race between truncate() and
3714 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3716 inode_dio_begin(inode
);
3718 /* If we do a overwrite dio, i_mutex locking can be released */
3719 overwrite
= *((int *)iocb
->private);
3722 inode_unlock(inode
);
3725 * For extent mapped files we could direct write to holes and fallocate.
3727 * Allocated blocks to fill the hole are marked as unwritten to prevent
3728 * parallel buffered read to expose the stale data before DIO complete
3731 * As to previously fallocated extents, ext4 get_block will just simply
3732 * mark the buffer mapped but still keep the extents unwritten.
3734 * For non AIO case, we will convert those unwritten extents to written
3735 * after return back from blockdev_direct_IO. That way we save us from
3736 * allocating io_end structure and also the overhead of offloading
3737 * the extent convertion to a workqueue.
3739 * For async DIO, the conversion needs to be deferred when the
3740 * IO is completed. The ext4 end_io callback function will be
3741 * called to take care of the conversion work. Here for async
3742 * case, we allocate an io_end structure to hook to the iocb.
3744 iocb
->private = NULL
;
3746 get_block_func
= ext4_dio_get_block_overwrite
;
3747 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3748 round_down(offset
, i_blocksize(inode
)) >= inode
->i_size
) {
3749 get_block_func
= ext4_dio_get_block
;
3750 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3751 } else if (is_sync_kiocb(iocb
)) {
3752 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3753 dio_flags
= DIO_LOCKING
;
3755 get_block_func
= ext4_dio_get_block_unwritten_async
;
3756 dio_flags
= DIO_LOCKING
;
3758 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3759 get_block_func
, ext4_end_io_dio
, NULL
,
3762 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3763 EXT4_STATE_DIO_UNWRITTEN
)) {
3766 * for non AIO case, since the IO is already
3767 * completed, we could do the conversion right here
3769 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3773 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3776 inode_dio_end(inode
);
3777 /* take i_mutex locking again if we do a ovewrite dio */
3781 if (ret
< 0 && final_size
> inode
->i_size
)
3782 ext4_truncate_failed_write(inode
);
3784 /* Handle extending of i_size after direct IO write */
3788 /* Credits for sb + inode write */
3789 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3790 if (IS_ERR(handle
)) {
3792 * We wrote the data but cannot extend
3793 * i_size. Bail out. In async io case, we do
3794 * not return error here because we have
3795 * already submmitted the corresponding
3796 * bio. Returning error here makes the caller
3797 * think that this IO is done and failed
3798 * resulting in race with bio's completion
3802 ret
= PTR_ERR(handle
);
3804 ext4_orphan_del(NULL
, inode
);
3809 ext4_orphan_del(handle
, inode
);
3811 loff_t end
= offset
+ ret
;
3812 if (end
> inode
->i_size
) {
3813 ext4_update_i_disksize(inode
, end
);
3814 i_size_write(inode
, end
);
3816 * We're going to return a positive `ret'
3817 * here due to non-zero-length I/O, so there's
3818 * no way of reporting error returns from
3819 * ext4_mark_inode_dirty() to userspace. So
3822 ext4_mark_inode_dirty(handle
, inode
);
3825 err
= ext4_journal_stop(handle
);
3833 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3835 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3836 struct inode
*inode
= mapping
->host
;
3837 size_t count
= iov_iter_count(iter
);
3841 * Shared inode_lock is enough for us - it protects against concurrent
3842 * writes & truncates and since we take care of writing back page cache,
3843 * we are protected against page writeback as well.
3845 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3846 if (!inode_trylock_shared(inode
))
3849 inode_lock_shared(inode
);
3852 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3853 iocb
->ki_pos
+ count
- 1);
3856 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3857 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3859 inode_unlock_shared(inode
);
3863 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3865 struct file
*file
= iocb
->ki_filp
;
3866 struct inode
*inode
= file
->f_mapping
->host
;
3867 size_t count
= iov_iter_count(iter
);
3868 loff_t offset
= iocb
->ki_pos
;
3871 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3872 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3877 * If we are doing data journalling we don't support O_DIRECT
3879 if (ext4_should_journal_data(inode
))
3882 /* Let buffer I/O handle the inline data case. */
3883 if (ext4_has_inline_data(inode
))
3886 /* DAX uses iomap path now */
3887 if (WARN_ON_ONCE(IS_DAX(inode
)))
3890 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3891 if (iov_iter_rw(iter
) == READ
)
3892 ret
= ext4_direct_IO_read(iocb
, iter
);
3894 ret
= ext4_direct_IO_write(iocb
, iter
);
3895 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3900 * Pages can be marked dirty completely asynchronously from ext4's journalling
3901 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3902 * much here because ->set_page_dirty is called under VFS locks. The page is
3903 * not necessarily locked.
3905 * We cannot just dirty the page and leave attached buffers clean, because the
3906 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3907 * or jbddirty because all the journalling code will explode.
3909 * So what we do is to mark the page "pending dirty" and next time writepage
3910 * is called, propagate that into the buffers appropriately.
3912 static int ext4_journalled_set_page_dirty(struct page
*page
)
3914 SetPageChecked(page
);
3915 return __set_page_dirty_nobuffers(page
);
3918 static int ext4_set_page_dirty(struct page
*page
)
3920 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3921 WARN_ON_ONCE(!page_has_buffers(page
));
3922 return __set_page_dirty_buffers(page
);
3925 static const struct address_space_operations ext4_aops
= {
3926 .readpage
= ext4_readpage
,
3927 .readpages
= ext4_readpages
,
3928 .writepage
= ext4_writepage
,
3929 .writepages
= ext4_writepages
,
3930 .write_begin
= ext4_write_begin
,
3931 .write_end
= ext4_write_end
,
3932 .set_page_dirty
= ext4_set_page_dirty
,
3934 .invalidatepage
= ext4_invalidatepage
,
3935 .releasepage
= ext4_releasepage
,
3936 .direct_IO
= ext4_direct_IO
,
3937 .migratepage
= buffer_migrate_page
,
3938 .is_partially_uptodate
= block_is_partially_uptodate
,
3939 .error_remove_page
= generic_error_remove_page
,
3942 static const struct address_space_operations ext4_journalled_aops
= {
3943 .readpage
= ext4_readpage
,
3944 .readpages
= ext4_readpages
,
3945 .writepage
= ext4_writepage
,
3946 .writepages
= ext4_writepages
,
3947 .write_begin
= ext4_write_begin
,
3948 .write_end
= ext4_journalled_write_end
,
3949 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3951 .invalidatepage
= ext4_journalled_invalidatepage
,
3952 .releasepage
= ext4_releasepage
,
3953 .direct_IO
= ext4_direct_IO
,
3954 .is_partially_uptodate
= block_is_partially_uptodate
,
3955 .error_remove_page
= generic_error_remove_page
,
3958 static const struct address_space_operations ext4_da_aops
= {
3959 .readpage
= ext4_readpage
,
3960 .readpages
= ext4_readpages
,
3961 .writepage
= ext4_writepage
,
3962 .writepages
= ext4_writepages
,
3963 .write_begin
= ext4_da_write_begin
,
3964 .write_end
= ext4_da_write_end
,
3965 .set_page_dirty
= ext4_set_page_dirty
,
3967 .invalidatepage
= ext4_da_invalidatepage
,
3968 .releasepage
= ext4_releasepage
,
3969 .direct_IO
= ext4_direct_IO
,
3970 .migratepage
= buffer_migrate_page
,
3971 .is_partially_uptodate
= block_is_partially_uptodate
,
3972 .error_remove_page
= generic_error_remove_page
,
3975 void ext4_set_aops(struct inode
*inode
)
3977 switch (ext4_inode_journal_mode(inode
)) {
3978 case EXT4_INODE_ORDERED_DATA_MODE
:
3979 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3981 case EXT4_INODE_JOURNAL_DATA_MODE
:
3982 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3987 if (test_opt(inode
->i_sb
, DELALLOC
))
3988 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3990 inode
->i_mapping
->a_ops
= &ext4_aops
;
3993 static int __ext4_block_zero_page_range(handle_t
*handle
,
3994 struct address_space
*mapping
, loff_t from
, loff_t length
)
3996 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3997 unsigned offset
= from
& (PAGE_SIZE
-1);
3998 unsigned blocksize
, pos
;
4000 struct inode
*inode
= mapping
->host
;
4001 struct buffer_head
*bh
;
4005 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
4006 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
4010 blocksize
= inode
->i_sb
->s_blocksize
;
4012 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4014 if (!page_has_buffers(page
))
4015 create_empty_buffers(page
, blocksize
, 0);
4017 /* Find the buffer that contains "offset" */
4018 bh
= page_buffers(page
);
4020 while (offset
>= pos
) {
4021 bh
= bh
->b_this_page
;
4025 if (buffer_freed(bh
)) {
4026 BUFFER_TRACE(bh
, "freed: skip");
4029 if (!buffer_mapped(bh
)) {
4030 BUFFER_TRACE(bh
, "unmapped");
4031 ext4_get_block(inode
, iblock
, bh
, 0);
4032 /* unmapped? It's a hole - nothing to do */
4033 if (!buffer_mapped(bh
)) {
4034 BUFFER_TRACE(bh
, "still unmapped");
4039 /* Ok, it's mapped. Make sure it's up-to-date */
4040 if (PageUptodate(page
))
4041 set_buffer_uptodate(bh
);
4043 if (!buffer_uptodate(bh
)) {
4045 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
4047 /* Uhhuh. Read error. Complain and punt. */
4048 if (!buffer_uptodate(bh
))
4050 if (S_ISREG(inode
->i_mode
) &&
4051 ext4_encrypted_inode(inode
)) {
4052 /* We expect the key to be set. */
4053 BUG_ON(!fscrypt_has_encryption_key(inode
));
4054 BUG_ON(blocksize
!= PAGE_SIZE
);
4055 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
4056 page
, PAGE_SIZE
, 0, page
->index
));
4059 if (ext4_should_journal_data(inode
)) {
4060 BUFFER_TRACE(bh
, "get write access");
4061 err
= ext4_journal_get_write_access(handle
, bh
);
4065 zero_user(page
, offset
, length
);
4066 BUFFER_TRACE(bh
, "zeroed end of block");
4068 if (ext4_should_journal_data(inode
)) {
4069 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4072 mark_buffer_dirty(bh
);
4073 if (ext4_should_order_data(inode
))
4074 err
= ext4_jbd2_inode_add_write(handle
, inode
, from
,
4085 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4086 * starting from file offset 'from'. The range to be zero'd must
4087 * be contained with in one block. If the specified range exceeds
4088 * the end of the block it will be shortened to end of the block
4089 * that cooresponds to 'from'
4091 static int ext4_block_zero_page_range(handle_t
*handle
,
4092 struct address_space
*mapping
, loff_t from
, loff_t length
)
4094 struct inode
*inode
= mapping
->host
;
4095 unsigned offset
= from
& (PAGE_SIZE
-1);
4096 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4097 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
4100 * correct length if it does not fall between
4101 * 'from' and the end of the block
4103 if (length
> max
|| length
< 0)
4106 if (IS_DAX(inode
)) {
4107 return iomap_zero_range(inode
, from
, length
, NULL
,
4110 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4114 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4115 * up to the end of the block which corresponds to `from'.
4116 * This required during truncate. We need to physically zero the tail end
4117 * of that block so it doesn't yield old data if the file is later grown.
4119 static int ext4_block_truncate_page(handle_t
*handle
,
4120 struct address_space
*mapping
, loff_t from
)
4122 unsigned offset
= from
& (PAGE_SIZE
-1);
4125 struct inode
*inode
= mapping
->host
;
4127 /* If we are processing an encrypted inode during orphan list handling */
4128 if (ext4_encrypted_inode(inode
) && !fscrypt_has_encryption_key(inode
))
4131 blocksize
= inode
->i_sb
->s_blocksize
;
4132 length
= blocksize
- (offset
& (blocksize
- 1));
4134 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4137 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
4138 loff_t lstart
, loff_t length
)
4140 struct super_block
*sb
= inode
->i_sb
;
4141 struct address_space
*mapping
= inode
->i_mapping
;
4142 unsigned partial_start
, partial_end
;
4143 ext4_fsblk_t start
, end
;
4144 loff_t byte_end
= (lstart
+ length
- 1);
4147 partial_start
= lstart
& (sb
->s_blocksize
- 1);
4148 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
4150 start
= lstart
>> sb
->s_blocksize_bits
;
4151 end
= byte_end
>> sb
->s_blocksize_bits
;
4153 /* Handle partial zero within the single block */
4155 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
4156 err
= ext4_block_zero_page_range(handle
, mapping
,
4160 /* Handle partial zero out on the start of the range */
4161 if (partial_start
) {
4162 err
= ext4_block_zero_page_range(handle
, mapping
,
4163 lstart
, sb
->s_blocksize
);
4167 /* Handle partial zero out on the end of the range */
4168 if (partial_end
!= sb
->s_blocksize
- 1)
4169 err
= ext4_block_zero_page_range(handle
, mapping
,
4170 byte_end
- partial_end
,
4175 int ext4_can_truncate(struct inode
*inode
)
4177 if (S_ISREG(inode
->i_mode
))
4179 if (S_ISDIR(inode
->i_mode
))
4181 if (S_ISLNK(inode
->i_mode
))
4182 return !ext4_inode_is_fast_symlink(inode
);
4187 * We have to make sure i_disksize gets properly updated before we truncate
4188 * page cache due to hole punching or zero range. Otherwise i_disksize update
4189 * can get lost as it may have been postponed to submission of writeback but
4190 * that will never happen after we truncate page cache.
4192 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
4196 loff_t size
= i_size_read(inode
);
4198 WARN_ON(!inode_is_locked(inode
));
4199 if (offset
> size
|| offset
+ len
< size
)
4202 if (EXT4_I(inode
)->i_disksize
>= size
)
4205 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4207 return PTR_ERR(handle
);
4208 ext4_update_i_disksize(inode
, size
);
4209 ext4_mark_inode_dirty(handle
, inode
);
4210 ext4_journal_stop(handle
);
4216 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4217 * associated with the given offset and length
4219 * @inode: File inode
4220 * @offset: The offset where the hole will begin
4221 * @len: The length of the hole
4223 * Returns: 0 on success or negative on failure
4226 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4228 struct super_block
*sb
= inode
->i_sb
;
4229 ext4_lblk_t first_block
, stop_block
;
4230 struct address_space
*mapping
= inode
->i_mapping
;
4231 loff_t first_block_offset
, last_block_offset
;
4233 unsigned int credits
;
4236 if (!S_ISREG(inode
->i_mode
))
4239 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4241 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
4242 if (ext4_has_inline_data(inode
)) {
4243 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4244 ret
= ext4_convert_inline_data(inode
);
4245 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4251 * Write out all dirty pages to avoid race conditions
4252 * Then release them.
4254 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4255 ret
= filemap_write_and_wait_range(mapping
, offset
,
4256 offset
+ length
- 1);
4263 /* No need to punch hole beyond i_size */
4264 if (offset
>= inode
->i_size
)
4268 * If the hole extends beyond i_size, set the hole
4269 * to end after the page that contains i_size
4271 if (offset
+ length
> inode
->i_size
) {
4272 length
= inode
->i_size
+
4273 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4277 if (offset
& (sb
->s_blocksize
- 1) ||
4278 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4280 * Attach jinode to inode for jbd2 if we do any zeroing of
4283 ret
= ext4_inode_attach_jinode(inode
);
4289 /* Wait all existing dio workers, newcomers will block on i_mutex */
4290 ext4_inode_block_unlocked_dio(inode
);
4291 inode_dio_wait(inode
);
4294 * Prevent page faults from reinstantiating pages we have released from
4297 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4298 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4299 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4301 /* Now release the pages and zero block aligned part of pages*/
4302 if (last_block_offset
> first_block_offset
) {
4303 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4306 truncate_pagecache_range(inode
, first_block_offset
,
4310 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4311 credits
= ext4_writepage_trans_blocks(inode
);
4313 credits
= ext4_blocks_for_truncate(inode
);
4314 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4315 if (IS_ERR(handle
)) {
4316 ret
= PTR_ERR(handle
);
4317 ext4_std_error(sb
, ret
);
4321 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4326 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4327 EXT4_BLOCK_SIZE_BITS(sb
);
4328 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4330 /* If there are blocks to remove, do it */
4331 if (stop_block
> first_block
) {
4333 down_write(&EXT4_I(inode
)->i_data_sem
);
4334 ext4_discard_preallocations(inode
);
4336 ret
= ext4_es_remove_extent(inode
, first_block
,
4337 stop_block
- first_block
);
4339 up_write(&EXT4_I(inode
)->i_data_sem
);
4343 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4344 ret
= ext4_ext_remove_space(inode
, first_block
,
4347 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4350 up_write(&EXT4_I(inode
)->i_data_sem
);
4353 ext4_handle_sync(handle
);
4355 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4356 ext4_mark_inode_dirty(handle
, inode
);
4358 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4360 ext4_journal_stop(handle
);
4362 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4363 ext4_inode_resume_unlocked_dio(inode
);
4365 inode_unlock(inode
);
4369 int ext4_inode_attach_jinode(struct inode
*inode
)
4371 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4372 struct jbd2_inode
*jinode
;
4374 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4377 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4378 spin_lock(&inode
->i_lock
);
4381 spin_unlock(&inode
->i_lock
);
4384 ei
->jinode
= jinode
;
4385 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4388 spin_unlock(&inode
->i_lock
);
4389 if (unlikely(jinode
!= NULL
))
4390 jbd2_free_inode(jinode
);
4397 * We block out ext4_get_block() block instantiations across the entire
4398 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4399 * simultaneously on behalf of the same inode.
4401 * As we work through the truncate and commit bits of it to the journal there
4402 * is one core, guiding principle: the file's tree must always be consistent on
4403 * disk. We must be able to restart the truncate after a crash.
4405 * The file's tree may be transiently inconsistent in memory (although it
4406 * probably isn't), but whenever we close off and commit a journal transaction,
4407 * the contents of (the filesystem + the journal) must be consistent and
4408 * restartable. It's pretty simple, really: bottom up, right to left (although
4409 * left-to-right works OK too).
4411 * Note that at recovery time, journal replay occurs *before* the restart of
4412 * truncate against the orphan inode list.
4414 * The committed inode has the new, desired i_size (which is the same as
4415 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4416 * that this inode's truncate did not complete and it will again call
4417 * ext4_truncate() to have another go. So there will be instantiated blocks
4418 * to the right of the truncation point in a crashed ext4 filesystem. But
4419 * that's fine - as long as they are linked from the inode, the post-crash
4420 * ext4_truncate() run will find them and release them.
4422 int ext4_truncate(struct inode
*inode
)
4424 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4425 unsigned int credits
;
4428 struct address_space
*mapping
= inode
->i_mapping
;
4431 * There is a possibility that we're either freeing the inode
4432 * or it's a completely new inode. In those cases we might not
4433 * have i_mutex locked because it's not necessary.
4435 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4436 WARN_ON(!inode_is_locked(inode
));
4437 trace_ext4_truncate_enter(inode
);
4439 if (!ext4_can_truncate(inode
))
4442 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4444 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4445 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4447 if (ext4_has_inline_data(inode
)) {
4450 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4457 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4458 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4459 if (ext4_inode_attach_jinode(inode
) < 0)
4463 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4464 credits
= ext4_writepage_trans_blocks(inode
);
4466 credits
= ext4_blocks_for_truncate(inode
);
4468 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4470 return PTR_ERR(handle
);
4472 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4473 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4476 * We add the inode to the orphan list, so that if this
4477 * truncate spans multiple transactions, and we crash, we will
4478 * resume the truncate when the filesystem recovers. It also
4479 * marks the inode dirty, to catch the new size.
4481 * Implication: the file must always be in a sane, consistent
4482 * truncatable state while each transaction commits.
4484 err
= ext4_orphan_add(handle
, inode
);
4488 down_write(&EXT4_I(inode
)->i_data_sem
);
4490 ext4_discard_preallocations(inode
);
4492 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4493 err
= ext4_ext_truncate(handle
, inode
);
4495 ext4_ind_truncate(handle
, inode
);
4497 up_write(&ei
->i_data_sem
);
4502 ext4_handle_sync(handle
);
4506 * If this was a simple ftruncate() and the file will remain alive,
4507 * then we need to clear up the orphan record which we created above.
4508 * However, if this was a real unlink then we were called by
4509 * ext4_evict_inode(), and we allow that function to clean up the
4510 * orphan info for us.
4513 ext4_orphan_del(handle
, inode
);
4515 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4516 ext4_mark_inode_dirty(handle
, inode
);
4517 ext4_journal_stop(handle
);
4519 trace_ext4_truncate_exit(inode
);
4524 * ext4_get_inode_loc returns with an extra refcount against the inode's
4525 * underlying buffer_head on success. If 'in_mem' is true, we have all
4526 * data in memory that is needed to recreate the on-disk version of this
4529 static int __ext4_get_inode_loc(struct inode
*inode
,
4530 struct ext4_iloc
*iloc
, int in_mem
)
4532 struct ext4_group_desc
*gdp
;
4533 struct buffer_head
*bh
;
4534 struct super_block
*sb
= inode
->i_sb
;
4536 int inodes_per_block
, inode_offset
;
4539 if (inode
->i_ino
< EXT4_ROOT_INO
||
4540 inode
->i_ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
4541 return -EFSCORRUPTED
;
4543 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4544 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4549 * Figure out the offset within the block group inode table
4551 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4552 inode_offset
= ((inode
->i_ino
- 1) %
4553 EXT4_INODES_PER_GROUP(sb
));
4554 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4555 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4557 bh
= sb_getblk(sb
, block
);
4560 if (!buffer_uptodate(bh
)) {
4564 * If the buffer has the write error flag, we have failed
4565 * to write out another inode in the same block. In this
4566 * case, we don't have to read the block because we may
4567 * read the old inode data successfully.
4569 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4570 set_buffer_uptodate(bh
);
4572 if (buffer_uptodate(bh
)) {
4573 /* someone brought it uptodate while we waited */
4579 * If we have all information of the inode in memory and this
4580 * is the only valid inode in the block, we need not read the
4584 struct buffer_head
*bitmap_bh
;
4587 start
= inode_offset
& ~(inodes_per_block
- 1);
4589 /* Is the inode bitmap in cache? */
4590 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4591 if (unlikely(!bitmap_bh
))
4595 * If the inode bitmap isn't in cache then the
4596 * optimisation may end up performing two reads instead
4597 * of one, so skip it.
4599 if (!buffer_uptodate(bitmap_bh
)) {
4603 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4604 if (i
== inode_offset
)
4606 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4610 if (i
== start
+ inodes_per_block
) {
4611 /* all other inodes are free, so skip I/O */
4612 memset(bh
->b_data
, 0, bh
->b_size
);
4613 set_buffer_uptodate(bh
);
4621 * If we need to do any I/O, try to pre-readahead extra
4622 * blocks from the inode table.
4624 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4625 ext4_fsblk_t b
, end
, table
;
4627 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4629 table
= ext4_inode_table(sb
, gdp
);
4630 /* s_inode_readahead_blks is always a power of 2 */
4631 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4635 num
= EXT4_INODES_PER_GROUP(sb
);
4636 if (ext4_has_group_desc_csum(sb
))
4637 num
-= ext4_itable_unused_count(sb
, gdp
);
4638 table
+= num
/ inodes_per_block
;
4642 sb_breadahead(sb
, b
++);
4646 * There are other valid inodes in the buffer, this inode
4647 * has in-inode xattrs, or we don't have this inode in memory.
4648 * Read the block from disk.
4650 trace_ext4_load_inode(inode
);
4652 bh
->b_end_io
= end_buffer_read_sync
;
4653 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4655 if (!buffer_uptodate(bh
)) {
4656 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4657 "unable to read itable block");
4667 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4669 /* We have all inode data except xattrs in memory here. */
4670 return __ext4_get_inode_loc(inode
, iloc
,
4671 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4674 static bool ext4_should_use_dax(struct inode
*inode
)
4676 if (!test_opt(inode
->i_sb
, DAX
))
4678 if (!S_ISREG(inode
->i_mode
))
4680 if (ext4_should_journal_data(inode
))
4682 if (ext4_has_inline_data(inode
))
4684 if (ext4_encrypted_inode(inode
))
4689 void ext4_set_inode_flags(struct inode
*inode
)
4691 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4692 unsigned int new_fl
= 0;
4694 if (flags
& EXT4_SYNC_FL
)
4696 if (flags
& EXT4_APPEND_FL
)
4698 if (flags
& EXT4_IMMUTABLE_FL
)
4699 new_fl
|= S_IMMUTABLE
;
4700 if (flags
& EXT4_NOATIME_FL
)
4701 new_fl
|= S_NOATIME
;
4702 if (flags
& EXT4_DIRSYNC_FL
)
4703 new_fl
|= S_DIRSYNC
;
4704 if (ext4_should_use_dax(inode
))
4706 if (flags
& EXT4_ENCRYPT_FL
)
4707 new_fl
|= S_ENCRYPTED
;
4708 inode_set_flags(inode
, new_fl
,
4709 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
|
4713 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4714 struct ext4_inode_info
*ei
)
4717 struct inode
*inode
= &(ei
->vfs_inode
);
4718 struct super_block
*sb
= inode
->i_sb
;
4720 if (ext4_has_feature_huge_file(sb
)) {
4721 /* we are using combined 48 bit field */
4722 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4723 le32_to_cpu(raw_inode
->i_blocks_lo
);
4724 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4725 /* i_blocks represent file system block size */
4726 return i_blocks
<< (inode
->i_blkbits
- 9);
4731 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4735 static inline int ext4_iget_extra_inode(struct inode
*inode
,
4736 struct ext4_inode
*raw_inode
,
4737 struct ext4_inode_info
*ei
)
4739 __le32
*magic
= (void *)raw_inode
+
4740 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4742 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4743 EXT4_INODE_SIZE(inode
->i_sb
) &&
4744 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4745 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4746 return ext4_find_inline_data_nolock(inode
);
4748 EXT4_I(inode
)->i_inline_off
= 0;
4752 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4754 if (!ext4_has_feature_project(inode
->i_sb
))
4756 *projid
= EXT4_I(inode
)->i_projid
;
4760 struct inode
*__ext4_iget(struct super_block
*sb
, unsigned long ino
,
4761 ext4_iget_flags flags
, const char *function
,
4764 struct ext4_iloc iloc
;
4765 struct ext4_inode
*raw_inode
;
4766 struct ext4_inode_info
*ei
;
4767 struct inode
*inode
;
4768 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4776 if ((!(flags
& EXT4_IGET_SPECIAL
) &&
4777 (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)) ||
4778 (ino
< EXT4_ROOT_INO
) ||
4779 (ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))) {
4780 if (flags
& EXT4_IGET_HANDLE
)
4781 return ERR_PTR(-ESTALE
);
4782 __ext4_error(sb
, function
, line
,
4783 "inode #%lu: comm %s: iget: illegal inode #",
4784 ino
, current
->comm
);
4785 return ERR_PTR(-EFSCORRUPTED
);
4788 inode
= iget_locked(sb
, ino
);
4790 return ERR_PTR(-ENOMEM
);
4791 if (!(inode
->i_state
& I_NEW
))
4797 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4800 raw_inode
= ext4_raw_inode(&iloc
);
4802 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4803 ext4_error_inode(inode
, function
, line
, 0,
4804 "iget: root inode unallocated");
4805 ret
= -EFSCORRUPTED
;
4809 if ((flags
& EXT4_IGET_HANDLE
) &&
4810 (raw_inode
->i_links_count
== 0) && (raw_inode
->i_mode
== 0)) {
4815 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4816 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4817 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4818 EXT4_INODE_SIZE(inode
->i_sb
) ||
4819 (ei
->i_extra_isize
& 3)) {
4820 ext4_error_inode(inode
, function
, line
, 0,
4821 "iget: bad extra_isize %u "
4824 EXT4_INODE_SIZE(inode
->i_sb
));
4825 ret
= -EFSCORRUPTED
;
4829 ei
->i_extra_isize
= 0;
4831 /* Precompute checksum seed for inode metadata */
4832 if (ext4_has_metadata_csum(sb
)) {
4833 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4835 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4836 __le32 gen
= raw_inode
->i_generation
;
4837 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4839 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4843 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4844 ext4_error_inode(inode
, function
, line
, 0,
4845 "iget: checksum invalid");
4850 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4851 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4852 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4853 if (ext4_has_feature_project(sb
) &&
4854 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4855 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4856 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4858 i_projid
= EXT4_DEF_PROJID
;
4860 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4861 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4862 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4864 i_uid_write(inode
, i_uid
);
4865 i_gid_write(inode
, i_gid
);
4866 ei
->i_projid
= make_kprojid(sb
->s_user_ns
, i_projid
);
4867 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4869 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4870 ei
->i_inline_off
= 0;
4871 ei
->i_dir_start_lookup
= 0;
4872 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4873 /* We now have enough fields to check if the inode was active or not.
4874 * This is needed because nfsd might try to access dead inodes
4875 * the test is that same one that e2fsck uses
4876 * NeilBrown 1999oct15
4878 if (inode
->i_nlink
== 0) {
4879 if ((inode
->i_mode
== 0 ||
4880 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4881 ino
!= EXT4_BOOT_LOADER_INO
) {
4882 /* this inode is deleted */
4886 /* The only unlinked inodes we let through here have
4887 * valid i_mode and are being read by the orphan
4888 * recovery code: that's fine, we're about to complete
4889 * the process of deleting those.
4890 * OR it is the EXT4_BOOT_LOADER_INO which is
4891 * not initialized on a new filesystem. */
4893 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4894 ext4_set_inode_flags(inode
);
4895 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4896 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4897 if (ext4_has_feature_64bit(sb
))
4899 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4900 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4901 if ((size
= i_size_read(inode
)) < 0) {
4902 ext4_error_inode(inode
, function
, line
, 0,
4903 "iget: bad i_size value: %lld", size
);
4904 ret
= -EFSCORRUPTED
;
4907 ei
->i_disksize
= inode
->i_size
;
4909 ei
->i_reserved_quota
= 0;
4911 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4912 ei
->i_block_group
= iloc
.block_group
;
4913 ei
->i_last_alloc_group
= ~0;
4915 * NOTE! The in-memory inode i_data array is in little-endian order
4916 * even on big-endian machines: we do NOT byteswap the block numbers!
4918 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4919 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4920 INIT_LIST_HEAD(&ei
->i_orphan
);
4923 * Set transaction id's of transactions that have to be committed
4924 * to finish f[data]sync. We set them to currently running transaction
4925 * as we cannot be sure that the inode or some of its metadata isn't
4926 * part of the transaction - the inode could have been reclaimed and
4927 * now it is reread from disk.
4930 transaction_t
*transaction
;
4933 read_lock(&journal
->j_state_lock
);
4934 if (journal
->j_running_transaction
)
4935 transaction
= journal
->j_running_transaction
;
4937 transaction
= journal
->j_committing_transaction
;
4939 tid
= transaction
->t_tid
;
4941 tid
= journal
->j_commit_sequence
;
4942 read_unlock(&journal
->j_state_lock
);
4943 ei
->i_sync_tid
= tid
;
4944 ei
->i_datasync_tid
= tid
;
4947 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4948 if (ei
->i_extra_isize
== 0) {
4949 /* The extra space is currently unused. Use it. */
4950 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4951 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4952 EXT4_GOOD_OLD_INODE_SIZE
;
4954 ret
= ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4960 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4961 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4962 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4963 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4965 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4966 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4967 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4968 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4970 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4975 if (ei
->i_file_acl
&&
4976 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4977 ext4_error_inode(inode
, function
, line
, 0,
4978 "iget: bad extended attribute block %llu",
4980 ret
= -EFSCORRUPTED
;
4982 } else if (!ext4_has_inline_data(inode
)) {
4983 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4984 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4985 (S_ISLNK(inode
->i_mode
) &&
4986 !ext4_inode_is_fast_symlink(inode
))))
4987 /* Validate extent which is part of inode */
4988 ret
= ext4_ext_check_inode(inode
);
4989 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4990 (S_ISLNK(inode
->i_mode
) &&
4991 !ext4_inode_is_fast_symlink(inode
))) {
4992 /* Validate block references which are part of inode */
4993 ret
= ext4_ind_check_inode(inode
);
4999 if (S_ISREG(inode
->i_mode
)) {
5000 inode
->i_op
= &ext4_file_inode_operations
;
5001 inode
->i_fop
= &ext4_file_operations
;
5002 ext4_set_aops(inode
);
5003 } else if (S_ISDIR(inode
->i_mode
)) {
5004 inode
->i_op
= &ext4_dir_inode_operations
;
5005 inode
->i_fop
= &ext4_dir_operations
;
5006 } else if (S_ISLNK(inode
->i_mode
)) {
5007 /* VFS does not allow setting these so must be corruption */
5008 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
5009 ext4_error_inode(inode
, function
, line
, 0,
5010 "iget: immutable or append flags "
5011 "not allowed on symlinks");
5012 ret
= -EFSCORRUPTED
;
5015 if (ext4_encrypted_inode(inode
)) {
5016 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
5017 ext4_set_aops(inode
);
5018 } else if (ext4_inode_is_fast_symlink(inode
)) {
5019 inode
->i_link
= (char *)ei
->i_data
;
5020 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5021 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5022 sizeof(ei
->i_data
) - 1);
5024 inode
->i_op
= &ext4_symlink_inode_operations
;
5025 ext4_set_aops(inode
);
5027 inode_nohighmem(inode
);
5028 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5029 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5030 inode
->i_op
= &ext4_special_inode_operations
;
5031 if (raw_inode
->i_block
[0])
5032 init_special_inode(inode
, inode
->i_mode
,
5033 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5035 init_special_inode(inode
, inode
->i_mode
,
5036 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5037 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
5038 make_bad_inode(inode
);
5040 ret
= -EFSCORRUPTED
;
5041 ext4_error_inode(inode
, function
, line
, 0,
5042 "iget: bogus i_mode (%o)", inode
->i_mode
);
5047 unlock_new_inode(inode
);
5053 return ERR_PTR(ret
);
5056 static int ext4_inode_blocks_set(handle_t
*handle
,
5057 struct ext4_inode
*raw_inode
,
5058 struct ext4_inode_info
*ei
)
5060 struct inode
*inode
= &(ei
->vfs_inode
);
5061 u64 i_blocks
= inode
->i_blocks
;
5062 struct super_block
*sb
= inode
->i_sb
;
5064 if (i_blocks
<= ~0U) {
5066 * i_blocks can be represented in a 32 bit variable
5067 * as multiple of 512 bytes
5069 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5070 raw_inode
->i_blocks_high
= 0;
5071 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5074 if (!ext4_has_feature_huge_file(sb
))
5077 if (i_blocks
<= 0xffffffffffffULL
) {
5079 * i_blocks can be represented in a 48 bit variable
5080 * as multiple of 512 bytes
5082 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5083 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5084 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5086 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5087 /* i_block is stored in file system block size */
5088 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5089 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5090 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5095 struct other_inode
{
5096 unsigned long orig_ino
;
5097 struct ext4_inode
*raw_inode
;
5100 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
5103 struct other_inode
*oi
= (struct other_inode
*) data
;
5105 if ((inode
->i_ino
!= ino
) ||
5106 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
5107 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
5108 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
5110 spin_lock(&inode
->i_lock
);
5111 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
5112 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
5113 (inode
->i_state
& I_DIRTY_TIME
)) {
5114 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5116 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
5117 spin_unlock(&inode
->i_lock
);
5119 spin_lock(&ei
->i_raw_lock
);
5120 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
5121 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
5122 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
5123 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
5124 spin_unlock(&ei
->i_raw_lock
);
5125 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
5128 spin_unlock(&inode
->i_lock
);
5133 * Opportunistically update the other time fields for other inodes in
5134 * the same inode table block.
5136 static void ext4_update_other_inodes_time(struct super_block
*sb
,
5137 unsigned long orig_ino
, char *buf
)
5139 struct other_inode oi
;
5141 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
5142 int inode_size
= EXT4_INODE_SIZE(sb
);
5144 oi
.orig_ino
= orig_ino
;
5146 * Calculate the first inode in the inode table block. Inode
5147 * numbers are one-based. That is, the first inode in a block
5148 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5150 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
5151 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
5152 if (ino
== orig_ino
)
5154 oi
.raw_inode
= (struct ext4_inode
*) buf
;
5155 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
5160 * Post the struct inode info into an on-disk inode location in the
5161 * buffer-cache. This gobbles the caller's reference to the
5162 * buffer_head in the inode location struct.
5164 * The caller must have write access to iloc->bh.
5166 static int ext4_do_update_inode(handle_t
*handle
,
5167 struct inode
*inode
,
5168 struct ext4_iloc
*iloc
)
5170 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5171 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5172 struct buffer_head
*bh
= iloc
->bh
;
5173 struct super_block
*sb
= inode
->i_sb
;
5174 int err
= 0, rc
, block
;
5175 int need_datasync
= 0, set_large_file
= 0;
5180 spin_lock(&ei
->i_raw_lock
);
5182 /* For fields not tracked in the in-memory inode,
5183 * initialise them to zero for new inodes. */
5184 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5185 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5187 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5188 i_uid
= i_uid_read(inode
);
5189 i_gid
= i_gid_read(inode
);
5190 i_projid
= from_kprojid(sb
->s_user_ns
, ei
->i_projid
);
5191 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5192 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
5193 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
5195 * Fix up interoperability with old kernels. Otherwise, old inodes get
5196 * re-used with the upper 16 bits of the uid/gid intact
5198 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
5199 raw_inode
->i_uid_high
= 0;
5200 raw_inode
->i_gid_high
= 0;
5202 raw_inode
->i_uid_high
=
5203 cpu_to_le16(high_16_bits(i_uid
));
5204 raw_inode
->i_gid_high
=
5205 cpu_to_le16(high_16_bits(i_gid
));
5208 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
5209 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
5210 raw_inode
->i_uid_high
= 0;
5211 raw_inode
->i_gid_high
= 0;
5213 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5215 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5216 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5217 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5218 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5220 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5222 spin_unlock(&ei
->i_raw_lock
);
5225 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5226 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5227 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5228 raw_inode
->i_file_acl_high
=
5229 cpu_to_le16(ei
->i_file_acl
>> 32);
5230 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5231 if (ei
->i_disksize
!= ext4_isize(inode
->i_sb
, raw_inode
)) {
5232 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5235 if (ei
->i_disksize
> 0x7fffffffULL
) {
5236 if (!ext4_has_feature_large_file(sb
) ||
5237 EXT4_SB(sb
)->s_es
->s_rev_level
==
5238 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5241 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5242 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5243 if (old_valid_dev(inode
->i_rdev
)) {
5244 raw_inode
->i_block
[0] =
5245 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5246 raw_inode
->i_block
[1] = 0;
5248 raw_inode
->i_block
[0] = 0;
5249 raw_inode
->i_block
[1] =
5250 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5251 raw_inode
->i_block
[2] = 0;
5253 } else if (!ext4_has_inline_data(inode
)) {
5254 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5255 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5258 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5259 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5260 if (ei
->i_extra_isize
) {
5261 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5262 raw_inode
->i_version_hi
=
5263 cpu_to_le32(inode
->i_version
>> 32);
5264 raw_inode
->i_extra_isize
=
5265 cpu_to_le16(ei
->i_extra_isize
);
5269 if (i_projid
!= (projid_t
)-1) {
5270 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5271 i_projid
!= EXT4_DEF_PROJID
);
5273 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5274 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5275 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5278 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5279 spin_unlock(&ei
->i_raw_lock
);
5280 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
)
5281 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5284 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5285 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5288 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5289 if (set_large_file
) {
5290 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5291 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5294 ext4_set_feature_large_file(sb
);
5295 ext4_handle_sync(handle
);
5296 err
= ext4_handle_dirty_super(handle
, sb
);
5298 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5301 ext4_std_error(inode
->i_sb
, err
);
5306 * ext4_write_inode()
5308 * We are called from a few places:
5310 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5311 * Here, there will be no transaction running. We wait for any running
5312 * transaction to commit.
5314 * - Within flush work (sys_sync(), kupdate and such).
5315 * We wait on commit, if told to.
5317 * - Within iput_final() -> write_inode_now()
5318 * We wait on commit, if told to.
5320 * In all cases it is actually safe for us to return without doing anything,
5321 * because the inode has been copied into a raw inode buffer in
5322 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5325 * Note that we are absolutely dependent upon all inode dirtiers doing the
5326 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5327 * which we are interested.
5329 * It would be a bug for them to not do this. The code:
5331 * mark_inode_dirty(inode)
5333 * inode->i_size = expr;
5335 * is in error because write_inode() could occur while `stuff()' is running,
5336 * and the new i_size will be lost. Plus the inode will no longer be on the
5337 * superblock's dirty inode list.
5339 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5343 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
) ||
5344 sb_rdonly(inode
->i_sb
))
5347 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5350 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5351 if (ext4_journal_current_handle()) {
5352 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5358 * No need to force transaction in WB_SYNC_NONE mode. Also
5359 * ext4_sync_fs() will force the commit after everything is
5362 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5365 err
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5366 EXT4_I(inode
)->i_sync_tid
);
5368 struct ext4_iloc iloc
;
5370 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5374 * sync(2) will flush the whole buffer cache. No need to do
5375 * it here separately for each inode.
5377 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5378 sync_dirty_buffer(iloc
.bh
);
5379 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5380 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5381 "IO error syncing inode");
5390 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5391 * buffers that are attached to a page stradding i_size and are undergoing
5392 * commit. In that case we have to wait for commit to finish and try again.
5394 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5398 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5399 tid_t commit_tid
= 0;
5402 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5404 * If the page is fully truncated, we don't need to wait for any commit
5405 * (and we even should not as __ext4_journalled_invalidatepage() may
5406 * strip all buffers from the page but keep the page dirty which can then
5407 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5408 * buffers). Also we don't need to wait for any commit if all buffers in
5409 * the page remain valid. This is most beneficial for the common case of
5410 * blocksize == PAGESIZE.
5412 if (!offset
|| offset
> (PAGE_SIZE
- i_blocksize(inode
)))
5415 page
= find_lock_page(inode
->i_mapping
,
5416 inode
->i_size
>> PAGE_SHIFT
);
5419 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5420 PAGE_SIZE
- offset
);
5426 read_lock(&journal
->j_state_lock
);
5427 if (journal
->j_committing_transaction
)
5428 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5429 read_unlock(&journal
->j_state_lock
);
5431 jbd2_log_wait_commit(journal
, commit_tid
);
5438 * Called from notify_change.
5440 * We want to trap VFS attempts to truncate the file as soon as
5441 * possible. In particular, we want to make sure that when the VFS
5442 * shrinks i_size, we put the inode on the orphan list and modify
5443 * i_disksize immediately, so that during the subsequent flushing of
5444 * dirty pages and freeing of disk blocks, we can guarantee that any
5445 * commit will leave the blocks being flushed in an unused state on
5446 * disk. (On recovery, the inode will get truncated and the blocks will
5447 * be freed, so we have a strong guarantee that no future commit will
5448 * leave these blocks visible to the user.)
5450 * Another thing we have to assure is that if we are in ordered mode
5451 * and inode is still attached to the committing transaction, we must
5452 * we start writeout of all the dirty pages which are being truncated.
5453 * This way we are sure that all the data written in the previous
5454 * transaction are already on disk (truncate waits for pages under
5457 * Called with inode->i_mutex down.
5459 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5461 struct inode
*inode
= d_inode(dentry
);
5464 const unsigned int ia_valid
= attr
->ia_valid
;
5466 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5469 if (unlikely(IS_IMMUTABLE(inode
)))
5472 if (unlikely(IS_APPEND(inode
) &&
5473 (ia_valid
& (ATTR_MODE
| ATTR_UID
|
5474 ATTR_GID
| ATTR_TIMES_SET
))))
5477 error
= setattr_prepare(dentry
, attr
);
5481 error
= fscrypt_prepare_setattr(dentry
, attr
);
5485 if (is_quota_modification(inode
, attr
)) {
5486 error
= dquot_initialize(inode
);
5490 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5491 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5494 /* (user+group)*(old+new) structure, inode write (sb,
5495 * inode block, ? - but truncate inode update has it) */
5496 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5497 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5498 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5499 if (IS_ERR(handle
)) {
5500 error
= PTR_ERR(handle
);
5504 /* dquot_transfer() calls back ext4_get_inode_usage() which
5505 * counts xattr inode references.
5507 down_read(&EXT4_I(inode
)->xattr_sem
);
5508 error
= dquot_transfer(inode
, attr
);
5509 up_read(&EXT4_I(inode
)->xattr_sem
);
5512 ext4_journal_stop(handle
);
5515 /* Update corresponding info in inode so that everything is in
5516 * one transaction */
5517 if (attr
->ia_valid
& ATTR_UID
)
5518 inode
->i_uid
= attr
->ia_uid
;
5519 if (attr
->ia_valid
& ATTR_GID
)
5520 inode
->i_gid
= attr
->ia_gid
;
5521 error
= ext4_mark_inode_dirty(handle
, inode
);
5522 ext4_journal_stop(handle
);
5525 if (attr
->ia_valid
& ATTR_SIZE
) {
5527 loff_t oldsize
= inode
->i_size
;
5528 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5530 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5531 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5533 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5536 if (!S_ISREG(inode
->i_mode
))
5539 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5540 inode_inc_iversion(inode
);
5542 if (ext4_should_order_data(inode
) &&
5543 (attr
->ia_size
< inode
->i_size
)) {
5544 error
= ext4_begin_ordered_truncate(inode
,
5549 if (attr
->ia_size
!= inode
->i_size
) {
5550 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5551 if (IS_ERR(handle
)) {
5552 error
= PTR_ERR(handle
);
5555 if (ext4_handle_valid(handle
) && shrink
) {
5556 error
= ext4_orphan_add(handle
, inode
);
5560 * Update c/mtime on truncate up, ext4_truncate() will
5561 * update c/mtime in shrink case below
5564 inode
->i_mtime
= current_time(inode
);
5565 inode
->i_ctime
= inode
->i_mtime
;
5567 down_write(&EXT4_I(inode
)->i_data_sem
);
5568 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5569 rc
= ext4_mark_inode_dirty(handle
, inode
);
5573 * We have to update i_size under i_data_sem together
5574 * with i_disksize to avoid races with writeback code
5575 * running ext4_wb_update_i_disksize().
5578 i_size_write(inode
, attr
->ia_size
);
5579 up_write(&EXT4_I(inode
)->i_data_sem
);
5580 ext4_journal_stop(handle
);
5582 if (orphan
&& inode
->i_nlink
)
5583 ext4_orphan_del(NULL
, inode
);
5588 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5591 * Blocks are going to be removed from the inode. Wait
5592 * for dio in flight.
5594 inode_dio_wait(inode
);
5596 if (orphan
&& ext4_should_journal_data(inode
))
5597 ext4_wait_for_tail_page_commit(inode
);
5598 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5600 * Truncate pagecache after we've waited for commit
5601 * in data=journal mode to make pages freeable.
5603 truncate_pagecache(inode
, inode
->i_size
);
5605 rc
= ext4_truncate(inode
);
5609 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5613 setattr_copy(inode
, attr
);
5614 mark_inode_dirty(inode
);
5618 * If the call to ext4_truncate failed to get a transaction handle at
5619 * all, we need to clean up the in-core orphan list manually.
5621 if (orphan
&& inode
->i_nlink
)
5622 ext4_orphan_del(NULL
, inode
);
5624 if (!error
&& (ia_valid
& ATTR_MODE
))
5625 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5628 ext4_std_error(inode
->i_sb
, error
);
5634 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5635 u32 request_mask
, unsigned int query_flags
)
5637 struct inode
*inode
= d_inode(path
->dentry
);
5638 struct ext4_inode
*raw_inode
;
5639 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5642 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5643 stat
->result_mask
|= STATX_BTIME
;
5644 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5645 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5648 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5649 if (flags
& EXT4_APPEND_FL
)
5650 stat
->attributes
|= STATX_ATTR_APPEND
;
5651 if (flags
& EXT4_COMPR_FL
)
5652 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5653 if (flags
& EXT4_ENCRYPT_FL
)
5654 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5655 if (flags
& EXT4_IMMUTABLE_FL
)
5656 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5657 if (flags
& EXT4_NODUMP_FL
)
5658 stat
->attributes
|= STATX_ATTR_NODUMP
;
5660 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5661 STATX_ATTR_COMPRESSED
|
5662 STATX_ATTR_ENCRYPTED
|
5663 STATX_ATTR_IMMUTABLE
|
5666 generic_fillattr(inode
, stat
);
5670 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5671 u32 request_mask
, unsigned int query_flags
)
5673 struct inode
*inode
= d_inode(path
->dentry
);
5674 u64 delalloc_blocks
;
5676 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5679 * If there is inline data in the inode, the inode will normally not
5680 * have data blocks allocated (it may have an external xattr block).
5681 * Report at least one sector for such files, so tools like tar, rsync,
5682 * others don't incorrectly think the file is completely sparse.
5684 if (unlikely(ext4_has_inline_data(inode
)))
5685 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5688 * We can't update i_blocks if the block allocation is delayed
5689 * otherwise in the case of system crash before the real block
5690 * allocation is done, we will have i_blocks inconsistent with
5691 * on-disk file blocks.
5692 * We always keep i_blocks updated together with real
5693 * allocation. But to not confuse with user, stat
5694 * will return the blocks that include the delayed allocation
5695 * blocks for this file.
5697 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5698 EXT4_I(inode
)->i_reserved_data_blocks
);
5699 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5703 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5706 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5707 return ext4_ind_trans_blocks(inode
, lblocks
);
5708 return ext4_ext_index_trans_blocks(inode
, pextents
);
5712 * Account for index blocks, block groups bitmaps and block group
5713 * descriptor blocks if modify datablocks and index blocks
5714 * worse case, the indexs blocks spread over different block groups
5716 * If datablocks are discontiguous, they are possible to spread over
5717 * different block groups too. If they are contiguous, with flexbg,
5718 * they could still across block group boundary.
5720 * Also account for superblock, inode, quota and xattr blocks
5722 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5725 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5731 * How many index blocks need to touch to map @lblocks logical blocks
5732 * to @pextents physical extents?
5734 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5739 * Now let's see how many group bitmaps and group descriptors need
5742 groups
= idxblocks
+ pextents
;
5744 if (groups
> ngroups
)
5746 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5747 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5749 /* bitmaps and block group descriptor blocks */
5750 ret
+= groups
+ gdpblocks
;
5752 /* Blocks for super block, inode, quota and xattr blocks */
5753 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5759 * Calculate the total number of credits to reserve to fit
5760 * the modification of a single pages into a single transaction,
5761 * which may include multiple chunks of block allocations.
5763 * This could be called via ext4_write_begin()
5765 * We need to consider the worse case, when
5766 * one new block per extent.
5768 int ext4_writepage_trans_blocks(struct inode
*inode
)
5770 int bpp
= ext4_journal_blocks_per_page(inode
);
5773 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5775 /* Account for data blocks for journalled mode */
5776 if (ext4_should_journal_data(inode
))
5782 * Calculate the journal credits for a chunk of data modification.
5784 * This is called from DIO, fallocate or whoever calling
5785 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5787 * journal buffers for data blocks are not included here, as DIO
5788 * and fallocate do no need to journal data buffers.
5790 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5792 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5796 * The caller must have previously called ext4_reserve_inode_write().
5797 * Give this, we know that the caller already has write access to iloc->bh.
5799 int ext4_mark_iloc_dirty(handle_t
*handle
,
5800 struct inode
*inode
, struct ext4_iloc
*iloc
)
5804 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
5808 if (IS_I_VERSION(inode
))
5809 inode_inc_iversion(inode
);
5811 /* the do_update_inode consumes one bh->b_count */
5814 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5815 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5821 * On success, We end up with an outstanding reference count against
5822 * iloc->bh. This _must_ be cleaned up later.
5826 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5827 struct ext4_iloc
*iloc
)
5831 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5834 err
= ext4_get_inode_loc(inode
, iloc
);
5836 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5837 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5843 ext4_std_error(inode
->i_sb
, err
);
5847 static int __ext4_expand_extra_isize(struct inode
*inode
,
5848 unsigned int new_extra_isize
,
5849 struct ext4_iloc
*iloc
,
5850 handle_t
*handle
, int *no_expand
)
5852 struct ext4_inode
*raw_inode
;
5853 struct ext4_xattr_ibody_header
*header
;
5854 unsigned int inode_size
= EXT4_INODE_SIZE(inode
->i_sb
);
5855 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5858 /* this was checked at iget time, but double check for good measure */
5859 if ((EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
> inode_size
) ||
5860 (ei
->i_extra_isize
& 3)) {
5861 EXT4_ERROR_INODE(inode
, "bad extra_isize %u (inode size %u)",
5863 EXT4_INODE_SIZE(inode
->i_sb
));
5864 return -EFSCORRUPTED
;
5866 if ((new_extra_isize
< ei
->i_extra_isize
) ||
5867 (new_extra_isize
< 4) ||
5868 (new_extra_isize
> inode_size
- EXT4_GOOD_OLD_INODE_SIZE
))
5869 return -EINVAL
; /* Should never happen */
5871 raw_inode
= ext4_raw_inode(iloc
);
5873 header
= IHDR(inode
, raw_inode
);
5875 /* No extended attributes present */
5876 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5877 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5878 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5879 EXT4_I(inode
)->i_extra_isize
, 0,
5880 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5881 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5885 /* try to expand with EAs present */
5886 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5890 * Inode size expansion failed; don't try again
5899 * Expand an inode by new_extra_isize bytes.
5900 * Returns 0 on success or negative error number on failure.
5902 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5903 unsigned int new_extra_isize
,
5904 struct ext4_iloc iloc
,
5910 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5914 * In nojournal mode, we can immediately attempt to expand
5915 * the inode. When journaled, we first need to obtain extra
5916 * buffer credits since we may write into the EA block
5917 * with this same handle. If journal_extend fails, then it will
5918 * only result in a minor loss of functionality for that inode.
5919 * If this is felt to be critical, then e2fsck should be run to
5920 * force a large enough s_min_extra_isize.
5922 if (ext4_handle_valid(handle
) &&
5923 jbd2_journal_extend(handle
,
5924 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) != 0)
5927 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5930 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5931 handle
, &no_expand
);
5932 ext4_write_unlock_xattr(inode
, &no_expand
);
5937 int ext4_expand_extra_isize(struct inode
*inode
,
5938 unsigned int new_extra_isize
,
5939 struct ext4_iloc
*iloc
)
5945 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5950 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5951 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5952 if (IS_ERR(handle
)) {
5953 error
= PTR_ERR(handle
);
5958 ext4_write_lock_xattr(inode
, &no_expand
);
5960 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5961 error
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5967 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
5968 handle
, &no_expand
);
5970 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
5975 ext4_write_unlock_xattr(inode
, &no_expand
);
5976 ext4_journal_stop(handle
);
5981 * What we do here is to mark the in-core inode as clean with respect to inode
5982 * dirtiness (it may still be data-dirty).
5983 * This means that the in-core inode may be reaped by prune_icache
5984 * without having to perform any I/O. This is a very good thing,
5985 * because *any* task may call prune_icache - even ones which
5986 * have a transaction open against a different journal.
5988 * Is this cheating? Not really. Sure, we haven't written the
5989 * inode out, but prune_icache isn't a user-visible syncing function.
5990 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5991 * we start and wait on commits.
5993 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5995 struct ext4_iloc iloc
;
5996 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
6000 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
6001 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
6005 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
6006 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
6009 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
6013 * ext4_dirty_inode() is called from __mark_inode_dirty()
6015 * We're really interested in the case where a file is being extended.
6016 * i_size has been changed by generic_commit_write() and we thus need
6017 * to include the updated inode in the current transaction.
6019 * Also, dquot_alloc_block() will always dirty the inode when blocks
6020 * are allocated to the file.
6022 * If the inode is marked synchronous, we don't honour that here - doing
6023 * so would cause a commit on atime updates, which we don't bother doing.
6024 * We handle synchronous inodes at the highest possible level.
6026 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6027 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6028 * to copy into the on-disk inode structure are the timestamp files.
6030 void ext4_dirty_inode(struct inode
*inode
, int flags
)
6034 if (flags
== I_DIRTY_TIME
)
6036 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
6040 ext4_mark_inode_dirty(handle
, inode
);
6042 ext4_journal_stop(handle
);
6049 * Bind an inode's backing buffer_head into this transaction, to prevent
6050 * it from being flushed to disk early. Unlike
6051 * ext4_reserve_inode_write, this leaves behind no bh reference and
6052 * returns no iloc structure, so the caller needs to repeat the iloc
6053 * lookup to mark the inode dirty later.
6055 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
6057 struct ext4_iloc iloc
;
6061 err
= ext4_get_inode_loc(inode
, &iloc
);
6063 BUFFER_TRACE(iloc
.bh
, "get_write_access");
6064 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
6066 err
= ext4_handle_dirty_metadata(handle
,
6072 ext4_std_error(inode
->i_sb
, err
);
6077 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
6082 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
6085 * We have to be very careful here: changing a data block's
6086 * journaling status dynamically is dangerous. If we write a
6087 * data block to the journal, change the status and then delete
6088 * that block, we risk forgetting to revoke the old log record
6089 * from the journal and so a subsequent replay can corrupt data.
6090 * So, first we make sure that the journal is empty and that
6091 * nobody is changing anything.
6094 journal
= EXT4_JOURNAL(inode
);
6097 if (is_journal_aborted(journal
))
6100 /* Wait for all existing dio workers */
6101 ext4_inode_block_unlocked_dio(inode
);
6102 inode_dio_wait(inode
);
6105 * Before flushing the journal and switching inode's aops, we have
6106 * to flush all dirty data the inode has. There can be outstanding
6107 * delayed allocations, there can be unwritten extents created by
6108 * fallocate or buffered writes in dioread_nolock mode covered by
6109 * dirty data which can be converted only after flushing the dirty
6110 * data (and journalled aops don't know how to handle these cases).
6113 down_write(&EXT4_I(inode
)->i_mmap_sem
);
6114 err
= filemap_write_and_wait(inode
->i_mapping
);
6116 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6117 ext4_inode_resume_unlocked_dio(inode
);
6122 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
6123 jbd2_journal_lock_updates(journal
);
6126 * OK, there are no updates running now, and all cached data is
6127 * synced to disk. We are now in a completely consistent state
6128 * which doesn't have anything in the journal, and we know that
6129 * no filesystem updates are running, so it is safe to modify
6130 * the inode's in-core data-journaling state flag now.
6134 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6136 err
= jbd2_journal_flush(journal
);
6138 jbd2_journal_unlock_updates(journal
);
6139 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6140 ext4_inode_resume_unlocked_dio(inode
);
6143 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6145 ext4_set_aops(inode
);
6147 jbd2_journal_unlock_updates(journal
);
6148 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6151 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6152 ext4_inode_resume_unlocked_dio(inode
);
6154 /* Finally we can mark the inode as dirty. */
6156 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
6158 return PTR_ERR(handle
);
6160 err
= ext4_mark_inode_dirty(handle
, inode
);
6161 ext4_handle_sync(handle
);
6162 ext4_journal_stop(handle
);
6163 ext4_std_error(inode
->i_sb
, err
);
6168 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
6170 return !buffer_mapped(bh
);
6173 int ext4_page_mkwrite(struct vm_fault
*vmf
)
6175 struct vm_area_struct
*vma
= vmf
->vma
;
6176 struct page
*page
= vmf
->page
;
6180 struct file
*file
= vma
->vm_file
;
6181 struct inode
*inode
= file_inode(file
);
6182 struct address_space
*mapping
= inode
->i_mapping
;
6184 get_block_t
*get_block
;
6187 if (unlikely(IS_IMMUTABLE(inode
)))
6188 return VM_FAULT_SIGBUS
;
6190 sb_start_pagefault(inode
->i_sb
);
6191 file_update_time(vma
->vm_file
);
6193 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6195 ret
= ext4_convert_inline_data(inode
);
6199 /* Delalloc case is easy... */
6200 if (test_opt(inode
->i_sb
, DELALLOC
) &&
6201 !ext4_should_journal_data(inode
) &&
6202 !ext4_nonda_switch(inode
->i_sb
)) {
6204 ret
= block_page_mkwrite(vma
, vmf
,
6205 ext4_da_get_block_prep
);
6206 } while (ret
== -ENOSPC
&&
6207 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
6212 size
= i_size_read(inode
);
6213 /* Page got truncated from under us? */
6214 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6216 ret
= VM_FAULT_NOPAGE
;
6220 if (page
->index
== size
>> PAGE_SHIFT
)
6221 len
= size
& ~PAGE_MASK
;
6225 * Return if we have all the buffers mapped. This avoids the need to do
6226 * journal_start/journal_stop which can block and take a long time
6228 if (page_has_buffers(page
)) {
6229 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
6231 ext4_bh_unmapped
)) {
6232 /* Wait so that we don't change page under IO */
6233 wait_for_stable_page(page
);
6234 ret
= VM_FAULT_LOCKED
;
6239 /* OK, we need to fill the hole... */
6240 if (ext4_should_dioread_nolock(inode
))
6241 get_block
= ext4_get_block_unwritten
;
6243 get_block
= ext4_get_block
;
6245 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6246 ext4_writepage_trans_blocks(inode
));
6247 if (IS_ERR(handle
)) {
6248 ret
= VM_FAULT_SIGBUS
;
6251 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
6252 if (!ret
&& ext4_should_journal_data(inode
)) {
6253 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
6254 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
6256 ret
= VM_FAULT_SIGBUS
;
6257 ext4_journal_stop(handle
);
6260 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6262 ext4_journal_stop(handle
);
6263 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6266 ret
= block_page_mkwrite_return(ret
);
6268 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6269 sb_end_pagefault(inode
->i_sb
);
6273 int ext4_filemap_fault(struct vm_fault
*vmf
)
6275 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
6278 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6279 err
= filemap_fault(vmf
);
6280 up_read(&EXT4_I(inode
)->i_mmap_sem
);