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>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
54 struct ext4_inode_info
*ei
)
56 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
59 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
60 unsigned int csum_size
= sizeof(dummy_csum
);
62 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
63 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
65 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
66 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
68 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
69 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
70 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
71 EXT4_GOOD_OLD_INODE_SIZE
,
72 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
73 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
74 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
78 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
79 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
85 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
86 struct ext4_inode_info
*ei
)
88 __u32 provided
, calculated
;
90 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
91 cpu_to_le32(EXT4_OS_LINUX
) ||
92 !ext4_has_metadata_csum(inode
->i_sb
))
95 provided
= le16_to_cpu(raw
->i_checksum_lo
);
96 calculated
= ext4_inode_csum(inode
, raw
, ei
);
97 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
98 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
99 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
101 calculated
&= 0xFFFF;
103 return provided
== calculated
;
106 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
107 struct ext4_inode_info
*ei
)
111 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
112 cpu_to_le32(EXT4_OS_LINUX
) ||
113 !ext4_has_metadata_csum(inode
->i_sb
))
116 csum
= ext4_inode_csum(inode
, raw
, ei
);
117 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
118 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
119 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
120 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
123 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
126 trace_ext4_begin_ordered_truncate(inode
, new_size
);
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
133 if (!EXT4_I(inode
)->jinode
)
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
136 EXT4_I(inode
)->jinode
,
140 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
141 unsigned int length
);
142 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
143 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
144 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
151 int ext4_inode_is_fast_symlink(struct inode
*inode
)
153 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
154 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
155 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
157 if (ext4_has_inline_data(inode
))
160 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
162 return S_ISLNK(inode
->i_mode
) && inode
->i_size
&&
163 (inode
->i_size
< EXT4_N_BLOCKS
* 4);
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
182 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
183 jbd_debug(2, "restarting handle %p\n", handle
);
184 up_write(&EXT4_I(inode
)->i_data_sem
);
185 ret
= ext4_journal_restart(handle
, nblocks
);
186 down_write(&EXT4_I(inode
)->i_data_sem
);
187 ext4_discard_preallocations(inode
);
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_evict_inode(struct inode
*inode
)
199 int extra_credits
= 3;
200 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
202 trace_ext4_evict_inode(inode
);
204 if (inode
->i_nlink
) {
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
220 * Note that directories do not have this problem because they
221 * don't use page cache.
223 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
224 ext4_should_journal_data(inode
) &&
225 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
226 inode
->i_data
.nrpages
) {
227 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
228 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
230 jbd2_complete_transaction(journal
, commit_tid
);
231 filemap_write_and_wait(&inode
->i_data
);
233 truncate_inode_pages_final(&inode
->i_data
);
238 if (is_bad_inode(inode
))
240 dquot_initialize(inode
);
242 if (ext4_should_order_data(inode
))
243 ext4_begin_ordered_truncate(inode
, 0);
244 truncate_inode_pages_final(&inode
->i_data
);
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
250 sb_start_intwrite(inode
->i_sb
);
252 if (!IS_NOQUOTA(inode
))
253 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
255 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
256 ext4_blocks_for_truncate(inode
)+extra_credits
);
257 if (IS_ERR(handle
)) {
258 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
264 ext4_orphan_del(NULL
, inode
);
265 sb_end_intwrite(inode
->i_sb
);
270 ext4_handle_sync(handle
);
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
279 if (ext4_inode_is_fast_symlink(inode
))
280 memset(EXT4_I(inode
)->i_data
, 0, sizeof(EXT4_I(inode
)->i_data
));
282 err
= ext4_mark_inode_dirty(handle
, inode
);
284 ext4_warning(inode
->i_sb
,
285 "couldn't mark inode dirty (err %d)", err
);
288 if (inode
->i_blocks
) {
289 err
= ext4_truncate(inode
);
291 ext4_error(inode
->i_sb
,
292 "couldn't truncate inode %lu (err %d)",
298 /* Remove xattr references. */
299 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
302 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
304 ext4_journal_stop(handle
);
305 ext4_orphan_del(NULL
, inode
);
306 sb_end_intwrite(inode
->i_sb
);
307 ext4_xattr_inode_array_free(ea_inode_array
);
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
319 ext4_orphan_del(handle
, inode
);
320 EXT4_I(inode
)->i_dtime
= (__u32
)ktime_get_real_seconds();
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
329 if (ext4_mark_inode_dirty(handle
, inode
))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode
);
333 ext4_free_inode(handle
, inode
);
334 ext4_journal_stop(handle
);
335 sb_end_intwrite(inode
->i_sb
);
336 ext4_xattr_inode_array_free(ea_inode_array
);
339 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
343 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
345 return &EXT4_I(inode
)->i_reserved_quota
;
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
353 void ext4_da_update_reserve_space(struct inode
*inode
,
354 int used
, int quota_claim
)
356 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
357 struct ext4_inode_info
*ei
= EXT4_I(inode
);
359 spin_lock(&ei
->i_block_reservation_lock
);
360 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
361 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
362 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__
, inode
->i_ino
, used
,
365 ei
->i_reserved_data_blocks
);
367 used
= ei
->i_reserved_data_blocks
;
370 /* Update per-inode reservations */
371 ei
->i_reserved_data_blocks
-= used
;
372 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
374 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
376 /* Update quota subsystem for data blocks */
378 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
385 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
393 if ((ei
->i_reserved_data_blocks
== 0) &&
394 (atomic_read(&inode
->i_writecount
) == 0))
395 ext4_discard_preallocations(inode
);
398 static int __check_block_validity(struct inode
*inode
, const char *func
,
400 struct ext4_map_blocks
*map
)
402 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
404 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
405 "lblock %lu mapped to illegal pblock %llu "
406 "(length %d)", (unsigned long) map
->m_lblk
,
407 map
->m_pblk
, map
->m_len
);
408 return -EFSCORRUPTED
;
413 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
418 if (ext4_encrypted_inode(inode
))
419 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
421 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
428 #define check_block_validity(inode, map) \
429 __check_block_validity((inode), __func__, __LINE__, (map))
431 #ifdef ES_AGGRESSIVE_TEST
432 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
434 struct ext4_map_blocks
*es_map
,
435 struct ext4_map_blocks
*map
,
442 * There is a race window that the result is not the same.
443 * e.g. xfstests #223 when dioread_nolock enables. The reason
444 * is that we lookup a block mapping in extent status tree with
445 * out taking i_data_sem. So at the time the unwritten extent
446 * could be converted.
448 down_read(&EXT4_I(inode
)->i_data_sem
);
449 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
450 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
451 EXT4_GET_BLOCKS_KEEP_SIZE
);
453 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
454 EXT4_GET_BLOCKS_KEEP_SIZE
);
456 up_read((&EXT4_I(inode
)->i_data_sem
));
459 * We don't check m_len because extent will be collpased in status
460 * tree. So the m_len might not equal.
462 if (es_map
->m_lblk
!= map
->m_lblk
||
463 es_map
->m_flags
!= map
->m_flags
||
464 es_map
->m_pblk
!= map
->m_pblk
) {
465 printk("ES cache assertion failed for inode: %lu "
466 "es_cached ex [%d/%d/%llu/%x] != "
467 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
468 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
469 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
470 map
->m_len
, map
->m_pblk
, map
->m_flags
,
474 #endif /* ES_AGGRESSIVE_TEST */
477 * The ext4_map_blocks() function tries to look up the requested blocks,
478 * and returns if the blocks are already mapped.
480 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
481 * and store the allocated blocks in the result buffer head and mark it
484 * If file type is extents based, it will call ext4_ext_map_blocks(),
485 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
488 * On success, it returns the number of blocks being mapped or allocated. if
489 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
490 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
492 * It returns 0 if plain look up failed (blocks have not been allocated), in
493 * that case, @map is returned as unmapped but we still do fill map->m_len to
494 * indicate the length of a hole starting at map->m_lblk.
496 * It returns the error in case of allocation failure.
498 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
499 struct ext4_map_blocks
*map
, int flags
)
501 struct extent_status es
;
504 #ifdef ES_AGGRESSIVE_TEST
505 struct ext4_map_blocks orig_map
;
507 memcpy(&orig_map
, map
, sizeof(*map
));
511 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
512 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
513 (unsigned long) map
->m_lblk
);
516 * ext4_map_blocks returns an int, and m_len is an unsigned int
518 if (unlikely(map
->m_len
> INT_MAX
))
519 map
->m_len
= INT_MAX
;
521 /* We can handle the block number less than EXT_MAX_BLOCKS */
522 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
523 return -EFSCORRUPTED
;
525 /* Lookup extent status tree firstly */
526 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
527 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
528 map
->m_pblk
= ext4_es_pblock(&es
) +
529 map
->m_lblk
- es
.es_lblk
;
530 map
->m_flags
|= ext4_es_is_written(&es
) ?
531 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
532 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
533 if (retval
> map
->m_len
)
536 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
538 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
539 if (retval
> map
->m_len
)
546 #ifdef ES_AGGRESSIVE_TEST
547 ext4_map_blocks_es_recheck(handle
, inode
, map
,
554 * Try to see if we can get the block without requesting a new
557 down_read(&EXT4_I(inode
)->i_data_sem
);
558 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
559 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
560 EXT4_GET_BLOCKS_KEEP_SIZE
);
562 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
563 EXT4_GET_BLOCKS_KEEP_SIZE
);
568 if (unlikely(retval
!= map
->m_len
)) {
569 ext4_warning(inode
->i_sb
,
570 "ES len assertion failed for inode "
571 "%lu: retval %d != map->m_len %d",
572 inode
->i_ino
, retval
, map
->m_len
);
576 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
577 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
578 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
579 !(status
& EXTENT_STATUS_WRITTEN
) &&
580 ext4_find_delalloc_range(inode
, map
->m_lblk
,
581 map
->m_lblk
+ map
->m_len
- 1))
582 status
|= EXTENT_STATUS_DELAYED
;
583 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
584 map
->m_len
, map
->m_pblk
, status
);
588 up_read((&EXT4_I(inode
)->i_data_sem
));
591 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
592 ret
= check_block_validity(inode
, map
);
597 /* If it is only a block(s) look up */
598 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
602 * Returns if the blocks have already allocated
604 * Note that if blocks have been preallocated
605 * ext4_ext_get_block() returns the create = 0
606 * with buffer head unmapped.
608 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
610 * If we need to convert extent to unwritten
611 * we continue and do the actual work in
612 * ext4_ext_map_blocks()
614 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
618 * Here we clear m_flags because after allocating an new extent,
619 * it will be set again.
621 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
624 * New blocks allocate and/or writing to unwritten extent
625 * will possibly result in updating i_data, so we take
626 * the write lock of i_data_sem, and call get_block()
627 * with create == 1 flag.
629 down_write(&EXT4_I(inode
)->i_data_sem
);
632 * We need to check for EXT4 here because migrate
633 * could have changed the inode type in between
635 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
636 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
638 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
640 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
642 * We allocated new blocks which will result in
643 * i_data's format changing. Force the migrate
644 * to fail by clearing migrate flags
646 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
650 * Update reserved blocks/metadata blocks after successful
651 * block allocation which had been deferred till now. We don't
652 * support fallocate for non extent files. So we can update
653 * reserve space here.
656 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
657 ext4_da_update_reserve_space(inode
, retval
, 1);
663 if (unlikely(retval
!= map
->m_len
)) {
664 ext4_warning(inode
->i_sb
,
665 "ES len assertion failed for inode "
666 "%lu: retval %d != map->m_len %d",
667 inode
->i_ino
, retval
, map
->m_len
);
672 * We have to zeroout blocks before inserting them into extent
673 * status tree. Otherwise someone could look them up there and
674 * use them before they are really zeroed. We also have to
675 * unmap metadata before zeroing as otherwise writeback can
676 * overwrite zeros with stale data from block device.
678 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
679 map
->m_flags
& EXT4_MAP_MAPPED
&&
680 map
->m_flags
& EXT4_MAP_NEW
) {
681 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
683 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
684 map
->m_pblk
, map
->m_len
);
692 * If the extent has been zeroed out, we don't need to update
693 * extent status tree.
695 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
696 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
697 if (ext4_es_is_written(&es
))
700 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
701 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
702 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
703 !(status
& EXTENT_STATUS_WRITTEN
) &&
704 ext4_find_delalloc_range(inode
, map
->m_lblk
,
705 map
->m_lblk
+ map
->m_len
- 1))
706 status
|= EXTENT_STATUS_DELAYED
;
707 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
708 map
->m_pblk
, status
);
716 up_write((&EXT4_I(inode
)->i_data_sem
));
717 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
718 ret
= check_block_validity(inode
, map
);
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
727 if (map
->m_flags
& EXT4_MAP_NEW
&&
728 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
729 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
730 !ext4_is_quota_file(inode
) &&
731 ext4_should_order_data(inode
)) {
732 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
733 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
735 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
744 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
745 * we have to be careful as someone else may be manipulating b_state as well.
747 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
749 unsigned long old_state
;
750 unsigned long new_state
;
752 flags
&= EXT4_MAP_FLAGS
;
754 /* Dummy buffer_head? Set non-atomically. */
756 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
760 * Someone else may be modifying b_state. Be careful! This is ugly but
761 * once we get rid of using bh as a container for mapping information
762 * to pass to / from get_block functions, this can go away.
765 old_state
= READ_ONCE(bh
->b_state
);
766 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
768 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
771 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
772 struct buffer_head
*bh
, int flags
)
774 struct ext4_map_blocks map
;
777 if (ext4_has_inline_data(inode
))
781 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
783 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
786 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
787 ext4_update_bh_state(bh
, map
.m_flags
);
788 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
790 } else if (ret
== 0) {
791 /* hole case, need to fill in bh->b_size */
792 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
797 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
798 struct buffer_head
*bh
, int create
)
800 return _ext4_get_block(inode
, iblock
, bh
,
801 create
? EXT4_GET_BLOCKS_CREATE
: 0);
805 * Get block function used when preparing for buffered write if we require
806 * creating an unwritten extent if blocks haven't been allocated. The extent
807 * will be converted to written after the IO is complete.
809 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
810 struct buffer_head
*bh_result
, int create
)
812 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
813 inode
->i_ino
, create
);
814 return _ext4_get_block(inode
, iblock
, bh_result
,
815 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
818 /* Maximum number of blocks we map for direct IO at once. */
819 #define DIO_MAX_BLOCKS 4096
822 * Get blocks function for the cases that need to start a transaction -
823 * generally difference cases of direct IO and DAX IO. It also handles retries
826 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
827 struct buffer_head
*bh_result
, int flags
)
834 /* Trim mapping request to maximum we can map at once for DIO */
835 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
836 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
837 dio_credits
= ext4_chunk_trans_blocks(inode
,
838 bh_result
->b_size
>> inode
->i_blkbits
);
840 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
842 return PTR_ERR(handle
);
844 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
845 ext4_journal_stop(handle
);
847 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
852 /* Get block function for DIO reads and writes to inodes without extents */
853 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
854 struct buffer_head
*bh
, int create
)
856 /* We don't expect handle for direct IO */
857 WARN_ON_ONCE(ext4_journal_current_handle());
860 return _ext4_get_block(inode
, iblock
, bh
, 0);
861 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
865 * Get block function for AIO DIO writes when we create unwritten extent if
866 * blocks are not allocated yet. The extent will be converted to written
867 * after IO is complete.
869 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
870 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
874 /* We don't expect handle for direct IO */
875 WARN_ON_ONCE(ext4_journal_current_handle());
877 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
878 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
881 * When doing DIO using unwritten extents, we need io_end to convert
882 * unwritten extents to written on IO completion. We allocate io_end
883 * once we spot unwritten extent and store it in b_private. Generic
884 * DIO code keeps b_private set and furthermore passes the value to
885 * our completion callback in 'private' argument.
887 if (!ret
&& buffer_unwritten(bh_result
)) {
888 if (!bh_result
->b_private
) {
889 ext4_io_end_t
*io_end
;
891 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
894 bh_result
->b_private
= io_end
;
895 ext4_set_io_unwritten_flag(inode
, io_end
);
897 set_buffer_defer_completion(bh_result
);
904 * Get block function for non-AIO DIO writes when we create unwritten extent if
905 * blocks are not allocated yet. The extent will be converted to written
906 * after IO is complete by ext4_direct_IO_write().
908 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
909 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
913 /* We don't expect handle for direct IO */
914 WARN_ON_ONCE(ext4_journal_current_handle());
916 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
917 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
920 * Mark inode as having pending DIO writes to unwritten extents.
921 * ext4_direct_IO_write() checks this flag and converts extents to
924 if (!ret
&& buffer_unwritten(bh_result
))
925 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
930 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
931 struct buffer_head
*bh_result
, int create
)
935 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
936 inode
->i_ino
, create
);
937 /* We don't expect handle for direct IO */
938 WARN_ON_ONCE(ext4_journal_current_handle());
940 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
942 * Blocks should have been preallocated! ext4_file_write_iter() checks
945 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
952 * `handle' can be NULL if create is zero
954 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
955 ext4_lblk_t block
, int map_flags
)
957 struct ext4_map_blocks map
;
958 struct buffer_head
*bh
;
959 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
962 J_ASSERT(handle
!= NULL
|| create
== 0);
966 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
969 return create
? ERR_PTR(-ENOSPC
) : NULL
;
973 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
975 return ERR_PTR(-ENOMEM
);
976 if (map
.m_flags
& EXT4_MAP_NEW
) {
977 J_ASSERT(create
!= 0);
978 J_ASSERT(handle
!= NULL
);
981 * Now that we do not always journal data, we should
982 * keep in mind whether this should always journal the
983 * new buffer as metadata. For now, regular file
984 * writes use ext4_get_block instead, so it's not a
988 BUFFER_TRACE(bh
, "call get_create_access");
989 err
= ext4_journal_get_create_access(handle
, bh
);
994 if (!buffer_uptodate(bh
)) {
995 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
996 set_buffer_uptodate(bh
);
999 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1000 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1004 BUFFER_TRACE(bh
, "not a new buffer");
1008 return ERR_PTR(err
);
1011 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1012 ext4_lblk_t block
, int map_flags
)
1014 struct buffer_head
*bh
;
1016 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1019 if (!bh
|| buffer_uptodate(bh
))
1021 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1023 if (buffer_uptodate(bh
))
1026 return ERR_PTR(-EIO
);
1029 /* Read a contiguous batch of blocks. */
1030 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
1031 bool wait
, struct buffer_head
**bhs
)
1035 for (i
= 0; i
< bh_count
; i
++) {
1036 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
1037 if (IS_ERR(bhs
[i
])) {
1038 err
= PTR_ERR(bhs
[i
]);
1044 for (i
= 0; i
< bh_count
; i
++)
1045 /* Note that NULL bhs[i] is valid because of holes. */
1046 if (bhs
[i
] && !buffer_uptodate(bhs
[i
]))
1047 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1,
1053 for (i
= 0; i
< bh_count
; i
++)
1055 wait_on_buffer(bhs
[i
]);
1057 for (i
= 0; i
< bh_count
; i
++) {
1058 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
1066 for (i
= 0; i
< bh_count
; i
++) {
1073 int ext4_walk_page_buffers(handle_t
*handle
,
1074 struct buffer_head
*head
,
1078 int (*fn
)(handle_t
*handle
,
1079 struct buffer_head
*bh
))
1081 struct buffer_head
*bh
;
1082 unsigned block_start
, block_end
;
1083 unsigned blocksize
= head
->b_size
;
1085 struct buffer_head
*next
;
1087 for (bh
= head
, block_start
= 0;
1088 ret
== 0 && (bh
!= head
|| !block_start
);
1089 block_start
= block_end
, bh
= next
) {
1090 next
= bh
->b_this_page
;
1091 block_end
= block_start
+ blocksize
;
1092 if (block_end
<= from
|| block_start
>= to
) {
1093 if (partial
&& !buffer_uptodate(bh
))
1097 err
= (*fn
)(handle
, bh
);
1105 * To preserve ordering, it is essential that the hole instantiation and
1106 * the data write be encapsulated in a single transaction. We cannot
1107 * close off a transaction and start a new one between the ext4_get_block()
1108 * and the commit_write(). So doing the jbd2_journal_start at the start of
1109 * prepare_write() is the right place.
1111 * Also, this function can nest inside ext4_writepage(). In that case, we
1112 * *know* that ext4_writepage() has generated enough buffer credits to do the
1113 * whole page. So we won't block on the journal in that case, which is good,
1114 * because the caller may be PF_MEMALLOC.
1116 * By accident, ext4 can be reentered when a transaction is open via
1117 * quota file writes. If we were to commit the transaction while thus
1118 * reentered, there can be a deadlock - we would be holding a quota
1119 * lock, and the commit would never complete if another thread had a
1120 * transaction open and was blocking on the quota lock - a ranking
1123 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1124 * will _not_ run commit under these circumstances because handle->h_ref
1125 * is elevated. We'll still have enough credits for the tiny quotafile
1128 int do_journal_get_write_access(handle_t
*handle
,
1129 struct buffer_head
*bh
)
1131 int dirty
= buffer_dirty(bh
);
1134 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1137 * __block_write_begin() could have dirtied some buffers. Clean
1138 * the dirty bit as jbd2_journal_get_write_access() could complain
1139 * otherwise about fs integrity issues. Setting of the dirty bit
1140 * by __block_write_begin() isn't a real problem here as we clear
1141 * the bit before releasing a page lock and thus writeback cannot
1142 * ever write the buffer.
1145 clear_buffer_dirty(bh
);
1146 BUFFER_TRACE(bh
, "get write access");
1147 ret
= ext4_journal_get_write_access(handle
, bh
);
1149 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1153 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1154 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1155 get_block_t
*get_block
)
1157 unsigned from
= pos
& (PAGE_SIZE
- 1);
1158 unsigned to
= from
+ len
;
1159 struct inode
*inode
= page
->mapping
->host
;
1160 unsigned block_start
, block_end
;
1163 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1165 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1166 bool decrypt
= false;
1168 BUG_ON(!PageLocked(page
));
1169 BUG_ON(from
> PAGE_SIZE
);
1170 BUG_ON(to
> PAGE_SIZE
);
1173 if (!page_has_buffers(page
))
1174 create_empty_buffers(page
, blocksize
, 0);
1175 head
= page_buffers(page
);
1176 bbits
= ilog2(blocksize
);
1177 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1179 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1180 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1181 block_end
= block_start
+ blocksize
;
1182 if (block_end
<= from
|| block_start
>= to
) {
1183 if (PageUptodate(page
)) {
1184 if (!buffer_uptodate(bh
))
1185 set_buffer_uptodate(bh
);
1190 clear_buffer_new(bh
);
1191 if (!buffer_mapped(bh
)) {
1192 WARN_ON(bh
->b_size
!= blocksize
);
1193 err
= get_block(inode
, block
, bh
, 1);
1196 if (buffer_new(bh
)) {
1197 clean_bdev_bh_alias(bh
);
1198 if (PageUptodate(page
)) {
1199 clear_buffer_new(bh
);
1200 set_buffer_uptodate(bh
);
1201 mark_buffer_dirty(bh
);
1204 if (block_end
> to
|| block_start
< from
)
1205 zero_user_segments(page
, to
, block_end
,
1210 if (PageUptodate(page
)) {
1211 if (!buffer_uptodate(bh
))
1212 set_buffer_uptodate(bh
);
1215 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1216 !buffer_unwritten(bh
) &&
1217 (block_start
< from
|| block_end
> to
)) {
1218 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1220 decrypt
= ext4_encrypted_inode(inode
) &&
1221 S_ISREG(inode
->i_mode
);
1225 * If we issued read requests, let them complete.
1227 while (wait_bh
> wait
) {
1228 wait_on_buffer(*--wait_bh
);
1229 if (!buffer_uptodate(*wait_bh
))
1233 page_zero_new_buffers(page
, from
, to
);
1235 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1236 PAGE_SIZE
, 0, page
->index
);
1241 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1242 loff_t pos
, unsigned len
, unsigned flags
,
1243 struct page
**pagep
, void **fsdata
)
1245 struct inode
*inode
= mapping
->host
;
1246 int ret
, needed_blocks
;
1253 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1256 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1258 * Reserve one block more for addition to orphan list in case
1259 * we allocate blocks but write fails for some reason
1261 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1262 index
= pos
>> PAGE_SHIFT
;
1263 from
= pos
& (PAGE_SIZE
- 1);
1266 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1267 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1276 * grab_cache_page_write_begin() can take a long time if the
1277 * system is thrashing due to memory pressure, or if the page
1278 * is being written back. So grab it first before we start
1279 * the transaction handle. This also allows us to allocate
1280 * the page (if needed) without using GFP_NOFS.
1283 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1289 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1290 if (IS_ERR(handle
)) {
1292 return PTR_ERR(handle
);
1296 if (page
->mapping
!= mapping
) {
1297 /* The page got truncated from under us */
1300 ext4_journal_stop(handle
);
1303 /* In case writeback began while the page was unlocked */
1304 wait_for_stable_page(page
);
1306 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1307 if (ext4_should_dioread_nolock(inode
))
1308 ret
= ext4_block_write_begin(page
, pos
, len
,
1309 ext4_get_block_unwritten
);
1311 ret
= ext4_block_write_begin(page
, pos
, len
,
1314 if (ext4_should_dioread_nolock(inode
))
1315 ret
= __block_write_begin(page
, pos
, len
,
1316 ext4_get_block_unwritten
);
1318 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1320 if (!ret
&& ext4_should_journal_data(inode
)) {
1321 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1323 do_journal_get_write_access
);
1329 * __block_write_begin may have instantiated a few blocks
1330 * outside i_size. Trim these off again. Don't need
1331 * i_size_read because we hold i_mutex.
1333 * Add inode to orphan list in case we crash before
1336 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1337 ext4_orphan_add(handle
, inode
);
1339 ext4_journal_stop(handle
);
1340 if (pos
+ len
> inode
->i_size
) {
1341 ext4_truncate_failed_write(inode
);
1343 * If truncate failed early the inode might
1344 * still be on the orphan list; we need to
1345 * make sure the inode is removed from the
1346 * orphan list in that case.
1349 ext4_orphan_del(NULL
, inode
);
1352 if (ret
== -ENOSPC
&&
1353 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1362 /* For write_end() in data=journal mode */
1363 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1366 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1368 set_buffer_uptodate(bh
);
1369 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1370 clear_buffer_meta(bh
);
1371 clear_buffer_prio(bh
);
1376 * We need to pick up the new inode size which generic_commit_write gave us
1377 * `file' can be NULL - eg, when called from page_symlink().
1379 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1380 * buffers are managed internally.
1382 static int ext4_write_end(struct file
*file
,
1383 struct address_space
*mapping
,
1384 loff_t pos
, unsigned len
, unsigned copied
,
1385 struct page
*page
, void *fsdata
)
1387 handle_t
*handle
= ext4_journal_current_handle();
1388 struct inode
*inode
= mapping
->host
;
1389 loff_t old_size
= inode
->i_size
;
1391 int i_size_changed
= 0;
1392 int inline_data
= ext4_has_inline_data(inode
);
1394 trace_ext4_write_end(inode
, pos
, len
, copied
);
1396 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1405 copied
= block_write_end(file
, mapping
, pos
,
1406 len
, copied
, page
, fsdata
);
1408 * it's important to update i_size while still holding page lock:
1409 * page writeout could otherwise come in and zero beyond i_size.
1411 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1416 pagecache_isize_extended(inode
, old_size
, pos
);
1418 * Don't mark the inode dirty under page lock. First, it unnecessarily
1419 * makes the holding time of page lock longer. Second, it forces lock
1420 * ordering of page lock and transaction start for journaling
1423 if (i_size_changed
|| inline_data
)
1424 ext4_mark_inode_dirty(handle
, inode
);
1426 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1427 /* if we have allocated more blocks and copied
1428 * less. We will have blocks allocated outside
1429 * inode->i_size. So truncate them
1431 ext4_orphan_add(handle
, inode
);
1433 ret2
= ext4_journal_stop(handle
);
1437 if (pos
+ len
> inode
->i_size
) {
1438 ext4_truncate_failed_write(inode
);
1440 * If truncate failed early the inode might still be
1441 * on the orphan list; we need to make sure the inode
1442 * is removed from the orphan list in that case.
1445 ext4_orphan_del(NULL
, inode
);
1448 return ret
? ret
: copied
;
1452 * This is a private version of page_zero_new_buffers() which doesn't
1453 * set the buffer to be dirty, since in data=journalled mode we need
1454 * to call ext4_handle_dirty_metadata() instead.
1456 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1458 unsigned from
, unsigned to
)
1460 unsigned int block_start
= 0, block_end
;
1461 struct buffer_head
*head
, *bh
;
1463 bh
= head
= page_buffers(page
);
1465 block_end
= block_start
+ bh
->b_size
;
1466 if (buffer_new(bh
)) {
1467 if (block_end
> from
&& block_start
< to
) {
1468 if (!PageUptodate(page
)) {
1469 unsigned start
, size
;
1471 start
= max(from
, block_start
);
1472 size
= min(to
, block_end
) - start
;
1474 zero_user(page
, start
, size
);
1475 write_end_fn(handle
, bh
);
1477 clear_buffer_new(bh
);
1480 block_start
= block_end
;
1481 bh
= bh
->b_this_page
;
1482 } while (bh
!= head
);
1485 static int ext4_journalled_write_end(struct file
*file
,
1486 struct address_space
*mapping
,
1487 loff_t pos
, unsigned len
, unsigned copied
,
1488 struct page
*page
, void *fsdata
)
1490 handle_t
*handle
= ext4_journal_current_handle();
1491 struct inode
*inode
= mapping
->host
;
1492 loff_t old_size
= inode
->i_size
;
1496 int size_changed
= 0;
1497 int inline_data
= ext4_has_inline_data(inode
);
1499 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1500 from
= pos
& (PAGE_SIZE
- 1);
1503 BUG_ON(!ext4_handle_valid(handle
));
1506 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1514 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1516 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1518 if (unlikely(copied
< len
))
1519 ext4_journalled_zero_new_buffers(handle
, page
,
1521 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1522 from
+ copied
, &partial
,
1525 SetPageUptodate(page
);
1527 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1528 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1529 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1534 pagecache_isize_extended(inode
, old_size
, pos
);
1536 if (size_changed
|| inline_data
) {
1537 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1542 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1543 /* if we have allocated more blocks and copied
1544 * less. We will have blocks allocated outside
1545 * inode->i_size. So truncate them
1547 ext4_orphan_add(handle
, inode
);
1550 ret2
= ext4_journal_stop(handle
);
1553 if (pos
+ len
> inode
->i_size
) {
1554 ext4_truncate_failed_write(inode
);
1556 * If truncate failed early the inode might still be
1557 * on the orphan list; we need to make sure the inode
1558 * is removed from the orphan list in that case.
1561 ext4_orphan_del(NULL
, inode
);
1564 return ret
? ret
: copied
;
1568 * Reserve space for a single cluster
1570 static int ext4_da_reserve_space(struct inode
*inode
)
1572 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1573 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1577 * We will charge metadata quota at writeout time; this saves
1578 * us from metadata over-estimation, though we may go over by
1579 * a small amount in the end. Here we just reserve for data.
1581 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1585 spin_lock(&ei
->i_block_reservation_lock
);
1586 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1587 spin_unlock(&ei
->i_block_reservation_lock
);
1588 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1591 ei
->i_reserved_data_blocks
++;
1592 trace_ext4_da_reserve_space(inode
);
1593 spin_unlock(&ei
->i_block_reservation_lock
);
1595 return 0; /* success */
1598 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1600 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1601 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1604 return; /* Nothing to release, exit */
1606 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1608 trace_ext4_da_release_space(inode
, to_free
);
1609 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1611 * if there aren't enough reserved blocks, then the
1612 * counter is messed up somewhere. Since this
1613 * function is called from invalidate page, it's
1614 * harmless to return without any action.
1616 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1617 "ino %lu, to_free %d with only %d reserved "
1618 "data blocks", inode
->i_ino
, to_free
,
1619 ei
->i_reserved_data_blocks
);
1621 to_free
= ei
->i_reserved_data_blocks
;
1623 ei
->i_reserved_data_blocks
-= to_free
;
1625 /* update fs dirty data blocks counter */
1626 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1628 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1630 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1633 static void ext4_da_page_release_reservation(struct page
*page
,
1634 unsigned int offset
,
1635 unsigned int length
)
1637 int to_release
= 0, contiguous_blks
= 0;
1638 struct buffer_head
*head
, *bh
;
1639 unsigned int curr_off
= 0;
1640 struct inode
*inode
= page
->mapping
->host
;
1641 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1642 unsigned int stop
= offset
+ length
;
1646 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1648 head
= page_buffers(page
);
1651 unsigned int next_off
= curr_off
+ bh
->b_size
;
1653 if (next_off
> stop
)
1656 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1659 clear_buffer_delay(bh
);
1660 } else if (contiguous_blks
) {
1661 lblk
= page
->index
<<
1662 (PAGE_SHIFT
- inode
->i_blkbits
);
1663 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1665 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1666 contiguous_blks
= 0;
1668 curr_off
= next_off
;
1669 } while ((bh
= bh
->b_this_page
) != head
);
1671 if (contiguous_blks
) {
1672 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1673 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1674 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1677 /* If we have released all the blocks belonging to a cluster, then we
1678 * need to release the reserved space for that cluster. */
1679 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1680 while (num_clusters
> 0) {
1681 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1682 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1683 if (sbi
->s_cluster_ratio
== 1 ||
1684 !ext4_find_delalloc_cluster(inode
, lblk
))
1685 ext4_da_release_space(inode
, 1);
1692 * Delayed allocation stuff
1695 struct mpage_da_data
{
1696 struct inode
*inode
;
1697 struct writeback_control
*wbc
;
1699 pgoff_t first_page
; /* The first page to write */
1700 pgoff_t next_page
; /* Current page to examine */
1701 pgoff_t last_page
; /* Last page to examine */
1703 * Extent to map - this can be after first_page because that can be
1704 * fully mapped. We somewhat abuse m_flags to store whether the extent
1705 * is delalloc or unwritten.
1707 struct ext4_map_blocks map
;
1708 struct ext4_io_submit io_submit
; /* IO submission data */
1709 unsigned int do_map
:1;
1712 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1717 struct pagevec pvec
;
1718 struct inode
*inode
= mpd
->inode
;
1719 struct address_space
*mapping
= inode
->i_mapping
;
1721 /* This is necessary when next_page == 0. */
1722 if (mpd
->first_page
>= mpd
->next_page
)
1725 index
= mpd
->first_page
;
1726 end
= mpd
->next_page
- 1;
1728 ext4_lblk_t start
, last
;
1729 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1730 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1731 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1734 pagevec_init(&pvec
);
1735 while (index
<= end
) {
1736 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1739 for (i
= 0; i
< nr_pages
; i
++) {
1740 struct page
*page
= pvec
.pages
[i
];
1742 BUG_ON(!PageLocked(page
));
1743 BUG_ON(PageWriteback(page
));
1745 if (page_mapped(page
))
1746 clear_page_dirty_for_io(page
);
1747 block_invalidatepage(page
, 0, PAGE_SIZE
);
1748 ClearPageUptodate(page
);
1752 pagevec_release(&pvec
);
1756 static void ext4_print_free_blocks(struct inode
*inode
)
1758 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1759 struct super_block
*sb
= inode
->i_sb
;
1760 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1762 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1763 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1764 ext4_count_free_clusters(sb
)));
1765 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1766 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1767 (long long) EXT4_C2B(EXT4_SB(sb
),
1768 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1769 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1770 (long long) EXT4_C2B(EXT4_SB(sb
),
1771 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1772 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1773 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1774 ei
->i_reserved_data_blocks
);
1778 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1780 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1784 * This function is grabs code from the very beginning of
1785 * ext4_map_blocks, but assumes that the caller is from delayed write
1786 * time. This function looks up the requested blocks and sets the
1787 * buffer delay bit under the protection of i_data_sem.
1789 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1790 struct ext4_map_blocks
*map
,
1791 struct buffer_head
*bh
)
1793 struct extent_status es
;
1795 sector_t invalid_block
= ~((sector_t
) 0xffff);
1796 #ifdef ES_AGGRESSIVE_TEST
1797 struct ext4_map_blocks orig_map
;
1799 memcpy(&orig_map
, map
, sizeof(*map
));
1802 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1806 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1807 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1808 (unsigned long) map
->m_lblk
);
1810 /* Lookup extent status tree firstly */
1811 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1812 if (ext4_es_is_hole(&es
)) {
1814 down_read(&EXT4_I(inode
)->i_data_sem
);
1819 * Delayed extent could be allocated by fallocate.
1820 * So we need to check it.
1822 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1823 map_bh(bh
, inode
->i_sb
, invalid_block
);
1825 set_buffer_delay(bh
);
1829 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1830 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1831 if (retval
> map
->m_len
)
1832 retval
= map
->m_len
;
1833 map
->m_len
= retval
;
1834 if (ext4_es_is_written(&es
))
1835 map
->m_flags
|= EXT4_MAP_MAPPED
;
1836 else if (ext4_es_is_unwritten(&es
))
1837 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1841 #ifdef ES_AGGRESSIVE_TEST
1842 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1848 * Try to see if we can get the block without requesting a new
1849 * file system block.
1851 down_read(&EXT4_I(inode
)->i_data_sem
);
1852 if (ext4_has_inline_data(inode
))
1854 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1855 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1857 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1863 * XXX: __block_prepare_write() unmaps passed block,
1867 * If the block was allocated from previously allocated cluster,
1868 * then we don't need to reserve it again. However we still need
1869 * to reserve metadata for every block we're going to write.
1871 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1872 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1873 ret
= ext4_da_reserve_space(inode
);
1875 /* not enough space to reserve */
1881 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1882 ~0, EXTENT_STATUS_DELAYED
);
1888 map_bh(bh
, inode
->i_sb
, invalid_block
);
1890 set_buffer_delay(bh
);
1891 } else if (retval
> 0) {
1893 unsigned int status
;
1895 if (unlikely(retval
!= map
->m_len
)) {
1896 ext4_warning(inode
->i_sb
,
1897 "ES len assertion failed for inode "
1898 "%lu: retval %d != map->m_len %d",
1899 inode
->i_ino
, retval
, map
->m_len
);
1903 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1904 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1905 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1906 map
->m_pblk
, status
);
1912 up_read((&EXT4_I(inode
)->i_data_sem
));
1918 * This is a special get_block_t callback which is used by
1919 * ext4_da_write_begin(). It will either return mapped block or
1920 * reserve space for a single block.
1922 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1923 * We also have b_blocknr = -1 and b_bdev initialized properly
1925 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1926 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1927 * initialized properly.
1929 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1930 struct buffer_head
*bh
, int create
)
1932 struct ext4_map_blocks map
;
1935 BUG_ON(create
== 0);
1936 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1938 map
.m_lblk
= iblock
;
1942 * first, we need to know whether the block is allocated already
1943 * preallocated blocks are unmapped but should treated
1944 * the same as allocated blocks.
1946 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1950 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1951 ext4_update_bh_state(bh
, map
.m_flags
);
1953 if (buffer_unwritten(bh
)) {
1954 /* A delayed write to unwritten bh should be marked
1955 * new and mapped. Mapped ensures that we don't do
1956 * get_block multiple times when we write to the same
1957 * offset and new ensures that we do proper zero out
1958 * for partial write.
1961 set_buffer_mapped(bh
);
1966 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1972 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1978 static int __ext4_journalled_writepage(struct page
*page
,
1981 struct address_space
*mapping
= page
->mapping
;
1982 struct inode
*inode
= mapping
->host
;
1983 struct buffer_head
*page_bufs
= NULL
;
1984 handle_t
*handle
= NULL
;
1985 int ret
= 0, err
= 0;
1986 int inline_data
= ext4_has_inline_data(inode
);
1987 struct buffer_head
*inode_bh
= NULL
;
1989 ClearPageChecked(page
);
1992 BUG_ON(page
->index
!= 0);
1993 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1994 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1995 if (inode_bh
== NULL
)
1998 page_bufs
= page_buffers(page
);
2003 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2007 * We need to release the page lock before we start the
2008 * journal, so grab a reference so the page won't disappear
2009 * out from under us.
2014 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2015 ext4_writepage_trans_blocks(inode
));
2016 if (IS_ERR(handle
)) {
2017 ret
= PTR_ERR(handle
);
2019 goto out_no_pagelock
;
2021 BUG_ON(!ext4_handle_valid(handle
));
2025 if (page
->mapping
!= mapping
) {
2026 /* The page got truncated from under us */
2027 ext4_journal_stop(handle
);
2033 ret
= ext4_mark_inode_dirty(handle
, inode
);
2035 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2036 do_journal_get_write_access
);
2038 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2043 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2044 err
= ext4_journal_stop(handle
);
2048 if (!ext4_has_inline_data(inode
))
2049 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
2051 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2060 * Note that we don't need to start a transaction unless we're journaling data
2061 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2062 * need to file the inode to the transaction's list in ordered mode because if
2063 * we are writing back data added by write(), the inode is already there and if
2064 * we are writing back data modified via mmap(), no one guarantees in which
2065 * transaction the data will hit the disk. In case we are journaling data, we
2066 * cannot start transaction directly because transaction start ranks above page
2067 * lock so we have to do some magic.
2069 * This function can get called via...
2070 * - ext4_writepages after taking page lock (have journal handle)
2071 * - journal_submit_inode_data_buffers (no journal handle)
2072 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2073 * - grab_page_cache when doing write_begin (have journal handle)
2075 * We don't do any block allocation in this function. If we have page with
2076 * multiple blocks we need to write those buffer_heads that are mapped. This
2077 * is important for mmaped based write. So if we do with blocksize 1K
2078 * truncate(f, 1024);
2079 * a = mmap(f, 0, 4096);
2081 * truncate(f, 4096);
2082 * we have in the page first buffer_head mapped via page_mkwrite call back
2083 * but other buffer_heads would be unmapped but dirty (dirty done via the
2084 * do_wp_page). So writepage should write the first block. If we modify
2085 * the mmap area beyond 1024 we will again get a page_fault and the
2086 * page_mkwrite callback will do the block allocation and mark the
2087 * buffer_heads mapped.
2089 * We redirty the page if we have any buffer_heads that is either delay or
2090 * unwritten in the page.
2092 * We can get recursively called as show below.
2094 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2097 * But since we don't do any block allocation we should not deadlock.
2098 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2100 static int ext4_writepage(struct page
*page
,
2101 struct writeback_control
*wbc
)
2106 struct buffer_head
*page_bufs
= NULL
;
2107 struct inode
*inode
= page
->mapping
->host
;
2108 struct ext4_io_submit io_submit
;
2109 bool keep_towrite
= false;
2111 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
2112 ext4_invalidatepage(page
, 0, PAGE_SIZE
);
2117 trace_ext4_writepage(page
);
2118 size
= i_size_read(inode
);
2119 if (page
->index
== size
>> PAGE_SHIFT
)
2120 len
= size
& ~PAGE_MASK
;
2124 page_bufs
= page_buffers(page
);
2126 * We cannot do block allocation or other extent handling in this
2127 * function. If there are buffers needing that, we have to redirty
2128 * the page. But we may reach here when we do a journal commit via
2129 * journal_submit_inode_data_buffers() and in that case we must write
2130 * allocated buffers to achieve data=ordered mode guarantees.
2132 * Also, if there is only one buffer per page (the fs block
2133 * size == the page size), if one buffer needs block
2134 * allocation or needs to modify the extent tree to clear the
2135 * unwritten flag, we know that the page can't be written at
2136 * all, so we might as well refuse the write immediately.
2137 * Unfortunately if the block size != page size, we can't as
2138 * easily detect this case using ext4_walk_page_buffers(), but
2139 * for the extremely common case, this is an optimization that
2140 * skips a useless round trip through ext4_bio_write_page().
2142 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2143 ext4_bh_delay_or_unwritten
)) {
2144 redirty_page_for_writepage(wbc
, page
);
2145 if ((current
->flags
& PF_MEMALLOC
) ||
2146 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2148 * For memory cleaning there's no point in writing only
2149 * some buffers. So just bail out. Warn if we came here
2150 * from direct reclaim.
2152 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2157 keep_towrite
= true;
2160 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2162 * It's mmapped pagecache. Add buffers and journal it. There
2163 * doesn't seem much point in redirtying the page here.
2165 return __ext4_journalled_writepage(page
, len
);
2167 ext4_io_submit_init(&io_submit
, wbc
);
2168 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2169 if (!io_submit
.io_end
) {
2170 redirty_page_for_writepage(wbc
, page
);
2174 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2175 ext4_io_submit(&io_submit
);
2176 /* Drop io_end reference we got from init */
2177 ext4_put_io_end_defer(io_submit
.io_end
);
2181 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2187 BUG_ON(page
->index
!= mpd
->first_page
);
2188 clear_page_dirty_for_io(page
);
2190 * We have to be very careful here! Nothing protects writeback path
2191 * against i_size changes and the page can be writeably mapped into
2192 * page tables. So an application can be growing i_size and writing
2193 * data through mmap while writeback runs. clear_page_dirty_for_io()
2194 * write-protects our page in page tables and the page cannot get
2195 * written to again until we release page lock. So only after
2196 * clear_page_dirty_for_io() we are safe to sample i_size for
2197 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2198 * on the barrier provided by TestClearPageDirty in
2199 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2200 * after page tables are updated.
2202 size
= i_size_read(mpd
->inode
);
2203 if (page
->index
== size
>> PAGE_SHIFT
)
2204 len
= size
& ~PAGE_MASK
;
2207 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2209 mpd
->wbc
->nr_to_write
--;
2215 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2218 * mballoc gives us at most this number of blocks...
2219 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2220 * The rest of mballoc seems to handle chunks up to full group size.
2222 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2225 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2227 * @mpd - extent of blocks
2228 * @lblk - logical number of the block in the file
2229 * @bh - buffer head we want to add to the extent
2231 * The function is used to collect contig. blocks in the same state. If the
2232 * buffer doesn't require mapping for writeback and we haven't started the
2233 * extent of buffers to map yet, the function returns 'true' immediately - the
2234 * caller can write the buffer right away. Otherwise the function returns true
2235 * if the block has been added to the extent, false if the block couldn't be
2238 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2239 struct buffer_head
*bh
)
2241 struct ext4_map_blocks
*map
= &mpd
->map
;
2243 /* Buffer that doesn't need mapping for writeback? */
2244 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2245 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2246 /* So far no extent to map => we write the buffer right away */
2247 if (map
->m_len
== 0)
2252 /* First block in the extent? */
2253 if (map
->m_len
== 0) {
2254 /* We cannot map unless handle is started... */
2259 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2263 /* Don't go larger than mballoc is willing to allocate */
2264 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2267 /* Can we merge the block to our big extent? */
2268 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2269 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2277 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2279 * @mpd - extent of blocks for mapping
2280 * @head - the first buffer in the page
2281 * @bh - buffer we should start processing from
2282 * @lblk - logical number of the block in the file corresponding to @bh
2284 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2285 * the page for IO if all buffers in this page were mapped and there's no
2286 * accumulated extent of buffers to map or add buffers in the page to the
2287 * extent of buffers to map. The function returns 1 if the caller can continue
2288 * by processing the next page, 0 if it should stop adding buffers to the
2289 * extent to map because we cannot extend it anymore. It can also return value
2290 * < 0 in case of error during IO submission.
2292 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2293 struct buffer_head
*head
,
2294 struct buffer_head
*bh
,
2297 struct inode
*inode
= mpd
->inode
;
2299 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2300 >> inode
->i_blkbits
;
2303 BUG_ON(buffer_locked(bh
));
2305 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2306 /* Found extent to map? */
2309 /* Buffer needs mapping and handle is not started? */
2312 /* Everything mapped so far and we hit EOF */
2315 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2316 /* So far everything mapped? Submit the page for IO. */
2317 if (mpd
->map
.m_len
== 0) {
2318 err
= mpage_submit_page(mpd
, head
->b_page
);
2322 return lblk
< blocks
;
2326 * mpage_map_buffers - update buffers corresponding to changed extent and
2327 * submit fully mapped pages for IO
2329 * @mpd - description of extent to map, on return next extent to map
2331 * Scan buffers corresponding to changed extent (we expect corresponding pages
2332 * to be already locked) and update buffer state according to new extent state.
2333 * We map delalloc buffers to their physical location, clear unwritten bits,
2334 * and mark buffers as uninit when we perform writes to unwritten extents
2335 * and do extent conversion after IO is finished. If the last page is not fully
2336 * mapped, we update @map to the next extent in the last page that needs
2337 * mapping. Otherwise we submit the page for IO.
2339 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2341 struct pagevec pvec
;
2343 struct inode
*inode
= mpd
->inode
;
2344 struct buffer_head
*head
, *bh
;
2345 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2351 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2352 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2353 lblk
= start
<< bpp_bits
;
2354 pblock
= mpd
->map
.m_pblk
;
2356 pagevec_init(&pvec
);
2357 while (start
<= end
) {
2358 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2362 for (i
= 0; i
< nr_pages
; i
++) {
2363 struct page
*page
= pvec
.pages
[i
];
2365 bh
= head
= page_buffers(page
);
2367 if (lblk
< mpd
->map
.m_lblk
)
2369 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2371 * Buffer after end of mapped extent.
2372 * Find next buffer in the page to map.
2375 mpd
->map
.m_flags
= 0;
2377 * FIXME: If dioread_nolock supports
2378 * blocksize < pagesize, we need to make
2379 * sure we add size mapped so far to
2380 * io_end->size as the following call
2381 * can submit the page for IO.
2383 err
= mpage_process_page_bufs(mpd
, head
,
2385 pagevec_release(&pvec
);
2390 if (buffer_delay(bh
)) {
2391 clear_buffer_delay(bh
);
2392 bh
->b_blocknr
= pblock
++;
2394 clear_buffer_unwritten(bh
);
2395 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2398 * FIXME: This is going to break if dioread_nolock
2399 * supports blocksize < pagesize as we will try to
2400 * convert potentially unmapped parts of inode.
2402 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2403 /* Page fully mapped - let IO run! */
2404 err
= mpage_submit_page(mpd
, page
);
2406 pagevec_release(&pvec
);
2410 pagevec_release(&pvec
);
2412 /* Extent fully mapped and matches with page boundary. We are done. */
2414 mpd
->map
.m_flags
= 0;
2418 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2420 struct inode
*inode
= mpd
->inode
;
2421 struct ext4_map_blocks
*map
= &mpd
->map
;
2422 int get_blocks_flags
;
2423 int err
, dioread_nolock
;
2425 trace_ext4_da_write_pages_extent(inode
, map
);
2427 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2428 * to convert an unwritten extent to be initialized (in the case
2429 * where we have written into one or more preallocated blocks). It is
2430 * possible that we're going to need more metadata blocks than
2431 * previously reserved. However we must not fail because we're in
2432 * writeback and there is nothing we can do about it so it might result
2433 * in data loss. So use reserved blocks to allocate metadata if
2436 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2437 * the blocks in question are delalloc blocks. This indicates
2438 * that the blocks and quotas has already been checked when
2439 * the data was copied into the page cache.
2441 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2442 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2443 EXT4_GET_BLOCKS_IO_SUBMIT
;
2444 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2446 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2447 if (map
->m_flags
& (1 << BH_Delay
))
2448 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2450 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2453 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2454 if (!mpd
->io_submit
.io_end
->handle
&&
2455 ext4_handle_valid(handle
)) {
2456 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2457 handle
->h_rsv_handle
= NULL
;
2459 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2462 BUG_ON(map
->m_len
== 0);
2463 if (map
->m_flags
& EXT4_MAP_NEW
) {
2464 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
2471 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2472 * mpd->len and submit pages underlying it for IO
2474 * @handle - handle for journal operations
2475 * @mpd - extent to map
2476 * @give_up_on_write - we set this to true iff there is a fatal error and there
2477 * is no hope of writing the data. The caller should discard
2478 * dirty pages to avoid infinite loops.
2480 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2481 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2482 * them to initialized or split the described range from larger unwritten
2483 * extent. Note that we need not map all the described range since allocation
2484 * can return less blocks or the range is covered by more unwritten extents. We
2485 * cannot map more because we are limited by reserved transaction credits. On
2486 * the other hand we always make sure that the last touched page is fully
2487 * mapped so that it can be written out (and thus forward progress is
2488 * guaranteed). After mapping we submit all mapped pages for IO.
2490 static int mpage_map_and_submit_extent(handle_t
*handle
,
2491 struct mpage_da_data
*mpd
,
2492 bool *give_up_on_write
)
2494 struct inode
*inode
= mpd
->inode
;
2495 struct ext4_map_blocks
*map
= &mpd
->map
;
2500 mpd
->io_submit
.io_end
->offset
=
2501 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2503 err
= mpage_map_one_extent(handle
, mpd
);
2505 struct super_block
*sb
= inode
->i_sb
;
2507 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2508 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2509 goto invalidate_dirty_pages
;
2511 * Let the uper layers retry transient errors.
2512 * In the case of ENOSPC, if ext4_count_free_blocks()
2513 * is non-zero, a commit should free up blocks.
2515 if ((err
== -ENOMEM
) ||
2516 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2518 goto update_disksize
;
2521 ext4_msg(sb
, KERN_CRIT
,
2522 "Delayed block allocation failed for "
2523 "inode %lu at logical offset %llu with"
2524 " max blocks %u with error %d",
2526 (unsigned long long)map
->m_lblk
,
2527 (unsigned)map
->m_len
, -err
);
2528 ext4_msg(sb
, KERN_CRIT
,
2529 "This should not happen!! Data will "
2532 ext4_print_free_blocks(inode
);
2533 invalidate_dirty_pages
:
2534 *give_up_on_write
= true;
2539 * Update buffer state, submit mapped pages, and get us new
2542 err
= mpage_map_and_submit_buffers(mpd
);
2544 goto update_disksize
;
2545 } while (map
->m_len
);
2549 * Update on-disk size after IO is submitted. Races with
2550 * truncate are avoided by checking i_size under i_data_sem.
2552 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2553 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2557 down_write(&EXT4_I(inode
)->i_data_sem
);
2558 i_size
= i_size_read(inode
);
2559 if (disksize
> i_size
)
2561 if (disksize
> EXT4_I(inode
)->i_disksize
)
2562 EXT4_I(inode
)->i_disksize
= disksize
;
2563 up_write(&EXT4_I(inode
)->i_data_sem
);
2564 err2
= ext4_mark_inode_dirty(handle
, inode
);
2566 ext4_error(inode
->i_sb
,
2567 "Failed to mark inode %lu dirty",
2576 * Calculate the total number of credits to reserve for one writepages
2577 * iteration. This is called from ext4_writepages(). We map an extent of
2578 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2579 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2580 * bpp - 1 blocks in bpp different extents.
2582 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2584 int bpp
= ext4_journal_blocks_per_page(inode
);
2586 return ext4_meta_trans_blocks(inode
,
2587 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2591 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2592 * and underlying extent to map
2594 * @mpd - where to look for pages
2596 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2597 * IO immediately. When we find a page which isn't mapped we start accumulating
2598 * extent of buffers underlying these pages that needs mapping (formed by
2599 * either delayed or unwritten buffers). We also lock the pages containing
2600 * these buffers. The extent found is returned in @mpd structure (starting at
2601 * mpd->lblk with length mpd->len blocks).
2603 * Note that this function can attach bios to one io_end structure which are
2604 * neither logically nor physically contiguous. Although it may seem as an
2605 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2606 * case as we need to track IO to all buffers underlying a page in one io_end.
2608 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2610 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2611 struct pagevec pvec
;
2612 unsigned int nr_pages
;
2613 long left
= mpd
->wbc
->nr_to_write
;
2614 pgoff_t index
= mpd
->first_page
;
2615 pgoff_t end
= mpd
->last_page
;
2618 int blkbits
= mpd
->inode
->i_blkbits
;
2620 struct buffer_head
*head
;
2622 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2623 tag
= PAGECACHE_TAG_TOWRITE
;
2625 tag
= PAGECACHE_TAG_DIRTY
;
2627 pagevec_init(&pvec
);
2629 mpd
->next_page
= index
;
2630 while (index
<= end
) {
2631 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
, end
,
2636 for (i
= 0; i
< nr_pages
; i
++) {
2637 struct page
*page
= pvec
.pages
[i
];
2640 * Accumulated enough dirty pages? This doesn't apply
2641 * to WB_SYNC_ALL mode. For integrity sync we have to
2642 * keep going because someone may be concurrently
2643 * dirtying pages, and we might have synced a lot of
2644 * newly appeared dirty pages, but have not synced all
2645 * of the old dirty pages.
2647 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2650 /* If we can't merge this page, we are done. */
2651 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2656 * If the page is no longer dirty, or its mapping no
2657 * longer corresponds to inode we are writing (which
2658 * means it has been truncated or invalidated), or the
2659 * page is already under writeback and we are not doing
2660 * a data integrity writeback, skip the page
2662 if (!PageDirty(page
) ||
2663 (PageWriteback(page
) &&
2664 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2665 unlikely(page
->mapping
!= mapping
)) {
2670 wait_on_page_writeback(page
);
2671 BUG_ON(PageWriteback(page
));
2673 if (mpd
->map
.m_len
== 0)
2674 mpd
->first_page
= page
->index
;
2675 mpd
->next_page
= page
->index
+ 1;
2676 /* Add all dirty buffers to mpd */
2677 lblk
= ((ext4_lblk_t
)page
->index
) <<
2678 (PAGE_SHIFT
- blkbits
);
2679 head
= page_buffers(page
);
2680 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2686 pagevec_release(&pvec
);
2691 pagevec_release(&pvec
);
2695 static int ext4_writepages(struct address_space
*mapping
,
2696 struct writeback_control
*wbc
)
2698 pgoff_t writeback_index
= 0;
2699 long nr_to_write
= wbc
->nr_to_write
;
2700 int range_whole
= 0;
2702 handle_t
*handle
= NULL
;
2703 struct mpage_da_data mpd
;
2704 struct inode
*inode
= mapping
->host
;
2705 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2706 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2708 struct blk_plug plug
;
2709 bool give_up_on_write
= false;
2711 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2714 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2715 trace_ext4_writepages(inode
, wbc
);
2718 * No pages to write? This is mainly a kludge to avoid starting
2719 * a transaction for special inodes like journal inode on last iput()
2720 * because that could violate lock ordering on umount
2722 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2723 goto out_writepages
;
2725 if (ext4_should_journal_data(inode
)) {
2726 ret
= generic_writepages(mapping
, wbc
);
2727 goto out_writepages
;
2731 * If the filesystem has aborted, it is read-only, so return
2732 * right away instead of dumping stack traces later on that
2733 * will obscure the real source of the problem. We test
2734 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2735 * the latter could be true if the filesystem is mounted
2736 * read-only, and in that case, ext4_writepages should
2737 * *never* be called, so if that ever happens, we would want
2740 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2741 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2743 goto out_writepages
;
2746 if (ext4_should_dioread_nolock(inode
)) {
2748 * We may need to convert up to one extent per block in
2749 * the page and we may dirty the inode.
2751 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2755 * If we have inline data and arrive here, it means that
2756 * we will soon create the block for the 1st page, so
2757 * we'd better clear the inline data here.
2759 if (ext4_has_inline_data(inode
)) {
2760 /* Just inode will be modified... */
2761 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2762 if (IS_ERR(handle
)) {
2763 ret
= PTR_ERR(handle
);
2764 goto out_writepages
;
2766 BUG_ON(ext4_test_inode_state(inode
,
2767 EXT4_STATE_MAY_INLINE_DATA
));
2768 ext4_destroy_inline_data(handle
, inode
);
2769 ext4_journal_stop(handle
);
2772 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2775 if (wbc
->range_cyclic
) {
2776 writeback_index
= mapping
->writeback_index
;
2777 if (writeback_index
)
2779 mpd
.first_page
= writeback_index
;
2782 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2783 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2788 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2790 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2791 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2793 blk_start_plug(&plug
);
2796 * First writeback pages that don't need mapping - we can avoid
2797 * starting a transaction unnecessarily and also avoid being blocked
2798 * in the block layer on device congestion while having transaction
2802 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2803 if (!mpd
.io_submit
.io_end
) {
2807 ret
= mpage_prepare_extent_to_map(&mpd
);
2808 /* Submit prepared bio */
2809 ext4_io_submit(&mpd
.io_submit
);
2810 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2811 mpd
.io_submit
.io_end
= NULL
;
2812 /* Unlock pages we didn't use */
2813 mpage_release_unused_pages(&mpd
, false);
2817 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2818 /* For each extent of pages we use new io_end */
2819 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2820 if (!mpd
.io_submit
.io_end
) {
2826 * We have two constraints: We find one extent to map and we
2827 * must always write out whole page (makes a difference when
2828 * blocksize < pagesize) so that we don't block on IO when we
2829 * try to write out the rest of the page. Journalled mode is
2830 * not supported by delalloc.
2832 BUG_ON(ext4_should_journal_data(inode
));
2833 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2835 /* start a new transaction */
2836 handle
= ext4_journal_start_with_reserve(inode
,
2837 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2838 if (IS_ERR(handle
)) {
2839 ret
= PTR_ERR(handle
);
2840 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2841 "%ld pages, ino %lu; err %d", __func__
,
2842 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2843 /* Release allocated io_end */
2844 ext4_put_io_end(mpd
.io_submit
.io_end
);
2845 mpd
.io_submit
.io_end
= NULL
;
2850 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2851 ret
= mpage_prepare_extent_to_map(&mpd
);
2854 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2858 * We scanned the whole range (or exhausted
2859 * nr_to_write), submitted what was mapped and
2860 * didn't find anything needing mapping. We are
2867 * Caution: If the handle is synchronous,
2868 * ext4_journal_stop() can wait for transaction commit
2869 * to finish which may depend on writeback of pages to
2870 * complete or on page lock to be released. In that
2871 * case, we have to wait until after after we have
2872 * submitted all the IO, released page locks we hold,
2873 * and dropped io_end reference (for extent conversion
2874 * to be able to complete) before stopping the handle.
2876 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2877 ext4_journal_stop(handle
);
2881 /* Submit prepared bio */
2882 ext4_io_submit(&mpd
.io_submit
);
2883 /* Unlock pages we didn't use */
2884 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2886 * Drop our io_end reference we got from init. We have
2887 * to be careful and use deferred io_end finishing if
2888 * we are still holding the transaction as we can
2889 * release the last reference to io_end which may end
2890 * up doing unwritten extent conversion.
2893 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2894 ext4_journal_stop(handle
);
2896 ext4_put_io_end(mpd
.io_submit
.io_end
);
2897 mpd
.io_submit
.io_end
= NULL
;
2899 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2901 * Commit the transaction which would
2902 * free blocks released in the transaction
2905 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2909 /* Fatal error - ENOMEM, EIO... */
2914 blk_finish_plug(&plug
);
2915 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2917 mpd
.last_page
= writeback_index
- 1;
2923 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2925 * Set the writeback_index so that range_cyclic
2926 * mode will write it back later
2928 mapping
->writeback_index
= mpd
.first_page
;
2931 trace_ext4_writepages_result(inode
, wbc
, ret
,
2932 nr_to_write
- wbc
->nr_to_write
);
2933 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2937 static int ext4_dax_writepages(struct address_space
*mapping
,
2938 struct writeback_control
*wbc
)
2941 long nr_to_write
= wbc
->nr_to_write
;
2942 struct inode
*inode
= mapping
->host
;
2943 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2945 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2948 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2949 trace_ext4_writepages(inode
, wbc
);
2951 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
, wbc
);
2952 trace_ext4_writepages_result(inode
, wbc
, ret
,
2953 nr_to_write
- wbc
->nr_to_write
);
2954 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2958 static int ext4_nonda_switch(struct super_block
*sb
)
2960 s64 free_clusters
, dirty_clusters
;
2961 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2964 * switch to non delalloc mode if we are running low
2965 * on free block. The free block accounting via percpu
2966 * counters can get slightly wrong with percpu_counter_batch getting
2967 * accumulated on each CPU without updating global counters
2968 * Delalloc need an accurate free block accounting. So switch
2969 * to non delalloc when we are near to error range.
2972 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2974 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2976 * Start pushing delalloc when 1/2 of free blocks are dirty.
2978 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2979 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2981 if (2 * free_clusters
< 3 * dirty_clusters
||
2982 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2984 * free block count is less than 150% of dirty blocks
2985 * or free blocks is less than watermark
2992 /* We always reserve for an inode update; the superblock could be there too */
2993 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2995 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2998 if (pos
+ len
<= 0x7fffffffULL
)
3001 /* We might need to update the superblock to set LARGE_FILE */
3005 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3006 loff_t pos
, unsigned len
, unsigned flags
,
3007 struct page
**pagep
, void **fsdata
)
3009 int ret
, retries
= 0;
3012 struct inode
*inode
= mapping
->host
;
3015 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
3018 index
= pos
>> PAGE_SHIFT
;
3020 if (ext4_nonda_switch(inode
->i_sb
) ||
3021 S_ISLNK(inode
->i_mode
)) {
3022 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3023 return ext4_write_begin(file
, mapping
, pos
,
3024 len
, flags
, pagep
, fsdata
);
3026 *fsdata
= (void *)0;
3027 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3029 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
3030 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
3040 * grab_cache_page_write_begin() can take a long time if the
3041 * system is thrashing due to memory pressure, or if the page
3042 * is being written back. So grab it first before we start
3043 * the transaction handle. This also allows us to allocate
3044 * the page (if needed) without using GFP_NOFS.
3047 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3053 * With delayed allocation, we don't log the i_disksize update
3054 * if there is delayed block allocation. But we still need
3055 * to journalling the i_disksize update if writes to the end
3056 * of file which has an already mapped buffer.
3059 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
3060 ext4_da_write_credits(inode
, pos
, len
));
3061 if (IS_ERR(handle
)) {
3063 return PTR_ERR(handle
);
3067 if (page
->mapping
!= mapping
) {
3068 /* The page got truncated from under us */
3071 ext4_journal_stop(handle
);
3074 /* In case writeback began while the page was unlocked */
3075 wait_for_stable_page(page
);
3077 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3078 ret
= ext4_block_write_begin(page
, pos
, len
,
3079 ext4_da_get_block_prep
);
3081 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3085 ext4_journal_stop(handle
);
3087 * block_write_begin may have instantiated a few blocks
3088 * outside i_size. Trim these off again. Don't need
3089 * i_size_read because we hold i_mutex.
3091 if (pos
+ len
> inode
->i_size
)
3092 ext4_truncate_failed_write(inode
);
3094 if (ret
== -ENOSPC
&&
3095 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3107 * Check if we should update i_disksize
3108 * when write to the end of file but not require block allocation
3110 static int ext4_da_should_update_i_disksize(struct page
*page
,
3111 unsigned long offset
)
3113 struct buffer_head
*bh
;
3114 struct inode
*inode
= page
->mapping
->host
;
3118 bh
= page_buffers(page
);
3119 idx
= offset
>> inode
->i_blkbits
;
3121 for (i
= 0; i
< idx
; i
++)
3122 bh
= bh
->b_this_page
;
3124 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3129 static int ext4_da_write_end(struct file
*file
,
3130 struct address_space
*mapping
,
3131 loff_t pos
, unsigned len
, unsigned copied
,
3132 struct page
*page
, void *fsdata
)
3134 struct inode
*inode
= mapping
->host
;
3136 handle_t
*handle
= ext4_journal_current_handle();
3138 unsigned long start
, end
;
3139 int write_mode
= (int)(unsigned long)fsdata
;
3141 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3142 return ext4_write_end(file
, mapping
, pos
,
3143 len
, copied
, page
, fsdata
);
3145 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3146 start
= pos
& (PAGE_SIZE
- 1);
3147 end
= start
+ copied
- 1;
3150 * generic_write_end() will run mark_inode_dirty() if i_size
3151 * changes. So let's piggyback the i_disksize mark_inode_dirty
3154 new_i_size
= pos
+ copied
;
3155 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3156 if (ext4_has_inline_data(inode
) ||
3157 ext4_da_should_update_i_disksize(page
, end
)) {
3158 ext4_update_i_disksize(inode
, new_i_size
);
3159 /* We need to mark inode dirty even if
3160 * new_i_size is less that inode->i_size
3161 * bu greater than i_disksize.(hint delalloc)
3163 ext4_mark_inode_dirty(handle
, inode
);
3167 if (write_mode
!= CONVERT_INLINE_DATA
&&
3168 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3169 ext4_has_inline_data(inode
))
3170 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3173 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3179 ret2
= ext4_journal_stop(handle
);
3183 return ret
? ret
: copied
;
3186 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3187 unsigned int length
)
3190 * Drop reserved blocks
3192 BUG_ON(!PageLocked(page
));
3193 if (!page_has_buffers(page
))
3196 ext4_da_page_release_reservation(page
, offset
, length
);
3199 ext4_invalidatepage(page
, offset
, length
);
3205 * Force all delayed allocation blocks to be allocated for a given inode.
3207 int ext4_alloc_da_blocks(struct inode
*inode
)
3209 trace_ext4_alloc_da_blocks(inode
);
3211 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3215 * We do something simple for now. The filemap_flush() will
3216 * also start triggering a write of the data blocks, which is
3217 * not strictly speaking necessary (and for users of
3218 * laptop_mode, not even desirable). However, to do otherwise
3219 * would require replicating code paths in:
3221 * ext4_writepages() ->
3222 * write_cache_pages() ---> (via passed in callback function)
3223 * __mpage_da_writepage() -->
3224 * mpage_add_bh_to_extent()
3225 * mpage_da_map_blocks()
3227 * The problem is that write_cache_pages(), located in
3228 * mm/page-writeback.c, marks pages clean in preparation for
3229 * doing I/O, which is not desirable if we're not planning on
3232 * We could call write_cache_pages(), and then redirty all of
3233 * the pages by calling redirty_page_for_writepage() but that
3234 * would be ugly in the extreme. So instead we would need to
3235 * replicate parts of the code in the above functions,
3236 * simplifying them because we wouldn't actually intend to
3237 * write out the pages, but rather only collect contiguous
3238 * logical block extents, call the multi-block allocator, and
3239 * then update the buffer heads with the block allocations.
3241 * For now, though, we'll cheat by calling filemap_flush(),
3242 * which will map the blocks, and start the I/O, but not
3243 * actually wait for the I/O to complete.
3245 return filemap_flush(inode
->i_mapping
);
3249 * bmap() is special. It gets used by applications such as lilo and by
3250 * the swapper to find the on-disk block of a specific piece of data.
3252 * Naturally, this is dangerous if the block concerned is still in the
3253 * journal. If somebody makes a swapfile on an ext4 data-journaling
3254 * filesystem and enables swap, then they may get a nasty shock when the
3255 * data getting swapped to that swapfile suddenly gets overwritten by
3256 * the original zero's written out previously to the journal and
3257 * awaiting writeback in the kernel's buffer cache.
3259 * So, if we see any bmap calls here on a modified, data-journaled file,
3260 * take extra steps to flush any blocks which might be in the cache.
3262 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3264 struct inode
*inode
= mapping
->host
;
3269 * We can get here for an inline file via the FIBMAP ioctl
3271 if (ext4_has_inline_data(inode
))
3274 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3275 test_opt(inode
->i_sb
, DELALLOC
)) {
3277 * With delalloc we want to sync the file
3278 * so that we can make sure we allocate
3281 filemap_write_and_wait(mapping
);
3284 if (EXT4_JOURNAL(inode
) &&
3285 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3287 * This is a REALLY heavyweight approach, but the use of
3288 * bmap on dirty files is expected to be extremely rare:
3289 * only if we run lilo or swapon on a freshly made file
3290 * do we expect this to happen.
3292 * (bmap requires CAP_SYS_RAWIO so this does not
3293 * represent an unprivileged user DOS attack --- we'd be
3294 * in trouble if mortal users could trigger this path at
3297 * NB. EXT4_STATE_JDATA is not set on files other than
3298 * regular files. If somebody wants to bmap a directory
3299 * or symlink and gets confused because the buffer
3300 * hasn't yet been flushed to disk, they deserve
3301 * everything they get.
3304 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3305 journal
= EXT4_JOURNAL(inode
);
3306 jbd2_journal_lock_updates(journal
);
3307 err
= jbd2_journal_flush(journal
);
3308 jbd2_journal_unlock_updates(journal
);
3314 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3317 static int ext4_readpage(struct file
*file
, struct page
*page
)
3320 struct inode
*inode
= page
->mapping
->host
;
3322 trace_ext4_readpage(page
);
3324 if (ext4_has_inline_data(inode
))
3325 ret
= ext4_readpage_inline(inode
, page
);
3328 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1,
3335 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3336 struct list_head
*pages
, unsigned nr_pages
)
3338 struct inode
*inode
= mapping
->host
;
3340 /* If the file has inline data, no need to do readpages. */
3341 if (ext4_has_inline_data(inode
))
3344 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
, true);
3347 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3348 unsigned int length
)
3350 trace_ext4_invalidatepage(page
, offset
, length
);
3352 /* No journalling happens on data buffers when this function is used */
3353 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3355 block_invalidatepage(page
, offset
, length
);
3358 static int __ext4_journalled_invalidatepage(struct page
*page
,
3359 unsigned int offset
,
3360 unsigned int length
)
3362 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3364 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3367 * If it's a full truncate we just forget about the pending dirtying
3369 if (offset
== 0 && length
== PAGE_SIZE
)
3370 ClearPageChecked(page
);
3372 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3375 /* Wrapper for aops... */
3376 static void ext4_journalled_invalidatepage(struct page
*page
,
3377 unsigned int offset
,
3378 unsigned int length
)
3380 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3383 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3385 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3387 trace_ext4_releasepage(page
);
3389 /* Page has dirty journalled data -> cannot release */
3390 if (PageChecked(page
))
3393 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3395 return try_to_free_buffers(page
);
3398 static bool ext4_inode_datasync_dirty(struct inode
*inode
)
3400 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
3403 return !jbd2_transaction_committed(journal
,
3404 EXT4_I(inode
)->i_datasync_tid
);
3405 /* Any metadata buffers to write? */
3406 if (!list_empty(&inode
->i_mapping
->private_list
))
3408 return inode
->i_state
& I_DIRTY_DATASYNC
;
3411 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3412 unsigned flags
, struct iomap
*iomap
)
3414 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3415 unsigned int blkbits
= inode
->i_blkbits
;
3416 unsigned long first_block
= offset
>> blkbits
;
3417 unsigned long last_block
= (offset
+ length
- 1) >> blkbits
;
3418 struct ext4_map_blocks map
;
3419 bool delalloc
= false;
3423 if (flags
& IOMAP_REPORT
) {
3424 if (ext4_has_inline_data(inode
)) {
3425 ret
= ext4_inline_data_iomap(inode
, iomap
);
3426 if (ret
!= -EAGAIN
) {
3427 if (ret
== 0 && offset
>= iomap
->length
)
3433 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3437 map
.m_lblk
= first_block
;
3438 map
.m_len
= last_block
- first_block
+ 1;
3440 if (flags
& IOMAP_REPORT
) {
3441 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3446 ext4_lblk_t end
= map
.m_lblk
+ map
.m_len
- 1;
3447 struct extent_status es
;
3449 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
, end
, &es
);
3451 if (!es
.es_len
|| es
.es_lblk
> end
) {
3452 /* entire range is a hole */
3453 } else if (es
.es_lblk
> map
.m_lblk
) {
3454 /* range starts with a hole */
3455 map
.m_len
= es
.es_lblk
- map
.m_lblk
;
3457 ext4_lblk_t offs
= 0;
3459 if (es
.es_lblk
< map
.m_lblk
)
3460 offs
= map
.m_lblk
- es
.es_lblk
;
3461 map
.m_lblk
= es
.es_lblk
+ offs
;
3462 map
.m_len
= es
.es_len
- offs
;
3466 } else if (flags
& IOMAP_WRITE
) {
3471 /* Trim mapping request to maximum we can map at once for DIO */
3472 if (map
.m_len
> DIO_MAX_BLOCKS
)
3473 map
.m_len
= DIO_MAX_BLOCKS
;
3474 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3477 * Either we allocate blocks and then we don't get unwritten
3478 * extent so we have reserved enough credits, or the blocks
3479 * are already allocated and unwritten and in that case
3480 * extent conversion fits in the credits as well.
3482 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3485 return PTR_ERR(handle
);
3487 ret
= ext4_map_blocks(handle
, inode
, &map
,
3488 EXT4_GET_BLOCKS_CREATE_ZERO
);
3490 ext4_journal_stop(handle
);
3491 if (ret
== -ENOSPC
&&
3492 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3498 * If we added blocks beyond i_size, we need to make sure they
3499 * will get truncated if we crash before updating i_size in
3500 * ext4_iomap_end(). For faults we don't need to do that (and
3501 * even cannot because for orphan list operations inode_lock is
3502 * required) - if we happen to instantiate block beyond i_size,
3503 * it is because we race with truncate which has already added
3504 * the inode to the orphan list.
3506 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3507 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3510 err
= ext4_orphan_add(handle
, inode
);
3512 ext4_journal_stop(handle
);
3516 ext4_journal_stop(handle
);
3518 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3524 if (ext4_inode_datasync_dirty(inode
))
3525 iomap
->flags
|= IOMAP_F_DIRTY
;
3526 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3527 iomap
->dax_dev
= sbi
->s_daxdev
;
3528 iomap
->offset
= (u64
)first_block
<< blkbits
;
3529 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3532 iomap
->type
= delalloc
? IOMAP_DELALLOC
: IOMAP_HOLE
;
3533 iomap
->addr
= IOMAP_NULL_ADDR
;
3535 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3536 iomap
->type
= IOMAP_MAPPED
;
3537 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3538 iomap
->type
= IOMAP_UNWRITTEN
;
3543 iomap
->addr
= (u64
)map
.m_pblk
<< blkbits
;
3546 if (map
.m_flags
& EXT4_MAP_NEW
)
3547 iomap
->flags
|= IOMAP_F_NEW
;
3552 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3553 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3557 int blkbits
= inode
->i_blkbits
;
3558 bool truncate
= false;
3560 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3563 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3564 if (IS_ERR(handle
)) {
3565 ret
= PTR_ERR(handle
);
3568 if (ext4_update_inode_size(inode
, offset
+ written
))
3569 ext4_mark_inode_dirty(handle
, inode
);
3571 * We may need to truncate allocated but not written blocks beyond EOF.
3573 if (iomap
->offset
+ iomap
->length
>
3574 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3575 ext4_lblk_t written_blk
, end_blk
;
3577 written_blk
= (offset
+ written
) >> blkbits
;
3578 end_blk
= (offset
+ length
) >> blkbits
;
3579 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3583 * Remove inode from orphan list if we were extending a inode and
3584 * everything went fine.
3586 if (!truncate
&& inode
->i_nlink
&&
3587 !list_empty(&EXT4_I(inode
)->i_orphan
))
3588 ext4_orphan_del(handle
, inode
);
3589 ext4_journal_stop(handle
);
3591 ext4_truncate_failed_write(inode
);
3594 * If truncate failed early the inode might still be on the
3595 * orphan list; we need to make sure the inode is removed from
3596 * the orphan list in that case.
3599 ext4_orphan_del(NULL
, inode
);
3604 const struct iomap_ops ext4_iomap_ops
= {
3605 .iomap_begin
= ext4_iomap_begin
,
3606 .iomap_end
= ext4_iomap_end
,
3609 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3610 ssize_t size
, void *private)
3612 ext4_io_end_t
*io_end
= private;
3614 /* if not async direct IO just return */
3618 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3619 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3620 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3623 * Error during AIO DIO. We cannot convert unwritten extents as the
3624 * data was not written. Just clear the unwritten flag and drop io_end.
3627 ext4_clear_io_unwritten_flag(io_end
);
3630 io_end
->offset
= offset
;
3631 io_end
->size
= size
;
3632 ext4_put_io_end(io_end
);
3638 * Handling of direct IO writes.
3640 * For ext4 extent files, ext4 will do direct-io write even to holes,
3641 * preallocated extents, and those write extend the file, no need to
3642 * fall back to buffered IO.
3644 * For holes, we fallocate those blocks, mark them as unwritten
3645 * If those blocks were preallocated, we mark sure they are split, but
3646 * still keep the range to write as unwritten.
3648 * The unwritten extents will be converted to written when DIO is completed.
3649 * For async direct IO, since the IO may still pending when return, we
3650 * set up an end_io call back function, which will do the conversion
3651 * when async direct IO completed.
3653 * If the O_DIRECT write will extend the file then add this inode to the
3654 * orphan list. So recovery will truncate it back to the original size
3655 * if the machine crashes during the write.
3658 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3660 struct file
*file
= iocb
->ki_filp
;
3661 struct inode
*inode
= file
->f_mapping
->host
;
3662 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3664 loff_t offset
= iocb
->ki_pos
;
3665 size_t count
= iov_iter_count(iter
);
3667 get_block_t
*get_block_func
= NULL
;
3669 loff_t final_size
= offset
+ count
;
3673 if (final_size
> inode
->i_size
|| final_size
> ei
->i_disksize
) {
3674 /* Credits for sb + inode write */
3675 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3676 if (IS_ERR(handle
)) {
3677 ret
= PTR_ERR(handle
);
3680 ret
= ext4_orphan_add(handle
, inode
);
3682 ext4_journal_stop(handle
);
3686 ext4_update_i_disksize(inode
, inode
->i_size
);
3687 ext4_journal_stop(handle
);
3690 BUG_ON(iocb
->private == NULL
);
3693 * Make all waiters for direct IO properly wait also for extent
3694 * conversion. This also disallows race between truncate() and
3695 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3697 inode_dio_begin(inode
);
3699 /* If we do a overwrite dio, i_mutex locking can be released */
3700 overwrite
= *((int *)iocb
->private);
3703 inode_unlock(inode
);
3706 * For extent mapped files we could direct write to holes and fallocate.
3708 * Allocated blocks to fill the hole are marked as unwritten to prevent
3709 * parallel buffered read to expose the stale data before DIO complete
3712 * As to previously fallocated extents, ext4 get_block will just simply
3713 * mark the buffer mapped but still keep the extents unwritten.
3715 * For non AIO case, we will convert those unwritten extents to written
3716 * after return back from blockdev_direct_IO. That way we save us from
3717 * allocating io_end structure and also the overhead of offloading
3718 * the extent convertion to a workqueue.
3720 * For async DIO, the conversion needs to be deferred when the
3721 * IO is completed. The ext4 end_io callback function will be
3722 * called to take care of the conversion work. Here for async
3723 * case, we allocate an io_end structure to hook to the iocb.
3725 iocb
->private = NULL
;
3727 get_block_func
= ext4_dio_get_block_overwrite
;
3728 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3729 round_down(offset
, i_blocksize(inode
)) >= inode
->i_size
) {
3730 get_block_func
= ext4_dio_get_block
;
3731 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3732 } else if (is_sync_kiocb(iocb
)) {
3733 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3734 dio_flags
= DIO_LOCKING
;
3736 get_block_func
= ext4_dio_get_block_unwritten_async
;
3737 dio_flags
= DIO_LOCKING
;
3739 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3740 get_block_func
, ext4_end_io_dio
, NULL
,
3743 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3744 EXT4_STATE_DIO_UNWRITTEN
)) {
3747 * for non AIO case, since the IO is already
3748 * completed, we could do the conversion right here
3750 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3754 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3757 inode_dio_end(inode
);
3758 /* take i_mutex locking again if we do a ovewrite dio */
3762 if (ret
< 0 && final_size
> inode
->i_size
)
3763 ext4_truncate_failed_write(inode
);
3765 /* Handle extending of i_size after direct IO write */
3769 /* Credits for sb + inode write */
3770 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3771 if (IS_ERR(handle
)) {
3773 * We wrote the data but cannot extend
3774 * i_size. Bail out. In async io case, we do
3775 * not return error here because we have
3776 * already submmitted the corresponding
3777 * bio. Returning error here makes the caller
3778 * think that this IO is done and failed
3779 * resulting in race with bio's completion
3783 ret
= PTR_ERR(handle
);
3785 ext4_orphan_del(NULL
, inode
);
3790 ext4_orphan_del(handle
, inode
);
3792 loff_t end
= offset
+ ret
;
3793 if (end
> inode
->i_size
|| end
> ei
->i_disksize
) {
3794 ext4_update_i_disksize(inode
, end
);
3795 if (end
> inode
->i_size
)
3796 i_size_write(inode
, end
);
3798 * We're going to return a positive `ret'
3799 * here due to non-zero-length I/O, so there's
3800 * no way of reporting error returns from
3801 * ext4_mark_inode_dirty() to userspace. So
3804 ext4_mark_inode_dirty(handle
, inode
);
3807 err
= ext4_journal_stop(handle
);
3815 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3817 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3818 struct inode
*inode
= mapping
->host
;
3819 size_t count
= iov_iter_count(iter
);
3823 * Shared inode_lock is enough for us - it protects against concurrent
3824 * writes & truncates and since we take care of writing back page cache,
3825 * we are protected against page writeback as well.
3827 inode_lock_shared(inode
);
3828 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3829 iocb
->ki_pos
+ count
- 1);
3832 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3833 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3835 inode_unlock_shared(inode
);
3839 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3841 struct file
*file
= iocb
->ki_filp
;
3842 struct inode
*inode
= file
->f_mapping
->host
;
3843 size_t count
= iov_iter_count(iter
);
3844 loff_t offset
= iocb
->ki_pos
;
3847 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3848 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3853 * If we are doing data journalling we don't support O_DIRECT
3855 if (ext4_should_journal_data(inode
))
3858 /* Let buffer I/O handle the inline data case. */
3859 if (ext4_has_inline_data(inode
))
3862 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3863 if (iov_iter_rw(iter
) == READ
)
3864 ret
= ext4_direct_IO_read(iocb
, iter
);
3866 ret
= ext4_direct_IO_write(iocb
, iter
);
3867 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3872 * Pages can be marked dirty completely asynchronously from ext4's journalling
3873 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3874 * much here because ->set_page_dirty is called under VFS locks. The page is
3875 * not necessarily locked.
3877 * We cannot just dirty the page and leave attached buffers clean, because the
3878 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3879 * or jbddirty because all the journalling code will explode.
3881 * So what we do is to mark the page "pending dirty" and next time writepage
3882 * is called, propagate that into the buffers appropriately.
3884 static int ext4_journalled_set_page_dirty(struct page
*page
)
3886 SetPageChecked(page
);
3887 return __set_page_dirty_nobuffers(page
);
3890 static int ext4_set_page_dirty(struct page
*page
)
3892 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3893 WARN_ON_ONCE(!page_has_buffers(page
));
3894 return __set_page_dirty_buffers(page
);
3897 static const struct address_space_operations ext4_aops
= {
3898 .readpage
= ext4_readpage
,
3899 .readpages
= ext4_readpages
,
3900 .writepage
= ext4_writepage
,
3901 .writepages
= ext4_writepages
,
3902 .write_begin
= ext4_write_begin
,
3903 .write_end
= ext4_write_end
,
3904 .set_page_dirty
= ext4_set_page_dirty
,
3906 .invalidatepage
= ext4_invalidatepage
,
3907 .releasepage
= ext4_releasepage
,
3908 .direct_IO
= ext4_direct_IO
,
3909 .migratepage
= buffer_migrate_page
,
3910 .is_partially_uptodate
= block_is_partially_uptodate
,
3911 .error_remove_page
= generic_error_remove_page
,
3914 static const struct address_space_operations ext4_journalled_aops
= {
3915 .readpage
= ext4_readpage
,
3916 .readpages
= ext4_readpages
,
3917 .writepage
= ext4_writepage
,
3918 .writepages
= ext4_writepages
,
3919 .write_begin
= ext4_write_begin
,
3920 .write_end
= ext4_journalled_write_end
,
3921 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3923 .invalidatepage
= ext4_journalled_invalidatepage
,
3924 .releasepage
= ext4_releasepage
,
3925 .direct_IO
= ext4_direct_IO
,
3926 .is_partially_uptodate
= block_is_partially_uptodate
,
3927 .error_remove_page
= generic_error_remove_page
,
3930 static const struct address_space_operations ext4_da_aops
= {
3931 .readpage
= ext4_readpage
,
3932 .readpages
= ext4_readpages
,
3933 .writepage
= ext4_writepage
,
3934 .writepages
= ext4_writepages
,
3935 .write_begin
= ext4_da_write_begin
,
3936 .write_end
= ext4_da_write_end
,
3937 .set_page_dirty
= ext4_set_page_dirty
,
3939 .invalidatepage
= ext4_da_invalidatepage
,
3940 .releasepage
= ext4_releasepage
,
3941 .direct_IO
= ext4_direct_IO
,
3942 .migratepage
= buffer_migrate_page
,
3943 .is_partially_uptodate
= block_is_partially_uptodate
,
3944 .error_remove_page
= generic_error_remove_page
,
3947 static const struct address_space_operations ext4_dax_aops
= {
3948 .writepages
= ext4_dax_writepages
,
3949 .direct_IO
= noop_direct_IO
,
3950 .set_page_dirty
= noop_set_page_dirty
,
3951 .invalidatepage
= noop_invalidatepage
,
3954 void ext4_set_aops(struct inode
*inode
)
3956 switch (ext4_inode_journal_mode(inode
)) {
3957 case EXT4_INODE_ORDERED_DATA_MODE
:
3958 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3960 case EXT4_INODE_JOURNAL_DATA_MODE
:
3961 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3967 inode
->i_mapping
->a_ops
= &ext4_dax_aops
;
3968 else if (test_opt(inode
->i_sb
, DELALLOC
))
3969 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3971 inode
->i_mapping
->a_ops
= &ext4_aops
;
3974 static int __ext4_block_zero_page_range(handle_t
*handle
,
3975 struct address_space
*mapping
, loff_t from
, loff_t length
)
3977 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3978 unsigned offset
= from
& (PAGE_SIZE
-1);
3979 unsigned blocksize
, pos
;
3981 struct inode
*inode
= mapping
->host
;
3982 struct buffer_head
*bh
;
3986 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3987 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3991 blocksize
= inode
->i_sb
->s_blocksize
;
3993 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3995 if (!page_has_buffers(page
))
3996 create_empty_buffers(page
, blocksize
, 0);
3998 /* Find the buffer that contains "offset" */
3999 bh
= page_buffers(page
);
4001 while (offset
>= pos
) {
4002 bh
= bh
->b_this_page
;
4006 if (buffer_freed(bh
)) {
4007 BUFFER_TRACE(bh
, "freed: skip");
4010 if (!buffer_mapped(bh
)) {
4011 BUFFER_TRACE(bh
, "unmapped");
4012 ext4_get_block(inode
, iblock
, bh
, 0);
4013 /* unmapped? It's a hole - nothing to do */
4014 if (!buffer_mapped(bh
)) {
4015 BUFFER_TRACE(bh
, "still unmapped");
4020 /* Ok, it's mapped. Make sure it's up-to-date */
4021 if (PageUptodate(page
))
4022 set_buffer_uptodate(bh
);
4024 if (!buffer_uptodate(bh
)) {
4026 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
4028 /* Uhhuh. Read error. Complain and punt. */
4029 if (!buffer_uptodate(bh
))
4031 if (S_ISREG(inode
->i_mode
) &&
4032 ext4_encrypted_inode(inode
)) {
4033 /* We expect the key to be set. */
4034 BUG_ON(!fscrypt_has_encryption_key(inode
));
4035 BUG_ON(blocksize
!= PAGE_SIZE
);
4036 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
4037 page
, PAGE_SIZE
, 0, page
->index
));
4040 if (ext4_should_journal_data(inode
)) {
4041 BUFFER_TRACE(bh
, "get write access");
4042 err
= ext4_journal_get_write_access(handle
, bh
);
4046 zero_user(page
, offset
, length
);
4047 BUFFER_TRACE(bh
, "zeroed end of block");
4049 if (ext4_should_journal_data(inode
)) {
4050 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4053 mark_buffer_dirty(bh
);
4054 if (ext4_should_order_data(inode
))
4055 err
= ext4_jbd2_inode_add_write(handle
, inode
);
4065 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4066 * starting from file offset 'from'. The range to be zero'd must
4067 * be contained with in one block. If the specified range exceeds
4068 * the end of the block it will be shortened to end of the block
4069 * that cooresponds to 'from'
4071 static int ext4_block_zero_page_range(handle_t
*handle
,
4072 struct address_space
*mapping
, loff_t from
, loff_t length
)
4074 struct inode
*inode
= mapping
->host
;
4075 unsigned offset
= from
& (PAGE_SIZE
-1);
4076 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4077 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
4080 * correct length if it does not fall between
4081 * 'from' and the end of the block
4083 if (length
> max
|| length
< 0)
4086 if (IS_DAX(inode
)) {
4087 return iomap_zero_range(inode
, from
, length
, NULL
,
4090 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4094 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4095 * up to the end of the block which corresponds to `from'.
4096 * This required during truncate. We need to physically zero the tail end
4097 * of that block so it doesn't yield old data if the file is later grown.
4099 static int ext4_block_truncate_page(handle_t
*handle
,
4100 struct address_space
*mapping
, loff_t from
)
4102 unsigned offset
= from
& (PAGE_SIZE
-1);
4105 struct inode
*inode
= mapping
->host
;
4107 /* If we are processing an encrypted inode during orphan list handling */
4108 if (ext4_encrypted_inode(inode
) && !fscrypt_has_encryption_key(inode
))
4111 blocksize
= inode
->i_sb
->s_blocksize
;
4112 length
= blocksize
- (offset
& (blocksize
- 1));
4114 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4117 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
4118 loff_t lstart
, loff_t length
)
4120 struct super_block
*sb
= inode
->i_sb
;
4121 struct address_space
*mapping
= inode
->i_mapping
;
4122 unsigned partial_start
, partial_end
;
4123 ext4_fsblk_t start
, end
;
4124 loff_t byte_end
= (lstart
+ length
- 1);
4127 partial_start
= lstart
& (sb
->s_blocksize
- 1);
4128 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
4130 start
= lstart
>> sb
->s_blocksize_bits
;
4131 end
= byte_end
>> sb
->s_blocksize_bits
;
4133 /* Handle partial zero within the single block */
4135 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
4136 err
= ext4_block_zero_page_range(handle
, mapping
,
4140 /* Handle partial zero out on the start of the range */
4141 if (partial_start
) {
4142 err
= ext4_block_zero_page_range(handle
, mapping
,
4143 lstart
, sb
->s_blocksize
);
4147 /* Handle partial zero out on the end of the range */
4148 if (partial_end
!= sb
->s_blocksize
- 1)
4149 err
= ext4_block_zero_page_range(handle
, mapping
,
4150 byte_end
- partial_end
,
4155 int ext4_can_truncate(struct inode
*inode
)
4157 if (S_ISREG(inode
->i_mode
))
4159 if (S_ISDIR(inode
->i_mode
))
4161 if (S_ISLNK(inode
->i_mode
))
4162 return !ext4_inode_is_fast_symlink(inode
);
4167 * We have to make sure i_disksize gets properly updated before we truncate
4168 * page cache due to hole punching or zero range. Otherwise i_disksize update
4169 * can get lost as it may have been postponed to submission of writeback but
4170 * that will never happen after we truncate page cache.
4172 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
4176 loff_t size
= i_size_read(inode
);
4178 WARN_ON(!inode_is_locked(inode
));
4179 if (offset
> size
|| offset
+ len
< size
)
4182 if (EXT4_I(inode
)->i_disksize
>= size
)
4185 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4187 return PTR_ERR(handle
);
4188 ext4_update_i_disksize(inode
, size
);
4189 ext4_mark_inode_dirty(handle
, inode
);
4190 ext4_journal_stop(handle
);
4195 static void ext4_wait_dax_page(struct ext4_inode_info
*ei
, bool *did_unlock
)
4198 up_write(&ei
->i_mmap_sem
);
4200 down_write(&ei
->i_mmap_sem
);
4203 int ext4_break_layouts(struct inode
*inode
)
4205 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4210 if (WARN_ON_ONCE(!rwsem_is_locked(&ei
->i_mmap_sem
)))
4215 page
= dax_layout_busy_page(inode
->i_mapping
);
4219 error
= ___wait_var_event(&page
->_refcount
,
4220 atomic_read(&page
->_refcount
) == 1,
4221 TASK_INTERRUPTIBLE
, 0, 0,
4222 ext4_wait_dax_page(ei
, &retry
));
4223 } while (error
== 0 && retry
);
4229 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4230 * associated with the given offset and length
4232 * @inode: File inode
4233 * @offset: The offset where the hole will begin
4234 * @len: The length of the hole
4236 * Returns: 0 on success or negative on failure
4239 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4241 struct super_block
*sb
= inode
->i_sb
;
4242 ext4_lblk_t first_block
, stop_block
;
4243 struct address_space
*mapping
= inode
->i_mapping
;
4244 loff_t first_block_offset
, last_block_offset
;
4246 unsigned int credits
;
4249 if (!S_ISREG(inode
->i_mode
))
4252 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4255 * Write out all dirty pages to avoid race conditions
4256 * Then release them.
4258 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4259 ret
= filemap_write_and_wait_range(mapping
, offset
,
4260 offset
+ length
- 1);
4267 /* No need to punch hole beyond i_size */
4268 if (offset
>= inode
->i_size
)
4272 * If the hole extends beyond i_size, set the hole
4273 * to end after the page that contains i_size
4275 if (offset
+ length
> inode
->i_size
) {
4276 length
= inode
->i_size
+
4277 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4281 if (offset
& (sb
->s_blocksize
- 1) ||
4282 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4284 * Attach jinode to inode for jbd2 if we do any zeroing of
4287 ret
= ext4_inode_attach_jinode(inode
);
4293 /* Wait all existing dio workers, newcomers will block on i_mutex */
4294 inode_dio_wait(inode
);
4297 * Prevent page faults from reinstantiating pages we have released from
4300 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4302 ret
= ext4_break_layouts(inode
);
4306 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4307 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4309 /* Now release the pages and zero block aligned part of pages*/
4310 if (last_block_offset
> first_block_offset
) {
4311 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4314 truncate_pagecache_range(inode
, first_block_offset
,
4318 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4319 credits
= ext4_writepage_trans_blocks(inode
);
4321 credits
= ext4_blocks_for_truncate(inode
);
4322 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4323 if (IS_ERR(handle
)) {
4324 ret
= PTR_ERR(handle
);
4325 ext4_std_error(sb
, ret
);
4329 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4334 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4335 EXT4_BLOCK_SIZE_BITS(sb
);
4336 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4338 /* If there are blocks to remove, do it */
4339 if (stop_block
> first_block
) {
4341 down_write(&EXT4_I(inode
)->i_data_sem
);
4342 ext4_discard_preallocations(inode
);
4344 ret
= ext4_es_remove_extent(inode
, first_block
,
4345 stop_block
- first_block
);
4347 up_write(&EXT4_I(inode
)->i_data_sem
);
4351 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4352 ret
= ext4_ext_remove_space(inode
, first_block
,
4355 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4358 up_write(&EXT4_I(inode
)->i_data_sem
);
4361 ext4_handle_sync(handle
);
4363 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4364 ext4_mark_inode_dirty(handle
, inode
);
4366 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4368 ext4_journal_stop(handle
);
4370 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4372 inode_unlock(inode
);
4376 int ext4_inode_attach_jinode(struct inode
*inode
)
4378 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4379 struct jbd2_inode
*jinode
;
4381 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4384 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4385 spin_lock(&inode
->i_lock
);
4388 spin_unlock(&inode
->i_lock
);
4391 ei
->jinode
= jinode
;
4392 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4395 spin_unlock(&inode
->i_lock
);
4396 if (unlikely(jinode
!= NULL
))
4397 jbd2_free_inode(jinode
);
4404 * We block out ext4_get_block() block instantiations across the entire
4405 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4406 * simultaneously on behalf of the same inode.
4408 * As we work through the truncate and commit bits of it to the journal there
4409 * is one core, guiding principle: the file's tree must always be consistent on
4410 * disk. We must be able to restart the truncate after a crash.
4412 * The file's tree may be transiently inconsistent in memory (although it
4413 * probably isn't), but whenever we close off and commit a journal transaction,
4414 * the contents of (the filesystem + the journal) must be consistent and
4415 * restartable. It's pretty simple, really: bottom up, right to left (although
4416 * left-to-right works OK too).
4418 * Note that at recovery time, journal replay occurs *before* the restart of
4419 * truncate against the orphan inode list.
4421 * The committed inode has the new, desired i_size (which is the same as
4422 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4423 * that this inode's truncate did not complete and it will again call
4424 * ext4_truncate() to have another go. So there will be instantiated blocks
4425 * to the right of the truncation point in a crashed ext4 filesystem. But
4426 * that's fine - as long as they are linked from the inode, the post-crash
4427 * ext4_truncate() run will find them and release them.
4429 int ext4_truncate(struct inode
*inode
)
4431 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4432 unsigned int credits
;
4435 struct address_space
*mapping
= inode
->i_mapping
;
4438 * There is a possibility that we're either freeing the inode
4439 * or it's a completely new inode. In those cases we might not
4440 * have i_mutex locked because it's not necessary.
4442 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4443 WARN_ON(!inode_is_locked(inode
));
4444 trace_ext4_truncate_enter(inode
);
4446 if (!ext4_can_truncate(inode
))
4449 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4451 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4452 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4454 if (ext4_has_inline_data(inode
)) {
4457 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4464 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4465 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4466 if (ext4_inode_attach_jinode(inode
) < 0)
4470 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4471 credits
= ext4_writepage_trans_blocks(inode
);
4473 credits
= ext4_blocks_for_truncate(inode
);
4475 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4477 return PTR_ERR(handle
);
4479 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4480 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4483 * We add the inode to the orphan list, so that if this
4484 * truncate spans multiple transactions, and we crash, we will
4485 * resume the truncate when the filesystem recovers. It also
4486 * marks the inode dirty, to catch the new size.
4488 * Implication: the file must always be in a sane, consistent
4489 * truncatable state while each transaction commits.
4491 err
= ext4_orphan_add(handle
, inode
);
4495 down_write(&EXT4_I(inode
)->i_data_sem
);
4497 ext4_discard_preallocations(inode
);
4499 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4500 err
= ext4_ext_truncate(handle
, inode
);
4502 ext4_ind_truncate(handle
, inode
);
4504 up_write(&ei
->i_data_sem
);
4509 ext4_handle_sync(handle
);
4513 * If this was a simple ftruncate() and the file will remain alive,
4514 * then we need to clear up the orphan record which we created above.
4515 * However, if this was a real unlink then we were called by
4516 * ext4_evict_inode(), and we allow that function to clean up the
4517 * orphan info for us.
4520 ext4_orphan_del(handle
, inode
);
4522 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4523 ext4_mark_inode_dirty(handle
, inode
);
4524 ext4_journal_stop(handle
);
4526 trace_ext4_truncate_exit(inode
);
4531 * ext4_get_inode_loc returns with an extra refcount against the inode's
4532 * underlying buffer_head on success. If 'in_mem' is true, we have all
4533 * data in memory that is needed to recreate the on-disk version of this
4536 static int __ext4_get_inode_loc(struct inode
*inode
,
4537 struct ext4_iloc
*iloc
, int in_mem
)
4539 struct ext4_group_desc
*gdp
;
4540 struct buffer_head
*bh
;
4541 struct super_block
*sb
= inode
->i_sb
;
4543 int inodes_per_block
, inode_offset
;
4546 if (inode
->i_ino
< EXT4_ROOT_INO
||
4547 inode
->i_ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
4548 return -EFSCORRUPTED
;
4550 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4551 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4556 * Figure out the offset within the block group inode table
4558 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4559 inode_offset
= ((inode
->i_ino
- 1) %
4560 EXT4_INODES_PER_GROUP(sb
));
4561 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4562 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4564 bh
= sb_getblk(sb
, block
);
4567 if (!buffer_uptodate(bh
)) {
4571 * If the buffer has the write error flag, we have failed
4572 * to write out another inode in the same block. In this
4573 * case, we don't have to read the block because we may
4574 * read the old inode data successfully.
4576 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4577 set_buffer_uptodate(bh
);
4579 if (buffer_uptodate(bh
)) {
4580 /* someone brought it uptodate while we waited */
4586 * If we have all information of the inode in memory and this
4587 * is the only valid inode in the block, we need not read the
4591 struct buffer_head
*bitmap_bh
;
4594 start
= inode_offset
& ~(inodes_per_block
- 1);
4596 /* Is the inode bitmap in cache? */
4597 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4598 if (unlikely(!bitmap_bh
))
4602 * If the inode bitmap isn't in cache then the
4603 * optimisation may end up performing two reads instead
4604 * of one, so skip it.
4606 if (!buffer_uptodate(bitmap_bh
)) {
4610 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4611 if (i
== inode_offset
)
4613 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4617 if (i
== start
+ inodes_per_block
) {
4618 /* all other inodes are free, so skip I/O */
4619 memset(bh
->b_data
, 0, bh
->b_size
);
4620 set_buffer_uptodate(bh
);
4628 * If we need to do any I/O, try to pre-readahead extra
4629 * blocks from the inode table.
4631 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4632 ext4_fsblk_t b
, end
, table
;
4634 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4636 table
= ext4_inode_table(sb
, gdp
);
4637 /* s_inode_readahead_blks is always a power of 2 */
4638 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4642 num
= EXT4_INODES_PER_GROUP(sb
);
4643 if (ext4_has_group_desc_csum(sb
))
4644 num
-= ext4_itable_unused_count(sb
, gdp
);
4645 table
+= num
/ inodes_per_block
;
4649 sb_breadahead(sb
, b
++);
4653 * There are other valid inodes in the buffer, this inode
4654 * has in-inode xattrs, or we don't have this inode in memory.
4655 * Read the block from disk.
4657 trace_ext4_load_inode(inode
);
4659 bh
->b_end_io
= end_buffer_read_sync
;
4660 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4662 if (!buffer_uptodate(bh
)) {
4663 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4664 "unable to read itable block");
4674 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4676 /* We have all inode data except xattrs in memory here. */
4677 return __ext4_get_inode_loc(inode
, iloc
,
4678 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4681 static bool ext4_should_use_dax(struct inode
*inode
)
4683 if (!test_opt(inode
->i_sb
, DAX
))
4685 if (!S_ISREG(inode
->i_mode
))
4687 if (ext4_should_journal_data(inode
))
4689 if (ext4_has_inline_data(inode
))
4691 if (ext4_encrypted_inode(inode
))
4696 void ext4_set_inode_flags(struct inode
*inode
)
4698 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4699 unsigned int new_fl
= 0;
4701 if (flags
& EXT4_SYNC_FL
)
4703 if (flags
& EXT4_APPEND_FL
)
4705 if (flags
& EXT4_IMMUTABLE_FL
)
4706 new_fl
|= S_IMMUTABLE
;
4707 if (flags
& EXT4_NOATIME_FL
)
4708 new_fl
|= S_NOATIME
;
4709 if (flags
& EXT4_DIRSYNC_FL
)
4710 new_fl
|= S_DIRSYNC
;
4711 if (ext4_should_use_dax(inode
))
4713 if (flags
& EXT4_ENCRYPT_FL
)
4714 new_fl
|= S_ENCRYPTED
;
4715 inode_set_flags(inode
, new_fl
,
4716 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
|
4720 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4721 struct ext4_inode_info
*ei
)
4724 struct inode
*inode
= &(ei
->vfs_inode
);
4725 struct super_block
*sb
= inode
->i_sb
;
4727 if (ext4_has_feature_huge_file(sb
)) {
4728 /* we are using combined 48 bit field */
4729 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4730 le32_to_cpu(raw_inode
->i_blocks_lo
);
4731 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4732 /* i_blocks represent file system block size */
4733 return i_blocks
<< (inode
->i_blkbits
- 9);
4738 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4742 static inline int ext4_iget_extra_inode(struct inode
*inode
,
4743 struct ext4_inode
*raw_inode
,
4744 struct ext4_inode_info
*ei
)
4746 __le32
*magic
= (void *)raw_inode
+
4747 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4749 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4750 EXT4_INODE_SIZE(inode
->i_sb
) &&
4751 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4752 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4753 return ext4_find_inline_data_nolock(inode
);
4755 EXT4_I(inode
)->i_inline_off
= 0;
4759 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4761 if (!ext4_has_feature_project(inode
->i_sb
))
4763 *projid
= EXT4_I(inode
)->i_projid
;
4768 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4769 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4772 static inline void ext4_inode_set_iversion_queried(struct inode
*inode
, u64 val
)
4774 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4775 inode_set_iversion_raw(inode
, val
);
4777 inode_set_iversion_queried(inode
, val
);
4779 static inline u64
ext4_inode_peek_iversion(const struct inode
*inode
)
4781 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4782 return inode_peek_iversion_raw(inode
);
4784 return inode_peek_iversion(inode
);
4787 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4789 struct ext4_iloc iloc
;
4790 struct ext4_inode
*raw_inode
;
4791 struct ext4_inode_info
*ei
;
4792 struct inode
*inode
;
4793 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4801 inode
= iget_locked(sb
, ino
);
4803 return ERR_PTR(-ENOMEM
);
4804 if (!(inode
->i_state
& I_NEW
))
4810 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4813 raw_inode
= ext4_raw_inode(&iloc
);
4815 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4816 EXT4_ERROR_INODE(inode
, "root inode unallocated");
4817 ret
= -EFSCORRUPTED
;
4821 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4822 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4823 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4824 EXT4_INODE_SIZE(inode
->i_sb
) ||
4825 (ei
->i_extra_isize
& 3)) {
4826 EXT4_ERROR_INODE(inode
,
4827 "bad extra_isize %u (inode size %u)",
4829 EXT4_INODE_SIZE(inode
->i_sb
));
4830 ret
= -EFSCORRUPTED
;
4834 ei
->i_extra_isize
= 0;
4836 /* Precompute checksum seed for inode metadata */
4837 if (ext4_has_metadata_csum(sb
)) {
4838 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4840 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4841 __le32 gen
= raw_inode
->i_generation
;
4842 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4844 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4848 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4849 EXT4_ERROR_INODE(inode
, "checksum invalid");
4854 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4855 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4856 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4857 if (ext4_has_feature_project(sb
) &&
4858 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4859 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4860 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4862 i_projid
= EXT4_DEF_PROJID
;
4864 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4865 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4866 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4868 i_uid_write(inode
, i_uid
);
4869 i_gid_write(inode
, i_gid
);
4870 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4871 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4873 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4874 ei
->i_inline_off
= 0;
4875 ei
->i_dir_start_lookup
= 0;
4876 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4877 /* We now have enough fields to check if the inode was active or not.
4878 * This is needed because nfsd might try to access dead inodes
4879 * the test is that same one that e2fsck uses
4880 * NeilBrown 1999oct15
4882 if (inode
->i_nlink
== 0) {
4883 if ((inode
->i_mode
== 0 ||
4884 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4885 ino
!= EXT4_BOOT_LOADER_INO
) {
4886 /* this inode is deleted */
4890 /* The only unlinked inodes we let through here have
4891 * valid i_mode and are being read by the orphan
4892 * recovery code: that's fine, we're about to complete
4893 * the process of deleting those.
4894 * OR it is the EXT4_BOOT_LOADER_INO which is
4895 * not initialized on a new filesystem. */
4897 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4898 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4899 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4900 if (ext4_has_feature_64bit(sb
))
4902 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4903 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4904 if ((size
= i_size_read(inode
)) < 0) {
4905 EXT4_ERROR_INODE(inode
, "bad i_size value: %lld", size
);
4906 ret
= -EFSCORRUPTED
;
4909 ei
->i_disksize
= inode
->i_size
;
4911 ei
->i_reserved_quota
= 0;
4913 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4914 ei
->i_block_group
= iloc
.block_group
;
4915 ei
->i_last_alloc_group
= ~0;
4917 * NOTE! The in-memory inode i_data array is in little-endian order
4918 * even on big-endian machines: we do NOT byteswap the block numbers!
4920 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4921 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4922 INIT_LIST_HEAD(&ei
->i_orphan
);
4925 * Set transaction id's of transactions that have to be committed
4926 * to finish f[data]sync. We set them to currently running transaction
4927 * as we cannot be sure that the inode or some of its metadata isn't
4928 * part of the transaction - the inode could have been reclaimed and
4929 * now it is reread from disk.
4932 transaction_t
*transaction
;
4935 read_lock(&journal
->j_state_lock
);
4936 if (journal
->j_running_transaction
)
4937 transaction
= journal
->j_running_transaction
;
4939 transaction
= journal
->j_committing_transaction
;
4941 tid
= transaction
->t_tid
;
4943 tid
= journal
->j_commit_sequence
;
4944 read_unlock(&journal
->j_state_lock
);
4945 ei
->i_sync_tid
= tid
;
4946 ei
->i_datasync_tid
= tid
;
4949 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4950 if (ei
->i_extra_isize
== 0) {
4951 /* The extra space is currently unused. Use it. */
4952 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4953 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4954 EXT4_GOOD_OLD_INODE_SIZE
;
4956 ret
= ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4962 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4963 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4964 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4965 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4967 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4968 u64 ivers
= le32_to_cpu(raw_inode
->i_disk_version
);
4970 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4971 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4973 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4975 ext4_inode_set_iversion_queried(inode
, ivers
);
4979 if (ei
->i_file_acl
&&
4980 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4981 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4983 ret
= -EFSCORRUPTED
;
4985 } else if (!ext4_has_inline_data(inode
)) {
4986 /* validate the block references in the inode */
4987 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4988 (S_ISLNK(inode
->i_mode
) &&
4989 !ext4_inode_is_fast_symlink(inode
))) {
4990 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4991 ret
= ext4_ext_check_inode(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
,
5010 "immutable or append flags not allowed on symlinks");
5011 ret
= -EFSCORRUPTED
;
5014 if (ext4_encrypted_inode(inode
)) {
5015 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
5016 ext4_set_aops(inode
);
5017 } else if (ext4_inode_is_fast_symlink(inode
)) {
5018 inode
->i_link
= (char *)ei
->i_data
;
5019 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5020 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5021 sizeof(ei
->i_data
) - 1);
5023 inode
->i_op
= &ext4_symlink_inode_operations
;
5024 ext4_set_aops(inode
);
5026 inode_nohighmem(inode
);
5027 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5028 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5029 inode
->i_op
= &ext4_special_inode_operations
;
5030 if (raw_inode
->i_block
[0])
5031 init_special_inode(inode
, inode
->i_mode
,
5032 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5034 init_special_inode(inode
, inode
->i_mode
,
5035 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5036 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
5037 make_bad_inode(inode
);
5039 ret
= -EFSCORRUPTED
;
5040 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
5044 ext4_set_inode_flags(inode
);
5046 unlock_new_inode(inode
);
5052 return ERR_PTR(ret
);
5055 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
5057 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
5058 return ERR_PTR(-EFSCORRUPTED
);
5059 return ext4_iget(sb
, ino
);
5062 static int ext4_inode_blocks_set(handle_t
*handle
,
5063 struct ext4_inode
*raw_inode
,
5064 struct ext4_inode_info
*ei
)
5066 struct inode
*inode
= &(ei
->vfs_inode
);
5067 u64 i_blocks
= inode
->i_blocks
;
5068 struct super_block
*sb
= inode
->i_sb
;
5070 if (i_blocks
<= ~0U) {
5072 * i_blocks can be represented in a 32 bit variable
5073 * as multiple of 512 bytes
5075 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5076 raw_inode
->i_blocks_high
= 0;
5077 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5080 if (!ext4_has_feature_huge_file(sb
))
5083 if (i_blocks
<= 0xffffffffffffULL
) {
5085 * i_blocks can be represented in a 48 bit variable
5086 * as multiple of 512 bytes
5088 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5089 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5090 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5092 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5093 /* i_block is stored in file system block size */
5094 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5095 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5096 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5101 struct other_inode
{
5102 unsigned long orig_ino
;
5103 struct ext4_inode
*raw_inode
;
5106 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
5109 struct other_inode
*oi
= (struct other_inode
*) data
;
5111 if ((inode
->i_ino
!= ino
) ||
5112 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
5114 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
5116 spin_lock(&inode
->i_lock
);
5117 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
5118 I_DIRTY_INODE
)) == 0) &&
5119 (inode
->i_state
& I_DIRTY_TIME
)) {
5120 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5122 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
5123 spin_unlock(&inode
->i_lock
);
5125 spin_lock(&ei
->i_raw_lock
);
5126 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
5127 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
5128 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
5129 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
5130 spin_unlock(&ei
->i_raw_lock
);
5131 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
5134 spin_unlock(&inode
->i_lock
);
5139 * Opportunistically update the other time fields for other inodes in
5140 * the same inode table block.
5142 static void ext4_update_other_inodes_time(struct super_block
*sb
,
5143 unsigned long orig_ino
, char *buf
)
5145 struct other_inode oi
;
5147 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
5148 int inode_size
= EXT4_INODE_SIZE(sb
);
5150 oi
.orig_ino
= orig_ino
;
5152 * Calculate the first inode in the inode table block. Inode
5153 * numbers are one-based. That is, the first inode in a block
5154 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5156 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
5157 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
5158 if (ino
== orig_ino
)
5160 oi
.raw_inode
= (struct ext4_inode
*) buf
;
5161 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
5166 * Post the struct inode info into an on-disk inode location in the
5167 * buffer-cache. This gobbles the caller's reference to the
5168 * buffer_head in the inode location struct.
5170 * The caller must have write access to iloc->bh.
5172 static int ext4_do_update_inode(handle_t
*handle
,
5173 struct inode
*inode
,
5174 struct ext4_iloc
*iloc
)
5176 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5177 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5178 struct buffer_head
*bh
= iloc
->bh
;
5179 struct super_block
*sb
= inode
->i_sb
;
5180 int err
= 0, rc
, block
;
5181 int need_datasync
= 0, set_large_file
= 0;
5186 spin_lock(&ei
->i_raw_lock
);
5188 /* For fields not tracked in the in-memory inode,
5189 * initialise them to zero for new inodes. */
5190 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5191 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5193 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5194 i_uid
= i_uid_read(inode
);
5195 i_gid
= i_gid_read(inode
);
5196 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
5197 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5198 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
5199 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
5201 * Fix up interoperability with old kernels. Otherwise, old inodes get
5202 * re-used with the upper 16 bits of the uid/gid intact
5204 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
5205 raw_inode
->i_uid_high
= 0;
5206 raw_inode
->i_gid_high
= 0;
5208 raw_inode
->i_uid_high
=
5209 cpu_to_le16(high_16_bits(i_uid
));
5210 raw_inode
->i_gid_high
=
5211 cpu_to_le16(high_16_bits(i_gid
));
5214 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
5215 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
5216 raw_inode
->i_uid_high
= 0;
5217 raw_inode
->i_gid_high
= 0;
5219 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5221 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5222 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5223 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5224 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5226 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5228 spin_unlock(&ei
->i_raw_lock
);
5231 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5232 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5233 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5234 raw_inode
->i_file_acl_high
=
5235 cpu_to_le16(ei
->i_file_acl
>> 32);
5236 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5237 if (ei
->i_disksize
!= ext4_isize(inode
->i_sb
, raw_inode
)) {
5238 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5241 if (ei
->i_disksize
> 0x7fffffffULL
) {
5242 if (!ext4_has_feature_large_file(sb
) ||
5243 EXT4_SB(sb
)->s_es
->s_rev_level
==
5244 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5247 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5248 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5249 if (old_valid_dev(inode
->i_rdev
)) {
5250 raw_inode
->i_block
[0] =
5251 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5252 raw_inode
->i_block
[1] = 0;
5254 raw_inode
->i_block
[0] = 0;
5255 raw_inode
->i_block
[1] =
5256 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5257 raw_inode
->i_block
[2] = 0;
5259 } else if (!ext4_has_inline_data(inode
)) {
5260 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5261 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5264 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5265 u64 ivers
= ext4_inode_peek_iversion(inode
);
5267 raw_inode
->i_disk_version
= cpu_to_le32(ivers
);
5268 if (ei
->i_extra_isize
) {
5269 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5270 raw_inode
->i_version_hi
=
5271 cpu_to_le32(ivers
>> 32);
5272 raw_inode
->i_extra_isize
=
5273 cpu_to_le16(ei
->i_extra_isize
);
5277 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5278 i_projid
!= EXT4_DEF_PROJID
);
5280 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5281 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5282 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5284 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5285 spin_unlock(&ei
->i_raw_lock
);
5286 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
)
5287 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5290 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5291 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5294 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5295 if (set_large_file
) {
5296 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5297 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5300 ext4_update_dynamic_rev(sb
);
5301 ext4_set_feature_large_file(sb
);
5302 ext4_handle_sync(handle
);
5303 err
= ext4_handle_dirty_super(handle
, sb
);
5305 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5308 ext4_std_error(inode
->i_sb
, err
);
5313 * ext4_write_inode()
5315 * We are called from a few places:
5317 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5318 * Here, there will be no transaction running. We wait for any running
5319 * transaction to commit.
5321 * - Within flush work (sys_sync(), kupdate and such).
5322 * We wait on commit, if told to.
5324 * - Within iput_final() -> write_inode_now()
5325 * We wait on commit, if told to.
5327 * In all cases it is actually safe for us to return without doing anything,
5328 * because the inode has been copied into a raw inode buffer in
5329 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5332 * Note that we are absolutely dependent upon all inode dirtiers doing the
5333 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5334 * which we are interested.
5336 * It would be a bug for them to not do this. The code:
5338 * mark_inode_dirty(inode)
5340 * inode->i_size = expr;
5342 * is in error because write_inode() could occur while `stuff()' is running,
5343 * and the new i_size will be lost. Plus the inode will no longer be on the
5344 * superblock's dirty inode list.
5346 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5350 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
5353 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5354 if (ext4_journal_current_handle()) {
5355 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5361 * No need to force transaction in WB_SYNC_NONE mode. Also
5362 * ext4_sync_fs() will force the commit after everything is
5365 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5368 err
= ext4_force_commit(inode
->i_sb
);
5370 struct ext4_iloc iloc
;
5372 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5376 * sync(2) will flush the whole buffer cache. No need to do
5377 * it here separately for each inode.
5379 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5380 sync_dirty_buffer(iloc
.bh
);
5381 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5382 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5383 "IO error syncing inode");
5392 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5393 * buffers that are attached to a page stradding i_size and are undergoing
5394 * commit. In that case we have to wait for commit to finish and try again.
5396 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5400 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5401 tid_t commit_tid
= 0;
5404 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5406 * All buffers in the last page remain valid? Then there's nothing to
5407 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5410 if (offset
> PAGE_SIZE
- i_blocksize(inode
))
5413 page
= find_lock_page(inode
->i_mapping
,
5414 inode
->i_size
>> PAGE_SHIFT
);
5417 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5418 PAGE_SIZE
- offset
);
5424 read_lock(&journal
->j_state_lock
);
5425 if (journal
->j_committing_transaction
)
5426 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5427 read_unlock(&journal
->j_state_lock
);
5429 jbd2_log_wait_commit(journal
, commit_tid
);
5436 * Called from notify_change.
5438 * We want to trap VFS attempts to truncate the file as soon as
5439 * possible. In particular, we want to make sure that when the VFS
5440 * shrinks i_size, we put the inode on the orphan list and modify
5441 * i_disksize immediately, so that during the subsequent flushing of
5442 * dirty pages and freeing of disk blocks, we can guarantee that any
5443 * commit will leave the blocks being flushed in an unused state on
5444 * disk. (On recovery, the inode will get truncated and the blocks will
5445 * be freed, so we have a strong guarantee that no future commit will
5446 * leave these blocks visible to the user.)
5448 * Another thing we have to assure is that if we are in ordered mode
5449 * and inode is still attached to the committing transaction, we must
5450 * we start writeout of all the dirty pages which are being truncated.
5451 * This way we are sure that all the data written in the previous
5452 * transaction are already on disk (truncate waits for pages under
5455 * Called with inode->i_mutex down.
5457 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5459 struct inode
*inode
= d_inode(dentry
);
5462 const unsigned int ia_valid
= attr
->ia_valid
;
5464 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5467 error
= setattr_prepare(dentry
, attr
);
5471 error
= fscrypt_prepare_setattr(dentry
, attr
);
5475 if (is_quota_modification(inode
, attr
)) {
5476 error
= dquot_initialize(inode
);
5480 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5481 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5484 /* (user+group)*(old+new) structure, inode write (sb,
5485 * inode block, ? - but truncate inode update has it) */
5486 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5487 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5488 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5489 if (IS_ERR(handle
)) {
5490 error
= PTR_ERR(handle
);
5494 /* dquot_transfer() calls back ext4_get_inode_usage() which
5495 * counts xattr inode references.
5497 down_read(&EXT4_I(inode
)->xattr_sem
);
5498 error
= dquot_transfer(inode
, attr
);
5499 up_read(&EXT4_I(inode
)->xattr_sem
);
5502 ext4_journal_stop(handle
);
5505 /* Update corresponding info in inode so that everything is in
5506 * one transaction */
5507 if (attr
->ia_valid
& ATTR_UID
)
5508 inode
->i_uid
= attr
->ia_uid
;
5509 if (attr
->ia_valid
& ATTR_GID
)
5510 inode
->i_gid
= attr
->ia_gid
;
5511 error
= ext4_mark_inode_dirty(handle
, inode
);
5512 ext4_journal_stop(handle
);
5515 if (attr
->ia_valid
& ATTR_SIZE
) {
5517 loff_t oldsize
= inode
->i_size
;
5518 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5520 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5521 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5523 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5526 if (!S_ISREG(inode
->i_mode
))
5529 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5530 inode_inc_iversion(inode
);
5532 if (ext4_should_order_data(inode
) &&
5533 (attr
->ia_size
< inode
->i_size
)) {
5534 error
= ext4_begin_ordered_truncate(inode
,
5539 if (attr
->ia_size
!= inode
->i_size
) {
5540 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5541 if (IS_ERR(handle
)) {
5542 error
= PTR_ERR(handle
);
5545 if (ext4_handle_valid(handle
) && shrink
) {
5546 error
= ext4_orphan_add(handle
, inode
);
5550 * Update c/mtime on truncate up, ext4_truncate() will
5551 * update c/mtime in shrink case below
5554 inode
->i_mtime
= current_time(inode
);
5555 inode
->i_ctime
= inode
->i_mtime
;
5557 down_write(&EXT4_I(inode
)->i_data_sem
);
5558 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5559 rc
= ext4_mark_inode_dirty(handle
, inode
);
5563 * We have to update i_size under i_data_sem together
5564 * with i_disksize to avoid races with writeback code
5565 * running ext4_wb_update_i_disksize().
5568 i_size_write(inode
, attr
->ia_size
);
5569 up_write(&EXT4_I(inode
)->i_data_sem
);
5570 ext4_journal_stop(handle
);
5573 ext4_orphan_del(NULL
, inode
);
5578 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5581 * Blocks are going to be removed from the inode. Wait
5582 * for dio in flight. Temporarily disable
5583 * dioread_nolock to prevent livelock.
5586 if (!ext4_should_journal_data(inode
)) {
5587 inode_dio_wait(inode
);
5589 ext4_wait_for_tail_page_commit(inode
);
5591 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5593 rc
= ext4_break_layouts(inode
);
5595 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5601 * Truncate pagecache after we've waited for commit
5602 * in data=journal mode to make pages freeable.
5604 truncate_pagecache(inode
, inode
->i_size
);
5606 rc
= ext4_truncate(inode
);
5610 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5614 setattr_copy(inode
, attr
);
5615 mark_inode_dirty(inode
);
5619 * If the call to ext4_truncate failed to get a transaction handle at
5620 * all, we need to clean up the in-core orphan list manually.
5622 if (orphan
&& inode
->i_nlink
)
5623 ext4_orphan_del(NULL
, inode
);
5625 if (!error
&& (ia_valid
& ATTR_MODE
))
5626 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5629 ext4_std_error(inode
->i_sb
, error
);
5635 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5636 u32 request_mask
, unsigned int query_flags
)
5638 struct inode
*inode
= d_inode(path
->dentry
);
5639 struct ext4_inode
*raw_inode
;
5640 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5643 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5644 stat
->result_mask
|= STATX_BTIME
;
5645 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5646 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5649 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5650 if (flags
& EXT4_APPEND_FL
)
5651 stat
->attributes
|= STATX_ATTR_APPEND
;
5652 if (flags
& EXT4_COMPR_FL
)
5653 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5654 if (flags
& EXT4_ENCRYPT_FL
)
5655 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5656 if (flags
& EXT4_IMMUTABLE_FL
)
5657 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5658 if (flags
& EXT4_NODUMP_FL
)
5659 stat
->attributes
|= STATX_ATTR_NODUMP
;
5661 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5662 STATX_ATTR_COMPRESSED
|
5663 STATX_ATTR_ENCRYPTED
|
5664 STATX_ATTR_IMMUTABLE
|
5667 generic_fillattr(inode
, stat
);
5671 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5672 u32 request_mask
, unsigned int query_flags
)
5674 struct inode
*inode
= d_inode(path
->dentry
);
5675 u64 delalloc_blocks
;
5677 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5680 * If there is inline data in the inode, the inode will normally not
5681 * have data blocks allocated (it may have an external xattr block).
5682 * Report at least one sector for such files, so tools like tar, rsync,
5683 * others don't incorrectly think the file is completely sparse.
5685 if (unlikely(ext4_has_inline_data(inode
)))
5686 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5689 * We can't update i_blocks if the block allocation is delayed
5690 * otherwise in the case of system crash before the real block
5691 * allocation is done, we will have i_blocks inconsistent with
5692 * on-disk file blocks.
5693 * We always keep i_blocks updated together with real
5694 * allocation. But to not confuse with user, stat
5695 * will return the blocks that include the delayed allocation
5696 * blocks for this file.
5698 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5699 EXT4_I(inode
)->i_reserved_data_blocks
);
5700 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5704 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5707 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5708 return ext4_ind_trans_blocks(inode
, lblocks
);
5709 return ext4_ext_index_trans_blocks(inode
, pextents
);
5713 * Account for index blocks, block groups bitmaps and block group
5714 * descriptor blocks if modify datablocks and index blocks
5715 * worse case, the indexs blocks spread over different block groups
5717 * If datablocks are discontiguous, they are possible to spread over
5718 * different block groups too. If they are contiguous, with flexbg,
5719 * they could still across block group boundary.
5721 * Also account for superblock, inode, quota and xattr blocks
5723 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5726 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5732 * How many index blocks need to touch to map @lblocks logical blocks
5733 * to @pextents physical extents?
5735 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5740 * Now let's see how many group bitmaps and group descriptors need
5743 groups
= idxblocks
+ pextents
;
5745 if (groups
> ngroups
)
5747 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5748 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5750 /* bitmaps and block group descriptor blocks */
5751 ret
+= groups
+ gdpblocks
;
5753 /* Blocks for super block, inode, quota and xattr blocks */
5754 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5760 * Calculate the total number of credits to reserve to fit
5761 * the modification of a single pages into a single transaction,
5762 * which may include multiple chunks of block allocations.
5764 * This could be called via ext4_write_begin()
5766 * We need to consider the worse case, when
5767 * one new block per extent.
5769 int ext4_writepage_trans_blocks(struct inode
*inode
)
5771 int bpp
= ext4_journal_blocks_per_page(inode
);
5774 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5776 /* Account for data blocks for journalled mode */
5777 if (ext4_should_journal_data(inode
))
5783 * Calculate the journal credits for a chunk of data modification.
5785 * This is called from DIO, fallocate or whoever calling
5786 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5788 * journal buffers for data blocks are not included here, as DIO
5789 * and fallocate do no need to journal data buffers.
5791 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5793 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5797 * The caller must have previously called ext4_reserve_inode_write().
5798 * Give this, we know that the caller already has write access to iloc->bh.
5800 int ext4_mark_iloc_dirty(handle_t
*handle
,
5801 struct inode
*inode
, struct ext4_iloc
*iloc
)
5805 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
;
5856 raw_inode
= ext4_raw_inode(iloc
);
5858 header
= IHDR(inode
, raw_inode
);
5860 /* No extended attributes present */
5861 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5862 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5863 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5864 EXT4_I(inode
)->i_extra_isize
, 0,
5865 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5866 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5870 /* try to expand with EAs present */
5871 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5875 * Inode size expansion failed; don't try again
5884 * Expand an inode by new_extra_isize bytes.
5885 * Returns 0 on success or negative error number on failure.
5887 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5888 unsigned int new_extra_isize
,
5889 struct ext4_iloc iloc
,
5895 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5899 * In nojournal mode, we can immediately attempt to expand
5900 * the inode. When journaled, we first need to obtain extra
5901 * buffer credits since we may write into the EA block
5902 * with this same handle. If journal_extend fails, then it will
5903 * only result in a minor loss of functionality for that inode.
5904 * If this is felt to be critical, then e2fsck should be run to
5905 * force a large enough s_min_extra_isize.
5907 if (ext4_handle_valid(handle
) &&
5908 jbd2_journal_extend(handle
,
5909 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) != 0)
5912 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5915 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5916 handle
, &no_expand
);
5917 ext4_write_unlock_xattr(inode
, &no_expand
);
5922 int ext4_expand_extra_isize(struct inode
*inode
,
5923 unsigned int new_extra_isize
,
5924 struct ext4_iloc
*iloc
)
5930 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5935 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5936 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5937 if (IS_ERR(handle
)) {
5938 error
= PTR_ERR(handle
);
5943 ext4_write_lock_xattr(inode
, &no_expand
);
5945 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5946 error
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5952 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
5953 handle
, &no_expand
);
5955 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
5959 ext4_write_unlock_xattr(inode
, &no_expand
);
5961 ext4_journal_stop(handle
);
5966 * What we do here is to mark the in-core inode as clean with respect to inode
5967 * dirtiness (it may still be data-dirty).
5968 * This means that the in-core inode may be reaped by prune_icache
5969 * without having to perform any I/O. This is a very good thing,
5970 * because *any* task may call prune_icache - even ones which
5971 * have a transaction open against a different journal.
5973 * Is this cheating? Not really. Sure, we haven't written the
5974 * inode out, but prune_icache isn't a user-visible syncing function.
5975 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5976 * we start and wait on commits.
5978 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5980 struct ext4_iloc iloc
;
5981 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5985 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5986 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5990 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
5991 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
5994 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5998 * ext4_dirty_inode() is called from __mark_inode_dirty()
6000 * We're really interested in the case where a file is being extended.
6001 * i_size has been changed by generic_commit_write() and we thus need
6002 * to include the updated inode in the current transaction.
6004 * Also, dquot_alloc_block() will always dirty the inode when blocks
6005 * are allocated to the file.
6007 * If the inode is marked synchronous, we don't honour that here - doing
6008 * so would cause a commit on atime updates, which we don't bother doing.
6009 * We handle synchronous inodes at the highest possible level.
6011 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6012 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6013 * to copy into the on-disk inode structure are the timestamp files.
6015 void ext4_dirty_inode(struct inode
*inode
, int flags
)
6019 if (flags
== I_DIRTY_TIME
)
6021 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
6025 ext4_mark_inode_dirty(handle
, inode
);
6027 ext4_journal_stop(handle
);
6034 * Bind an inode's backing buffer_head into this transaction, to prevent
6035 * it from being flushed to disk early. Unlike
6036 * ext4_reserve_inode_write, this leaves behind no bh reference and
6037 * returns no iloc structure, so the caller needs to repeat the iloc
6038 * lookup to mark the inode dirty later.
6040 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
6042 struct ext4_iloc iloc
;
6046 err
= ext4_get_inode_loc(inode
, &iloc
);
6048 BUFFER_TRACE(iloc
.bh
, "get_write_access");
6049 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
6051 err
= ext4_handle_dirty_metadata(handle
,
6057 ext4_std_error(inode
->i_sb
, err
);
6062 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
6067 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
6070 * We have to be very careful here: changing a data block's
6071 * journaling status dynamically is dangerous. If we write a
6072 * data block to the journal, change the status and then delete
6073 * that block, we risk forgetting to revoke the old log record
6074 * from the journal and so a subsequent replay can corrupt data.
6075 * So, first we make sure that the journal is empty and that
6076 * nobody is changing anything.
6079 journal
= EXT4_JOURNAL(inode
);
6082 if (is_journal_aborted(journal
))
6085 /* Wait for all existing dio workers */
6086 inode_dio_wait(inode
);
6089 * Before flushing the journal and switching inode's aops, we have
6090 * to flush all dirty data the inode has. There can be outstanding
6091 * delayed allocations, there can be unwritten extents created by
6092 * fallocate or buffered writes in dioread_nolock mode covered by
6093 * dirty data which can be converted only after flushing the dirty
6094 * data (and journalled aops don't know how to handle these cases).
6097 down_write(&EXT4_I(inode
)->i_mmap_sem
);
6098 err
= filemap_write_and_wait(inode
->i_mapping
);
6100 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6105 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
6106 jbd2_journal_lock_updates(journal
);
6109 * OK, there are no updates running now, and all cached data is
6110 * synced to disk. We are now in a completely consistent state
6111 * which doesn't have anything in the journal, and we know that
6112 * no filesystem updates are running, so it is safe to modify
6113 * the inode's in-core data-journaling state flag now.
6117 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6119 err
= jbd2_journal_flush(journal
);
6121 jbd2_journal_unlock_updates(journal
);
6122 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6125 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6127 ext4_set_aops(inode
);
6129 jbd2_journal_unlock_updates(journal
);
6130 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6133 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6135 /* Finally we can mark the inode as dirty. */
6137 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
6139 return PTR_ERR(handle
);
6141 err
= ext4_mark_inode_dirty(handle
, inode
);
6142 ext4_handle_sync(handle
);
6143 ext4_journal_stop(handle
);
6144 ext4_std_error(inode
->i_sb
, err
);
6149 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
6151 return !buffer_mapped(bh
);
6154 int ext4_page_mkwrite(struct vm_fault
*vmf
)
6156 struct vm_area_struct
*vma
= vmf
->vma
;
6157 struct page
*page
= vmf
->page
;
6161 struct file
*file
= vma
->vm_file
;
6162 struct inode
*inode
= file_inode(file
);
6163 struct address_space
*mapping
= inode
->i_mapping
;
6165 get_block_t
*get_block
;
6168 sb_start_pagefault(inode
->i_sb
);
6169 file_update_time(vma
->vm_file
);
6171 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6173 ret
= ext4_convert_inline_data(inode
);
6177 /* Delalloc case is easy... */
6178 if (test_opt(inode
->i_sb
, DELALLOC
) &&
6179 !ext4_should_journal_data(inode
) &&
6180 !ext4_nonda_switch(inode
->i_sb
)) {
6182 ret
= block_page_mkwrite(vma
, vmf
,
6183 ext4_da_get_block_prep
);
6184 } while (ret
== -ENOSPC
&&
6185 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
6190 size
= i_size_read(inode
);
6191 /* Page got truncated from under us? */
6192 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6194 ret
= VM_FAULT_NOPAGE
;
6198 if (page
->index
== size
>> PAGE_SHIFT
)
6199 len
= size
& ~PAGE_MASK
;
6203 * Return if we have all the buffers mapped. This avoids the need to do
6204 * journal_start/journal_stop which can block and take a long time
6206 if (page_has_buffers(page
)) {
6207 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
6209 ext4_bh_unmapped
)) {
6210 /* Wait so that we don't change page under IO */
6211 wait_for_stable_page(page
);
6212 ret
= VM_FAULT_LOCKED
;
6217 /* OK, we need to fill the hole... */
6218 if (ext4_should_dioread_nolock(inode
))
6219 get_block
= ext4_get_block_unwritten
;
6221 get_block
= ext4_get_block
;
6223 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6224 ext4_writepage_trans_blocks(inode
));
6225 if (IS_ERR(handle
)) {
6226 ret
= VM_FAULT_SIGBUS
;
6229 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
6230 if (!ret
&& ext4_should_journal_data(inode
)) {
6231 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
6232 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
6234 ret
= VM_FAULT_SIGBUS
;
6235 ext4_journal_stop(handle
);
6238 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6240 ext4_journal_stop(handle
);
6241 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6244 ret
= block_page_mkwrite_return(ret
);
6246 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6247 sb_end_pagefault(inode
->i_sb
);
6251 int ext4_filemap_fault(struct vm_fault
*vmf
)
6253 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
6256 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6257 err
= filemap_fault(vmf
);
6258 up_read(&EXT4_I(inode
)->i_mmap_sem
);