2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
40 #include <linux/iomap.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
52 struct ext4_inode_info
*ei
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
57 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
58 unsigned int csum_size
= sizeof(dummy_csum
);
60 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
61 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
63 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
64 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
66 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
67 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
68 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
69 EXT4_GOOD_OLD_INODE_SIZE
,
70 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
71 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
72 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
76 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
77 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
83 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
84 struct ext4_inode_info
*ei
)
86 __u32 provided
, calculated
;
88 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
89 cpu_to_le32(EXT4_OS_LINUX
) ||
90 !ext4_has_metadata_csum(inode
->i_sb
))
93 provided
= le16_to_cpu(raw
->i_checksum_lo
);
94 calculated
= ext4_inode_csum(inode
, raw
, ei
);
95 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
96 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
97 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
101 return provided
== calculated
;
104 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
105 struct ext4_inode_info
*ei
)
109 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
110 cpu_to_le32(EXT4_OS_LINUX
) ||
111 !ext4_has_metadata_csum(inode
->i_sb
))
114 csum
= ext4_inode_csum(inode
, raw
, ei
);
115 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
116 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
117 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
118 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
124 trace_ext4_begin_ordered_truncate(inode
, new_size
);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode
)->jinode
)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
134 EXT4_I(inode
)->jinode
,
138 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
139 unsigned int length
);
140 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
141 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
142 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
146 * Test whether an inode is a fast symlink.
148 int ext4_inode_is_fast_symlink(struct inode
*inode
)
150 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
151 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
153 if (ext4_has_inline_data(inode
))
156 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
160 * Restart the transaction associated with *handle. This does a commit,
161 * so before we call here everything must be consistently dirtied against
164 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
170 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
171 * moment, get_block can be called only for blocks inside i_size since
172 * page cache has been already dropped and writes are blocked by
173 * i_mutex. So we can safely drop the i_data_sem here.
175 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
176 jbd_debug(2, "restarting handle %p\n", handle
);
177 up_write(&EXT4_I(inode
)->i_data_sem
);
178 ret
= ext4_journal_restart(handle
, nblocks
);
179 down_write(&EXT4_I(inode
)->i_data_sem
);
180 ext4_discard_preallocations(inode
);
186 * Called at the last iput() if i_nlink is zero.
188 void ext4_evict_inode(struct inode
*inode
)
192 int extra_credits
= 3;
193 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
195 trace_ext4_evict_inode(inode
);
197 if (inode
->i_nlink
) {
199 * When journalling data dirty buffers are tracked only in the
200 * journal. So although mm thinks everything is clean and
201 * ready for reaping the inode might still have some pages to
202 * write in the running transaction or waiting to be
203 * checkpointed. Thus calling jbd2_journal_invalidatepage()
204 * (via truncate_inode_pages()) to discard these buffers can
205 * cause data loss. Also even if we did not discard these
206 * buffers, we would have no way to find them after the inode
207 * is reaped and thus user could see stale data if he tries to
208 * read them before the transaction is checkpointed. So be
209 * careful and force everything to disk here... We use
210 * ei->i_datasync_tid to store the newest transaction
211 * containing inode's data.
213 * Note that directories do not have this problem because they
214 * don't use page cache.
216 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
217 ext4_should_journal_data(inode
) &&
218 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
219 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
220 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
222 jbd2_complete_transaction(journal
, commit_tid
);
223 filemap_write_and_wait(&inode
->i_data
);
225 truncate_inode_pages_final(&inode
->i_data
);
230 if (is_bad_inode(inode
))
232 dquot_initialize(inode
);
234 if (ext4_should_order_data(inode
))
235 ext4_begin_ordered_truncate(inode
, 0);
236 truncate_inode_pages_final(&inode
->i_data
);
239 * Protect us against freezing - iput() caller didn't have to have any
240 * protection against it
242 sb_start_intwrite(inode
->i_sb
);
244 if (!IS_NOQUOTA(inode
))
245 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
247 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
248 ext4_blocks_for_truncate(inode
)+extra_credits
);
249 if (IS_ERR(handle
)) {
250 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
252 * If we're going to skip the normal cleanup, we still need to
253 * make sure that the in-core orphan linked list is properly
256 ext4_orphan_del(NULL
, inode
);
257 sb_end_intwrite(inode
->i_sb
);
262 ext4_handle_sync(handle
);
264 err
= ext4_mark_inode_dirty(handle
, inode
);
266 ext4_warning(inode
->i_sb
,
267 "couldn't mark inode dirty (err %d)", err
);
270 if (inode
->i_blocks
) {
271 err
= ext4_truncate(inode
);
273 ext4_error(inode
->i_sb
,
274 "couldn't truncate inode %lu (err %d)",
280 /* Remove xattr references. */
281 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
284 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
286 ext4_journal_stop(handle
);
287 ext4_orphan_del(NULL
, inode
);
288 sb_end_intwrite(inode
->i_sb
);
289 ext4_xattr_inode_array_free(ea_inode_array
);
294 * Kill off the orphan record which ext4_truncate created.
295 * AKPM: I think this can be inside the above `if'.
296 * Note that ext4_orphan_del() has to be able to cope with the
297 * deletion of a non-existent orphan - this is because we don't
298 * know if ext4_truncate() actually created an orphan record.
299 * (Well, we could do this if we need to, but heck - it works)
301 ext4_orphan_del(handle
, inode
);
302 EXT4_I(inode
)->i_dtime
= get_seconds();
305 * One subtle ordering requirement: if anything has gone wrong
306 * (transaction abort, IO errors, whatever), then we can still
307 * do these next steps (the fs will already have been marked as
308 * having errors), but we can't free the inode if the mark_dirty
311 if (ext4_mark_inode_dirty(handle
, inode
))
312 /* If that failed, just do the required in-core inode clear. */
313 ext4_clear_inode(inode
);
315 ext4_free_inode(handle
, inode
);
316 ext4_journal_stop(handle
);
317 sb_end_intwrite(inode
->i_sb
);
318 ext4_xattr_inode_array_free(ea_inode_array
);
321 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
325 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
327 return &EXT4_I(inode
)->i_reserved_quota
;
332 * Called with i_data_sem down, which is important since we can call
333 * ext4_discard_preallocations() from here.
335 void ext4_da_update_reserve_space(struct inode
*inode
,
336 int used
, int quota_claim
)
338 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
339 struct ext4_inode_info
*ei
= EXT4_I(inode
);
341 spin_lock(&ei
->i_block_reservation_lock
);
342 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
343 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
344 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
345 "with only %d reserved data blocks",
346 __func__
, inode
->i_ino
, used
,
347 ei
->i_reserved_data_blocks
);
349 used
= ei
->i_reserved_data_blocks
;
352 /* Update per-inode reservations */
353 ei
->i_reserved_data_blocks
-= used
;
354 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
356 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
358 /* Update quota subsystem for data blocks */
360 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
363 * We did fallocate with an offset that is already delayed
364 * allocated. So on delayed allocated writeback we should
365 * not re-claim the quota for fallocated blocks.
367 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
371 * If we have done all the pending block allocations and if
372 * there aren't any writers on the inode, we can discard the
373 * inode's preallocations.
375 if ((ei
->i_reserved_data_blocks
== 0) &&
376 (atomic_read(&inode
->i_writecount
) == 0))
377 ext4_discard_preallocations(inode
);
380 static int __check_block_validity(struct inode
*inode
, const char *func
,
382 struct ext4_map_blocks
*map
)
384 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
386 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
387 "lblock %lu mapped to illegal pblock "
388 "(length %d)", (unsigned long) map
->m_lblk
,
390 return -EFSCORRUPTED
;
395 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
400 if (ext4_encrypted_inode(inode
))
401 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
403 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
410 #define check_block_validity(inode, map) \
411 __check_block_validity((inode), __func__, __LINE__, (map))
413 #ifdef ES_AGGRESSIVE_TEST
414 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
416 struct ext4_map_blocks
*es_map
,
417 struct ext4_map_blocks
*map
,
424 * There is a race window that the result is not the same.
425 * e.g. xfstests #223 when dioread_nolock enables. The reason
426 * is that we lookup a block mapping in extent status tree with
427 * out taking i_data_sem. So at the time the unwritten extent
428 * could be converted.
430 down_read(&EXT4_I(inode
)->i_data_sem
);
431 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
432 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
433 EXT4_GET_BLOCKS_KEEP_SIZE
);
435 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
436 EXT4_GET_BLOCKS_KEEP_SIZE
);
438 up_read((&EXT4_I(inode
)->i_data_sem
));
441 * We don't check m_len because extent will be collpased in status
442 * tree. So the m_len might not equal.
444 if (es_map
->m_lblk
!= map
->m_lblk
||
445 es_map
->m_flags
!= map
->m_flags
||
446 es_map
->m_pblk
!= map
->m_pblk
) {
447 printk("ES cache assertion failed for inode: %lu "
448 "es_cached ex [%d/%d/%llu/%x] != "
449 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
450 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
451 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
452 map
->m_len
, map
->m_pblk
, map
->m_flags
,
456 #endif /* ES_AGGRESSIVE_TEST */
459 * The ext4_map_blocks() function tries to look up the requested blocks,
460 * and returns if the blocks are already mapped.
462 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
463 * and store the allocated blocks in the result buffer head and mark it
466 * If file type is extents based, it will call ext4_ext_map_blocks(),
467 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
470 * On success, it returns the number of blocks being mapped or allocated. if
471 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
472 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
474 * It returns 0 if plain look up failed (blocks have not been allocated), in
475 * that case, @map is returned as unmapped but we still do fill map->m_len to
476 * indicate the length of a hole starting at map->m_lblk.
478 * It returns the error in case of allocation failure.
480 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
481 struct ext4_map_blocks
*map
, int flags
)
483 struct extent_status es
;
486 #ifdef ES_AGGRESSIVE_TEST
487 struct ext4_map_blocks orig_map
;
489 memcpy(&orig_map
, map
, sizeof(*map
));
493 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
494 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
495 (unsigned long) map
->m_lblk
);
498 * ext4_map_blocks returns an int, and m_len is an unsigned int
500 if (unlikely(map
->m_len
> INT_MAX
))
501 map
->m_len
= INT_MAX
;
503 /* We can handle the block number less than EXT_MAX_BLOCKS */
504 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
505 return -EFSCORRUPTED
;
507 /* Lookup extent status tree firstly */
508 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
509 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
510 map
->m_pblk
= ext4_es_pblock(&es
) +
511 map
->m_lblk
- es
.es_lblk
;
512 map
->m_flags
|= ext4_es_is_written(&es
) ?
513 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
514 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
515 if (retval
> map
->m_len
)
518 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
520 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
521 if (retval
> map
->m_len
)
528 #ifdef ES_AGGRESSIVE_TEST
529 ext4_map_blocks_es_recheck(handle
, inode
, map
,
536 * Try to see if we can get the block without requesting a new
539 down_read(&EXT4_I(inode
)->i_data_sem
);
540 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
541 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
542 EXT4_GET_BLOCKS_KEEP_SIZE
);
544 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
545 EXT4_GET_BLOCKS_KEEP_SIZE
);
550 if (unlikely(retval
!= map
->m_len
)) {
551 ext4_warning(inode
->i_sb
,
552 "ES len assertion failed for inode "
553 "%lu: retval %d != map->m_len %d",
554 inode
->i_ino
, retval
, map
->m_len
);
558 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
559 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
560 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
561 !(status
& EXTENT_STATUS_WRITTEN
) &&
562 ext4_find_delalloc_range(inode
, map
->m_lblk
,
563 map
->m_lblk
+ map
->m_len
- 1))
564 status
|= EXTENT_STATUS_DELAYED
;
565 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
566 map
->m_len
, map
->m_pblk
, status
);
570 up_read((&EXT4_I(inode
)->i_data_sem
));
573 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
574 ret
= check_block_validity(inode
, map
);
579 /* If it is only a block(s) look up */
580 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
584 * Returns if the blocks have already allocated
586 * Note that if blocks have been preallocated
587 * ext4_ext_get_block() returns the create = 0
588 * with buffer head unmapped.
590 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
592 * If we need to convert extent to unwritten
593 * we continue and do the actual work in
594 * ext4_ext_map_blocks()
596 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
600 * Here we clear m_flags because after allocating an new extent,
601 * it will be set again.
603 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
606 * New blocks allocate and/or writing to unwritten extent
607 * will possibly result in updating i_data, so we take
608 * the write lock of i_data_sem, and call get_block()
609 * with create == 1 flag.
611 down_write(&EXT4_I(inode
)->i_data_sem
);
614 * We need to check for EXT4 here because migrate
615 * could have changed the inode type in between
617 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
618 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
620 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
622 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
624 * We allocated new blocks which will result in
625 * i_data's format changing. Force the migrate
626 * to fail by clearing migrate flags
628 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
632 * Update reserved blocks/metadata blocks after successful
633 * block allocation which had been deferred till now. We don't
634 * support fallocate for non extent files. So we can update
635 * reserve space here.
638 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
639 ext4_da_update_reserve_space(inode
, retval
, 1);
645 if (unlikely(retval
!= map
->m_len
)) {
646 ext4_warning(inode
->i_sb
,
647 "ES len assertion failed for inode "
648 "%lu: retval %d != map->m_len %d",
649 inode
->i_ino
, retval
, map
->m_len
);
654 * We have to zeroout blocks before inserting them into extent
655 * status tree. Otherwise someone could look them up there and
656 * use them before they are really zeroed. We also have to
657 * unmap metadata before zeroing as otherwise writeback can
658 * overwrite zeros with stale data from block device.
660 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
661 map
->m_flags
& EXT4_MAP_MAPPED
&&
662 map
->m_flags
& EXT4_MAP_NEW
) {
663 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
665 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
666 map
->m_pblk
, map
->m_len
);
674 * If the extent has been zeroed out, we don't need to update
675 * extent status tree.
677 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
678 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
679 if (ext4_es_is_written(&es
))
682 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
683 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
684 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
685 !(status
& EXTENT_STATUS_WRITTEN
) &&
686 ext4_find_delalloc_range(inode
, map
->m_lblk
,
687 map
->m_lblk
+ map
->m_len
- 1))
688 status
|= EXTENT_STATUS_DELAYED
;
689 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
690 map
->m_pblk
, status
);
698 up_write((&EXT4_I(inode
)->i_data_sem
));
699 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
700 ret
= check_block_validity(inode
, map
);
705 * Inodes with freshly allocated blocks where contents will be
706 * visible after transaction commit must be on transaction's
709 if (map
->m_flags
& EXT4_MAP_NEW
&&
710 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
711 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
712 !ext4_is_quota_file(inode
) &&
713 ext4_should_order_data(inode
)) {
714 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
715 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
717 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
726 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
727 * we have to be careful as someone else may be manipulating b_state as well.
729 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
731 unsigned long old_state
;
732 unsigned long new_state
;
734 flags
&= EXT4_MAP_FLAGS
;
736 /* Dummy buffer_head? Set non-atomically. */
738 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
742 * Someone else may be modifying b_state. Be careful! This is ugly but
743 * once we get rid of using bh as a container for mapping information
744 * to pass to / from get_block functions, this can go away.
747 old_state
= READ_ONCE(bh
->b_state
);
748 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
750 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
753 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
754 struct buffer_head
*bh
, int flags
)
756 struct ext4_map_blocks map
;
759 if (ext4_has_inline_data(inode
))
763 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
765 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
768 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
769 ext4_update_bh_state(bh
, map
.m_flags
);
770 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
772 } else if (ret
== 0) {
773 /* hole case, need to fill in bh->b_size */
774 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
779 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
780 struct buffer_head
*bh
, int create
)
782 return _ext4_get_block(inode
, iblock
, bh
,
783 create
? EXT4_GET_BLOCKS_CREATE
: 0);
787 * Get block function used when preparing for buffered write if we require
788 * creating an unwritten extent if blocks haven't been allocated. The extent
789 * will be converted to written after the IO is complete.
791 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
792 struct buffer_head
*bh_result
, int create
)
794 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
795 inode
->i_ino
, create
);
796 return _ext4_get_block(inode
, iblock
, bh_result
,
797 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
800 /* Maximum number of blocks we map for direct IO at once. */
801 #define DIO_MAX_BLOCKS 4096
804 * Get blocks function for the cases that need to start a transaction -
805 * generally difference cases of direct IO and DAX IO. It also handles retries
808 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
809 struct buffer_head
*bh_result
, int flags
)
816 /* Trim mapping request to maximum we can map at once for DIO */
817 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
818 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
819 dio_credits
= ext4_chunk_trans_blocks(inode
,
820 bh_result
->b_size
>> inode
->i_blkbits
);
822 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
824 return PTR_ERR(handle
);
826 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
827 ext4_journal_stop(handle
);
829 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
834 /* Get block function for DIO reads and writes to inodes without extents */
835 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
836 struct buffer_head
*bh
, int create
)
838 /* We don't expect handle for direct IO */
839 WARN_ON_ONCE(ext4_journal_current_handle());
842 return _ext4_get_block(inode
, iblock
, bh
, 0);
843 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
847 * Get block function for AIO DIO writes when we create unwritten extent if
848 * blocks are not allocated yet. The extent will be converted to written
849 * after IO is complete.
851 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
852 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
856 /* We don't expect handle for direct IO */
857 WARN_ON_ONCE(ext4_journal_current_handle());
859 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
860 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
863 * When doing DIO using unwritten extents, we need io_end to convert
864 * unwritten extents to written on IO completion. We allocate io_end
865 * once we spot unwritten extent and store it in b_private. Generic
866 * DIO code keeps b_private set and furthermore passes the value to
867 * our completion callback in 'private' argument.
869 if (!ret
&& buffer_unwritten(bh_result
)) {
870 if (!bh_result
->b_private
) {
871 ext4_io_end_t
*io_end
;
873 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
876 bh_result
->b_private
= io_end
;
877 ext4_set_io_unwritten_flag(inode
, io_end
);
879 set_buffer_defer_completion(bh_result
);
886 * Get block function for non-AIO DIO writes when we create unwritten extent if
887 * blocks are not allocated yet. The extent will be converted to written
888 * after IO is complete from ext4_ext_direct_IO() function.
890 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
891 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
895 /* We don't expect handle for direct IO */
896 WARN_ON_ONCE(ext4_journal_current_handle());
898 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
899 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
902 * Mark inode as having pending DIO writes to unwritten extents.
903 * ext4_ext_direct_IO() checks this flag and converts extents to
906 if (!ret
&& buffer_unwritten(bh_result
))
907 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
912 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
913 struct buffer_head
*bh_result
, int create
)
917 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
918 inode
->i_ino
, create
);
919 /* We don't expect handle for direct IO */
920 WARN_ON_ONCE(ext4_journal_current_handle());
922 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
924 * Blocks should have been preallocated! ext4_file_write_iter() checks
927 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
934 * `handle' can be NULL if create is zero
936 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
937 ext4_lblk_t block
, int map_flags
)
939 struct ext4_map_blocks map
;
940 struct buffer_head
*bh
;
941 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
944 J_ASSERT(handle
!= NULL
|| create
== 0);
948 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
951 return create
? ERR_PTR(-ENOSPC
) : NULL
;
955 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
957 return ERR_PTR(-ENOMEM
);
958 if (map
.m_flags
& EXT4_MAP_NEW
) {
959 J_ASSERT(create
!= 0);
960 J_ASSERT(handle
!= NULL
);
963 * Now that we do not always journal data, we should
964 * keep in mind whether this should always journal the
965 * new buffer as metadata. For now, regular file
966 * writes use ext4_get_block instead, so it's not a
970 BUFFER_TRACE(bh
, "call get_create_access");
971 err
= ext4_journal_get_create_access(handle
, bh
);
976 if (!buffer_uptodate(bh
)) {
977 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
978 set_buffer_uptodate(bh
);
981 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
982 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
986 BUFFER_TRACE(bh
, "not a new buffer");
993 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
994 ext4_lblk_t block
, int map_flags
)
996 struct buffer_head
*bh
;
998 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1001 if (!bh
|| buffer_uptodate(bh
))
1003 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1005 if (buffer_uptodate(bh
))
1008 return ERR_PTR(-EIO
);
1011 int ext4_walk_page_buffers(handle_t
*handle
,
1012 struct buffer_head
*head
,
1016 int (*fn
)(handle_t
*handle
,
1017 struct buffer_head
*bh
))
1019 struct buffer_head
*bh
;
1020 unsigned block_start
, block_end
;
1021 unsigned blocksize
= head
->b_size
;
1023 struct buffer_head
*next
;
1025 for (bh
= head
, block_start
= 0;
1026 ret
== 0 && (bh
!= head
|| !block_start
);
1027 block_start
= block_end
, bh
= next
) {
1028 next
= bh
->b_this_page
;
1029 block_end
= block_start
+ blocksize
;
1030 if (block_end
<= from
|| block_start
>= to
) {
1031 if (partial
&& !buffer_uptodate(bh
))
1035 err
= (*fn
)(handle
, bh
);
1043 * To preserve ordering, it is essential that the hole instantiation and
1044 * the data write be encapsulated in a single transaction. We cannot
1045 * close off a transaction and start a new one between the ext4_get_block()
1046 * and the commit_write(). So doing the jbd2_journal_start at the start of
1047 * prepare_write() is the right place.
1049 * Also, this function can nest inside ext4_writepage(). In that case, we
1050 * *know* that ext4_writepage() has generated enough buffer credits to do the
1051 * whole page. So we won't block on the journal in that case, which is good,
1052 * because the caller may be PF_MEMALLOC.
1054 * By accident, ext4 can be reentered when a transaction is open via
1055 * quota file writes. If we were to commit the transaction while thus
1056 * reentered, there can be a deadlock - we would be holding a quota
1057 * lock, and the commit would never complete if another thread had a
1058 * transaction open and was blocking on the quota lock - a ranking
1061 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1062 * will _not_ run commit under these circumstances because handle->h_ref
1063 * is elevated. We'll still have enough credits for the tiny quotafile
1066 int do_journal_get_write_access(handle_t
*handle
,
1067 struct buffer_head
*bh
)
1069 int dirty
= buffer_dirty(bh
);
1072 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1075 * __block_write_begin() could have dirtied some buffers. Clean
1076 * the dirty bit as jbd2_journal_get_write_access() could complain
1077 * otherwise about fs integrity issues. Setting of the dirty bit
1078 * by __block_write_begin() isn't a real problem here as we clear
1079 * the bit before releasing a page lock and thus writeback cannot
1080 * ever write the buffer.
1083 clear_buffer_dirty(bh
);
1084 BUFFER_TRACE(bh
, "get write access");
1085 ret
= ext4_journal_get_write_access(handle
, bh
);
1087 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1091 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1092 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1093 get_block_t
*get_block
)
1095 unsigned from
= pos
& (PAGE_SIZE
- 1);
1096 unsigned to
= from
+ len
;
1097 struct inode
*inode
= page
->mapping
->host
;
1098 unsigned block_start
, block_end
;
1101 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1103 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1104 bool decrypt
= false;
1106 BUG_ON(!PageLocked(page
));
1107 BUG_ON(from
> PAGE_SIZE
);
1108 BUG_ON(to
> PAGE_SIZE
);
1111 if (!page_has_buffers(page
))
1112 create_empty_buffers(page
, blocksize
, 0);
1113 head
= page_buffers(page
);
1114 bbits
= ilog2(blocksize
);
1115 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1117 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1118 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1119 block_end
= block_start
+ blocksize
;
1120 if (block_end
<= from
|| block_start
>= to
) {
1121 if (PageUptodate(page
)) {
1122 if (!buffer_uptodate(bh
))
1123 set_buffer_uptodate(bh
);
1128 clear_buffer_new(bh
);
1129 if (!buffer_mapped(bh
)) {
1130 WARN_ON(bh
->b_size
!= blocksize
);
1131 err
= get_block(inode
, block
, bh
, 1);
1134 if (buffer_new(bh
)) {
1135 clean_bdev_bh_alias(bh
);
1136 if (PageUptodate(page
)) {
1137 clear_buffer_new(bh
);
1138 set_buffer_uptodate(bh
);
1139 mark_buffer_dirty(bh
);
1142 if (block_end
> to
|| block_start
< from
)
1143 zero_user_segments(page
, to
, block_end
,
1148 if (PageUptodate(page
)) {
1149 if (!buffer_uptodate(bh
))
1150 set_buffer_uptodate(bh
);
1153 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1154 !buffer_unwritten(bh
) &&
1155 (block_start
< from
|| block_end
> to
)) {
1156 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1158 decrypt
= ext4_encrypted_inode(inode
) &&
1159 S_ISREG(inode
->i_mode
);
1163 * If we issued read requests, let them complete.
1165 while (wait_bh
> wait
) {
1166 wait_on_buffer(*--wait_bh
);
1167 if (!buffer_uptodate(*wait_bh
))
1171 page_zero_new_buffers(page
, from
, to
);
1173 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1174 PAGE_SIZE
, 0, page
->index
);
1179 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1180 loff_t pos
, unsigned len
, unsigned flags
,
1181 struct page
**pagep
, void **fsdata
)
1183 struct inode
*inode
= mapping
->host
;
1184 int ret
, needed_blocks
;
1191 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1194 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1196 * Reserve one block more for addition to orphan list in case
1197 * we allocate blocks but write fails for some reason
1199 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1200 index
= pos
>> PAGE_SHIFT
;
1201 from
= pos
& (PAGE_SIZE
- 1);
1204 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1205 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1214 * grab_cache_page_write_begin() can take a long time if the
1215 * system is thrashing due to memory pressure, or if the page
1216 * is being written back. So grab it first before we start
1217 * the transaction handle. This also allows us to allocate
1218 * the page (if needed) without using GFP_NOFS.
1221 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1227 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1228 if (IS_ERR(handle
)) {
1230 return PTR_ERR(handle
);
1234 if (page
->mapping
!= mapping
) {
1235 /* The page got truncated from under us */
1238 ext4_journal_stop(handle
);
1241 /* In case writeback began while the page was unlocked */
1242 wait_for_stable_page(page
);
1244 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1245 if (ext4_should_dioread_nolock(inode
))
1246 ret
= ext4_block_write_begin(page
, pos
, len
,
1247 ext4_get_block_unwritten
);
1249 ret
= ext4_block_write_begin(page
, pos
, len
,
1252 if (ext4_should_dioread_nolock(inode
))
1253 ret
= __block_write_begin(page
, pos
, len
,
1254 ext4_get_block_unwritten
);
1256 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1258 if (!ret
&& ext4_should_journal_data(inode
)) {
1259 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1261 do_journal_get_write_access
);
1267 * __block_write_begin may have instantiated a few blocks
1268 * outside i_size. Trim these off again. Don't need
1269 * i_size_read because we hold i_mutex.
1271 * Add inode to orphan list in case we crash before
1274 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1275 ext4_orphan_add(handle
, inode
);
1277 ext4_journal_stop(handle
);
1278 if (pos
+ len
> inode
->i_size
) {
1279 ext4_truncate_failed_write(inode
);
1281 * If truncate failed early the inode might
1282 * still be on the orphan list; we need to
1283 * make sure the inode is removed from the
1284 * orphan list in that case.
1287 ext4_orphan_del(NULL
, inode
);
1290 if (ret
== -ENOSPC
&&
1291 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1300 /* For write_end() in data=journal mode */
1301 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1304 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1306 set_buffer_uptodate(bh
);
1307 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1308 clear_buffer_meta(bh
);
1309 clear_buffer_prio(bh
);
1314 * We need to pick up the new inode size which generic_commit_write gave us
1315 * `file' can be NULL - eg, when called from page_symlink().
1317 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1318 * buffers are managed internally.
1320 static int ext4_write_end(struct file
*file
,
1321 struct address_space
*mapping
,
1322 loff_t pos
, unsigned len
, unsigned copied
,
1323 struct page
*page
, void *fsdata
)
1325 handle_t
*handle
= ext4_journal_current_handle();
1326 struct inode
*inode
= mapping
->host
;
1327 loff_t old_size
= inode
->i_size
;
1329 int i_size_changed
= 0;
1331 trace_ext4_write_end(inode
, pos
, len
, copied
);
1332 if (ext4_has_inline_data(inode
)) {
1333 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1342 copied
= block_write_end(file
, mapping
, pos
,
1343 len
, copied
, page
, fsdata
);
1345 * it's important to update i_size while still holding page lock:
1346 * page writeout could otherwise come in and zero beyond i_size.
1348 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1353 pagecache_isize_extended(inode
, old_size
, pos
);
1355 * Don't mark the inode dirty under page lock. First, it unnecessarily
1356 * makes the holding time of page lock longer. Second, it forces lock
1357 * ordering of page lock and transaction start for journaling
1361 ext4_mark_inode_dirty(handle
, inode
);
1363 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1364 /* if we have allocated more blocks and copied
1365 * less. We will have blocks allocated outside
1366 * inode->i_size. So truncate them
1368 ext4_orphan_add(handle
, inode
);
1370 ret2
= ext4_journal_stop(handle
);
1374 if (pos
+ len
> inode
->i_size
) {
1375 ext4_truncate_failed_write(inode
);
1377 * If truncate failed early the inode might still be
1378 * on the orphan list; we need to make sure the inode
1379 * is removed from the orphan list in that case.
1382 ext4_orphan_del(NULL
, inode
);
1385 return ret
? ret
: copied
;
1389 * This is a private version of page_zero_new_buffers() which doesn't
1390 * set the buffer to be dirty, since in data=journalled mode we need
1391 * to call ext4_handle_dirty_metadata() instead.
1393 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1395 unsigned from
, unsigned to
)
1397 unsigned int block_start
= 0, block_end
;
1398 struct buffer_head
*head
, *bh
;
1400 bh
= head
= page_buffers(page
);
1402 block_end
= block_start
+ bh
->b_size
;
1403 if (buffer_new(bh
)) {
1404 if (block_end
> from
&& block_start
< to
) {
1405 if (!PageUptodate(page
)) {
1406 unsigned start
, size
;
1408 start
= max(from
, block_start
);
1409 size
= min(to
, block_end
) - start
;
1411 zero_user(page
, start
, size
);
1412 write_end_fn(handle
, bh
);
1414 clear_buffer_new(bh
);
1417 block_start
= block_end
;
1418 bh
= bh
->b_this_page
;
1419 } while (bh
!= head
);
1422 static int ext4_journalled_write_end(struct file
*file
,
1423 struct address_space
*mapping
,
1424 loff_t pos
, unsigned len
, unsigned copied
,
1425 struct page
*page
, void *fsdata
)
1427 handle_t
*handle
= ext4_journal_current_handle();
1428 struct inode
*inode
= mapping
->host
;
1429 loff_t old_size
= inode
->i_size
;
1433 int size_changed
= 0;
1435 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1436 from
= pos
& (PAGE_SIZE
- 1);
1439 BUG_ON(!ext4_handle_valid(handle
));
1441 if (ext4_has_inline_data(inode
)) {
1442 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1450 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1452 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1454 if (unlikely(copied
< len
))
1455 ext4_journalled_zero_new_buffers(handle
, page
,
1457 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1458 from
+ copied
, &partial
,
1461 SetPageUptodate(page
);
1463 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1464 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1465 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1470 pagecache_isize_extended(inode
, old_size
, pos
);
1473 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1478 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1479 /* if we have allocated more blocks and copied
1480 * less. We will have blocks allocated outside
1481 * inode->i_size. So truncate them
1483 ext4_orphan_add(handle
, inode
);
1486 ret2
= ext4_journal_stop(handle
);
1489 if (pos
+ len
> inode
->i_size
) {
1490 ext4_truncate_failed_write(inode
);
1492 * If truncate failed early the inode might still be
1493 * on the orphan list; we need to make sure the inode
1494 * is removed from the orphan list in that case.
1497 ext4_orphan_del(NULL
, inode
);
1500 return ret
? ret
: copied
;
1504 * Reserve space for a single cluster
1506 static int ext4_da_reserve_space(struct inode
*inode
)
1508 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1509 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1513 * We will charge metadata quota at writeout time; this saves
1514 * us from metadata over-estimation, though we may go over by
1515 * a small amount in the end. Here we just reserve for data.
1517 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1521 spin_lock(&ei
->i_block_reservation_lock
);
1522 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1523 spin_unlock(&ei
->i_block_reservation_lock
);
1524 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1527 ei
->i_reserved_data_blocks
++;
1528 trace_ext4_da_reserve_space(inode
);
1529 spin_unlock(&ei
->i_block_reservation_lock
);
1531 return 0; /* success */
1534 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1536 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1537 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1540 return; /* Nothing to release, exit */
1542 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1544 trace_ext4_da_release_space(inode
, to_free
);
1545 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1547 * if there aren't enough reserved blocks, then the
1548 * counter is messed up somewhere. Since this
1549 * function is called from invalidate page, it's
1550 * harmless to return without any action.
1552 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1553 "ino %lu, to_free %d with only %d reserved "
1554 "data blocks", inode
->i_ino
, to_free
,
1555 ei
->i_reserved_data_blocks
);
1557 to_free
= ei
->i_reserved_data_blocks
;
1559 ei
->i_reserved_data_blocks
-= to_free
;
1561 /* update fs dirty data blocks counter */
1562 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1564 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1566 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1569 static void ext4_da_page_release_reservation(struct page
*page
,
1570 unsigned int offset
,
1571 unsigned int length
)
1573 int to_release
= 0, contiguous_blks
= 0;
1574 struct buffer_head
*head
, *bh
;
1575 unsigned int curr_off
= 0;
1576 struct inode
*inode
= page
->mapping
->host
;
1577 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1578 unsigned int stop
= offset
+ length
;
1582 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1584 head
= page_buffers(page
);
1587 unsigned int next_off
= curr_off
+ bh
->b_size
;
1589 if (next_off
> stop
)
1592 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1595 clear_buffer_delay(bh
);
1596 } else if (contiguous_blks
) {
1597 lblk
= page
->index
<<
1598 (PAGE_SHIFT
- inode
->i_blkbits
);
1599 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1601 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1602 contiguous_blks
= 0;
1604 curr_off
= next_off
;
1605 } while ((bh
= bh
->b_this_page
) != head
);
1607 if (contiguous_blks
) {
1608 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1609 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1610 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1613 /* If we have released all the blocks belonging to a cluster, then we
1614 * need to release the reserved space for that cluster. */
1615 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1616 while (num_clusters
> 0) {
1617 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1618 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1619 if (sbi
->s_cluster_ratio
== 1 ||
1620 !ext4_find_delalloc_cluster(inode
, lblk
))
1621 ext4_da_release_space(inode
, 1);
1628 * Delayed allocation stuff
1631 struct mpage_da_data
{
1632 struct inode
*inode
;
1633 struct writeback_control
*wbc
;
1635 pgoff_t first_page
; /* The first page to write */
1636 pgoff_t next_page
; /* Current page to examine */
1637 pgoff_t last_page
; /* Last page to examine */
1639 * Extent to map - this can be after first_page because that can be
1640 * fully mapped. We somewhat abuse m_flags to store whether the extent
1641 * is delalloc or unwritten.
1643 struct ext4_map_blocks map
;
1644 struct ext4_io_submit io_submit
; /* IO submission data */
1645 unsigned int do_map
:1;
1648 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1653 struct pagevec pvec
;
1654 struct inode
*inode
= mpd
->inode
;
1655 struct address_space
*mapping
= inode
->i_mapping
;
1657 /* This is necessary when next_page == 0. */
1658 if (mpd
->first_page
>= mpd
->next_page
)
1661 index
= mpd
->first_page
;
1662 end
= mpd
->next_page
- 1;
1664 ext4_lblk_t start
, last
;
1665 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1666 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1667 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1670 pagevec_init(&pvec
, 0);
1671 while (index
<= end
) {
1672 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1675 for (i
= 0; i
< nr_pages
; i
++) {
1676 struct page
*page
= pvec
.pages
[i
];
1677 if (page
->index
> end
)
1679 BUG_ON(!PageLocked(page
));
1680 BUG_ON(PageWriteback(page
));
1682 if (page_mapped(page
))
1683 clear_page_dirty_for_io(page
);
1684 block_invalidatepage(page
, 0, PAGE_SIZE
);
1685 ClearPageUptodate(page
);
1689 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1690 pagevec_release(&pvec
);
1694 static void ext4_print_free_blocks(struct inode
*inode
)
1696 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1697 struct super_block
*sb
= inode
->i_sb
;
1698 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1700 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1701 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1702 ext4_count_free_clusters(sb
)));
1703 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1704 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1705 (long long) EXT4_C2B(EXT4_SB(sb
),
1706 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1707 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1708 (long long) EXT4_C2B(EXT4_SB(sb
),
1709 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1710 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1711 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1712 ei
->i_reserved_data_blocks
);
1716 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1718 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1722 * This function is grabs code from the very beginning of
1723 * ext4_map_blocks, but assumes that the caller is from delayed write
1724 * time. This function looks up the requested blocks and sets the
1725 * buffer delay bit under the protection of i_data_sem.
1727 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1728 struct ext4_map_blocks
*map
,
1729 struct buffer_head
*bh
)
1731 struct extent_status es
;
1733 sector_t invalid_block
= ~((sector_t
) 0xffff);
1734 #ifdef ES_AGGRESSIVE_TEST
1735 struct ext4_map_blocks orig_map
;
1737 memcpy(&orig_map
, map
, sizeof(*map
));
1740 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1744 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1745 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1746 (unsigned long) map
->m_lblk
);
1748 /* Lookup extent status tree firstly */
1749 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1750 if (ext4_es_is_hole(&es
)) {
1752 down_read(&EXT4_I(inode
)->i_data_sem
);
1757 * Delayed extent could be allocated by fallocate.
1758 * So we need to check it.
1760 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1761 map_bh(bh
, inode
->i_sb
, invalid_block
);
1763 set_buffer_delay(bh
);
1767 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1768 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1769 if (retval
> map
->m_len
)
1770 retval
= map
->m_len
;
1771 map
->m_len
= retval
;
1772 if (ext4_es_is_written(&es
))
1773 map
->m_flags
|= EXT4_MAP_MAPPED
;
1774 else if (ext4_es_is_unwritten(&es
))
1775 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1779 #ifdef ES_AGGRESSIVE_TEST
1780 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1786 * Try to see if we can get the block without requesting a new
1787 * file system block.
1789 down_read(&EXT4_I(inode
)->i_data_sem
);
1790 if (ext4_has_inline_data(inode
))
1792 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1793 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1795 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1801 * XXX: __block_prepare_write() unmaps passed block,
1805 * If the block was allocated from previously allocated cluster,
1806 * then we don't need to reserve it again. However we still need
1807 * to reserve metadata for every block we're going to write.
1809 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1810 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1811 ret
= ext4_da_reserve_space(inode
);
1813 /* not enough space to reserve */
1819 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1820 ~0, EXTENT_STATUS_DELAYED
);
1826 map_bh(bh
, inode
->i_sb
, invalid_block
);
1828 set_buffer_delay(bh
);
1829 } else if (retval
> 0) {
1831 unsigned int status
;
1833 if (unlikely(retval
!= map
->m_len
)) {
1834 ext4_warning(inode
->i_sb
,
1835 "ES len assertion failed for inode "
1836 "%lu: retval %d != map->m_len %d",
1837 inode
->i_ino
, retval
, map
->m_len
);
1841 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1842 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1843 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1844 map
->m_pblk
, status
);
1850 up_read((&EXT4_I(inode
)->i_data_sem
));
1856 * This is a special get_block_t callback which is used by
1857 * ext4_da_write_begin(). It will either return mapped block or
1858 * reserve space for a single block.
1860 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1861 * We also have b_blocknr = -1 and b_bdev initialized properly
1863 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1864 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1865 * initialized properly.
1867 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1868 struct buffer_head
*bh
, int create
)
1870 struct ext4_map_blocks map
;
1873 BUG_ON(create
== 0);
1874 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1876 map
.m_lblk
= iblock
;
1880 * first, we need to know whether the block is allocated already
1881 * preallocated blocks are unmapped but should treated
1882 * the same as allocated blocks.
1884 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1888 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1889 ext4_update_bh_state(bh
, map
.m_flags
);
1891 if (buffer_unwritten(bh
)) {
1892 /* A delayed write to unwritten bh should be marked
1893 * new and mapped. Mapped ensures that we don't do
1894 * get_block multiple times when we write to the same
1895 * offset and new ensures that we do proper zero out
1896 * for partial write.
1899 set_buffer_mapped(bh
);
1904 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1910 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1916 static int __ext4_journalled_writepage(struct page
*page
,
1919 struct address_space
*mapping
= page
->mapping
;
1920 struct inode
*inode
= mapping
->host
;
1921 struct buffer_head
*page_bufs
= NULL
;
1922 handle_t
*handle
= NULL
;
1923 int ret
= 0, err
= 0;
1924 int inline_data
= ext4_has_inline_data(inode
);
1925 struct buffer_head
*inode_bh
= NULL
;
1927 ClearPageChecked(page
);
1930 BUG_ON(page
->index
!= 0);
1931 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1932 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1933 if (inode_bh
== NULL
)
1936 page_bufs
= page_buffers(page
);
1941 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1945 * We need to release the page lock before we start the
1946 * journal, so grab a reference so the page won't disappear
1947 * out from under us.
1952 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1953 ext4_writepage_trans_blocks(inode
));
1954 if (IS_ERR(handle
)) {
1955 ret
= PTR_ERR(handle
);
1957 goto out_no_pagelock
;
1959 BUG_ON(!ext4_handle_valid(handle
));
1963 if (page
->mapping
!= mapping
) {
1964 /* The page got truncated from under us */
1965 ext4_journal_stop(handle
);
1971 BUFFER_TRACE(inode_bh
, "get write access");
1972 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1974 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1977 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1978 do_journal_get_write_access
);
1980 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1985 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1986 err
= ext4_journal_stop(handle
);
1990 if (!ext4_has_inline_data(inode
))
1991 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1993 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2002 * Note that we don't need to start a transaction unless we're journaling data
2003 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2004 * need to file the inode to the transaction's list in ordered mode because if
2005 * we are writing back data added by write(), the inode is already there and if
2006 * we are writing back data modified via mmap(), no one guarantees in which
2007 * transaction the data will hit the disk. In case we are journaling data, we
2008 * cannot start transaction directly because transaction start ranks above page
2009 * lock so we have to do some magic.
2011 * This function can get called via...
2012 * - ext4_writepages after taking page lock (have journal handle)
2013 * - journal_submit_inode_data_buffers (no journal handle)
2014 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2015 * - grab_page_cache when doing write_begin (have journal handle)
2017 * We don't do any block allocation in this function. If we have page with
2018 * multiple blocks we need to write those buffer_heads that are mapped. This
2019 * is important for mmaped based write. So if we do with blocksize 1K
2020 * truncate(f, 1024);
2021 * a = mmap(f, 0, 4096);
2023 * truncate(f, 4096);
2024 * we have in the page first buffer_head mapped via page_mkwrite call back
2025 * but other buffer_heads would be unmapped but dirty (dirty done via the
2026 * do_wp_page). So writepage should write the first block. If we modify
2027 * the mmap area beyond 1024 we will again get a page_fault and the
2028 * page_mkwrite callback will do the block allocation and mark the
2029 * buffer_heads mapped.
2031 * We redirty the page if we have any buffer_heads that is either delay or
2032 * unwritten in the page.
2034 * We can get recursively called as show below.
2036 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2039 * But since we don't do any block allocation we should not deadlock.
2040 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2042 static int ext4_writepage(struct page
*page
,
2043 struct writeback_control
*wbc
)
2048 struct buffer_head
*page_bufs
= NULL
;
2049 struct inode
*inode
= page
->mapping
->host
;
2050 struct ext4_io_submit io_submit
;
2051 bool keep_towrite
= false;
2053 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
2054 ext4_invalidatepage(page
, 0, PAGE_SIZE
);
2059 trace_ext4_writepage(page
);
2060 size
= i_size_read(inode
);
2061 if (page
->index
== size
>> PAGE_SHIFT
)
2062 len
= size
& ~PAGE_MASK
;
2066 page_bufs
= page_buffers(page
);
2068 * We cannot do block allocation or other extent handling in this
2069 * function. If there are buffers needing that, we have to redirty
2070 * the page. But we may reach here when we do a journal commit via
2071 * journal_submit_inode_data_buffers() and in that case we must write
2072 * allocated buffers to achieve data=ordered mode guarantees.
2074 * Also, if there is only one buffer per page (the fs block
2075 * size == the page size), if one buffer needs block
2076 * allocation or needs to modify the extent tree to clear the
2077 * unwritten flag, we know that the page can't be written at
2078 * all, so we might as well refuse the write immediately.
2079 * Unfortunately if the block size != page size, we can't as
2080 * easily detect this case using ext4_walk_page_buffers(), but
2081 * for the extremely common case, this is an optimization that
2082 * skips a useless round trip through ext4_bio_write_page().
2084 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2085 ext4_bh_delay_or_unwritten
)) {
2086 redirty_page_for_writepage(wbc
, page
);
2087 if ((current
->flags
& PF_MEMALLOC
) ||
2088 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2090 * For memory cleaning there's no point in writing only
2091 * some buffers. So just bail out. Warn if we came here
2092 * from direct reclaim.
2094 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2099 keep_towrite
= true;
2102 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2104 * It's mmapped pagecache. Add buffers and journal it. There
2105 * doesn't seem much point in redirtying the page here.
2107 return __ext4_journalled_writepage(page
, len
);
2109 ext4_io_submit_init(&io_submit
, wbc
);
2110 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2111 if (!io_submit
.io_end
) {
2112 redirty_page_for_writepage(wbc
, page
);
2116 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2117 ext4_io_submit(&io_submit
);
2118 /* Drop io_end reference we got from init */
2119 ext4_put_io_end_defer(io_submit
.io_end
);
2123 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2129 BUG_ON(page
->index
!= mpd
->first_page
);
2130 clear_page_dirty_for_io(page
);
2132 * We have to be very careful here! Nothing protects writeback path
2133 * against i_size changes and the page can be writeably mapped into
2134 * page tables. So an application can be growing i_size and writing
2135 * data through mmap while writeback runs. clear_page_dirty_for_io()
2136 * write-protects our page in page tables and the page cannot get
2137 * written to again until we release page lock. So only after
2138 * clear_page_dirty_for_io() we are safe to sample i_size for
2139 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2140 * on the barrier provided by TestClearPageDirty in
2141 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2142 * after page tables are updated.
2144 size
= i_size_read(mpd
->inode
);
2145 if (page
->index
== size
>> PAGE_SHIFT
)
2146 len
= size
& ~PAGE_MASK
;
2149 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2151 mpd
->wbc
->nr_to_write
--;
2157 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2160 * mballoc gives us at most this number of blocks...
2161 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2162 * The rest of mballoc seems to handle chunks up to full group size.
2164 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2167 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2169 * @mpd - extent of blocks
2170 * @lblk - logical number of the block in the file
2171 * @bh - buffer head we want to add to the extent
2173 * The function is used to collect contig. blocks in the same state. If the
2174 * buffer doesn't require mapping for writeback and we haven't started the
2175 * extent of buffers to map yet, the function returns 'true' immediately - the
2176 * caller can write the buffer right away. Otherwise the function returns true
2177 * if the block has been added to the extent, false if the block couldn't be
2180 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2181 struct buffer_head
*bh
)
2183 struct ext4_map_blocks
*map
= &mpd
->map
;
2185 /* Buffer that doesn't need mapping for writeback? */
2186 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2187 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2188 /* So far no extent to map => we write the buffer right away */
2189 if (map
->m_len
== 0)
2194 /* First block in the extent? */
2195 if (map
->m_len
== 0) {
2196 /* We cannot map unless handle is started... */
2201 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2205 /* Don't go larger than mballoc is willing to allocate */
2206 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2209 /* Can we merge the block to our big extent? */
2210 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2211 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2219 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2221 * @mpd - extent of blocks for mapping
2222 * @head - the first buffer in the page
2223 * @bh - buffer we should start processing from
2224 * @lblk - logical number of the block in the file corresponding to @bh
2226 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2227 * the page for IO if all buffers in this page were mapped and there's no
2228 * accumulated extent of buffers to map or add buffers in the page to the
2229 * extent of buffers to map. The function returns 1 if the caller can continue
2230 * by processing the next page, 0 if it should stop adding buffers to the
2231 * extent to map because we cannot extend it anymore. It can also return value
2232 * < 0 in case of error during IO submission.
2234 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2235 struct buffer_head
*head
,
2236 struct buffer_head
*bh
,
2239 struct inode
*inode
= mpd
->inode
;
2241 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2242 >> inode
->i_blkbits
;
2245 BUG_ON(buffer_locked(bh
));
2247 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2248 /* Found extent to map? */
2251 /* Buffer needs mapping and handle is not started? */
2254 /* Everything mapped so far and we hit EOF */
2257 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2258 /* So far everything mapped? Submit the page for IO. */
2259 if (mpd
->map
.m_len
== 0) {
2260 err
= mpage_submit_page(mpd
, head
->b_page
);
2264 return lblk
< blocks
;
2268 * mpage_map_buffers - update buffers corresponding to changed extent and
2269 * submit fully mapped pages for IO
2271 * @mpd - description of extent to map, on return next extent to map
2273 * Scan buffers corresponding to changed extent (we expect corresponding pages
2274 * to be already locked) and update buffer state according to new extent state.
2275 * We map delalloc buffers to their physical location, clear unwritten bits,
2276 * and mark buffers as uninit when we perform writes to unwritten extents
2277 * and do extent conversion after IO is finished. If the last page is not fully
2278 * mapped, we update @map to the next extent in the last page that needs
2279 * mapping. Otherwise we submit the page for IO.
2281 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2283 struct pagevec pvec
;
2285 struct inode
*inode
= mpd
->inode
;
2286 struct buffer_head
*head
, *bh
;
2287 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2293 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2294 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2295 lblk
= start
<< bpp_bits
;
2296 pblock
= mpd
->map
.m_pblk
;
2298 pagevec_init(&pvec
, 0);
2299 while (start
<= end
) {
2300 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2304 for (i
= 0; i
< nr_pages
; i
++) {
2305 struct page
*page
= pvec
.pages
[i
];
2307 if (page
->index
> end
)
2309 /* Up to 'end' pages must be contiguous */
2310 BUG_ON(page
->index
!= start
);
2311 bh
= head
= page_buffers(page
);
2313 if (lblk
< mpd
->map
.m_lblk
)
2315 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2317 * Buffer after end of mapped extent.
2318 * Find next buffer in the page to map.
2321 mpd
->map
.m_flags
= 0;
2323 * FIXME: If dioread_nolock supports
2324 * blocksize < pagesize, we need to make
2325 * sure we add size mapped so far to
2326 * io_end->size as the following call
2327 * can submit the page for IO.
2329 err
= mpage_process_page_bufs(mpd
, head
,
2331 pagevec_release(&pvec
);
2336 if (buffer_delay(bh
)) {
2337 clear_buffer_delay(bh
);
2338 bh
->b_blocknr
= pblock
++;
2340 clear_buffer_unwritten(bh
);
2341 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2344 * FIXME: This is going to break if dioread_nolock
2345 * supports blocksize < pagesize as we will try to
2346 * convert potentially unmapped parts of inode.
2348 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2349 /* Page fully mapped - let IO run! */
2350 err
= mpage_submit_page(mpd
, page
);
2352 pagevec_release(&pvec
);
2357 pagevec_release(&pvec
);
2359 /* Extent fully mapped and matches with page boundary. We are done. */
2361 mpd
->map
.m_flags
= 0;
2365 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2367 struct inode
*inode
= mpd
->inode
;
2368 struct ext4_map_blocks
*map
= &mpd
->map
;
2369 int get_blocks_flags
;
2370 int err
, dioread_nolock
;
2372 trace_ext4_da_write_pages_extent(inode
, map
);
2374 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2375 * to convert an unwritten extent to be initialized (in the case
2376 * where we have written into one or more preallocated blocks). It is
2377 * possible that we're going to need more metadata blocks than
2378 * previously reserved. However we must not fail because we're in
2379 * writeback and there is nothing we can do about it so it might result
2380 * in data loss. So use reserved blocks to allocate metadata if
2383 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2384 * the blocks in question are delalloc blocks. This indicates
2385 * that the blocks and quotas has already been checked when
2386 * the data was copied into the page cache.
2388 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2389 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2390 EXT4_GET_BLOCKS_IO_SUBMIT
;
2391 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2393 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2394 if (map
->m_flags
& (1 << BH_Delay
))
2395 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2397 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2400 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2401 if (!mpd
->io_submit
.io_end
->handle
&&
2402 ext4_handle_valid(handle
)) {
2403 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2404 handle
->h_rsv_handle
= NULL
;
2406 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2409 BUG_ON(map
->m_len
== 0);
2410 if (map
->m_flags
& EXT4_MAP_NEW
) {
2411 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
2418 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2419 * mpd->len and submit pages underlying it for IO
2421 * @handle - handle for journal operations
2422 * @mpd - extent to map
2423 * @give_up_on_write - we set this to true iff there is a fatal error and there
2424 * is no hope of writing the data. The caller should discard
2425 * dirty pages to avoid infinite loops.
2427 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2428 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2429 * them to initialized or split the described range from larger unwritten
2430 * extent. Note that we need not map all the described range since allocation
2431 * can return less blocks or the range is covered by more unwritten extents. We
2432 * cannot map more because we are limited by reserved transaction credits. On
2433 * the other hand we always make sure that the last touched page is fully
2434 * mapped so that it can be written out (and thus forward progress is
2435 * guaranteed). After mapping we submit all mapped pages for IO.
2437 static int mpage_map_and_submit_extent(handle_t
*handle
,
2438 struct mpage_da_data
*mpd
,
2439 bool *give_up_on_write
)
2441 struct inode
*inode
= mpd
->inode
;
2442 struct ext4_map_blocks
*map
= &mpd
->map
;
2447 mpd
->io_submit
.io_end
->offset
=
2448 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2450 err
= mpage_map_one_extent(handle
, mpd
);
2452 struct super_block
*sb
= inode
->i_sb
;
2454 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2455 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2456 goto invalidate_dirty_pages
;
2458 * Let the uper layers retry transient errors.
2459 * In the case of ENOSPC, if ext4_count_free_blocks()
2460 * is non-zero, a commit should free up blocks.
2462 if ((err
== -ENOMEM
) ||
2463 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2465 goto update_disksize
;
2468 ext4_msg(sb
, KERN_CRIT
,
2469 "Delayed block allocation failed for "
2470 "inode %lu at logical offset %llu with"
2471 " max blocks %u with error %d",
2473 (unsigned long long)map
->m_lblk
,
2474 (unsigned)map
->m_len
, -err
);
2475 ext4_msg(sb
, KERN_CRIT
,
2476 "This should not happen!! Data will "
2479 ext4_print_free_blocks(inode
);
2480 invalidate_dirty_pages
:
2481 *give_up_on_write
= true;
2486 * Update buffer state, submit mapped pages, and get us new
2489 err
= mpage_map_and_submit_buffers(mpd
);
2491 goto update_disksize
;
2492 } while (map
->m_len
);
2496 * Update on-disk size after IO is submitted. Races with
2497 * truncate are avoided by checking i_size under i_data_sem.
2499 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2500 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2504 down_write(&EXT4_I(inode
)->i_data_sem
);
2505 i_size
= i_size_read(inode
);
2506 if (disksize
> i_size
)
2508 if (disksize
> EXT4_I(inode
)->i_disksize
)
2509 EXT4_I(inode
)->i_disksize
= disksize
;
2510 up_write(&EXT4_I(inode
)->i_data_sem
);
2511 err2
= ext4_mark_inode_dirty(handle
, inode
);
2513 ext4_error(inode
->i_sb
,
2514 "Failed to mark inode %lu dirty",
2523 * Calculate the total number of credits to reserve for one writepages
2524 * iteration. This is called from ext4_writepages(). We map an extent of
2525 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2526 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2527 * bpp - 1 blocks in bpp different extents.
2529 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2531 int bpp
= ext4_journal_blocks_per_page(inode
);
2533 return ext4_meta_trans_blocks(inode
,
2534 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2538 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2539 * and underlying extent to map
2541 * @mpd - where to look for pages
2543 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2544 * IO immediately. When we find a page which isn't mapped we start accumulating
2545 * extent of buffers underlying these pages that needs mapping (formed by
2546 * either delayed or unwritten buffers). We also lock the pages containing
2547 * these buffers. The extent found is returned in @mpd structure (starting at
2548 * mpd->lblk with length mpd->len blocks).
2550 * Note that this function can attach bios to one io_end structure which are
2551 * neither logically nor physically contiguous. Although it may seem as an
2552 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2553 * case as we need to track IO to all buffers underlying a page in one io_end.
2555 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2557 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2558 struct pagevec pvec
;
2559 unsigned int nr_pages
;
2560 long left
= mpd
->wbc
->nr_to_write
;
2561 pgoff_t index
= mpd
->first_page
;
2562 pgoff_t end
= mpd
->last_page
;
2565 int blkbits
= mpd
->inode
->i_blkbits
;
2567 struct buffer_head
*head
;
2569 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2570 tag
= PAGECACHE_TAG_TOWRITE
;
2572 tag
= PAGECACHE_TAG_DIRTY
;
2574 pagevec_init(&pvec
, 0);
2576 mpd
->next_page
= index
;
2577 while (index
<= end
) {
2578 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2579 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2583 for (i
= 0; i
< nr_pages
; i
++) {
2584 struct page
*page
= pvec
.pages
[i
];
2587 * At this point, the page may be truncated or
2588 * invalidated (changing page->mapping to NULL), or
2589 * even swizzled back from swapper_space to tmpfs file
2590 * mapping. However, page->index will not change
2591 * because we have a reference on the page.
2593 if (page
->index
> end
)
2597 * Accumulated enough dirty pages? This doesn't apply
2598 * to WB_SYNC_ALL mode. For integrity sync we have to
2599 * keep going because someone may be concurrently
2600 * dirtying pages, and we might have synced a lot of
2601 * newly appeared dirty pages, but have not synced all
2602 * of the old dirty pages.
2604 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2607 /* If we can't merge this page, we are done. */
2608 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2613 * If the page is no longer dirty, or its mapping no
2614 * longer corresponds to inode we are writing (which
2615 * means it has been truncated or invalidated), or the
2616 * page is already under writeback and we are not doing
2617 * a data integrity writeback, skip the page
2619 if (!PageDirty(page
) ||
2620 (PageWriteback(page
) &&
2621 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2622 unlikely(page
->mapping
!= mapping
)) {
2627 wait_on_page_writeback(page
);
2628 BUG_ON(PageWriteback(page
));
2630 if (mpd
->map
.m_len
== 0)
2631 mpd
->first_page
= page
->index
;
2632 mpd
->next_page
= page
->index
+ 1;
2633 /* Add all dirty buffers to mpd */
2634 lblk
= ((ext4_lblk_t
)page
->index
) <<
2635 (PAGE_SHIFT
- blkbits
);
2636 head
= page_buffers(page
);
2637 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2643 pagevec_release(&pvec
);
2648 pagevec_release(&pvec
);
2652 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2655 struct address_space
*mapping
= data
;
2656 int ret
= ext4_writepage(page
, wbc
);
2657 mapping_set_error(mapping
, ret
);
2661 static int ext4_writepages(struct address_space
*mapping
,
2662 struct writeback_control
*wbc
)
2664 pgoff_t writeback_index
= 0;
2665 long nr_to_write
= wbc
->nr_to_write
;
2666 int range_whole
= 0;
2668 handle_t
*handle
= NULL
;
2669 struct mpage_da_data mpd
;
2670 struct inode
*inode
= mapping
->host
;
2671 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2672 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2674 struct blk_plug plug
;
2675 bool give_up_on_write
= false;
2677 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2680 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2681 trace_ext4_writepages(inode
, wbc
);
2683 if (dax_mapping(mapping
)) {
2684 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2686 goto out_writepages
;
2690 * No pages to write? This is mainly a kludge to avoid starting
2691 * a transaction for special inodes like journal inode on last iput()
2692 * because that could violate lock ordering on umount
2694 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2695 goto out_writepages
;
2697 if (ext4_should_journal_data(inode
)) {
2698 struct blk_plug plug
;
2700 blk_start_plug(&plug
);
2701 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2702 blk_finish_plug(&plug
);
2703 goto out_writepages
;
2707 * If the filesystem has aborted, it is read-only, so return
2708 * right away instead of dumping stack traces later on that
2709 * will obscure the real source of the problem. We test
2710 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2711 * the latter could be true if the filesystem is mounted
2712 * read-only, and in that case, ext4_writepages should
2713 * *never* be called, so if that ever happens, we would want
2716 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2717 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2719 goto out_writepages
;
2722 if (ext4_should_dioread_nolock(inode
)) {
2724 * We may need to convert up to one extent per block in
2725 * the page and we may dirty the inode.
2727 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2731 * If we have inline data and arrive here, it means that
2732 * we will soon create the block for the 1st page, so
2733 * we'd better clear the inline data here.
2735 if (ext4_has_inline_data(inode
)) {
2736 /* Just inode will be modified... */
2737 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2738 if (IS_ERR(handle
)) {
2739 ret
= PTR_ERR(handle
);
2740 goto out_writepages
;
2742 BUG_ON(ext4_test_inode_state(inode
,
2743 EXT4_STATE_MAY_INLINE_DATA
));
2744 ext4_destroy_inline_data(handle
, inode
);
2745 ext4_journal_stop(handle
);
2748 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2751 if (wbc
->range_cyclic
) {
2752 writeback_index
= mapping
->writeback_index
;
2753 if (writeback_index
)
2755 mpd
.first_page
= writeback_index
;
2758 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2759 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2764 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2766 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2767 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2769 blk_start_plug(&plug
);
2772 * First writeback pages that don't need mapping - we can avoid
2773 * starting a transaction unnecessarily and also avoid being blocked
2774 * in the block layer on device congestion while having transaction
2778 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2779 if (!mpd
.io_submit
.io_end
) {
2783 ret
= mpage_prepare_extent_to_map(&mpd
);
2784 /* Submit prepared bio */
2785 ext4_io_submit(&mpd
.io_submit
);
2786 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2787 mpd
.io_submit
.io_end
= NULL
;
2788 /* Unlock pages we didn't use */
2789 mpage_release_unused_pages(&mpd
, false);
2793 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2794 /* For each extent of pages we use new io_end */
2795 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2796 if (!mpd
.io_submit
.io_end
) {
2802 * We have two constraints: We find one extent to map and we
2803 * must always write out whole page (makes a difference when
2804 * blocksize < pagesize) so that we don't block on IO when we
2805 * try to write out the rest of the page. Journalled mode is
2806 * not supported by delalloc.
2808 BUG_ON(ext4_should_journal_data(inode
));
2809 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2811 /* start a new transaction */
2812 handle
= ext4_journal_start_with_reserve(inode
,
2813 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2814 if (IS_ERR(handle
)) {
2815 ret
= PTR_ERR(handle
);
2816 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2817 "%ld pages, ino %lu; err %d", __func__
,
2818 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2819 /* Release allocated io_end */
2820 ext4_put_io_end(mpd
.io_submit
.io_end
);
2821 mpd
.io_submit
.io_end
= NULL
;
2826 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2827 ret
= mpage_prepare_extent_to_map(&mpd
);
2830 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2834 * We scanned the whole range (or exhausted
2835 * nr_to_write), submitted what was mapped and
2836 * didn't find anything needing mapping. We are
2843 * Caution: If the handle is synchronous,
2844 * ext4_journal_stop() can wait for transaction commit
2845 * to finish which may depend on writeback of pages to
2846 * complete or on page lock to be released. In that
2847 * case, we have to wait until after after we have
2848 * submitted all the IO, released page locks we hold,
2849 * and dropped io_end reference (for extent conversion
2850 * to be able to complete) before stopping the handle.
2852 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2853 ext4_journal_stop(handle
);
2857 /* Submit prepared bio */
2858 ext4_io_submit(&mpd
.io_submit
);
2859 /* Unlock pages we didn't use */
2860 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2862 * Drop our io_end reference we got from init. We have
2863 * to be careful and use deferred io_end finishing if
2864 * we are still holding the transaction as we can
2865 * release the last reference to io_end which may end
2866 * up doing unwritten extent conversion.
2869 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2870 ext4_journal_stop(handle
);
2872 ext4_put_io_end(mpd
.io_submit
.io_end
);
2873 mpd
.io_submit
.io_end
= NULL
;
2875 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2877 * Commit the transaction which would
2878 * free blocks released in the transaction
2881 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2885 /* Fatal error - ENOMEM, EIO... */
2890 blk_finish_plug(&plug
);
2891 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2893 mpd
.last_page
= writeback_index
- 1;
2899 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2901 * Set the writeback_index so that range_cyclic
2902 * mode will write it back later
2904 mapping
->writeback_index
= mpd
.first_page
;
2907 trace_ext4_writepages_result(inode
, wbc
, ret
,
2908 nr_to_write
- wbc
->nr_to_write
);
2909 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2913 static int ext4_nonda_switch(struct super_block
*sb
)
2915 s64 free_clusters
, dirty_clusters
;
2916 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2919 * switch to non delalloc mode if we are running low
2920 * on free block. The free block accounting via percpu
2921 * counters can get slightly wrong with percpu_counter_batch getting
2922 * accumulated on each CPU without updating global counters
2923 * Delalloc need an accurate free block accounting. So switch
2924 * to non delalloc when we are near to error range.
2927 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2929 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2931 * Start pushing delalloc when 1/2 of free blocks are dirty.
2933 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2934 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2936 if (2 * free_clusters
< 3 * dirty_clusters
||
2937 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2939 * free block count is less than 150% of dirty blocks
2940 * or free blocks is less than watermark
2947 /* We always reserve for an inode update; the superblock could be there too */
2948 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2950 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2953 if (pos
+ len
<= 0x7fffffffULL
)
2956 /* We might need to update the superblock to set LARGE_FILE */
2960 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2961 loff_t pos
, unsigned len
, unsigned flags
,
2962 struct page
**pagep
, void **fsdata
)
2964 int ret
, retries
= 0;
2967 struct inode
*inode
= mapping
->host
;
2970 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2973 index
= pos
>> PAGE_SHIFT
;
2975 if (ext4_nonda_switch(inode
->i_sb
) ||
2976 S_ISLNK(inode
->i_mode
)) {
2977 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2978 return ext4_write_begin(file
, mapping
, pos
,
2979 len
, flags
, pagep
, fsdata
);
2981 *fsdata
= (void *)0;
2982 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2984 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2985 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2995 * grab_cache_page_write_begin() can take a long time if the
2996 * system is thrashing due to memory pressure, or if the page
2997 * is being written back. So grab it first before we start
2998 * the transaction handle. This also allows us to allocate
2999 * the page (if needed) without using GFP_NOFS.
3002 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3008 * With delayed allocation, we don't log the i_disksize update
3009 * if there is delayed block allocation. But we still need
3010 * to journalling the i_disksize update if writes to the end
3011 * of file which has an already mapped buffer.
3014 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
3015 ext4_da_write_credits(inode
, pos
, len
));
3016 if (IS_ERR(handle
)) {
3018 return PTR_ERR(handle
);
3022 if (page
->mapping
!= mapping
) {
3023 /* The page got truncated from under us */
3026 ext4_journal_stop(handle
);
3029 /* In case writeback began while the page was unlocked */
3030 wait_for_stable_page(page
);
3032 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3033 ret
= ext4_block_write_begin(page
, pos
, len
,
3034 ext4_da_get_block_prep
);
3036 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3040 ext4_journal_stop(handle
);
3042 * block_write_begin may have instantiated a few blocks
3043 * outside i_size. Trim these off again. Don't need
3044 * i_size_read because we hold i_mutex.
3046 if (pos
+ len
> inode
->i_size
)
3047 ext4_truncate_failed_write(inode
);
3049 if (ret
== -ENOSPC
&&
3050 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3062 * Check if we should update i_disksize
3063 * when write to the end of file but not require block allocation
3065 static int ext4_da_should_update_i_disksize(struct page
*page
,
3066 unsigned long offset
)
3068 struct buffer_head
*bh
;
3069 struct inode
*inode
= page
->mapping
->host
;
3073 bh
= page_buffers(page
);
3074 idx
= offset
>> inode
->i_blkbits
;
3076 for (i
= 0; i
< idx
; i
++)
3077 bh
= bh
->b_this_page
;
3079 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3084 static int ext4_da_write_end(struct file
*file
,
3085 struct address_space
*mapping
,
3086 loff_t pos
, unsigned len
, unsigned copied
,
3087 struct page
*page
, void *fsdata
)
3089 struct inode
*inode
= mapping
->host
;
3091 handle_t
*handle
= ext4_journal_current_handle();
3093 unsigned long start
, end
;
3094 int write_mode
= (int)(unsigned long)fsdata
;
3096 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3097 return ext4_write_end(file
, mapping
, pos
,
3098 len
, copied
, page
, fsdata
);
3100 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3101 start
= pos
& (PAGE_SIZE
- 1);
3102 end
= start
+ copied
- 1;
3105 * generic_write_end() will run mark_inode_dirty() if i_size
3106 * changes. So let's piggyback the i_disksize mark_inode_dirty
3109 new_i_size
= pos
+ copied
;
3110 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3111 if (ext4_has_inline_data(inode
) ||
3112 ext4_da_should_update_i_disksize(page
, end
)) {
3113 ext4_update_i_disksize(inode
, new_i_size
);
3114 /* We need to mark inode dirty even if
3115 * new_i_size is less that inode->i_size
3116 * bu greater than i_disksize.(hint delalloc)
3118 ext4_mark_inode_dirty(handle
, inode
);
3122 if (write_mode
!= CONVERT_INLINE_DATA
&&
3123 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3124 ext4_has_inline_data(inode
))
3125 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3128 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3134 ret2
= ext4_journal_stop(handle
);
3138 return ret
? ret
: copied
;
3141 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3142 unsigned int length
)
3145 * Drop reserved blocks
3147 BUG_ON(!PageLocked(page
));
3148 if (!page_has_buffers(page
))
3151 ext4_da_page_release_reservation(page
, offset
, length
);
3154 ext4_invalidatepage(page
, offset
, length
);
3160 * Force all delayed allocation blocks to be allocated for a given inode.
3162 int ext4_alloc_da_blocks(struct inode
*inode
)
3164 trace_ext4_alloc_da_blocks(inode
);
3166 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3170 * We do something simple for now. The filemap_flush() will
3171 * also start triggering a write of the data blocks, which is
3172 * not strictly speaking necessary (and for users of
3173 * laptop_mode, not even desirable). However, to do otherwise
3174 * would require replicating code paths in:
3176 * ext4_writepages() ->
3177 * write_cache_pages() ---> (via passed in callback function)
3178 * __mpage_da_writepage() -->
3179 * mpage_add_bh_to_extent()
3180 * mpage_da_map_blocks()
3182 * The problem is that write_cache_pages(), located in
3183 * mm/page-writeback.c, marks pages clean in preparation for
3184 * doing I/O, which is not desirable if we're not planning on
3187 * We could call write_cache_pages(), and then redirty all of
3188 * the pages by calling redirty_page_for_writepage() but that
3189 * would be ugly in the extreme. So instead we would need to
3190 * replicate parts of the code in the above functions,
3191 * simplifying them because we wouldn't actually intend to
3192 * write out the pages, but rather only collect contiguous
3193 * logical block extents, call the multi-block allocator, and
3194 * then update the buffer heads with the block allocations.
3196 * For now, though, we'll cheat by calling filemap_flush(),
3197 * which will map the blocks, and start the I/O, but not
3198 * actually wait for the I/O to complete.
3200 return filemap_flush(inode
->i_mapping
);
3204 * bmap() is special. It gets used by applications such as lilo and by
3205 * the swapper to find the on-disk block of a specific piece of data.
3207 * Naturally, this is dangerous if the block concerned is still in the
3208 * journal. If somebody makes a swapfile on an ext4 data-journaling
3209 * filesystem and enables swap, then they may get a nasty shock when the
3210 * data getting swapped to that swapfile suddenly gets overwritten by
3211 * the original zero's written out previously to the journal and
3212 * awaiting writeback in the kernel's buffer cache.
3214 * So, if we see any bmap calls here on a modified, data-journaled file,
3215 * take extra steps to flush any blocks which might be in the cache.
3217 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3219 struct inode
*inode
= mapping
->host
;
3224 * We can get here for an inline file via the FIBMAP ioctl
3226 if (ext4_has_inline_data(inode
))
3229 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3230 test_opt(inode
->i_sb
, DELALLOC
)) {
3232 * With delalloc we want to sync the file
3233 * so that we can make sure we allocate
3236 filemap_write_and_wait(mapping
);
3239 if (EXT4_JOURNAL(inode
) &&
3240 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3242 * This is a REALLY heavyweight approach, but the use of
3243 * bmap on dirty files is expected to be extremely rare:
3244 * only if we run lilo or swapon on a freshly made file
3245 * do we expect this to happen.
3247 * (bmap requires CAP_SYS_RAWIO so this does not
3248 * represent an unprivileged user DOS attack --- we'd be
3249 * in trouble if mortal users could trigger this path at
3252 * NB. EXT4_STATE_JDATA is not set on files other than
3253 * regular files. If somebody wants to bmap a directory
3254 * or symlink and gets confused because the buffer
3255 * hasn't yet been flushed to disk, they deserve
3256 * everything they get.
3259 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3260 journal
= EXT4_JOURNAL(inode
);
3261 jbd2_journal_lock_updates(journal
);
3262 err
= jbd2_journal_flush(journal
);
3263 jbd2_journal_unlock_updates(journal
);
3269 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3272 static int ext4_readpage(struct file
*file
, struct page
*page
)
3275 struct inode
*inode
= page
->mapping
->host
;
3277 trace_ext4_readpage(page
);
3279 if (ext4_has_inline_data(inode
))
3280 ret
= ext4_readpage_inline(inode
, page
);
3283 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3289 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3290 struct list_head
*pages
, unsigned nr_pages
)
3292 struct inode
*inode
= mapping
->host
;
3294 /* If the file has inline data, no need to do readpages. */
3295 if (ext4_has_inline_data(inode
))
3298 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3301 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3302 unsigned int length
)
3304 trace_ext4_invalidatepage(page
, offset
, length
);
3306 /* No journalling happens on data buffers when this function is used */
3307 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3309 block_invalidatepage(page
, offset
, length
);
3312 static int __ext4_journalled_invalidatepage(struct page
*page
,
3313 unsigned int offset
,
3314 unsigned int length
)
3316 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3318 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3321 * If it's a full truncate we just forget about the pending dirtying
3323 if (offset
== 0 && length
== PAGE_SIZE
)
3324 ClearPageChecked(page
);
3326 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3329 /* Wrapper for aops... */
3330 static void ext4_journalled_invalidatepage(struct page
*page
,
3331 unsigned int offset
,
3332 unsigned int length
)
3334 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3337 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3339 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3341 trace_ext4_releasepage(page
);
3343 /* Page has dirty journalled data -> cannot release */
3344 if (PageChecked(page
))
3347 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3349 return try_to_free_buffers(page
);
3352 #ifdef CONFIG_FS_DAX
3353 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3354 unsigned flags
, struct iomap
*iomap
)
3356 struct block_device
*bdev
;
3357 unsigned int blkbits
= inode
->i_blkbits
;
3358 unsigned long first_block
= offset
>> blkbits
;
3359 unsigned long last_block
= (offset
+ length
- 1) >> blkbits
;
3360 struct ext4_map_blocks map
;
3363 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3366 map
.m_lblk
= first_block
;
3367 map
.m_len
= last_block
- first_block
+ 1;
3369 if (!(flags
& IOMAP_WRITE
)) {
3370 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3376 /* Trim mapping request to maximum we can map at once for DIO */
3377 if (map
.m_len
> DIO_MAX_BLOCKS
)
3378 map
.m_len
= DIO_MAX_BLOCKS
;
3379 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3382 * Either we allocate blocks and then we don't get unwritten
3383 * extent so we have reserved enough credits, or the blocks
3384 * are already allocated and unwritten and in that case
3385 * extent conversion fits in the credits as well.
3387 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3390 return PTR_ERR(handle
);
3392 ret
= ext4_map_blocks(handle
, inode
, &map
,
3393 EXT4_GET_BLOCKS_CREATE_ZERO
);
3395 ext4_journal_stop(handle
);
3396 if (ret
== -ENOSPC
&&
3397 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3403 * If we added blocks beyond i_size, we need to make sure they
3404 * will get truncated if we crash before updating i_size in
3405 * ext4_iomap_end(). For faults we don't need to do that (and
3406 * even cannot because for orphan list operations inode_lock is
3407 * required) - if we happen to instantiate block beyond i_size,
3408 * it is because we race with truncate which has already added
3409 * the inode to the orphan list.
3411 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3412 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3415 err
= ext4_orphan_add(handle
, inode
);
3417 ext4_journal_stop(handle
);
3421 ext4_journal_stop(handle
);
3425 bdev
= inode
->i_sb
->s_bdev
;
3427 if (blk_queue_dax(bdev
->bd_queue
))
3428 iomap
->dax_dev
= fs_dax_get_by_host(bdev
->bd_disk
->disk_name
);
3430 iomap
->dax_dev
= NULL
;
3431 iomap
->offset
= first_block
<< blkbits
;
3434 iomap
->type
= IOMAP_HOLE
;
3435 iomap
->blkno
= IOMAP_NULL_BLOCK
;
3436 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3438 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3439 iomap
->type
= IOMAP_MAPPED
;
3440 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3441 iomap
->type
= IOMAP_UNWRITTEN
;
3446 iomap
->blkno
= (sector_t
)map
.m_pblk
<< (blkbits
- 9);
3447 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3450 if (map
.m_flags
& EXT4_MAP_NEW
)
3451 iomap
->flags
|= IOMAP_F_NEW
;
3455 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3456 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3460 int blkbits
= inode
->i_blkbits
;
3461 bool truncate
= false;
3463 fs_put_dax(iomap
->dax_dev
);
3464 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3467 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3468 if (IS_ERR(handle
)) {
3469 ret
= PTR_ERR(handle
);
3472 if (ext4_update_inode_size(inode
, offset
+ written
))
3473 ext4_mark_inode_dirty(handle
, inode
);
3475 * We may need to truncate allocated but not written blocks beyond EOF.
3477 if (iomap
->offset
+ iomap
->length
>
3478 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3479 ext4_lblk_t written_blk
, end_blk
;
3481 written_blk
= (offset
+ written
) >> blkbits
;
3482 end_blk
= (offset
+ length
) >> blkbits
;
3483 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3487 * Remove inode from orphan list if we were extending a inode and
3488 * everything went fine.
3490 if (!truncate
&& inode
->i_nlink
&&
3491 !list_empty(&EXT4_I(inode
)->i_orphan
))
3492 ext4_orphan_del(handle
, inode
);
3493 ext4_journal_stop(handle
);
3495 ext4_truncate_failed_write(inode
);
3498 * If truncate failed early the inode might still be on the
3499 * orphan list; we need to make sure the inode is removed from
3500 * the orphan list in that case.
3503 ext4_orphan_del(NULL
, inode
);
3508 const struct iomap_ops ext4_iomap_ops
= {
3509 .iomap_begin
= ext4_iomap_begin
,
3510 .iomap_end
= ext4_iomap_end
,
3515 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3516 ssize_t size
, void *private)
3518 ext4_io_end_t
*io_end
= private;
3520 /* if not async direct IO just return */
3524 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3525 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3526 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3529 * Error during AIO DIO. We cannot convert unwritten extents as the
3530 * data was not written. Just clear the unwritten flag and drop io_end.
3533 ext4_clear_io_unwritten_flag(io_end
);
3536 io_end
->offset
= offset
;
3537 io_end
->size
= size
;
3538 ext4_put_io_end(io_end
);
3544 * Handling of direct IO writes.
3546 * For ext4 extent files, ext4 will do direct-io write even to holes,
3547 * preallocated extents, and those write extend the file, no need to
3548 * fall back to buffered IO.
3550 * For holes, we fallocate those blocks, mark them as unwritten
3551 * If those blocks were preallocated, we mark sure they are split, but
3552 * still keep the range to write as unwritten.
3554 * The unwritten extents will be converted to written when DIO is completed.
3555 * For async direct IO, since the IO may still pending when return, we
3556 * set up an end_io call back function, which will do the conversion
3557 * when async direct IO completed.
3559 * If the O_DIRECT write will extend the file then add this inode to the
3560 * orphan list. So recovery will truncate it back to the original size
3561 * if the machine crashes during the write.
3564 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3566 struct file
*file
= iocb
->ki_filp
;
3567 struct inode
*inode
= file
->f_mapping
->host
;
3568 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3570 loff_t offset
= iocb
->ki_pos
;
3571 size_t count
= iov_iter_count(iter
);
3573 get_block_t
*get_block_func
= NULL
;
3575 loff_t final_size
= offset
+ count
;
3579 if (final_size
> inode
->i_size
) {
3580 /* Credits for sb + inode write */
3581 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3582 if (IS_ERR(handle
)) {
3583 ret
= PTR_ERR(handle
);
3586 ret
= ext4_orphan_add(handle
, inode
);
3588 ext4_journal_stop(handle
);
3592 ei
->i_disksize
= inode
->i_size
;
3593 ext4_journal_stop(handle
);
3596 BUG_ON(iocb
->private == NULL
);
3599 * Make all waiters for direct IO properly wait also for extent
3600 * conversion. This also disallows race between truncate() and
3601 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3603 inode_dio_begin(inode
);
3605 /* If we do a overwrite dio, i_mutex locking can be released */
3606 overwrite
= *((int *)iocb
->private);
3609 inode_unlock(inode
);
3612 * For extent mapped files we could direct write to holes and fallocate.
3614 * Allocated blocks to fill the hole are marked as unwritten to prevent
3615 * parallel buffered read to expose the stale data before DIO complete
3618 * As to previously fallocated extents, ext4 get_block will just simply
3619 * mark the buffer mapped but still keep the extents unwritten.
3621 * For non AIO case, we will convert those unwritten extents to written
3622 * after return back from blockdev_direct_IO. That way we save us from
3623 * allocating io_end structure and also the overhead of offloading
3624 * the extent convertion to a workqueue.
3626 * For async DIO, the conversion needs to be deferred when the
3627 * IO is completed. The ext4 end_io callback function will be
3628 * called to take care of the conversion work. Here for async
3629 * case, we allocate an io_end structure to hook to the iocb.
3631 iocb
->private = NULL
;
3633 get_block_func
= ext4_dio_get_block_overwrite
;
3634 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3635 round_down(offset
, i_blocksize(inode
)) >= inode
->i_size
) {
3636 get_block_func
= ext4_dio_get_block
;
3637 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3638 } else if (is_sync_kiocb(iocb
)) {
3639 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3640 dio_flags
= DIO_LOCKING
;
3642 get_block_func
= ext4_dio_get_block_unwritten_async
;
3643 dio_flags
= DIO_LOCKING
;
3645 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3646 get_block_func
, ext4_end_io_dio
, NULL
,
3649 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3650 EXT4_STATE_DIO_UNWRITTEN
)) {
3653 * for non AIO case, since the IO is already
3654 * completed, we could do the conversion right here
3656 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3660 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3663 inode_dio_end(inode
);
3664 /* take i_mutex locking again if we do a ovewrite dio */
3668 if (ret
< 0 && final_size
> inode
->i_size
)
3669 ext4_truncate_failed_write(inode
);
3671 /* Handle extending of i_size after direct IO write */
3675 /* Credits for sb + inode write */
3676 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3677 if (IS_ERR(handle
)) {
3678 /* This is really bad luck. We've written the data
3679 * but cannot extend i_size. Bail out and pretend
3680 * the write failed... */
3681 ret
= PTR_ERR(handle
);
3683 ext4_orphan_del(NULL
, inode
);
3688 ext4_orphan_del(handle
, inode
);
3690 loff_t end
= offset
+ ret
;
3691 if (end
> inode
->i_size
) {
3692 ei
->i_disksize
= end
;
3693 i_size_write(inode
, end
);
3695 * We're going to return a positive `ret'
3696 * here due to non-zero-length I/O, so there's
3697 * no way of reporting error returns from
3698 * ext4_mark_inode_dirty() to userspace. So
3701 ext4_mark_inode_dirty(handle
, inode
);
3704 err
= ext4_journal_stop(handle
);
3712 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3714 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3715 struct inode
*inode
= mapping
->host
;
3716 size_t count
= iov_iter_count(iter
);
3720 * Shared inode_lock is enough for us - it protects against concurrent
3721 * writes & truncates and since we take care of writing back page cache,
3722 * we are protected against page writeback as well.
3724 inode_lock_shared(inode
);
3725 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3726 iocb
->ki_pos
+ count
- 1);
3729 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3730 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3732 inode_unlock_shared(inode
);
3736 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3738 struct file
*file
= iocb
->ki_filp
;
3739 struct inode
*inode
= file
->f_mapping
->host
;
3740 size_t count
= iov_iter_count(iter
);
3741 loff_t offset
= iocb
->ki_pos
;
3744 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3745 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3750 * If we are doing data journalling we don't support O_DIRECT
3752 if (ext4_should_journal_data(inode
))
3755 /* Let buffer I/O handle the inline data case. */
3756 if (ext4_has_inline_data(inode
))
3759 /* DAX uses iomap path now */
3760 if (WARN_ON_ONCE(IS_DAX(inode
)))
3763 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3764 if (iov_iter_rw(iter
) == READ
)
3765 ret
= ext4_direct_IO_read(iocb
, iter
);
3767 ret
= ext4_direct_IO_write(iocb
, iter
);
3768 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3773 * Pages can be marked dirty completely asynchronously from ext4's journalling
3774 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3775 * much here because ->set_page_dirty is called under VFS locks. The page is
3776 * not necessarily locked.
3778 * We cannot just dirty the page and leave attached buffers clean, because the
3779 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3780 * or jbddirty because all the journalling code will explode.
3782 * So what we do is to mark the page "pending dirty" and next time writepage
3783 * is called, propagate that into the buffers appropriately.
3785 static int ext4_journalled_set_page_dirty(struct page
*page
)
3787 SetPageChecked(page
);
3788 return __set_page_dirty_nobuffers(page
);
3791 static int ext4_set_page_dirty(struct page
*page
)
3793 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3794 WARN_ON_ONCE(!page_has_buffers(page
));
3795 return __set_page_dirty_buffers(page
);
3798 static const struct address_space_operations ext4_aops
= {
3799 .readpage
= ext4_readpage
,
3800 .readpages
= ext4_readpages
,
3801 .writepage
= ext4_writepage
,
3802 .writepages
= ext4_writepages
,
3803 .write_begin
= ext4_write_begin
,
3804 .write_end
= ext4_write_end
,
3805 .set_page_dirty
= ext4_set_page_dirty
,
3807 .invalidatepage
= ext4_invalidatepage
,
3808 .releasepage
= ext4_releasepage
,
3809 .direct_IO
= ext4_direct_IO
,
3810 .migratepage
= buffer_migrate_page
,
3811 .is_partially_uptodate
= block_is_partially_uptodate
,
3812 .error_remove_page
= generic_error_remove_page
,
3815 static const struct address_space_operations ext4_journalled_aops
= {
3816 .readpage
= ext4_readpage
,
3817 .readpages
= ext4_readpages
,
3818 .writepage
= ext4_writepage
,
3819 .writepages
= ext4_writepages
,
3820 .write_begin
= ext4_write_begin
,
3821 .write_end
= ext4_journalled_write_end
,
3822 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3824 .invalidatepage
= ext4_journalled_invalidatepage
,
3825 .releasepage
= ext4_releasepage
,
3826 .direct_IO
= ext4_direct_IO
,
3827 .is_partially_uptodate
= block_is_partially_uptodate
,
3828 .error_remove_page
= generic_error_remove_page
,
3831 static const struct address_space_operations ext4_da_aops
= {
3832 .readpage
= ext4_readpage
,
3833 .readpages
= ext4_readpages
,
3834 .writepage
= ext4_writepage
,
3835 .writepages
= ext4_writepages
,
3836 .write_begin
= ext4_da_write_begin
,
3837 .write_end
= ext4_da_write_end
,
3838 .set_page_dirty
= ext4_set_page_dirty
,
3840 .invalidatepage
= ext4_da_invalidatepage
,
3841 .releasepage
= ext4_releasepage
,
3842 .direct_IO
= ext4_direct_IO
,
3843 .migratepage
= buffer_migrate_page
,
3844 .is_partially_uptodate
= block_is_partially_uptodate
,
3845 .error_remove_page
= generic_error_remove_page
,
3848 void ext4_set_aops(struct inode
*inode
)
3850 switch (ext4_inode_journal_mode(inode
)) {
3851 case EXT4_INODE_ORDERED_DATA_MODE
:
3852 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3854 case EXT4_INODE_JOURNAL_DATA_MODE
:
3855 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3860 if (test_opt(inode
->i_sb
, DELALLOC
))
3861 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3863 inode
->i_mapping
->a_ops
= &ext4_aops
;
3866 static int __ext4_block_zero_page_range(handle_t
*handle
,
3867 struct address_space
*mapping
, loff_t from
, loff_t length
)
3869 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3870 unsigned offset
= from
& (PAGE_SIZE
-1);
3871 unsigned blocksize
, pos
;
3873 struct inode
*inode
= mapping
->host
;
3874 struct buffer_head
*bh
;
3878 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3879 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3883 blocksize
= inode
->i_sb
->s_blocksize
;
3885 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3887 if (!page_has_buffers(page
))
3888 create_empty_buffers(page
, blocksize
, 0);
3890 /* Find the buffer that contains "offset" */
3891 bh
= page_buffers(page
);
3893 while (offset
>= pos
) {
3894 bh
= bh
->b_this_page
;
3898 if (buffer_freed(bh
)) {
3899 BUFFER_TRACE(bh
, "freed: skip");
3902 if (!buffer_mapped(bh
)) {
3903 BUFFER_TRACE(bh
, "unmapped");
3904 ext4_get_block(inode
, iblock
, bh
, 0);
3905 /* unmapped? It's a hole - nothing to do */
3906 if (!buffer_mapped(bh
)) {
3907 BUFFER_TRACE(bh
, "still unmapped");
3912 /* Ok, it's mapped. Make sure it's up-to-date */
3913 if (PageUptodate(page
))
3914 set_buffer_uptodate(bh
);
3916 if (!buffer_uptodate(bh
)) {
3918 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3920 /* Uhhuh. Read error. Complain and punt. */
3921 if (!buffer_uptodate(bh
))
3923 if (S_ISREG(inode
->i_mode
) &&
3924 ext4_encrypted_inode(inode
)) {
3925 /* We expect the key to be set. */
3926 BUG_ON(!fscrypt_has_encryption_key(inode
));
3927 BUG_ON(blocksize
!= PAGE_SIZE
);
3928 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
3929 page
, PAGE_SIZE
, 0, page
->index
));
3932 if (ext4_should_journal_data(inode
)) {
3933 BUFFER_TRACE(bh
, "get write access");
3934 err
= ext4_journal_get_write_access(handle
, bh
);
3938 zero_user(page
, offset
, length
);
3939 BUFFER_TRACE(bh
, "zeroed end of block");
3941 if (ext4_should_journal_data(inode
)) {
3942 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3945 mark_buffer_dirty(bh
);
3946 if (ext4_should_order_data(inode
))
3947 err
= ext4_jbd2_inode_add_write(handle
, inode
);
3957 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3958 * starting from file offset 'from'. The range to be zero'd must
3959 * be contained with in one block. If the specified range exceeds
3960 * the end of the block it will be shortened to end of the block
3961 * that cooresponds to 'from'
3963 static int ext4_block_zero_page_range(handle_t
*handle
,
3964 struct address_space
*mapping
, loff_t from
, loff_t length
)
3966 struct inode
*inode
= mapping
->host
;
3967 unsigned offset
= from
& (PAGE_SIZE
-1);
3968 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3969 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3972 * correct length if it does not fall between
3973 * 'from' and the end of the block
3975 if (length
> max
|| length
< 0)
3978 if (IS_DAX(inode
)) {
3979 return iomap_zero_range(inode
, from
, length
, NULL
,
3982 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3986 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3987 * up to the end of the block which corresponds to `from'.
3988 * This required during truncate. We need to physically zero the tail end
3989 * of that block so it doesn't yield old data if the file is later grown.
3991 static int ext4_block_truncate_page(handle_t
*handle
,
3992 struct address_space
*mapping
, loff_t from
)
3994 unsigned offset
= from
& (PAGE_SIZE
-1);
3997 struct inode
*inode
= mapping
->host
;
3999 /* If we are processing an encrypted inode during orphan list handling */
4000 if (ext4_encrypted_inode(inode
) && !fscrypt_has_encryption_key(inode
))
4003 blocksize
= inode
->i_sb
->s_blocksize
;
4004 length
= blocksize
- (offset
& (blocksize
- 1));
4006 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4009 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
4010 loff_t lstart
, loff_t length
)
4012 struct super_block
*sb
= inode
->i_sb
;
4013 struct address_space
*mapping
= inode
->i_mapping
;
4014 unsigned partial_start
, partial_end
;
4015 ext4_fsblk_t start
, end
;
4016 loff_t byte_end
= (lstart
+ length
- 1);
4019 partial_start
= lstart
& (sb
->s_blocksize
- 1);
4020 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
4022 start
= lstart
>> sb
->s_blocksize_bits
;
4023 end
= byte_end
>> sb
->s_blocksize_bits
;
4025 /* Handle partial zero within the single block */
4027 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
4028 err
= ext4_block_zero_page_range(handle
, mapping
,
4032 /* Handle partial zero out on the start of the range */
4033 if (partial_start
) {
4034 err
= ext4_block_zero_page_range(handle
, mapping
,
4035 lstart
, sb
->s_blocksize
);
4039 /* Handle partial zero out on the end of the range */
4040 if (partial_end
!= sb
->s_blocksize
- 1)
4041 err
= ext4_block_zero_page_range(handle
, mapping
,
4042 byte_end
- partial_end
,
4047 int ext4_can_truncate(struct inode
*inode
)
4049 if (S_ISREG(inode
->i_mode
))
4051 if (S_ISDIR(inode
->i_mode
))
4053 if (S_ISLNK(inode
->i_mode
))
4054 return !ext4_inode_is_fast_symlink(inode
);
4059 * We have to make sure i_disksize gets properly updated before we truncate
4060 * page cache due to hole punching or zero range. Otherwise i_disksize update
4061 * can get lost as it may have been postponed to submission of writeback but
4062 * that will never happen after we truncate page cache.
4064 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
4068 loff_t size
= i_size_read(inode
);
4070 WARN_ON(!inode_is_locked(inode
));
4071 if (offset
> size
|| offset
+ len
< size
)
4074 if (EXT4_I(inode
)->i_disksize
>= size
)
4077 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4079 return PTR_ERR(handle
);
4080 ext4_update_i_disksize(inode
, size
);
4081 ext4_mark_inode_dirty(handle
, inode
);
4082 ext4_journal_stop(handle
);
4088 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4089 * associated with the given offset and length
4091 * @inode: File inode
4092 * @offset: The offset where the hole will begin
4093 * @len: The length of the hole
4095 * Returns: 0 on success or negative on failure
4098 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4100 struct super_block
*sb
= inode
->i_sb
;
4101 ext4_lblk_t first_block
, stop_block
;
4102 struct address_space
*mapping
= inode
->i_mapping
;
4103 loff_t first_block_offset
, last_block_offset
;
4105 unsigned int credits
;
4108 if (!S_ISREG(inode
->i_mode
))
4111 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4114 * Write out all dirty pages to avoid race conditions
4115 * Then release them.
4117 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4118 ret
= filemap_write_and_wait_range(mapping
, offset
,
4119 offset
+ length
- 1);
4126 /* No need to punch hole beyond i_size */
4127 if (offset
>= inode
->i_size
)
4131 * If the hole extends beyond i_size, set the hole
4132 * to end after the page that contains i_size
4134 if (offset
+ length
> inode
->i_size
) {
4135 length
= inode
->i_size
+
4136 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4140 if (offset
& (sb
->s_blocksize
- 1) ||
4141 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4143 * Attach jinode to inode for jbd2 if we do any zeroing of
4146 ret
= ext4_inode_attach_jinode(inode
);
4152 /* Wait all existing dio workers, newcomers will block on i_mutex */
4153 ext4_inode_block_unlocked_dio(inode
);
4154 inode_dio_wait(inode
);
4157 * Prevent page faults from reinstantiating pages we have released from
4160 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4161 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4162 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4164 /* Now release the pages and zero block aligned part of pages*/
4165 if (last_block_offset
> first_block_offset
) {
4166 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4169 truncate_pagecache_range(inode
, first_block_offset
,
4173 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4174 credits
= ext4_writepage_trans_blocks(inode
);
4176 credits
= ext4_blocks_for_truncate(inode
);
4177 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4178 if (IS_ERR(handle
)) {
4179 ret
= PTR_ERR(handle
);
4180 ext4_std_error(sb
, ret
);
4184 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4189 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4190 EXT4_BLOCK_SIZE_BITS(sb
);
4191 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4193 /* If there are no blocks to remove, return now */
4194 if (first_block
>= stop_block
)
4197 down_write(&EXT4_I(inode
)->i_data_sem
);
4198 ext4_discard_preallocations(inode
);
4200 ret
= ext4_es_remove_extent(inode
, first_block
,
4201 stop_block
- first_block
);
4203 up_write(&EXT4_I(inode
)->i_data_sem
);
4207 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4208 ret
= ext4_ext_remove_space(inode
, first_block
,
4211 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4214 up_write(&EXT4_I(inode
)->i_data_sem
);
4216 ext4_handle_sync(handle
);
4218 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4219 ext4_mark_inode_dirty(handle
, inode
);
4221 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4223 ext4_journal_stop(handle
);
4225 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4226 ext4_inode_resume_unlocked_dio(inode
);
4228 inode_unlock(inode
);
4232 int ext4_inode_attach_jinode(struct inode
*inode
)
4234 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4235 struct jbd2_inode
*jinode
;
4237 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4240 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4241 spin_lock(&inode
->i_lock
);
4244 spin_unlock(&inode
->i_lock
);
4247 ei
->jinode
= jinode
;
4248 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4251 spin_unlock(&inode
->i_lock
);
4252 if (unlikely(jinode
!= NULL
))
4253 jbd2_free_inode(jinode
);
4260 * We block out ext4_get_block() block instantiations across the entire
4261 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4262 * simultaneously on behalf of the same inode.
4264 * As we work through the truncate and commit bits of it to the journal there
4265 * is one core, guiding principle: the file's tree must always be consistent on
4266 * disk. We must be able to restart the truncate after a crash.
4268 * The file's tree may be transiently inconsistent in memory (although it
4269 * probably isn't), but whenever we close off and commit a journal transaction,
4270 * the contents of (the filesystem + the journal) must be consistent and
4271 * restartable. It's pretty simple, really: bottom up, right to left (although
4272 * left-to-right works OK too).
4274 * Note that at recovery time, journal replay occurs *before* the restart of
4275 * truncate against the orphan inode list.
4277 * The committed inode has the new, desired i_size (which is the same as
4278 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4279 * that this inode's truncate did not complete and it will again call
4280 * ext4_truncate() to have another go. So there will be instantiated blocks
4281 * to the right of the truncation point in a crashed ext4 filesystem. But
4282 * that's fine - as long as they are linked from the inode, the post-crash
4283 * ext4_truncate() run will find them and release them.
4285 int ext4_truncate(struct inode
*inode
)
4287 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4288 unsigned int credits
;
4291 struct address_space
*mapping
= inode
->i_mapping
;
4294 * There is a possibility that we're either freeing the inode
4295 * or it's a completely new inode. In those cases we might not
4296 * have i_mutex locked because it's not necessary.
4298 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4299 WARN_ON(!inode_is_locked(inode
));
4300 trace_ext4_truncate_enter(inode
);
4302 if (!ext4_can_truncate(inode
))
4305 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4307 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4308 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4310 if (ext4_has_inline_data(inode
)) {
4313 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4320 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4321 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4322 if (ext4_inode_attach_jinode(inode
) < 0)
4326 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4327 credits
= ext4_writepage_trans_blocks(inode
);
4329 credits
= ext4_blocks_for_truncate(inode
);
4331 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4333 return PTR_ERR(handle
);
4335 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4336 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4339 * We add the inode to the orphan list, so that if this
4340 * truncate spans multiple transactions, and we crash, we will
4341 * resume the truncate when the filesystem recovers. It also
4342 * marks the inode dirty, to catch the new size.
4344 * Implication: the file must always be in a sane, consistent
4345 * truncatable state while each transaction commits.
4347 err
= ext4_orphan_add(handle
, inode
);
4351 down_write(&EXT4_I(inode
)->i_data_sem
);
4353 ext4_discard_preallocations(inode
);
4355 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4356 err
= ext4_ext_truncate(handle
, inode
);
4358 ext4_ind_truncate(handle
, inode
);
4360 up_write(&ei
->i_data_sem
);
4365 ext4_handle_sync(handle
);
4369 * If this was a simple ftruncate() and the file will remain alive,
4370 * then we need to clear up the orphan record which we created above.
4371 * However, if this was a real unlink then we were called by
4372 * ext4_evict_inode(), and we allow that function to clean up the
4373 * orphan info for us.
4376 ext4_orphan_del(handle
, inode
);
4378 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4379 ext4_mark_inode_dirty(handle
, inode
);
4380 ext4_journal_stop(handle
);
4382 trace_ext4_truncate_exit(inode
);
4387 * ext4_get_inode_loc returns with an extra refcount against the inode's
4388 * underlying buffer_head on success. If 'in_mem' is true, we have all
4389 * data in memory that is needed to recreate the on-disk version of this
4392 static int __ext4_get_inode_loc(struct inode
*inode
,
4393 struct ext4_iloc
*iloc
, int in_mem
)
4395 struct ext4_group_desc
*gdp
;
4396 struct buffer_head
*bh
;
4397 struct super_block
*sb
= inode
->i_sb
;
4399 int inodes_per_block
, inode_offset
;
4402 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4403 return -EFSCORRUPTED
;
4405 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4406 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4411 * Figure out the offset within the block group inode table
4413 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4414 inode_offset
= ((inode
->i_ino
- 1) %
4415 EXT4_INODES_PER_GROUP(sb
));
4416 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4417 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4419 bh
= sb_getblk(sb
, block
);
4422 if (!buffer_uptodate(bh
)) {
4426 * If the buffer has the write error flag, we have failed
4427 * to write out another inode in the same block. In this
4428 * case, we don't have to read the block because we may
4429 * read the old inode data successfully.
4431 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4432 set_buffer_uptodate(bh
);
4434 if (buffer_uptodate(bh
)) {
4435 /* someone brought it uptodate while we waited */
4441 * If we have all information of the inode in memory and this
4442 * is the only valid inode in the block, we need not read the
4446 struct buffer_head
*bitmap_bh
;
4449 start
= inode_offset
& ~(inodes_per_block
- 1);
4451 /* Is the inode bitmap in cache? */
4452 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4453 if (unlikely(!bitmap_bh
))
4457 * If the inode bitmap isn't in cache then the
4458 * optimisation may end up performing two reads instead
4459 * of one, so skip it.
4461 if (!buffer_uptodate(bitmap_bh
)) {
4465 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4466 if (i
== inode_offset
)
4468 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4472 if (i
== start
+ inodes_per_block
) {
4473 /* all other inodes are free, so skip I/O */
4474 memset(bh
->b_data
, 0, bh
->b_size
);
4475 set_buffer_uptodate(bh
);
4483 * If we need to do any I/O, try to pre-readahead extra
4484 * blocks from the inode table.
4486 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4487 ext4_fsblk_t b
, end
, table
;
4489 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4491 table
= ext4_inode_table(sb
, gdp
);
4492 /* s_inode_readahead_blks is always a power of 2 */
4493 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4497 num
= EXT4_INODES_PER_GROUP(sb
);
4498 if (ext4_has_group_desc_csum(sb
))
4499 num
-= ext4_itable_unused_count(sb
, gdp
);
4500 table
+= num
/ inodes_per_block
;
4504 sb_breadahead(sb
, b
++);
4508 * There are other valid inodes in the buffer, this inode
4509 * has in-inode xattrs, or we don't have this inode in memory.
4510 * Read the block from disk.
4512 trace_ext4_load_inode(inode
);
4514 bh
->b_end_io
= end_buffer_read_sync
;
4515 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4517 if (!buffer_uptodate(bh
)) {
4518 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4519 "unable to read itable block");
4529 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4531 /* We have all inode data except xattrs in memory here. */
4532 return __ext4_get_inode_loc(inode
, iloc
,
4533 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4536 void ext4_set_inode_flags(struct inode
*inode
)
4538 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4539 unsigned int new_fl
= 0;
4541 if (flags
& EXT4_SYNC_FL
)
4543 if (flags
& EXT4_APPEND_FL
)
4545 if (flags
& EXT4_IMMUTABLE_FL
)
4546 new_fl
|= S_IMMUTABLE
;
4547 if (flags
& EXT4_NOATIME_FL
)
4548 new_fl
|= S_NOATIME
;
4549 if (flags
& EXT4_DIRSYNC_FL
)
4550 new_fl
|= S_DIRSYNC
;
4551 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
) &&
4552 !ext4_should_journal_data(inode
) && !ext4_has_inline_data(inode
) &&
4553 !ext4_encrypted_inode(inode
))
4555 inode_set_flags(inode
, new_fl
,
4556 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4559 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4560 struct ext4_inode_info
*ei
)
4563 struct inode
*inode
= &(ei
->vfs_inode
);
4564 struct super_block
*sb
= inode
->i_sb
;
4566 if (ext4_has_feature_huge_file(sb
)) {
4567 /* we are using combined 48 bit field */
4568 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4569 le32_to_cpu(raw_inode
->i_blocks_lo
);
4570 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4571 /* i_blocks represent file system block size */
4572 return i_blocks
<< (inode
->i_blkbits
- 9);
4577 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4581 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4582 struct ext4_inode
*raw_inode
,
4583 struct ext4_inode_info
*ei
)
4585 __le32
*magic
= (void *)raw_inode
+
4586 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4587 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4588 EXT4_INODE_SIZE(inode
->i_sb
) &&
4589 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4590 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4591 ext4_find_inline_data_nolock(inode
);
4593 EXT4_I(inode
)->i_inline_off
= 0;
4596 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4598 if (!ext4_has_feature_project(inode
->i_sb
))
4600 *projid
= EXT4_I(inode
)->i_projid
;
4604 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4606 struct ext4_iloc iloc
;
4607 struct ext4_inode
*raw_inode
;
4608 struct ext4_inode_info
*ei
;
4609 struct inode
*inode
;
4610 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4618 inode
= iget_locked(sb
, ino
);
4620 return ERR_PTR(-ENOMEM
);
4621 if (!(inode
->i_state
& I_NEW
))
4627 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4630 raw_inode
= ext4_raw_inode(&iloc
);
4632 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4633 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4634 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4635 EXT4_INODE_SIZE(inode
->i_sb
) ||
4636 (ei
->i_extra_isize
& 3)) {
4637 EXT4_ERROR_INODE(inode
,
4638 "bad extra_isize %u (inode size %u)",
4640 EXT4_INODE_SIZE(inode
->i_sb
));
4641 ret
= -EFSCORRUPTED
;
4645 ei
->i_extra_isize
= 0;
4647 /* Precompute checksum seed for inode metadata */
4648 if (ext4_has_metadata_csum(sb
)) {
4649 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4651 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4652 __le32 gen
= raw_inode
->i_generation
;
4653 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4655 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4659 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4660 EXT4_ERROR_INODE(inode
, "checksum invalid");
4665 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4666 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4667 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4668 if (ext4_has_feature_project(sb
) &&
4669 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4670 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4671 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4673 i_projid
= EXT4_DEF_PROJID
;
4675 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4676 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4677 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4679 i_uid_write(inode
, i_uid
);
4680 i_gid_write(inode
, i_gid
);
4681 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4682 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4684 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4685 ei
->i_inline_off
= 0;
4686 ei
->i_dir_start_lookup
= 0;
4687 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4688 /* We now have enough fields to check if the inode was active or not.
4689 * This is needed because nfsd might try to access dead inodes
4690 * the test is that same one that e2fsck uses
4691 * NeilBrown 1999oct15
4693 if (inode
->i_nlink
== 0) {
4694 if ((inode
->i_mode
== 0 ||
4695 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4696 ino
!= EXT4_BOOT_LOADER_INO
) {
4697 /* this inode is deleted */
4701 /* The only unlinked inodes we let through here have
4702 * valid i_mode and are being read by the orphan
4703 * recovery code: that's fine, we're about to complete
4704 * the process of deleting those.
4705 * OR it is the EXT4_BOOT_LOADER_INO which is
4706 * not initialized on a new filesystem. */
4708 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4709 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4710 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4711 if (ext4_has_feature_64bit(sb
))
4713 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4714 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4715 if ((size
= i_size_read(inode
)) < 0) {
4716 EXT4_ERROR_INODE(inode
, "bad i_size value: %lld", size
);
4717 ret
= -EFSCORRUPTED
;
4720 ei
->i_disksize
= inode
->i_size
;
4722 ei
->i_reserved_quota
= 0;
4724 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4725 ei
->i_block_group
= iloc
.block_group
;
4726 ei
->i_last_alloc_group
= ~0;
4728 * NOTE! The in-memory inode i_data array is in little-endian order
4729 * even on big-endian machines: we do NOT byteswap the block numbers!
4731 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4732 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4733 INIT_LIST_HEAD(&ei
->i_orphan
);
4736 * Set transaction id's of transactions that have to be committed
4737 * to finish f[data]sync. We set them to currently running transaction
4738 * as we cannot be sure that the inode or some of its metadata isn't
4739 * part of the transaction - the inode could have been reclaimed and
4740 * now it is reread from disk.
4743 transaction_t
*transaction
;
4746 read_lock(&journal
->j_state_lock
);
4747 if (journal
->j_running_transaction
)
4748 transaction
= journal
->j_running_transaction
;
4750 transaction
= journal
->j_committing_transaction
;
4752 tid
= transaction
->t_tid
;
4754 tid
= journal
->j_commit_sequence
;
4755 read_unlock(&journal
->j_state_lock
);
4756 ei
->i_sync_tid
= tid
;
4757 ei
->i_datasync_tid
= tid
;
4760 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4761 if (ei
->i_extra_isize
== 0) {
4762 /* The extra space is currently unused. Use it. */
4763 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4764 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4765 EXT4_GOOD_OLD_INODE_SIZE
;
4767 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4771 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4772 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4773 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4774 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4776 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4777 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4778 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4779 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4781 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4786 if (ei
->i_file_acl
&&
4787 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4788 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4790 ret
= -EFSCORRUPTED
;
4792 } else if (!ext4_has_inline_data(inode
)) {
4793 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4794 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4795 (S_ISLNK(inode
->i_mode
) &&
4796 !ext4_inode_is_fast_symlink(inode
))))
4797 /* Validate extent which is part of inode */
4798 ret
= ext4_ext_check_inode(inode
);
4799 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4800 (S_ISLNK(inode
->i_mode
) &&
4801 !ext4_inode_is_fast_symlink(inode
))) {
4802 /* Validate block references which are part of inode */
4803 ret
= ext4_ind_check_inode(inode
);
4809 if (S_ISREG(inode
->i_mode
)) {
4810 inode
->i_op
= &ext4_file_inode_operations
;
4811 inode
->i_fop
= &ext4_file_operations
;
4812 ext4_set_aops(inode
);
4813 } else if (S_ISDIR(inode
->i_mode
)) {
4814 inode
->i_op
= &ext4_dir_inode_operations
;
4815 inode
->i_fop
= &ext4_dir_operations
;
4816 } else if (S_ISLNK(inode
->i_mode
)) {
4817 if (ext4_encrypted_inode(inode
)) {
4818 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4819 ext4_set_aops(inode
);
4820 } else if (ext4_inode_is_fast_symlink(inode
)) {
4821 inode
->i_link
= (char *)ei
->i_data
;
4822 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4823 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4824 sizeof(ei
->i_data
) - 1);
4826 inode
->i_op
= &ext4_symlink_inode_operations
;
4827 ext4_set_aops(inode
);
4829 inode_nohighmem(inode
);
4830 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4831 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4832 inode
->i_op
= &ext4_special_inode_operations
;
4833 if (raw_inode
->i_block
[0])
4834 init_special_inode(inode
, inode
->i_mode
,
4835 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4837 init_special_inode(inode
, inode
->i_mode
,
4838 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4839 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4840 make_bad_inode(inode
);
4842 ret
= -EFSCORRUPTED
;
4843 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4847 ext4_set_inode_flags(inode
);
4849 if (ei
->i_flags
& EXT4_EA_INODE_FL
) {
4850 ext4_xattr_inode_set_class(inode
);
4853 inode
->i_flags
|= S_NOQUOTA
;
4854 inode_unlock(inode
);
4857 unlock_new_inode(inode
);
4863 return ERR_PTR(ret
);
4866 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4868 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4869 return ERR_PTR(-EFSCORRUPTED
);
4870 return ext4_iget(sb
, ino
);
4873 static int ext4_inode_blocks_set(handle_t
*handle
,
4874 struct ext4_inode
*raw_inode
,
4875 struct ext4_inode_info
*ei
)
4877 struct inode
*inode
= &(ei
->vfs_inode
);
4878 u64 i_blocks
= inode
->i_blocks
;
4879 struct super_block
*sb
= inode
->i_sb
;
4881 if (i_blocks
<= ~0U) {
4883 * i_blocks can be represented in a 32 bit variable
4884 * as multiple of 512 bytes
4886 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4887 raw_inode
->i_blocks_high
= 0;
4888 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4891 if (!ext4_has_feature_huge_file(sb
))
4894 if (i_blocks
<= 0xffffffffffffULL
) {
4896 * i_blocks can be represented in a 48 bit variable
4897 * as multiple of 512 bytes
4899 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4900 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4901 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4903 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4904 /* i_block is stored in file system block size */
4905 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4906 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4907 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4912 struct other_inode
{
4913 unsigned long orig_ino
;
4914 struct ext4_inode
*raw_inode
;
4917 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4920 struct other_inode
*oi
= (struct other_inode
*) data
;
4922 if ((inode
->i_ino
!= ino
) ||
4923 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4924 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4925 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4927 spin_lock(&inode
->i_lock
);
4928 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4929 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4930 (inode
->i_state
& I_DIRTY_TIME
)) {
4931 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4933 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4934 spin_unlock(&inode
->i_lock
);
4936 spin_lock(&ei
->i_raw_lock
);
4937 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4938 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4939 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4940 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4941 spin_unlock(&ei
->i_raw_lock
);
4942 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4945 spin_unlock(&inode
->i_lock
);
4950 * Opportunistically update the other time fields for other inodes in
4951 * the same inode table block.
4953 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4954 unsigned long orig_ino
, char *buf
)
4956 struct other_inode oi
;
4958 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4959 int inode_size
= EXT4_INODE_SIZE(sb
);
4961 oi
.orig_ino
= orig_ino
;
4963 * Calculate the first inode in the inode table block. Inode
4964 * numbers are one-based. That is, the first inode in a block
4965 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4967 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4968 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4969 if (ino
== orig_ino
)
4971 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4972 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4977 * Post the struct inode info into an on-disk inode location in the
4978 * buffer-cache. This gobbles the caller's reference to the
4979 * buffer_head in the inode location struct.
4981 * The caller must have write access to iloc->bh.
4983 static int ext4_do_update_inode(handle_t
*handle
,
4984 struct inode
*inode
,
4985 struct ext4_iloc
*iloc
)
4987 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4988 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4989 struct buffer_head
*bh
= iloc
->bh
;
4990 struct super_block
*sb
= inode
->i_sb
;
4991 int err
= 0, rc
, block
;
4992 int need_datasync
= 0, set_large_file
= 0;
4997 spin_lock(&ei
->i_raw_lock
);
4999 /* For fields not tracked in the in-memory inode,
5000 * initialise them to zero for new inodes. */
5001 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5002 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5004 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5005 i_uid
= i_uid_read(inode
);
5006 i_gid
= i_gid_read(inode
);
5007 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
5008 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5009 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
5010 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
5012 * Fix up interoperability with old kernels. Otherwise, old inodes get
5013 * re-used with the upper 16 bits of the uid/gid intact
5015 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
5016 raw_inode
->i_uid_high
= 0;
5017 raw_inode
->i_gid_high
= 0;
5019 raw_inode
->i_uid_high
=
5020 cpu_to_le16(high_16_bits(i_uid
));
5021 raw_inode
->i_gid_high
=
5022 cpu_to_le16(high_16_bits(i_gid
));
5025 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
5026 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
5027 raw_inode
->i_uid_high
= 0;
5028 raw_inode
->i_gid_high
= 0;
5030 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5032 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5033 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5034 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5035 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5037 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5039 spin_unlock(&ei
->i_raw_lock
);
5042 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5043 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5044 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5045 raw_inode
->i_file_acl_high
=
5046 cpu_to_le16(ei
->i_file_acl
>> 32);
5047 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5048 if (ei
->i_disksize
!= ext4_isize(inode
->i_sb
, raw_inode
)) {
5049 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5052 if (ei
->i_disksize
> 0x7fffffffULL
) {
5053 if (!ext4_has_feature_large_file(sb
) ||
5054 EXT4_SB(sb
)->s_es
->s_rev_level
==
5055 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5058 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5059 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5060 if (old_valid_dev(inode
->i_rdev
)) {
5061 raw_inode
->i_block
[0] =
5062 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5063 raw_inode
->i_block
[1] = 0;
5065 raw_inode
->i_block
[0] = 0;
5066 raw_inode
->i_block
[1] =
5067 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5068 raw_inode
->i_block
[2] = 0;
5070 } else if (!ext4_has_inline_data(inode
)) {
5071 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5072 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5075 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5076 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5077 if (ei
->i_extra_isize
) {
5078 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5079 raw_inode
->i_version_hi
=
5080 cpu_to_le32(inode
->i_version
>> 32);
5081 raw_inode
->i_extra_isize
=
5082 cpu_to_le16(ei
->i_extra_isize
);
5086 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5087 i_projid
!= EXT4_DEF_PROJID
);
5089 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5090 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5091 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5093 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5094 spin_unlock(&ei
->i_raw_lock
);
5095 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
5096 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5099 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5100 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5103 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5104 if (set_large_file
) {
5105 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5106 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5109 ext4_update_dynamic_rev(sb
);
5110 ext4_set_feature_large_file(sb
);
5111 ext4_handle_sync(handle
);
5112 err
= ext4_handle_dirty_super(handle
, sb
);
5114 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5117 ext4_std_error(inode
->i_sb
, err
);
5122 * ext4_write_inode()
5124 * We are called from a few places:
5126 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5127 * Here, there will be no transaction running. We wait for any running
5128 * transaction to commit.
5130 * - Within flush work (sys_sync(), kupdate and such).
5131 * We wait on commit, if told to.
5133 * - Within iput_final() -> write_inode_now()
5134 * We wait on commit, if told to.
5136 * In all cases it is actually safe for us to return without doing anything,
5137 * because the inode has been copied into a raw inode buffer in
5138 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5141 * Note that we are absolutely dependent upon all inode dirtiers doing the
5142 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5143 * which we are interested.
5145 * It would be a bug for them to not do this. The code:
5147 * mark_inode_dirty(inode)
5149 * inode->i_size = expr;
5151 * is in error because write_inode() could occur while `stuff()' is running,
5152 * and the new i_size will be lost. Plus the inode will no longer be on the
5153 * superblock's dirty inode list.
5155 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5159 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
5162 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5163 if (ext4_journal_current_handle()) {
5164 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5170 * No need to force transaction in WB_SYNC_NONE mode. Also
5171 * ext4_sync_fs() will force the commit after everything is
5174 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5177 err
= ext4_force_commit(inode
->i_sb
);
5179 struct ext4_iloc iloc
;
5181 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5185 * sync(2) will flush the whole buffer cache. No need to do
5186 * it here separately for each inode.
5188 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5189 sync_dirty_buffer(iloc
.bh
);
5190 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5191 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5192 "IO error syncing inode");
5201 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5202 * buffers that are attached to a page stradding i_size and are undergoing
5203 * commit. In that case we have to wait for commit to finish and try again.
5205 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5209 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5210 tid_t commit_tid
= 0;
5213 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5215 * All buffers in the last page remain valid? Then there's nothing to
5216 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5219 if (offset
> PAGE_SIZE
- i_blocksize(inode
))
5222 page
= find_lock_page(inode
->i_mapping
,
5223 inode
->i_size
>> PAGE_SHIFT
);
5226 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5227 PAGE_SIZE
- offset
);
5233 read_lock(&journal
->j_state_lock
);
5234 if (journal
->j_committing_transaction
)
5235 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5236 read_unlock(&journal
->j_state_lock
);
5238 jbd2_log_wait_commit(journal
, commit_tid
);
5245 * Called from notify_change.
5247 * We want to trap VFS attempts to truncate the file as soon as
5248 * possible. In particular, we want to make sure that when the VFS
5249 * shrinks i_size, we put the inode on the orphan list and modify
5250 * i_disksize immediately, so that during the subsequent flushing of
5251 * dirty pages and freeing of disk blocks, we can guarantee that any
5252 * commit will leave the blocks being flushed in an unused state on
5253 * disk. (On recovery, the inode will get truncated and the blocks will
5254 * be freed, so we have a strong guarantee that no future commit will
5255 * leave these blocks visible to the user.)
5257 * Another thing we have to assure is that if we are in ordered mode
5258 * and inode is still attached to the committing transaction, we must
5259 * we start writeout of all the dirty pages which are being truncated.
5260 * This way we are sure that all the data written in the previous
5261 * transaction are already on disk (truncate waits for pages under
5264 * Called with inode->i_mutex down.
5266 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5268 struct inode
*inode
= d_inode(dentry
);
5271 const unsigned int ia_valid
= attr
->ia_valid
;
5273 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5276 error
= setattr_prepare(dentry
, attr
);
5280 if (is_quota_modification(inode
, attr
)) {
5281 error
= dquot_initialize(inode
);
5285 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5286 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5289 /* (user+group)*(old+new) structure, inode write (sb,
5290 * inode block, ? - but truncate inode update has it) */
5291 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5292 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5293 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5294 if (IS_ERR(handle
)) {
5295 error
= PTR_ERR(handle
);
5298 error
= dquot_transfer(inode
, attr
);
5300 ext4_journal_stop(handle
);
5303 /* Update corresponding info in inode so that everything is in
5304 * one transaction */
5305 if (attr
->ia_valid
& ATTR_UID
)
5306 inode
->i_uid
= attr
->ia_uid
;
5307 if (attr
->ia_valid
& ATTR_GID
)
5308 inode
->i_gid
= attr
->ia_gid
;
5309 error
= ext4_mark_inode_dirty(handle
, inode
);
5310 ext4_journal_stop(handle
);
5313 if (attr
->ia_valid
& ATTR_SIZE
) {
5315 loff_t oldsize
= inode
->i_size
;
5316 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5318 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5319 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5321 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5324 if (!S_ISREG(inode
->i_mode
))
5327 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5328 inode_inc_iversion(inode
);
5330 if (ext4_should_order_data(inode
) &&
5331 (attr
->ia_size
< inode
->i_size
)) {
5332 error
= ext4_begin_ordered_truncate(inode
,
5337 if (attr
->ia_size
!= inode
->i_size
) {
5338 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5339 if (IS_ERR(handle
)) {
5340 error
= PTR_ERR(handle
);
5343 if (ext4_handle_valid(handle
) && shrink
) {
5344 error
= ext4_orphan_add(handle
, inode
);
5348 * Update c/mtime on truncate up, ext4_truncate() will
5349 * update c/mtime in shrink case below
5352 inode
->i_mtime
= current_time(inode
);
5353 inode
->i_ctime
= inode
->i_mtime
;
5355 down_write(&EXT4_I(inode
)->i_data_sem
);
5356 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5357 rc
= ext4_mark_inode_dirty(handle
, inode
);
5361 * We have to update i_size under i_data_sem together
5362 * with i_disksize to avoid races with writeback code
5363 * running ext4_wb_update_i_disksize().
5366 i_size_write(inode
, attr
->ia_size
);
5367 up_write(&EXT4_I(inode
)->i_data_sem
);
5368 ext4_journal_stop(handle
);
5371 ext4_orphan_del(NULL
, inode
);
5376 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5379 * Blocks are going to be removed from the inode. Wait
5380 * for dio in flight. Temporarily disable
5381 * dioread_nolock to prevent livelock.
5384 if (!ext4_should_journal_data(inode
)) {
5385 ext4_inode_block_unlocked_dio(inode
);
5386 inode_dio_wait(inode
);
5387 ext4_inode_resume_unlocked_dio(inode
);
5389 ext4_wait_for_tail_page_commit(inode
);
5391 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5393 * Truncate pagecache after we've waited for commit
5394 * in data=journal mode to make pages freeable.
5396 truncate_pagecache(inode
, inode
->i_size
);
5398 rc
= ext4_truncate(inode
);
5402 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5406 setattr_copy(inode
, attr
);
5407 mark_inode_dirty(inode
);
5411 * If the call to ext4_truncate failed to get a transaction handle at
5412 * all, we need to clean up the in-core orphan list manually.
5414 if (orphan
&& inode
->i_nlink
)
5415 ext4_orphan_del(NULL
, inode
);
5417 if (!error
&& (ia_valid
& ATTR_MODE
))
5418 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5421 ext4_std_error(inode
->i_sb
, error
);
5427 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5428 u32 request_mask
, unsigned int query_flags
)
5430 struct inode
*inode
= d_inode(path
->dentry
);
5431 struct ext4_inode
*raw_inode
;
5432 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5435 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5436 stat
->result_mask
|= STATX_BTIME
;
5437 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5438 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5441 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5442 if (flags
& EXT4_APPEND_FL
)
5443 stat
->attributes
|= STATX_ATTR_APPEND
;
5444 if (flags
& EXT4_COMPR_FL
)
5445 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5446 if (flags
& EXT4_ENCRYPT_FL
)
5447 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5448 if (flags
& EXT4_IMMUTABLE_FL
)
5449 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5450 if (flags
& EXT4_NODUMP_FL
)
5451 stat
->attributes
|= STATX_ATTR_NODUMP
;
5453 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5454 STATX_ATTR_COMPRESSED
|
5455 STATX_ATTR_ENCRYPTED
|
5456 STATX_ATTR_IMMUTABLE
|
5459 generic_fillattr(inode
, stat
);
5463 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5464 u32 request_mask
, unsigned int query_flags
)
5466 struct inode
*inode
= d_inode(path
->dentry
);
5467 u64 delalloc_blocks
;
5469 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5472 * If there is inline data in the inode, the inode will normally not
5473 * have data blocks allocated (it may have an external xattr block).
5474 * Report at least one sector for such files, so tools like tar, rsync,
5475 * others don't incorrectly think the file is completely sparse.
5477 if (unlikely(ext4_has_inline_data(inode
)))
5478 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5481 * We can't update i_blocks if the block allocation is delayed
5482 * otherwise in the case of system crash before the real block
5483 * allocation is done, we will have i_blocks inconsistent with
5484 * on-disk file blocks.
5485 * We always keep i_blocks updated together with real
5486 * allocation. But to not confuse with user, stat
5487 * will return the blocks that include the delayed allocation
5488 * blocks for this file.
5490 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5491 EXT4_I(inode
)->i_reserved_data_blocks
);
5492 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5496 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5499 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5500 return ext4_ind_trans_blocks(inode
, lblocks
);
5501 return ext4_ext_index_trans_blocks(inode
, pextents
);
5505 * Account for index blocks, block groups bitmaps and block group
5506 * descriptor blocks if modify datablocks and index blocks
5507 * worse case, the indexs blocks spread over different block groups
5509 * If datablocks are discontiguous, they are possible to spread over
5510 * different block groups too. If they are contiguous, with flexbg,
5511 * they could still across block group boundary.
5513 * Also account for superblock, inode, quota and xattr blocks
5515 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5518 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5524 * How many index blocks need to touch to map @lblocks logical blocks
5525 * to @pextents physical extents?
5527 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5532 * Now let's see how many group bitmaps and group descriptors need
5535 groups
= idxblocks
+ pextents
;
5537 if (groups
> ngroups
)
5539 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5540 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5542 /* bitmaps and block group descriptor blocks */
5543 ret
+= groups
+ gdpblocks
;
5545 /* Blocks for super block, inode, quota and xattr blocks */
5546 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5552 * Calculate the total number of credits to reserve to fit
5553 * the modification of a single pages into a single transaction,
5554 * which may include multiple chunks of block allocations.
5556 * This could be called via ext4_write_begin()
5558 * We need to consider the worse case, when
5559 * one new block per extent.
5561 int ext4_writepage_trans_blocks(struct inode
*inode
)
5563 int bpp
= ext4_journal_blocks_per_page(inode
);
5566 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5568 /* Account for data blocks for journalled mode */
5569 if (ext4_should_journal_data(inode
))
5575 * Calculate the journal credits for a chunk of data modification.
5577 * This is called from DIO, fallocate or whoever calling
5578 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5580 * journal buffers for data blocks are not included here, as DIO
5581 * and fallocate do no need to journal data buffers.
5583 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5585 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5589 * The caller must have previously called ext4_reserve_inode_write().
5590 * Give this, we know that the caller already has write access to iloc->bh.
5592 int ext4_mark_iloc_dirty(handle_t
*handle
,
5593 struct inode
*inode
, struct ext4_iloc
*iloc
)
5597 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5600 if (IS_I_VERSION(inode
))
5601 inode_inc_iversion(inode
);
5603 /* the do_update_inode consumes one bh->b_count */
5606 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5607 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5613 * On success, We end up with an outstanding reference count against
5614 * iloc->bh. This _must_ be cleaned up later.
5618 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5619 struct ext4_iloc
*iloc
)
5623 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5626 err
= ext4_get_inode_loc(inode
, iloc
);
5628 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5629 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5635 ext4_std_error(inode
->i_sb
, err
);
5640 * Expand an inode by new_extra_isize bytes.
5641 * Returns 0 on success or negative error number on failure.
5643 static int ext4_expand_extra_isize(struct inode
*inode
,
5644 unsigned int new_extra_isize
,
5645 struct ext4_iloc iloc
,
5648 struct ext4_inode
*raw_inode
;
5649 struct ext4_xattr_ibody_header
*header
;
5651 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5654 raw_inode
= ext4_raw_inode(&iloc
);
5656 header
= IHDR(inode
, raw_inode
);
5658 /* No extended attributes present */
5659 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5660 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5661 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5662 EXT4_I(inode
)->i_extra_isize
, 0,
5663 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5664 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5668 /* try to expand with EAs present */
5669 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5674 * What we do here is to mark the in-core inode as clean with respect to inode
5675 * dirtiness (it may still be data-dirty).
5676 * This means that the in-core inode may be reaped by prune_icache
5677 * without having to perform any I/O. This is a very good thing,
5678 * because *any* task may call prune_icache - even ones which
5679 * have a transaction open against a different journal.
5681 * Is this cheating? Not really. Sure, we haven't written the
5682 * inode out, but prune_icache isn't a user-visible syncing function.
5683 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5684 * we start and wait on commits.
5686 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5688 struct ext4_iloc iloc
;
5689 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5690 static unsigned int mnt_count
;
5694 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5695 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5698 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5699 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5701 * In nojournal mode, we can immediately attempt to expand
5702 * the inode. When journaled, we first need to obtain extra
5703 * buffer credits since we may write into the EA block
5704 * with this same handle. If journal_extend fails, then it will
5705 * only result in a minor loss of functionality for that inode.
5706 * If this is felt to be critical, then e2fsck should be run to
5707 * force a large enough s_min_extra_isize.
5709 if (!ext4_handle_valid(handle
) ||
5710 jbd2_journal_extend(handle
,
5711 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) == 0) {
5712 ret
= ext4_expand_extra_isize(inode
,
5713 sbi
->s_want_extra_isize
,
5717 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5718 ext4_warning(inode
->i_sb
,
5719 "Unable to expand inode %lu. Delete"
5720 " some EAs or run e2fsck.",
5723 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5728 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5732 * ext4_dirty_inode() is called from __mark_inode_dirty()
5734 * We're really interested in the case where a file is being extended.
5735 * i_size has been changed by generic_commit_write() and we thus need
5736 * to include the updated inode in the current transaction.
5738 * Also, dquot_alloc_block() will always dirty the inode when blocks
5739 * are allocated to the file.
5741 * If the inode is marked synchronous, we don't honour that here - doing
5742 * so would cause a commit on atime updates, which we don't bother doing.
5743 * We handle synchronous inodes at the highest possible level.
5745 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5746 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5747 * to copy into the on-disk inode structure are the timestamp files.
5749 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5753 if (flags
== I_DIRTY_TIME
)
5755 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5759 ext4_mark_inode_dirty(handle
, inode
);
5761 ext4_journal_stop(handle
);
5768 * Bind an inode's backing buffer_head into this transaction, to prevent
5769 * it from being flushed to disk early. Unlike
5770 * ext4_reserve_inode_write, this leaves behind no bh reference and
5771 * returns no iloc structure, so the caller needs to repeat the iloc
5772 * lookup to mark the inode dirty later.
5774 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5776 struct ext4_iloc iloc
;
5780 err
= ext4_get_inode_loc(inode
, &iloc
);
5782 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5783 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5785 err
= ext4_handle_dirty_metadata(handle
,
5791 ext4_std_error(inode
->i_sb
, err
);
5796 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5801 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5804 * We have to be very careful here: changing a data block's
5805 * journaling status dynamically is dangerous. If we write a
5806 * data block to the journal, change the status and then delete
5807 * that block, we risk forgetting to revoke the old log record
5808 * from the journal and so a subsequent replay can corrupt data.
5809 * So, first we make sure that the journal is empty and that
5810 * nobody is changing anything.
5813 journal
= EXT4_JOURNAL(inode
);
5816 if (is_journal_aborted(journal
))
5819 /* Wait for all existing dio workers */
5820 ext4_inode_block_unlocked_dio(inode
);
5821 inode_dio_wait(inode
);
5824 * Before flushing the journal and switching inode's aops, we have
5825 * to flush all dirty data the inode has. There can be outstanding
5826 * delayed allocations, there can be unwritten extents created by
5827 * fallocate or buffered writes in dioread_nolock mode covered by
5828 * dirty data which can be converted only after flushing the dirty
5829 * data (and journalled aops don't know how to handle these cases).
5832 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5833 err
= filemap_write_and_wait(inode
->i_mapping
);
5835 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5836 ext4_inode_resume_unlocked_dio(inode
);
5841 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5842 jbd2_journal_lock_updates(journal
);
5845 * OK, there are no updates running now, and all cached data is
5846 * synced to disk. We are now in a completely consistent state
5847 * which doesn't have anything in the journal, and we know that
5848 * no filesystem updates are running, so it is safe to modify
5849 * the inode's in-core data-journaling state flag now.
5853 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5855 err
= jbd2_journal_flush(journal
);
5857 jbd2_journal_unlock_updates(journal
);
5858 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5859 ext4_inode_resume_unlocked_dio(inode
);
5862 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5864 ext4_set_aops(inode
);
5866 * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated.
5867 * E.g. S_DAX may get cleared / set.
5869 ext4_set_inode_flags(inode
);
5871 jbd2_journal_unlock_updates(journal
);
5872 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5875 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5876 ext4_inode_resume_unlocked_dio(inode
);
5878 /* Finally we can mark the inode as dirty. */
5880 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5882 return PTR_ERR(handle
);
5884 err
= ext4_mark_inode_dirty(handle
, inode
);
5885 ext4_handle_sync(handle
);
5886 ext4_journal_stop(handle
);
5887 ext4_std_error(inode
->i_sb
, err
);
5892 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5894 return !buffer_mapped(bh
);
5897 int ext4_page_mkwrite(struct vm_fault
*vmf
)
5899 struct vm_area_struct
*vma
= vmf
->vma
;
5900 struct page
*page
= vmf
->page
;
5904 struct file
*file
= vma
->vm_file
;
5905 struct inode
*inode
= file_inode(file
);
5906 struct address_space
*mapping
= inode
->i_mapping
;
5908 get_block_t
*get_block
;
5911 sb_start_pagefault(inode
->i_sb
);
5912 file_update_time(vma
->vm_file
);
5914 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5916 ret
= ext4_convert_inline_data(inode
);
5920 /* Delalloc case is easy... */
5921 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5922 !ext4_should_journal_data(inode
) &&
5923 !ext4_nonda_switch(inode
->i_sb
)) {
5925 ret
= block_page_mkwrite(vma
, vmf
,
5926 ext4_da_get_block_prep
);
5927 } while (ret
== -ENOSPC
&&
5928 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5933 size
= i_size_read(inode
);
5934 /* Page got truncated from under us? */
5935 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5937 ret
= VM_FAULT_NOPAGE
;
5941 if (page
->index
== size
>> PAGE_SHIFT
)
5942 len
= size
& ~PAGE_MASK
;
5946 * Return if we have all the buffers mapped. This avoids the need to do
5947 * journal_start/journal_stop which can block and take a long time
5949 if (page_has_buffers(page
)) {
5950 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5952 ext4_bh_unmapped
)) {
5953 /* Wait so that we don't change page under IO */
5954 wait_for_stable_page(page
);
5955 ret
= VM_FAULT_LOCKED
;
5960 /* OK, we need to fill the hole... */
5961 if (ext4_should_dioread_nolock(inode
))
5962 get_block
= ext4_get_block_unwritten
;
5964 get_block
= ext4_get_block
;
5966 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5967 ext4_writepage_trans_blocks(inode
));
5968 if (IS_ERR(handle
)) {
5969 ret
= VM_FAULT_SIGBUS
;
5972 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5973 if (!ret
&& ext4_should_journal_data(inode
)) {
5974 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5975 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
5977 ret
= VM_FAULT_SIGBUS
;
5978 ext4_journal_stop(handle
);
5981 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5983 ext4_journal_stop(handle
);
5984 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5987 ret
= block_page_mkwrite_return(ret
);
5989 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5990 sb_end_pagefault(inode
->i_sb
);
5994 int ext4_filemap_fault(struct vm_fault
*vmf
)
5996 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
5999 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6000 err
= filemap_fault(vmf
);
6001 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6007 * Find the first extent at or after @lblk in an inode that is not a hole.
6008 * Search for @map_len blocks at most. The extent is returned in @result.
6010 * The function returns 1 if we found an extent. The function returns 0 in
6011 * case there is no extent at or after @lblk and in that case also sets
6012 * @result->es_len to 0. In case of error, the error code is returned.
6014 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
6015 unsigned int map_len
, struct extent_status
*result
)
6017 struct ext4_map_blocks map
;
6018 struct extent_status es
= {};
6022 map
.m_len
= map_len
;
6025 * For non-extent based files this loop may iterate several times since
6026 * we do not determine full hole size.
6028 while (map
.m_len
> 0) {
6029 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
6032 /* There's extent covering m_lblk? Just return it. */
6036 ext4_es_store_pblock(result
, map
.m_pblk
);
6037 result
->es_lblk
= map
.m_lblk
;
6038 result
->es_len
= map
.m_len
;
6039 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
6040 status
= EXTENT_STATUS_UNWRITTEN
;
6042 status
= EXTENT_STATUS_WRITTEN
;
6043 ext4_es_store_status(result
, status
);
6046 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
6047 map
.m_lblk
+ map
.m_len
- 1,
6049 /* Is delalloc data before next block in extent tree? */
6050 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
6051 ext4_lblk_t offset
= 0;
6053 if (es
.es_lblk
< lblk
)
6054 offset
= lblk
- es
.es_lblk
;
6055 result
->es_lblk
= es
.es_lblk
+ offset
;
6056 ext4_es_store_pblock(result
,
6057 ext4_es_pblock(&es
) + offset
);
6058 result
->es_len
= es
.es_len
- offset
;
6059 ext4_es_store_status(result
, ext4_es_status(&es
));
6063 /* There's a hole at m_lblk, advance us after it */
6064 map
.m_lblk
+= map
.m_len
;
6065 map_len
-= map
.m_len
;
6066 map
.m_len
= map_len
;