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.
147 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149 int ext4_inode_is_fast_symlink(struct inode
*inode
)
151 return S_ISLNK(inode
->i_mode
) && inode
->i_size
&&
152 (inode
->i_size
< EXT4_N_BLOCKS
* 4);
156 * Restart the transaction associated with *handle. This does a commit,
157 * so before we call here everything must be consistently dirtied against
160 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
166 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
167 * moment, get_block can be called only for blocks inside i_size since
168 * page cache has been already dropped and writes are blocked by
169 * i_mutex. So we can safely drop the i_data_sem here.
171 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
172 jbd_debug(2, "restarting handle %p\n", handle
);
173 up_write(&EXT4_I(inode
)->i_data_sem
);
174 ret
= ext4_journal_restart(handle
, nblocks
);
175 down_write(&EXT4_I(inode
)->i_data_sem
);
176 ext4_discard_preallocations(inode
);
182 * Called at the last iput() if i_nlink is zero.
184 void ext4_evict_inode(struct inode
*inode
)
188 int extra_credits
= 3;
189 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
191 trace_ext4_evict_inode(inode
);
193 if (inode
->i_nlink
) {
195 * When journalling data dirty buffers are tracked only in the
196 * journal. So although mm thinks everything is clean and
197 * ready for reaping the inode might still have some pages to
198 * write in the running transaction or waiting to be
199 * checkpointed. Thus calling jbd2_journal_invalidatepage()
200 * (via truncate_inode_pages()) to discard these buffers can
201 * cause data loss. Also even if we did not discard these
202 * buffers, we would have no way to find them after the inode
203 * is reaped and thus user could see stale data if he tries to
204 * read them before the transaction is checkpointed. So be
205 * careful and force everything to disk here... We use
206 * ei->i_datasync_tid to store the newest transaction
207 * containing inode's data.
209 * Note that directories do not have this problem because they
210 * don't use page cache.
212 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
213 ext4_should_journal_data(inode
) &&
214 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
215 inode
->i_data
.nrpages
) {
216 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
217 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
219 jbd2_complete_transaction(journal
, commit_tid
);
220 filemap_write_and_wait(&inode
->i_data
);
222 truncate_inode_pages_final(&inode
->i_data
);
227 if (is_bad_inode(inode
))
229 dquot_initialize(inode
);
231 if (ext4_should_order_data(inode
))
232 ext4_begin_ordered_truncate(inode
, 0);
233 truncate_inode_pages_final(&inode
->i_data
);
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode
->i_sb
);
241 if (!IS_NOQUOTA(inode
))
242 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
244 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
245 ext4_blocks_for_truncate(inode
)+extra_credits
);
246 if (IS_ERR(handle
)) {
247 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
249 * If we're going to skip the normal cleanup, we still need to
250 * make sure that the in-core orphan linked list is properly
253 ext4_orphan_del(NULL
, inode
);
254 sb_end_intwrite(inode
->i_sb
);
259 ext4_handle_sync(handle
);
262 * Set inode->i_size to 0 before calling ext4_truncate(). We need
263 * special handling of symlinks here because i_size is used to
264 * determine whether ext4_inode_info->i_data contains symlink data or
265 * block mappings. Setting i_size to 0 will remove its fast symlink
266 * status. Erase i_data so that it becomes a valid empty block map.
268 if (ext4_inode_is_fast_symlink(inode
))
269 memset(EXT4_I(inode
)->i_data
, 0, sizeof(EXT4_I(inode
)->i_data
));
271 err
= ext4_mark_inode_dirty(handle
, inode
);
273 ext4_warning(inode
->i_sb
,
274 "couldn't mark inode dirty (err %d)", err
);
277 if (inode
->i_blocks
) {
278 err
= ext4_truncate(inode
);
280 ext4_error(inode
->i_sb
,
281 "couldn't truncate inode %lu (err %d)",
287 /* Remove xattr references. */
288 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
291 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
293 ext4_journal_stop(handle
);
294 ext4_orphan_del(NULL
, inode
);
295 sb_end_intwrite(inode
->i_sb
);
296 ext4_xattr_inode_array_free(ea_inode_array
);
301 * Kill off the orphan record which ext4_truncate created.
302 * AKPM: I think this can be inside the above `if'.
303 * Note that ext4_orphan_del() has to be able to cope with the
304 * deletion of a non-existent orphan - this is because we don't
305 * know if ext4_truncate() actually created an orphan record.
306 * (Well, we could do this if we need to, but heck - it works)
308 ext4_orphan_del(handle
, inode
);
309 EXT4_I(inode
)->i_dtime
= get_seconds();
312 * One subtle ordering requirement: if anything has gone wrong
313 * (transaction abort, IO errors, whatever), then we can still
314 * do these next steps (the fs will already have been marked as
315 * having errors), but we can't free the inode if the mark_dirty
318 if (ext4_mark_inode_dirty(handle
, inode
))
319 /* If that failed, just do the required in-core inode clear. */
320 ext4_clear_inode(inode
);
322 ext4_free_inode(handle
, inode
);
323 ext4_journal_stop(handle
);
324 sb_end_intwrite(inode
->i_sb
);
325 ext4_xattr_inode_array_free(ea_inode_array
);
328 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
332 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
334 return &EXT4_I(inode
)->i_reserved_quota
;
339 * Called with i_data_sem down, which is important since we can call
340 * ext4_discard_preallocations() from here.
342 void ext4_da_update_reserve_space(struct inode
*inode
,
343 int used
, int quota_claim
)
345 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
346 struct ext4_inode_info
*ei
= EXT4_I(inode
);
348 spin_lock(&ei
->i_block_reservation_lock
);
349 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
350 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
351 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
352 "with only %d reserved data blocks",
353 __func__
, inode
->i_ino
, used
,
354 ei
->i_reserved_data_blocks
);
356 used
= ei
->i_reserved_data_blocks
;
359 /* Update per-inode reservations */
360 ei
->i_reserved_data_blocks
-= used
;
361 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
363 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
365 /* Update quota subsystem for data blocks */
367 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
370 * We did fallocate with an offset that is already delayed
371 * allocated. So on delayed allocated writeback we should
372 * not re-claim the quota for fallocated blocks.
374 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
378 * If we have done all the pending block allocations and if
379 * there aren't any writers on the inode, we can discard the
380 * inode's preallocations.
382 if ((ei
->i_reserved_data_blocks
== 0) &&
383 (atomic_read(&inode
->i_writecount
) == 0))
384 ext4_discard_preallocations(inode
);
387 static int __check_block_validity(struct inode
*inode
, const char *func
,
389 struct ext4_map_blocks
*map
)
391 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
393 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
394 "lblock %lu mapped to illegal pblock "
395 "(length %d)", (unsigned long) map
->m_lblk
,
397 return -EFSCORRUPTED
;
402 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
407 if (ext4_encrypted_inode(inode
))
408 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
410 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
417 #define check_block_validity(inode, map) \
418 __check_block_validity((inode), __func__, __LINE__, (map))
420 #ifdef ES_AGGRESSIVE_TEST
421 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
423 struct ext4_map_blocks
*es_map
,
424 struct ext4_map_blocks
*map
,
431 * There is a race window that the result is not the same.
432 * e.g. xfstests #223 when dioread_nolock enables. The reason
433 * is that we lookup a block mapping in extent status tree with
434 * out taking i_data_sem. So at the time the unwritten extent
435 * could be converted.
437 down_read(&EXT4_I(inode
)->i_data_sem
);
438 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
439 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
440 EXT4_GET_BLOCKS_KEEP_SIZE
);
442 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
443 EXT4_GET_BLOCKS_KEEP_SIZE
);
445 up_read((&EXT4_I(inode
)->i_data_sem
));
448 * We don't check m_len because extent will be collpased in status
449 * tree. So the m_len might not equal.
451 if (es_map
->m_lblk
!= map
->m_lblk
||
452 es_map
->m_flags
!= map
->m_flags
||
453 es_map
->m_pblk
!= map
->m_pblk
) {
454 printk("ES cache assertion failed for inode: %lu "
455 "es_cached ex [%d/%d/%llu/%x] != "
456 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
457 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
458 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
459 map
->m_len
, map
->m_pblk
, map
->m_flags
,
463 #endif /* ES_AGGRESSIVE_TEST */
466 * The ext4_map_blocks() function tries to look up the requested blocks,
467 * and returns if the blocks are already mapped.
469 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
470 * and store the allocated blocks in the result buffer head and mark it
473 * If file type is extents based, it will call ext4_ext_map_blocks(),
474 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
477 * On success, it returns the number of blocks being mapped or allocated. if
478 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
479 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
481 * It returns 0 if plain look up failed (blocks have not been allocated), in
482 * that case, @map is returned as unmapped but we still do fill map->m_len to
483 * indicate the length of a hole starting at map->m_lblk.
485 * It returns the error in case of allocation failure.
487 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
488 struct ext4_map_blocks
*map
, int flags
)
490 struct extent_status es
;
493 #ifdef ES_AGGRESSIVE_TEST
494 struct ext4_map_blocks orig_map
;
496 memcpy(&orig_map
, map
, sizeof(*map
));
500 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
501 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
502 (unsigned long) map
->m_lblk
);
505 * ext4_map_blocks returns an int, and m_len is an unsigned int
507 if (unlikely(map
->m_len
> INT_MAX
))
508 map
->m_len
= INT_MAX
;
510 /* We can handle the block number less than EXT_MAX_BLOCKS */
511 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
512 return -EFSCORRUPTED
;
514 /* Lookup extent status tree firstly */
515 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
516 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
517 map
->m_pblk
= ext4_es_pblock(&es
) +
518 map
->m_lblk
- es
.es_lblk
;
519 map
->m_flags
|= ext4_es_is_written(&es
) ?
520 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
521 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
522 if (retval
> map
->m_len
)
525 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
527 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
528 if (retval
> map
->m_len
)
535 #ifdef ES_AGGRESSIVE_TEST
536 ext4_map_blocks_es_recheck(handle
, inode
, map
,
543 * Try to see if we can get the block without requesting a new
546 down_read(&EXT4_I(inode
)->i_data_sem
);
547 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
548 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
549 EXT4_GET_BLOCKS_KEEP_SIZE
);
551 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
552 EXT4_GET_BLOCKS_KEEP_SIZE
);
557 if (unlikely(retval
!= map
->m_len
)) {
558 ext4_warning(inode
->i_sb
,
559 "ES len assertion failed for inode "
560 "%lu: retval %d != map->m_len %d",
561 inode
->i_ino
, retval
, map
->m_len
);
565 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
566 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
567 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
568 !(status
& EXTENT_STATUS_WRITTEN
) &&
569 ext4_find_delalloc_range(inode
, map
->m_lblk
,
570 map
->m_lblk
+ map
->m_len
- 1))
571 status
|= EXTENT_STATUS_DELAYED
;
572 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
573 map
->m_len
, map
->m_pblk
, status
);
577 up_read((&EXT4_I(inode
)->i_data_sem
));
580 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
581 ret
= check_block_validity(inode
, map
);
586 /* If it is only a block(s) look up */
587 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
591 * Returns if the blocks have already allocated
593 * Note that if blocks have been preallocated
594 * ext4_ext_get_block() returns the create = 0
595 * with buffer head unmapped.
597 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
599 * If we need to convert extent to unwritten
600 * we continue and do the actual work in
601 * ext4_ext_map_blocks()
603 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
607 * Here we clear m_flags because after allocating an new extent,
608 * it will be set again.
610 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
613 * New blocks allocate and/or writing to unwritten extent
614 * will possibly result in updating i_data, so we take
615 * the write lock of i_data_sem, and call get_block()
616 * with create == 1 flag.
618 down_write(&EXT4_I(inode
)->i_data_sem
);
621 * We need to check for EXT4 here because migrate
622 * could have changed the inode type in between
624 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
625 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
627 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
629 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
631 * We allocated new blocks which will result in
632 * i_data's format changing. Force the migrate
633 * to fail by clearing migrate flags
635 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
639 * Update reserved blocks/metadata blocks after successful
640 * block allocation which had been deferred till now. We don't
641 * support fallocate for non extent files. So we can update
642 * reserve space here.
645 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
646 ext4_da_update_reserve_space(inode
, retval
, 1);
652 if (unlikely(retval
!= map
->m_len
)) {
653 ext4_warning(inode
->i_sb
,
654 "ES len assertion failed for inode "
655 "%lu: retval %d != map->m_len %d",
656 inode
->i_ino
, retval
, map
->m_len
);
661 * We have to zeroout blocks before inserting them into extent
662 * status tree. Otherwise someone could look them up there and
663 * use them before they are really zeroed. We also have to
664 * unmap metadata before zeroing as otherwise writeback can
665 * overwrite zeros with stale data from block device.
667 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
668 map
->m_flags
& EXT4_MAP_MAPPED
&&
669 map
->m_flags
& EXT4_MAP_NEW
) {
670 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
672 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
673 map
->m_pblk
, map
->m_len
);
681 * If the extent has been zeroed out, we don't need to update
682 * extent status tree.
684 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
685 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
686 if (ext4_es_is_written(&es
))
689 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
690 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
691 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
692 !(status
& EXTENT_STATUS_WRITTEN
) &&
693 ext4_find_delalloc_range(inode
, map
->m_lblk
,
694 map
->m_lblk
+ map
->m_len
- 1))
695 status
|= EXTENT_STATUS_DELAYED
;
696 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
697 map
->m_pblk
, status
);
705 up_write((&EXT4_I(inode
)->i_data_sem
));
706 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
707 ret
= check_block_validity(inode
, map
);
712 * Inodes with freshly allocated blocks where contents will be
713 * visible after transaction commit must be on transaction's
716 if (map
->m_flags
& EXT4_MAP_NEW
&&
717 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
718 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
719 !ext4_is_quota_file(inode
) &&
720 ext4_should_order_data(inode
)) {
721 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
722 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
724 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
733 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
734 * we have to be careful as someone else may be manipulating b_state as well.
736 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
738 unsigned long old_state
;
739 unsigned long new_state
;
741 flags
&= EXT4_MAP_FLAGS
;
743 /* Dummy buffer_head? Set non-atomically. */
745 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
749 * Someone else may be modifying b_state. Be careful! This is ugly but
750 * once we get rid of using bh as a container for mapping information
751 * to pass to / from get_block functions, this can go away.
754 old_state
= READ_ONCE(bh
->b_state
);
755 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
757 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
760 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
761 struct buffer_head
*bh
, int flags
)
763 struct ext4_map_blocks map
;
766 if (ext4_has_inline_data(inode
))
770 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
772 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
775 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
776 ext4_update_bh_state(bh
, map
.m_flags
);
777 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
779 } else if (ret
== 0) {
780 /* hole case, need to fill in bh->b_size */
781 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
786 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
787 struct buffer_head
*bh
, int create
)
789 return _ext4_get_block(inode
, iblock
, bh
,
790 create
? EXT4_GET_BLOCKS_CREATE
: 0);
794 * Get block function used when preparing for buffered write if we require
795 * creating an unwritten extent if blocks haven't been allocated. The extent
796 * will be converted to written after the IO is complete.
798 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
799 struct buffer_head
*bh_result
, int create
)
801 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
802 inode
->i_ino
, create
);
803 return _ext4_get_block(inode
, iblock
, bh_result
,
804 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
807 /* Maximum number of blocks we map for direct IO at once. */
808 #define DIO_MAX_BLOCKS 4096
811 * Get blocks function for the cases that need to start a transaction -
812 * generally difference cases of direct IO and DAX IO. It also handles retries
815 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
816 struct buffer_head
*bh_result
, int flags
)
823 /* Trim mapping request to maximum we can map at once for DIO */
824 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
825 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
826 dio_credits
= ext4_chunk_trans_blocks(inode
,
827 bh_result
->b_size
>> inode
->i_blkbits
);
829 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
831 return PTR_ERR(handle
);
833 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
834 ext4_journal_stop(handle
);
836 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
841 /* Get block function for DIO reads and writes to inodes without extents */
842 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
843 struct buffer_head
*bh
, int create
)
845 /* We don't expect handle for direct IO */
846 WARN_ON_ONCE(ext4_journal_current_handle());
849 return _ext4_get_block(inode
, iblock
, bh
, 0);
850 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
854 * Get block function for AIO DIO writes when we create unwritten extent if
855 * blocks are not allocated yet. The extent will be converted to written
856 * after IO is complete.
858 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
859 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
863 /* We don't expect handle for direct IO */
864 WARN_ON_ONCE(ext4_journal_current_handle());
866 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
867 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
870 * When doing DIO using unwritten extents, we need io_end to convert
871 * unwritten extents to written on IO completion. We allocate io_end
872 * once we spot unwritten extent and store it in b_private. Generic
873 * DIO code keeps b_private set and furthermore passes the value to
874 * our completion callback in 'private' argument.
876 if (!ret
&& buffer_unwritten(bh_result
)) {
877 if (!bh_result
->b_private
) {
878 ext4_io_end_t
*io_end
;
880 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
883 bh_result
->b_private
= io_end
;
884 ext4_set_io_unwritten_flag(inode
, io_end
);
886 set_buffer_defer_completion(bh_result
);
893 * Get block function for non-AIO DIO writes when we create unwritten extent if
894 * blocks are not allocated yet. The extent will be converted to written
895 * after IO is complete by ext4_direct_IO_write().
897 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
898 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
902 /* We don't expect handle for direct IO */
903 WARN_ON_ONCE(ext4_journal_current_handle());
905 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
906 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
909 * Mark inode as having pending DIO writes to unwritten extents.
910 * ext4_direct_IO_write() checks this flag and converts extents to
913 if (!ret
&& buffer_unwritten(bh_result
))
914 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
919 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
920 struct buffer_head
*bh_result
, int create
)
924 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
925 inode
->i_ino
, create
);
926 /* We don't expect handle for direct IO */
927 WARN_ON_ONCE(ext4_journal_current_handle());
929 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
931 * Blocks should have been preallocated! ext4_file_write_iter() checks
934 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
941 * `handle' can be NULL if create is zero
943 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
944 ext4_lblk_t block
, int map_flags
)
946 struct ext4_map_blocks map
;
947 struct buffer_head
*bh
;
948 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
951 J_ASSERT(handle
!= NULL
|| create
== 0);
955 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
958 return create
? ERR_PTR(-ENOSPC
) : NULL
;
962 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
964 return ERR_PTR(-ENOMEM
);
965 if (map
.m_flags
& EXT4_MAP_NEW
) {
966 J_ASSERT(create
!= 0);
967 J_ASSERT(handle
!= NULL
);
970 * Now that we do not always journal data, we should
971 * keep in mind whether this should always journal the
972 * new buffer as metadata. For now, regular file
973 * writes use ext4_get_block instead, so it's not a
977 BUFFER_TRACE(bh
, "call get_create_access");
978 err
= ext4_journal_get_create_access(handle
, bh
);
983 if (!buffer_uptodate(bh
)) {
984 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
985 set_buffer_uptodate(bh
);
988 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
989 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
993 BUFFER_TRACE(bh
, "not a new buffer");
1000 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1001 ext4_lblk_t block
, int map_flags
)
1003 struct buffer_head
*bh
;
1005 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1008 if (!bh
|| buffer_uptodate(bh
))
1010 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1012 if (buffer_uptodate(bh
))
1015 return ERR_PTR(-EIO
);
1018 /* Read a contiguous batch of blocks. */
1019 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
1020 bool wait
, struct buffer_head
**bhs
)
1024 for (i
= 0; i
< bh_count
; i
++) {
1025 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
1026 if (IS_ERR(bhs
[i
])) {
1027 err
= PTR_ERR(bhs
[i
]);
1033 for (i
= 0; i
< bh_count
; i
++)
1034 /* Note that NULL bhs[i] is valid because of holes. */
1035 if (bhs
[i
] && !buffer_uptodate(bhs
[i
]))
1036 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1,
1042 for (i
= 0; i
< bh_count
; i
++)
1044 wait_on_buffer(bhs
[i
]);
1046 for (i
= 0; i
< bh_count
; i
++) {
1047 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
1055 for (i
= 0; i
< bh_count
; i
++) {
1062 int ext4_walk_page_buffers(handle_t
*handle
,
1063 struct buffer_head
*head
,
1067 int (*fn
)(handle_t
*handle
,
1068 struct buffer_head
*bh
))
1070 struct buffer_head
*bh
;
1071 unsigned block_start
, block_end
;
1072 unsigned blocksize
= head
->b_size
;
1074 struct buffer_head
*next
;
1076 for (bh
= head
, block_start
= 0;
1077 ret
== 0 && (bh
!= head
|| !block_start
);
1078 block_start
= block_end
, bh
= next
) {
1079 next
= bh
->b_this_page
;
1080 block_end
= block_start
+ blocksize
;
1081 if (block_end
<= from
|| block_start
>= to
) {
1082 if (partial
&& !buffer_uptodate(bh
))
1086 err
= (*fn
)(handle
, bh
);
1094 * To preserve ordering, it is essential that the hole instantiation and
1095 * the data write be encapsulated in a single transaction. We cannot
1096 * close off a transaction and start a new one between the ext4_get_block()
1097 * and the commit_write(). So doing the jbd2_journal_start at the start of
1098 * prepare_write() is the right place.
1100 * Also, this function can nest inside ext4_writepage(). In that case, we
1101 * *know* that ext4_writepage() has generated enough buffer credits to do the
1102 * whole page. So we won't block on the journal in that case, which is good,
1103 * because the caller may be PF_MEMALLOC.
1105 * By accident, ext4 can be reentered when a transaction is open via
1106 * quota file writes. If we were to commit the transaction while thus
1107 * reentered, there can be a deadlock - we would be holding a quota
1108 * lock, and the commit would never complete if another thread had a
1109 * transaction open and was blocking on the quota lock - a ranking
1112 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1113 * will _not_ run commit under these circumstances because handle->h_ref
1114 * is elevated. We'll still have enough credits for the tiny quotafile
1117 int do_journal_get_write_access(handle_t
*handle
,
1118 struct buffer_head
*bh
)
1120 int dirty
= buffer_dirty(bh
);
1123 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1126 * __block_write_begin() could have dirtied some buffers. Clean
1127 * the dirty bit as jbd2_journal_get_write_access() could complain
1128 * otherwise about fs integrity issues. Setting of the dirty bit
1129 * by __block_write_begin() isn't a real problem here as we clear
1130 * the bit before releasing a page lock and thus writeback cannot
1131 * ever write the buffer.
1134 clear_buffer_dirty(bh
);
1135 BUFFER_TRACE(bh
, "get write access");
1136 ret
= ext4_journal_get_write_access(handle
, bh
);
1138 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1142 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1143 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1144 get_block_t
*get_block
)
1146 unsigned from
= pos
& (PAGE_SIZE
- 1);
1147 unsigned to
= from
+ len
;
1148 struct inode
*inode
= page
->mapping
->host
;
1149 unsigned block_start
, block_end
;
1152 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1154 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1155 bool decrypt
= false;
1157 BUG_ON(!PageLocked(page
));
1158 BUG_ON(from
> PAGE_SIZE
);
1159 BUG_ON(to
> PAGE_SIZE
);
1162 if (!page_has_buffers(page
))
1163 create_empty_buffers(page
, blocksize
, 0);
1164 head
= page_buffers(page
);
1165 bbits
= ilog2(blocksize
);
1166 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1168 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1169 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1170 block_end
= block_start
+ blocksize
;
1171 if (block_end
<= from
|| block_start
>= to
) {
1172 if (PageUptodate(page
)) {
1173 if (!buffer_uptodate(bh
))
1174 set_buffer_uptodate(bh
);
1179 clear_buffer_new(bh
);
1180 if (!buffer_mapped(bh
)) {
1181 WARN_ON(bh
->b_size
!= blocksize
);
1182 err
= get_block(inode
, block
, bh
, 1);
1185 if (buffer_new(bh
)) {
1186 clean_bdev_bh_alias(bh
);
1187 if (PageUptodate(page
)) {
1188 clear_buffer_new(bh
);
1189 set_buffer_uptodate(bh
);
1190 mark_buffer_dirty(bh
);
1193 if (block_end
> to
|| block_start
< from
)
1194 zero_user_segments(page
, to
, block_end
,
1199 if (PageUptodate(page
)) {
1200 if (!buffer_uptodate(bh
))
1201 set_buffer_uptodate(bh
);
1204 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1205 !buffer_unwritten(bh
) &&
1206 (block_start
< from
|| block_end
> to
)) {
1207 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1209 decrypt
= ext4_encrypted_inode(inode
) &&
1210 S_ISREG(inode
->i_mode
);
1214 * If we issued read requests, let them complete.
1216 while (wait_bh
> wait
) {
1217 wait_on_buffer(*--wait_bh
);
1218 if (!buffer_uptodate(*wait_bh
))
1222 page_zero_new_buffers(page
, from
, to
);
1224 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1225 PAGE_SIZE
, 0, page
->index
);
1230 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1231 loff_t pos
, unsigned len
, unsigned flags
,
1232 struct page
**pagep
, void **fsdata
)
1234 struct inode
*inode
= mapping
->host
;
1235 int ret
, needed_blocks
;
1242 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1245 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1247 * Reserve one block more for addition to orphan list in case
1248 * we allocate blocks but write fails for some reason
1250 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1251 index
= pos
>> PAGE_SHIFT
;
1252 from
= pos
& (PAGE_SIZE
- 1);
1255 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1256 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1265 * grab_cache_page_write_begin() can take a long time if the
1266 * system is thrashing due to memory pressure, or if the page
1267 * is being written back. So grab it first before we start
1268 * the transaction handle. This also allows us to allocate
1269 * the page (if needed) without using GFP_NOFS.
1272 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1278 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1279 if (IS_ERR(handle
)) {
1281 return PTR_ERR(handle
);
1285 if (page
->mapping
!= mapping
) {
1286 /* The page got truncated from under us */
1289 ext4_journal_stop(handle
);
1292 /* In case writeback began while the page was unlocked */
1293 wait_for_stable_page(page
);
1295 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1296 if (ext4_should_dioread_nolock(inode
))
1297 ret
= ext4_block_write_begin(page
, pos
, len
,
1298 ext4_get_block_unwritten
);
1300 ret
= ext4_block_write_begin(page
, pos
, len
,
1303 if (ext4_should_dioread_nolock(inode
))
1304 ret
= __block_write_begin(page
, pos
, len
,
1305 ext4_get_block_unwritten
);
1307 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1309 if (!ret
&& ext4_should_journal_data(inode
)) {
1310 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1312 do_journal_get_write_access
);
1318 * __block_write_begin may have instantiated a few blocks
1319 * outside i_size. Trim these off again. Don't need
1320 * i_size_read because we hold i_mutex.
1322 * Add inode to orphan list in case we crash before
1325 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1326 ext4_orphan_add(handle
, inode
);
1328 ext4_journal_stop(handle
);
1329 if (pos
+ len
> inode
->i_size
) {
1330 ext4_truncate_failed_write(inode
);
1332 * If truncate failed early the inode might
1333 * still be on the orphan list; we need to
1334 * make sure the inode is removed from the
1335 * orphan list in that case.
1338 ext4_orphan_del(NULL
, inode
);
1341 if (ret
== -ENOSPC
&&
1342 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1351 /* For write_end() in data=journal mode */
1352 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1355 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1357 set_buffer_uptodate(bh
);
1358 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1359 clear_buffer_meta(bh
);
1360 clear_buffer_prio(bh
);
1365 * We need to pick up the new inode size which generic_commit_write gave us
1366 * `file' can be NULL - eg, when called from page_symlink().
1368 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1369 * buffers are managed internally.
1371 static int ext4_write_end(struct file
*file
,
1372 struct address_space
*mapping
,
1373 loff_t pos
, unsigned len
, unsigned copied
,
1374 struct page
*page
, void *fsdata
)
1376 handle_t
*handle
= ext4_journal_current_handle();
1377 struct inode
*inode
= mapping
->host
;
1378 loff_t old_size
= inode
->i_size
;
1380 int i_size_changed
= 0;
1382 trace_ext4_write_end(inode
, pos
, len
, copied
);
1383 if (ext4_has_inline_data(inode
)) {
1384 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1393 copied
= block_write_end(file
, mapping
, pos
,
1394 len
, copied
, page
, fsdata
);
1396 * it's important to update i_size while still holding page lock:
1397 * page writeout could otherwise come in and zero beyond i_size.
1399 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1404 pagecache_isize_extended(inode
, old_size
, pos
);
1406 * Don't mark the inode dirty under page lock. First, it unnecessarily
1407 * makes the holding time of page lock longer. Second, it forces lock
1408 * ordering of page lock and transaction start for journaling
1412 ext4_mark_inode_dirty(handle
, inode
);
1414 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1415 /* if we have allocated more blocks and copied
1416 * less. We will have blocks allocated outside
1417 * inode->i_size. So truncate them
1419 ext4_orphan_add(handle
, inode
);
1421 ret2
= ext4_journal_stop(handle
);
1425 if (pos
+ len
> inode
->i_size
) {
1426 ext4_truncate_failed_write(inode
);
1428 * If truncate failed early the inode might still be
1429 * on the orphan list; we need to make sure the inode
1430 * is removed from the orphan list in that case.
1433 ext4_orphan_del(NULL
, inode
);
1436 return ret
? ret
: copied
;
1440 * This is a private version of page_zero_new_buffers() which doesn't
1441 * set the buffer to be dirty, since in data=journalled mode we need
1442 * to call ext4_handle_dirty_metadata() instead.
1444 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1446 unsigned from
, unsigned to
)
1448 unsigned int block_start
= 0, block_end
;
1449 struct buffer_head
*head
, *bh
;
1451 bh
= head
= page_buffers(page
);
1453 block_end
= block_start
+ bh
->b_size
;
1454 if (buffer_new(bh
)) {
1455 if (block_end
> from
&& block_start
< to
) {
1456 if (!PageUptodate(page
)) {
1457 unsigned start
, size
;
1459 start
= max(from
, block_start
);
1460 size
= min(to
, block_end
) - start
;
1462 zero_user(page
, start
, size
);
1463 write_end_fn(handle
, bh
);
1465 clear_buffer_new(bh
);
1468 block_start
= block_end
;
1469 bh
= bh
->b_this_page
;
1470 } while (bh
!= head
);
1473 static int ext4_journalled_write_end(struct file
*file
,
1474 struct address_space
*mapping
,
1475 loff_t pos
, unsigned len
, unsigned copied
,
1476 struct page
*page
, void *fsdata
)
1478 handle_t
*handle
= ext4_journal_current_handle();
1479 struct inode
*inode
= mapping
->host
;
1480 loff_t old_size
= inode
->i_size
;
1484 int size_changed
= 0;
1486 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1487 from
= pos
& (PAGE_SIZE
- 1);
1490 BUG_ON(!ext4_handle_valid(handle
));
1492 if (ext4_has_inline_data(inode
)) {
1493 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1501 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1503 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1505 if (unlikely(copied
< len
))
1506 ext4_journalled_zero_new_buffers(handle
, page
,
1508 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1509 from
+ copied
, &partial
,
1512 SetPageUptodate(page
);
1514 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1515 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1516 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1521 pagecache_isize_extended(inode
, old_size
, pos
);
1524 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1529 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1530 /* if we have allocated more blocks and copied
1531 * less. We will have blocks allocated outside
1532 * inode->i_size. So truncate them
1534 ext4_orphan_add(handle
, inode
);
1537 ret2
= ext4_journal_stop(handle
);
1540 if (pos
+ len
> inode
->i_size
) {
1541 ext4_truncate_failed_write(inode
);
1543 * If truncate failed early the inode might still be
1544 * on the orphan list; we need to make sure the inode
1545 * is removed from the orphan list in that case.
1548 ext4_orphan_del(NULL
, inode
);
1551 return ret
? ret
: copied
;
1555 * Reserve space for a single cluster
1557 static int ext4_da_reserve_space(struct inode
*inode
)
1559 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1560 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1564 * We will charge metadata quota at writeout time; this saves
1565 * us from metadata over-estimation, though we may go over by
1566 * a small amount in the end. Here we just reserve for data.
1568 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1572 spin_lock(&ei
->i_block_reservation_lock
);
1573 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1574 spin_unlock(&ei
->i_block_reservation_lock
);
1575 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1578 ei
->i_reserved_data_blocks
++;
1579 trace_ext4_da_reserve_space(inode
);
1580 spin_unlock(&ei
->i_block_reservation_lock
);
1582 return 0; /* success */
1585 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1587 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1588 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1591 return; /* Nothing to release, exit */
1593 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1595 trace_ext4_da_release_space(inode
, to_free
);
1596 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1598 * if there aren't enough reserved blocks, then the
1599 * counter is messed up somewhere. Since this
1600 * function is called from invalidate page, it's
1601 * harmless to return without any action.
1603 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1604 "ino %lu, to_free %d with only %d reserved "
1605 "data blocks", inode
->i_ino
, to_free
,
1606 ei
->i_reserved_data_blocks
);
1608 to_free
= ei
->i_reserved_data_blocks
;
1610 ei
->i_reserved_data_blocks
-= to_free
;
1612 /* update fs dirty data blocks counter */
1613 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1615 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1617 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1620 static void ext4_da_page_release_reservation(struct page
*page
,
1621 unsigned int offset
,
1622 unsigned int length
)
1624 int to_release
= 0, contiguous_blks
= 0;
1625 struct buffer_head
*head
, *bh
;
1626 unsigned int curr_off
= 0;
1627 struct inode
*inode
= page
->mapping
->host
;
1628 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1629 unsigned int stop
= offset
+ length
;
1633 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1635 head
= page_buffers(page
);
1638 unsigned int next_off
= curr_off
+ bh
->b_size
;
1640 if (next_off
> stop
)
1643 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1646 clear_buffer_delay(bh
);
1647 } else if (contiguous_blks
) {
1648 lblk
= page
->index
<<
1649 (PAGE_SHIFT
- inode
->i_blkbits
);
1650 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1652 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1653 contiguous_blks
= 0;
1655 curr_off
= next_off
;
1656 } while ((bh
= bh
->b_this_page
) != head
);
1658 if (contiguous_blks
) {
1659 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1660 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1661 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1664 /* If we have released all the blocks belonging to a cluster, then we
1665 * need to release the reserved space for that cluster. */
1666 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1667 while (num_clusters
> 0) {
1668 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1669 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1670 if (sbi
->s_cluster_ratio
== 1 ||
1671 !ext4_find_delalloc_cluster(inode
, lblk
))
1672 ext4_da_release_space(inode
, 1);
1679 * Delayed allocation stuff
1682 struct mpage_da_data
{
1683 struct inode
*inode
;
1684 struct writeback_control
*wbc
;
1686 pgoff_t first_page
; /* The first page to write */
1687 pgoff_t next_page
; /* Current page to examine */
1688 pgoff_t last_page
; /* Last page to examine */
1690 * Extent to map - this can be after first_page because that can be
1691 * fully mapped. We somewhat abuse m_flags to store whether the extent
1692 * is delalloc or unwritten.
1694 struct ext4_map_blocks map
;
1695 struct ext4_io_submit io_submit
; /* IO submission data */
1696 unsigned int do_map
:1;
1699 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1704 struct pagevec pvec
;
1705 struct inode
*inode
= mpd
->inode
;
1706 struct address_space
*mapping
= inode
->i_mapping
;
1708 /* This is necessary when next_page == 0. */
1709 if (mpd
->first_page
>= mpd
->next_page
)
1712 index
= mpd
->first_page
;
1713 end
= mpd
->next_page
- 1;
1715 ext4_lblk_t start
, last
;
1716 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1717 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1718 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1721 pagevec_init(&pvec
, 0);
1722 while (index
<= end
) {
1723 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1726 for (i
= 0; i
< nr_pages
; i
++) {
1727 struct page
*page
= pvec
.pages
[i
];
1729 BUG_ON(!PageLocked(page
));
1730 BUG_ON(PageWriteback(page
));
1732 if (page_mapped(page
))
1733 clear_page_dirty_for_io(page
);
1734 block_invalidatepage(page
, 0, PAGE_SIZE
);
1735 ClearPageUptodate(page
);
1739 pagevec_release(&pvec
);
1743 static void ext4_print_free_blocks(struct inode
*inode
)
1745 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1746 struct super_block
*sb
= inode
->i_sb
;
1747 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1749 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1750 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1751 ext4_count_free_clusters(sb
)));
1752 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1753 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1754 (long long) EXT4_C2B(EXT4_SB(sb
),
1755 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1756 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1757 (long long) EXT4_C2B(EXT4_SB(sb
),
1758 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1759 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1760 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1761 ei
->i_reserved_data_blocks
);
1765 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1767 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1771 * This function is grabs code from the very beginning of
1772 * ext4_map_blocks, but assumes that the caller is from delayed write
1773 * time. This function looks up the requested blocks and sets the
1774 * buffer delay bit under the protection of i_data_sem.
1776 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1777 struct ext4_map_blocks
*map
,
1778 struct buffer_head
*bh
)
1780 struct extent_status es
;
1782 sector_t invalid_block
= ~((sector_t
) 0xffff);
1783 #ifdef ES_AGGRESSIVE_TEST
1784 struct ext4_map_blocks orig_map
;
1786 memcpy(&orig_map
, map
, sizeof(*map
));
1789 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1793 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1794 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1795 (unsigned long) map
->m_lblk
);
1797 /* Lookup extent status tree firstly */
1798 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1799 if (ext4_es_is_hole(&es
)) {
1801 down_read(&EXT4_I(inode
)->i_data_sem
);
1806 * Delayed extent could be allocated by fallocate.
1807 * So we need to check it.
1809 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1810 map_bh(bh
, inode
->i_sb
, invalid_block
);
1812 set_buffer_delay(bh
);
1816 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1817 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1818 if (retval
> map
->m_len
)
1819 retval
= map
->m_len
;
1820 map
->m_len
= retval
;
1821 if (ext4_es_is_written(&es
))
1822 map
->m_flags
|= EXT4_MAP_MAPPED
;
1823 else if (ext4_es_is_unwritten(&es
))
1824 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1828 #ifdef ES_AGGRESSIVE_TEST
1829 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1835 * Try to see if we can get the block without requesting a new
1836 * file system block.
1838 down_read(&EXT4_I(inode
)->i_data_sem
);
1839 if (ext4_has_inline_data(inode
))
1841 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1842 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1844 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1850 * XXX: __block_prepare_write() unmaps passed block,
1854 * If the block was allocated from previously allocated cluster,
1855 * then we don't need to reserve it again. However we still need
1856 * to reserve metadata for every block we're going to write.
1858 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1859 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1860 ret
= ext4_da_reserve_space(inode
);
1862 /* not enough space to reserve */
1868 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1869 ~0, EXTENT_STATUS_DELAYED
);
1875 map_bh(bh
, inode
->i_sb
, invalid_block
);
1877 set_buffer_delay(bh
);
1878 } else if (retval
> 0) {
1880 unsigned int status
;
1882 if (unlikely(retval
!= map
->m_len
)) {
1883 ext4_warning(inode
->i_sb
,
1884 "ES len assertion failed for inode "
1885 "%lu: retval %d != map->m_len %d",
1886 inode
->i_ino
, retval
, map
->m_len
);
1890 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1891 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1892 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1893 map
->m_pblk
, status
);
1899 up_read((&EXT4_I(inode
)->i_data_sem
));
1905 * This is a special get_block_t callback which is used by
1906 * ext4_da_write_begin(). It will either return mapped block or
1907 * reserve space for a single block.
1909 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1910 * We also have b_blocknr = -1 and b_bdev initialized properly
1912 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1913 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1914 * initialized properly.
1916 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1917 struct buffer_head
*bh
, int create
)
1919 struct ext4_map_blocks map
;
1922 BUG_ON(create
== 0);
1923 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1925 map
.m_lblk
= iblock
;
1929 * first, we need to know whether the block is allocated already
1930 * preallocated blocks are unmapped but should treated
1931 * the same as allocated blocks.
1933 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1937 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1938 ext4_update_bh_state(bh
, map
.m_flags
);
1940 if (buffer_unwritten(bh
)) {
1941 /* A delayed write to unwritten bh should be marked
1942 * new and mapped. Mapped ensures that we don't do
1943 * get_block multiple times when we write to the same
1944 * offset and new ensures that we do proper zero out
1945 * for partial write.
1948 set_buffer_mapped(bh
);
1953 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1959 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1965 static int __ext4_journalled_writepage(struct page
*page
,
1968 struct address_space
*mapping
= page
->mapping
;
1969 struct inode
*inode
= mapping
->host
;
1970 struct buffer_head
*page_bufs
= NULL
;
1971 handle_t
*handle
= NULL
;
1972 int ret
= 0, err
= 0;
1973 int inline_data
= ext4_has_inline_data(inode
);
1974 struct buffer_head
*inode_bh
= NULL
;
1976 ClearPageChecked(page
);
1979 BUG_ON(page
->index
!= 0);
1980 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1981 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1982 if (inode_bh
== NULL
)
1985 page_bufs
= page_buffers(page
);
1990 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1994 * We need to release the page lock before we start the
1995 * journal, so grab a reference so the page won't disappear
1996 * out from under us.
2001 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2002 ext4_writepage_trans_blocks(inode
));
2003 if (IS_ERR(handle
)) {
2004 ret
= PTR_ERR(handle
);
2006 goto out_no_pagelock
;
2008 BUG_ON(!ext4_handle_valid(handle
));
2012 if (page
->mapping
!= mapping
) {
2013 /* The page got truncated from under us */
2014 ext4_journal_stop(handle
);
2020 BUFFER_TRACE(inode_bh
, "get write access");
2021 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
2023 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
2026 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2027 do_journal_get_write_access
);
2029 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2034 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2035 err
= ext4_journal_stop(handle
);
2039 if (!ext4_has_inline_data(inode
))
2040 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
2042 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2051 * Note that we don't need to start a transaction unless we're journaling data
2052 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2053 * need to file the inode to the transaction's list in ordered mode because if
2054 * we are writing back data added by write(), the inode is already there and if
2055 * we are writing back data modified via mmap(), no one guarantees in which
2056 * transaction the data will hit the disk. In case we are journaling data, we
2057 * cannot start transaction directly because transaction start ranks above page
2058 * lock so we have to do some magic.
2060 * This function can get called via...
2061 * - ext4_writepages after taking page lock (have journal handle)
2062 * - journal_submit_inode_data_buffers (no journal handle)
2063 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2064 * - grab_page_cache when doing write_begin (have journal handle)
2066 * We don't do any block allocation in this function. If we have page with
2067 * multiple blocks we need to write those buffer_heads that are mapped. This
2068 * is important for mmaped based write. So if we do with blocksize 1K
2069 * truncate(f, 1024);
2070 * a = mmap(f, 0, 4096);
2072 * truncate(f, 4096);
2073 * we have in the page first buffer_head mapped via page_mkwrite call back
2074 * but other buffer_heads would be unmapped but dirty (dirty done via the
2075 * do_wp_page). So writepage should write the first block. If we modify
2076 * the mmap area beyond 1024 we will again get a page_fault and the
2077 * page_mkwrite callback will do the block allocation and mark the
2078 * buffer_heads mapped.
2080 * We redirty the page if we have any buffer_heads that is either delay or
2081 * unwritten in the page.
2083 * We can get recursively called as show below.
2085 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2088 * But since we don't do any block allocation we should not deadlock.
2089 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2091 static int ext4_writepage(struct page
*page
,
2092 struct writeback_control
*wbc
)
2097 struct buffer_head
*page_bufs
= NULL
;
2098 struct inode
*inode
= page
->mapping
->host
;
2099 struct ext4_io_submit io_submit
;
2100 bool keep_towrite
= false;
2102 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
2103 ext4_invalidatepage(page
, 0, PAGE_SIZE
);
2108 trace_ext4_writepage(page
);
2109 size
= i_size_read(inode
);
2110 if (page
->index
== size
>> PAGE_SHIFT
)
2111 len
= size
& ~PAGE_MASK
;
2115 page_bufs
= page_buffers(page
);
2117 * We cannot do block allocation or other extent handling in this
2118 * function. If there are buffers needing that, we have to redirty
2119 * the page. But we may reach here when we do a journal commit via
2120 * journal_submit_inode_data_buffers() and in that case we must write
2121 * allocated buffers to achieve data=ordered mode guarantees.
2123 * Also, if there is only one buffer per page (the fs block
2124 * size == the page size), if one buffer needs block
2125 * allocation or needs to modify the extent tree to clear the
2126 * unwritten flag, we know that the page can't be written at
2127 * all, so we might as well refuse the write immediately.
2128 * Unfortunately if the block size != page size, we can't as
2129 * easily detect this case using ext4_walk_page_buffers(), but
2130 * for the extremely common case, this is an optimization that
2131 * skips a useless round trip through ext4_bio_write_page().
2133 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2134 ext4_bh_delay_or_unwritten
)) {
2135 redirty_page_for_writepage(wbc
, page
);
2136 if ((current
->flags
& PF_MEMALLOC
) ||
2137 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2139 * For memory cleaning there's no point in writing only
2140 * some buffers. So just bail out. Warn if we came here
2141 * from direct reclaim.
2143 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2148 keep_towrite
= true;
2151 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2153 * It's mmapped pagecache. Add buffers and journal it. There
2154 * doesn't seem much point in redirtying the page here.
2156 return __ext4_journalled_writepage(page
, len
);
2158 ext4_io_submit_init(&io_submit
, wbc
);
2159 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2160 if (!io_submit
.io_end
) {
2161 redirty_page_for_writepage(wbc
, page
);
2165 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2166 ext4_io_submit(&io_submit
);
2167 /* Drop io_end reference we got from init */
2168 ext4_put_io_end_defer(io_submit
.io_end
);
2172 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2178 BUG_ON(page
->index
!= mpd
->first_page
);
2179 clear_page_dirty_for_io(page
);
2181 * We have to be very careful here! Nothing protects writeback path
2182 * against i_size changes and the page can be writeably mapped into
2183 * page tables. So an application can be growing i_size and writing
2184 * data through mmap while writeback runs. clear_page_dirty_for_io()
2185 * write-protects our page in page tables and the page cannot get
2186 * written to again until we release page lock. So only after
2187 * clear_page_dirty_for_io() we are safe to sample i_size for
2188 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2189 * on the barrier provided by TestClearPageDirty in
2190 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2191 * after page tables are updated.
2193 size
= i_size_read(mpd
->inode
);
2194 if (page
->index
== size
>> PAGE_SHIFT
)
2195 len
= size
& ~PAGE_MASK
;
2198 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2200 mpd
->wbc
->nr_to_write
--;
2206 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2209 * mballoc gives us at most this number of blocks...
2210 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2211 * The rest of mballoc seems to handle chunks up to full group size.
2213 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2216 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2218 * @mpd - extent of blocks
2219 * @lblk - logical number of the block in the file
2220 * @bh - buffer head we want to add to the extent
2222 * The function is used to collect contig. blocks in the same state. If the
2223 * buffer doesn't require mapping for writeback and we haven't started the
2224 * extent of buffers to map yet, the function returns 'true' immediately - the
2225 * caller can write the buffer right away. Otherwise the function returns true
2226 * if the block has been added to the extent, false if the block couldn't be
2229 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2230 struct buffer_head
*bh
)
2232 struct ext4_map_blocks
*map
= &mpd
->map
;
2234 /* Buffer that doesn't need mapping for writeback? */
2235 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2236 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2237 /* So far no extent to map => we write the buffer right away */
2238 if (map
->m_len
== 0)
2243 /* First block in the extent? */
2244 if (map
->m_len
== 0) {
2245 /* We cannot map unless handle is started... */
2250 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2254 /* Don't go larger than mballoc is willing to allocate */
2255 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2258 /* Can we merge the block to our big extent? */
2259 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2260 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2268 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2270 * @mpd - extent of blocks for mapping
2271 * @head - the first buffer in the page
2272 * @bh - buffer we should start processing from
2273 * @lblk - logical number of the block in the file corresponding to @bh
2275 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2276 * the page for IO if all buffers in this page were mapped and there's no
2277 * accumulated extent of buffers to map or add buffers in the page to the
2278 * extent of buffers to map. The function returns 1 if the caller can continue
2279 * by processing the next page, 0 if it should stop adding buffers to the
2280 * extent to map because we cannot extend it anymore. It can also return value
2281 * < 0 in case of error during IO submission.
2283 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2284 struct buffer_head
*head
,
2285 struct buffer_head
*bh
,
2288 struct inode
*inode
= mpd
->inode
;
2290 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2291 >> inode
->i_blkbits
;
2294 BUG_ON(buffer_locked(bh
));
2296 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2297 /* Found extent to map? */
2300 /* Buffer needs mapping and handle is not started? */
2303 /* Everything mapped so far and we hit EOF */
2306 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2307 /* So far everything mapped? Submit the page for IO. */
2308 if (mpd
->map
.m_len
== 0) {
2309 err
= mpage_submit_page(mpd
, head
->b_page
);
2313 return lblk
< blocks
;
2317 * mpage_map_buffers - update buffers corresponding to changed extent and
2318 * submit fully mapped pages for IO
2320 * @mpd - description of extent to map, on return next extent to map
2322 * Scan buffers corresponding to changed extent (we expect corresponding pages
2323 * to be already locked) and update buffer state according to new extent state.
2324 * We map delalloc buffers to their physical location, clear unwritten bits,
2325 * and mark buffers as uninit when we perform writes to unwritten extents
2326 * and do extent conversion after IO is finished. If the last page is not fully
2327 * mapped, we update @map to the next extent in the last page that needs
2328 * mapping. Otherwise we submit the page for IO.
2330 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2332 struct pagevec pvec
;
2334 struct inode
*inode
= mpd
->inode
;
2335 struct buffer_head
*head
, *bh
;
2336 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2342 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2343 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2344 lblk
= start
<< bpp_bits
;
2345 pblock
= mpd
->map
.m_pblk
;
2347 pagevec_init(&pvec
, 0);
2348 while (start
<= end
) {
2349 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2353 for (i
= 0; i
< nr_pages
; i
++) {
2354 struct page
*page
= pvec
.pages
[i
];
2356 bh
= head
= page_buffers(page
);
2358 if (lblk
< mpd
->map
.m_lblk
)
2360 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2362 * Buffer after end of mapped extent.
2363 * Find next buffer in the page to map.
2366 mpd
->map
.m_flags
= 0;
2368 * FIXME: If dioread_nolock supports
2369 * blocksize < pagesize, we need to make
2370 * sure we add size mapped so far to
2371 * io_end->size as the following call
2372 * can submit the page for IO.
2374 err
= mpage_process_page_bufs(mpd
, head
,
2376 pagevec_release(&pvec
);
2381 if (buffer_delay(bh
)) {
2382 clear_buffer_delay(bh
);
2383 bh
->b_blocknr
= pblock
++;
2385 clear_buffer_unwritten(bh
);
2386 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2389 * FIXME: This is going to break if dioread_nolock
2390 * supports blocksize < pagesize as we will try to
2391 * convert potentially unmapped parts of inode.
2393 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2394 /* Page fully mapped - let IO run! */
2395 err
= mpage_submit_page(mpd
, page
);
2397 pagevec_release(&pvec
);
2401 pagevec_release(&pvec
);
2403 /* Extent fully mapped and matches with page boundary. We are done. */
2405 mpd
->map
.m_flags
= 0;
2409 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2411 struct inode
*inode
= mpd
->inode
;
2412 struct ext4_map_blocks
*map
= &mpd
->map
;
2413 int get_blocks_flags
;
2414 int err
, dioread_nolock
;
2416 trace_ext4_da_write_pages_extent(inode
, map
);
2418 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2419 * to convert an unwritten extent to be initialized (in the case
2420 * where we have written into one or more preallocated blocks). It is
2421 * possible that we're going to need more metadata blocks than
2422 * previously reserved. However we must not fail because we're in
2423 * writeback and there is nothing we can do about it so it might result
2424 * in data loss. So use reserved blocks to allocate metadata if
2427 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2428 * the blocks in question are delalloc blocks. This indicates
2429 * that the blocks and quotas has already been checked when
2430 * the data was copied into the page cache.
2432 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2433 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2434 EXT4_GET_BLOCKS_IO_SUBMIT
;
2435 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2437 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2438 if (map
->m_flags
& (1 << BH_Delay
))
2439 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2441 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2444 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2445 if (!mpd
->io_submit
.io_end
->handle
&&
2446 ext4_handle_valid(handle
)) {
2447 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2448 handle
->h_rsv_handle
= NULL
;
2450 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2453 BUG_ON(map
->m_len
== 0);
2454 if (map
->m_flags
& EXT4_MAP_NEW
) {
2455 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
2462 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2463 * mpd->len and submit pages underlying it for IO
2465 * @handle - handle for journal operations
2466 * @mpd - extent to map
2467 * @give_up_on_write - we set this to true iff there is a fatal error and there
2468 * is no hope of writing the data. The caller should discard
2469 * dirty pages to avoid infinite loops.
2471 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2472 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2473 * them to initialized or split the described range from larger unwritten
2474 * extent. Note that we need not map all the described range since allocation
2475 * can return less blocks or the range is covered by more unwritten extents. We
2476 * cannot map more because we are limited by reserved transaction credits. On
2477 * the other hand we always make sure that the last touched page is fully
2478 * mapped so that it can be written out (and thus forward progress is
2479 * guaranteed). After mapping we submit all mapped pages for IO.
2481 static int mpage_map_and_submit_extent(handle_t
*handle
,
2482 struct mpage_da_data
*mpd
,
2483 bool *give_up_on_write
)
2485 struct inode
*inode
= mpd
->inode
;
2486 struct ext4_map_blocks
*map
= &mpd
->map
;
2491 mpd
->io_submit
.io_end
->offset
=
2492 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2494 err
= mpage_map_one_extent(handle
, mpd
);
2496 struct super_block
*sb
= inode
->i_sb
;
2498 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2499 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2500 goto invalidate_dirty_pages
;
2502 * Let the uper layers retry transient errors.
2503 * In the case of ENOSPC, if ext4_count_free_blocks()
2504 * is non-zero, a commit should free up blocks.
2506 if ((err
== -ENOMEM
) ||
2507 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2509 goto update_disksize
;
2512 ext4_msg(sb
, KERN_CRIT
,
2513 "Delayed block allocation failed for "
2514 "inode %lu at logical offset %llu with"
2515 " max blocks %u with error %d",
2517 (unsigned long long)map
->m_lblk
,
2518 (unsigned)map
->m_len
, -err
);
2519 ext4_msg(sb
, KERN_CRIT
,
2520 "This should not happen!! Data will "
2523 ext4_print_free_blocks(inode
);
2524 invalidate_dirty_pages
:
2525 *give_up_on_write
= true;
2530 * Update buffer state, submit mapped pages, and get us new
2533 err
= mpage_map_and_submit_buffers(mpd
);
2535 goto update_disksize
;
2536 } while (map
->m_len
);
2540 * Update on-disk size after IO is submitted. Races with
2541 * truncate are avoided by checking i_size under i_data_sem.
2543 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2544 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2548 down_write(&EXT4_I(inode
)->i_data_sem
);
2549 i_size
= i_size_read(inode
);
2550 if (disksize
> i_size
)
2552 if (disksize
> EXT4_I(inode
)->i_disksize
)
2553 EXT4_I(inode
)->i_disksize
= disksize
;
2554 up_write(&EXT4_I(inode
)->i_data_sem
);
2555 err2
= ext4_mark_inode_dirty(handle
, inode
);
2557 ext4_error(inode
->i_sb
,
2558 "Failed to mark inode %lu dirty",
2567 * Calculate the total number of credits to reserve for one writepages
2568 * iteration. This is called from ext4_writepages(). We map an extent of
2569 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2570 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2571 * bpp - 1 blocks in bpp different extents.
2573 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2575 int bpp
= ext4_journal_blocks_per_page(inode
);
2577 return ext4_meta_trans_blocks(inode
,
2578 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2582 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2583 * and underlying extent to map
2585 * @mpd - where to look for pages
2587 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2588 * IO immediately. When we find a page which isn't mapped we start accumulating
2589 * extent of buffers underlying these pages that needs mapping (formed by
2590 * either delayed or unwritten buffers). We also lock the pages containing
2591 * these buffers. The extent found is returned in @mpd structure (starting at
2592 * mpd->lblk with length mpd->len blocks).
2594 * Note that this function can attach bios to one io_end structure which are
2595 * neither logically nor physically contiguous. Although it may seem as an
2596 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2597 * case as we need to track IO to all buffers underlying a page in one io_end.
2599 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2601 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2602 struct pagevec pvec
;
2603 unsigned int nr_pages
;
2604 long left
= mpd
->wbc
->nr_to_write
;
2605 pgoff_t index
= mpd
->first_page
;
2606 pgoff_t end
= mpd
->last_page
;
2609 int blkbits
= mpd
->inode
->i_blkbits
;
2611 struct buffer_head
*head
;
2613 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2614 tag
= PAGECACHE_TAG_TOWRITE
;
2616 tag
= PAGECACHE_TAG_DIRTY
;
2618 pagevec_init(&pvec
, 0);
2620 mpd
->next_page
= index
;
2621 while (index
<= end
) {
2622 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2623 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2627 for (i
= 0; i
< nr_pages
; i
++) {
2628 struct page
*page
= pvec
.pages
[i
];
2631 * At this point, the page may be truncated or
2632 * invalidated (changing page->mapping to NULL), or
2633 * even swizzled back from swapper_space to tmpfs file
2634 * mapping. However, page->index will not change
2635 * because we have a reference on the page.
2637 if (page
->index
> end
)
2641 * Accumulated enough dirty pages? This doesn't apply
2642 * to WB_SYNC_ALL mode. For integrity sync we have to
2643 * keep going because someone may be concurrently
2644 * dirtying pages, and we might have synced a lot of
2645 * newly appeared dirty pages, but have not synced all
2646 * of the old dirty pages.
2648 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2651 /* If we can't merge this page, we are done. */
2652 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2657 * If the page is no longer dirty, or its mapping no
2658 * longer corresponds to inode we are writing (which
2659 * means it has been truncated or invalidated), or the
2660 * page is already under writeback and we are not doing
2661 * a data integrity writeback, skip the page
2663 if (!PageDirty(page
) ||
2664 (PageWriteback(page
) &&
2665 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2666 unlikely(page
->mapping
!= mapping
)) {
2671 wait_on_page_writeback(page
);
2672 BUG_ON(PageWriteback(page
));
2674 if (mpd
->map
.m_len
== 0)
2675 mpd
->first_page
= page
->index
;
2676 mpd
->next_page
= page
->index
+ 1;
2677 /* Add all dirty buffers to mpd */
2678 lblk
= ((ext4_lblk_t
)page
->index
) <<
2679 (PAGE_SHIFT
- blkbits
);
2680 head
= page_buffers(page
);
2681 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2687 pagevec_release(&pvec
);
2692 pagevec_release(&pvec
);
2696 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2699 struct address_space
*mapping
= data
;
2700 int ret
= ext4_writepage(page
, wbc
);
2701 mapping_set_error(mapping
, ret
);
2705 static int ext4_writepages(struct address_space
*mapping
,
2706 struct writeback_control
*wbc
)
2708 pgoff_t writeback_index
= 0;
2709 long nr_to_write
= wbc
->nr_to_write
;
2710 int range_whole
= 0;
2712 handle_t
*handle
= NULL
;
2713 struct mpage_da_data mpd
;
2714 struct inode
*inode
= mapping
->host
;
2715 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2716 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2718 struct blk_plug plug
;
2719 bool give_up_on_write
= false;
2721 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2724 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2725 trace_ext4_writepages(inode
, wbc
);
2727 if (dax_mapping(mapping
)) {
2728 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2730 goto out_writepages
;
2734 * No pages to write? This is mainly a kludge to avoid starting
2735 * a transaction for special inodes like journal inode on last iput()
2736 * because that could violate lock ordering on umount
2738 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2739 goto out_writepages
;
2741 if (ext4_should_journal_data(inode
)) {
2742 struct blk_plug plug
;
2744 blk_start_plug(&plug
);
2745 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2746 blk_finish_plug(&plug
);
2747 goto out_writepages
;
2751 * If the filesystem has aborted, it is read-only, so return
2752 * right away instead of dumping stack traces later on that
2753 * will obscure the real source of the problem. We test
2754 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2755 * the latter could be true if the filesystem is mounted
2756 * read-only, and in that case, ext4_writepages should
2757 * *never* be called, so if that ever happens, we would want
2760 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2761 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2763 goto out_writepages
;
2766 if (ext4_should_dioread_nolock(inode
)) {
2768 * We may need to convert up to one extent per block in
2769 * the page and we may dirty the inode.
2771 rsv_blocks
= 1 + (PAGE_SIZE
>> inode
->i_blkbits
);
2775 * If we have inline data and arrive here, it means that
2776 * we will soon create the block for the 1st page, so
2777 * we'd better clear the inline data here.
2779 if (ext4_has_inline_data(inode
)) {
2780 /* Just inode will be modified... */
2781 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2782 if (IS_ERR(handle
)) {
2783 ret
= PTR_ERR(handle
);
2784 goto out_writepages
;
2786 BUG_ON(ext4_test_inode_state(inode
,
2787 EXT4_STATE_MAY_INLINE_DATA
));
2788 ext4_destroy_inline_data(handle
, inode
);
2789 ext4_journal_stop(handle
);
2792 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2795 if (wbc
->range_cyclic
) {
2796 writeback_index
= mapping
->writeback_index
;
2797 if (writeback_index
)
2799 mpd
.first_page
= writeback_index
;
2802 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2803 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2808 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2810 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2811 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2813 blk_start_plug(&plug
);
2816 * First writeback pages that don't need mapping - we can avoid
2817 * starting a transaction unnecessarily and also avoid being blocked
2818 * in the block layer on device congestion while having transaction
2822 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2823 if (!mpd
.io_submit
.io_end
) {
2827 ret
= mpage_prepare_extent_to_map(&mpd
);
2828 /* Submit prepared bio */
2829 ext4_io_submit(&mpd
.io_submit
);
2830 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2831 mpd
.io_submit
.io_end
= NULL
;
2832 /* Unlock pages we didn't use */
2833 mpage_release_unused_pages(&mpd
, false);
2837 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2838 /* For each extent of pages we use new io_end */
2839 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2840 if (!mpd
.io_submit
.io_end
) {
2846 * We have two constraints: We find one extent to map and we
2847 * must always write out whole page (makes a difference when
2848 * blocksize < pagesize) so that we don't block on IO when we
2849 * try to write out the rest of the page. Journalled mode is
2850 * not supported by delalloc.
2852 BUG_ON(ext4_should_journal_data(inode
));
2853 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2855 /* start a new transaction */
2856 handle
= ext4_journal_start_with_reserve(inode
,
2857 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2858 if (IS_ERR(handle
)) {
2859 ret
= PTR_ERR(handle
);
2860 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2861 "%ld pages, ino %lu; err %d", __func__
,
2862 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2863 /* Release allocated io_end */
2864 ext4_put_io_end(mpd
.io_submit
.io_end
);
2865 mpd
.io_submit
.io_end
= NULL
;
2870 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2871 ret
= mpage_prepare_extent_to_map(&mpd
);
2874 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2878 * We scanned the whole range (or exhausted
2879 * nr_to_write), submitted what was mapped and
2880 * didn't find anything needing mapping. We are
2887 * Caution: If the handle is synchronous,
2888 * ext4_journal_stop() can wait for transaction commit
2889 * to finish which may depend on writeback of pages to
2890 * complete or on page lock to be released. In that
2891 * case, we have to wait until after after we have
2892 * submitted all the IO, released page locks we hold,
2893 * and dropped io_end reference (for extent conversion
2894 * to be able to complete) before stopping the handle.
2896 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2897 ext4_journal_stop(handle
);
2901 /* Submit prepared bio */
2902 ext4_io_submit(&mpd
.io_submit
);
2903 /* Unlock pages we didn't use */
2904 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2906 * Drop our io_end reference we got from init. We have
2907 * to be careful and use deferred io_end finishing if
2908 * we are still holding the transaction as we can
2909 * release the last reference to io_end which may end
2910 * up doing unwritten extent conversion.
2913 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2914 ext4_journal_stop(handle
);
2916 ext4_put_io_end(mpd
.io_submit
.io_end
);
2917 mpd
.io_submit
.io_end
= NULL
;
2919 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2921 * Commit the transaction which would
2922 * free blocks released in the transaction
2925 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2929 /* Fatal error - ENOMEM, EIO... */
2934 blk_finish_plug(&plug
);
2935 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2937 mpd
.last_page
= writeback_index
- 1;
2943 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2945 * Set the writeback_index so that range_cyclic
2946 * mode will write it back later
2948 mapping
->writeback_index
= mpd
.first_page
;
2951 trace_ext4_writepages_result(inode
, wbc
, ret
,
2952 nr_to_write
- wbc
->nr_to_write
);
2953 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2957 static int ext4_nonda_switch(struct super_block
*sb
)
2959 s64 free_clusters
, dirty_clusters
;
2960 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2963 * switch to non delalloc mode if we are running low
2964 * on free block. The free block accounting via percpu
2965 * counters can get slightly wrong with percpu_counter_batch getting
2966 * accumulated on each CPU without updating global counters
2967 * Delalloc need an accurate free block accounting. So switch
2968 * to non delalloc when we are near to error range.
2971 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2973 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2975 * Start pushing delalloc when 1/2 of free blocks are dirty.
2977 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2978 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2980 if (2 * free_clusters
< 3 * dirty_clusters
||
2981 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2983 * free block count is less than 150% of dirty blocks
2984 * or free blocks is less than watermark
2991 /* We always reserve for an inode update; the superblock could be there too */
2992 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2994 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2997 if (pos
+ len
<= 0x7fffffffULL
)
3000 /* We might need to update the superblock to set LARGE_FILE */
3004 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3005 loff_t pos
, unsigned len
, unsigned flags
,
3006 struct page
**pagep
, void **fsdata
)
3008 int ret
, retries
= 0;
3011 struct inode
*inode
= mapping
->host
;
3014 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
3017 index
= pos
>> PAGE_SHIFT
;
3019 if (ext4_nonda_switch(inode
->i_sb
) ||
3020 S_ISLNK(inode
->i_mode
)) {
3021 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3022 return ext4_write_begin(file
, mapping
, pos
,
3023 len
, flags
, pagep
, fsdata
);
3025 *fsdata
= (void *)0;
3026 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3028 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
3029 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
3039 * grab_cache_page_write_begin() can take a long time if the
3040 * system is thrashing due to memory pressure, or if the page
3041 * is being written back. So grab it first before we start
3042 * the transaction handle. This also allows us to allocate
3043 * the page (if needed) without using GFP_NOFS.
3046 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3052 * With delayed allocation, we don't log the i_disksize update
3053 * if there is delayed block allocation. But we still need
3054 * to journalling the i_disksize update if writes to the end
3055 * of file which has an already mapped buffer.
3058 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
3059 ext4_da_write_credits(inode
, pos
, len
));
3060 if (IS_ERR(handle
)) {
3062 return PTR_ERR(handle
);
3066 if (page
->mapping
!= mapping
) {
3067 /* The page got truncated from under us */
3070 ext4_journal_stop(handle
);
3073 /* In case writeback began while the page was unlocked */
3074 wait_for_stable_page(page
);
3076 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3077 ret
= ext4_block_write_begin(page
, pos
, len
,
3078 ext4_da_get_block_prep
);
3080 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3084 ext4_journal_stop(handle
);
3086 * block_write_begin may have instantiated a few blocks
3087 * outside i_size. Trim these off again. Don't need
3088 * i_size_read because we hold i_mutex.
3090 if (pos
+ len
> inode
->i_size
)
3091 ext4_truncate_failed_write(inode
);
3093 if (ret
== -ENOSPC
&&
3094 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3106 * Check if we should update i_disksize
3107 * when write to the end of file but not require block allocation
3109 static int ext4_da_should_update_i_disksize(struct page
*page
,
3110 unsigned long offset
)
3112 struct buffer_head
*bh
;
3113 struct inode
*inode
= page
->mapping
->host
;
3117 bh
= page_buffers(page
);
3118 idx
= offset
>> inode
->i_blkbits
;
3120 for (i
= 0; i
< idx
; i
++)
3121 bh
= bh
->b_this_page
;
3123 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3128 static int ext4_da_write_end(struct file
*file
,
3129 struct address_space
*mapping
,
3130 loff_t pos
, unsigned len
, unsigned copied
,
3131 struct page
*page
, void *fsdata
)
3133 struct inode
*inode
= mapping
->host
;
3135 handle_t
*handle
= ext4_journal_current_handle();
3137 unsigned long start
, end
;
3138 int write_mode
= (int)(unsigned long)fsdata
;
3140 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3141 return ext4_write_end(file
, mapping
, pos
,
3142 len
, copied
, page
, fsdata
);
3144 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3145 start
= pos
& (PAGE_SIZE
- 1);
3146 end
= start
+ copied
- 1;
3149 * generic_write_end() will run mark_inode_dirty() if i_size
3150 * changes. So let's piggyback the i_disksize mark_inode_dirty
3153 new_i_size
= pos
+ copied
;
3154 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3155 if (ext4_has_inline_data(inode
) ||
3156 ext4_da_should_update_i_disksize(page
, end
)) {
3157 ext4_update_i_disksize(inode
, new_i_size
);
3158 /* We need to mark inode dirty even if
3159 * new_i_size is less that inode->i_size
3160 * bu greater than i_disksize.(hint delalloc)
3162 ext4_mark_inode_dirty(handle
, inode
);
3166 if (write_mode
!= CONVERT_INLINE_DATA
&&
3167 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3168 ext4_has_inline_data(inode
))
3169 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3172 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3178 ret2
= ext4_journal_stop(handle
);
3182 return ret
? ret
: copied
;
3185 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3186 unsigned int length
)
3189 * Drop reserved blocks
3191 BUG_ON(!PageLocked(page
));
3192 if (!page_has_buffers(page
))
3195 ext4_da_page_release_reservation(page
, offset
, length
);
3198 ext4_invalidatepage(page
, offset
, length
);
3204 * Force all delayed allocation blocks to be allocated for a given inode.
3206 int ext4_alloc_da_blocks(struct inode
*inode
)
3208 trace_ext4_alloc_da_blocks(inode
);
3210 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3214 * We do something simple for now. The filemap_flush() will
3215 * also start triggering a write of the data blocks, which is
3216 * not strictly speaking necessary (and for users of
3217 * laptop_mode, not even desirable). However, to do otherwise
3218 * would require replicating code paths in:
3220 * ext4_writepages() ->
3221 * write_cache_pages() ---> (via passed in callback function)
3222 * __mpage_da_writepage() -->
3223 * mpage_add_bh_to_extent()
3224 * mpage_da_map_blocks()
3226 * The problem is that write_cache_pages(), located in
3227 * mm/page-writeback.c, marks pages clean in preparation for
3228 * doing I/O, which is not desirable if we're not planning on
3231 * We could call write_cache_pages(), and then redirty all of
3232 * the pages by calling redirty_page_for_writepage() but that
3233 * would be ugly in the extreme. So instead we would need to
3234 * replicate parts of the code in the above functions,
3235 * simplifying them because we wouldn't actually intend to
3236 * write out the pages, but rather only collect contiguous
3237 * logical block extents, call the multi-block allocator, and
3238 * then update the buffer heads with the block allocations.
3240 * For now, though, we'll cheat by calling filemap_flush(),
3241 * which will map the blocks, and start the I/O, but not
3242 * actually wait for the I/O to complete.
3244 return filemap_flush(inode
->i_mapping
);
3248 * bmap() is special. It gets used by applications such as lilo and by
3249 * the swapper to find the on-disk block of a specific piece of data.
3251 * Naturally, this is dangerous if the block concerned is still in the
3252 * journal. If somebody makes a swapfile on an ext4 data-journaling
3253 * filesystem and enables swap, then they may get a nasty shock when the
3254 * data getting swapped to that swapfile suddenly gets overwritten by
3255 * the original zero's written out previously to the journal and
3256 * awaiting writeback in the kernel's buffer cache.
3258 * So, if we see any bmap calls here on a modified, data-journaled file,
3259 * take extra steps to flush any blocks which might be in the cache.
3261 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3263 struct inode
*inode
= mapping
->host
;
3268 * We can get here for an inline file via the FIBMAP ioctl
3270 if (ext4_has_inline_data(inode
))
3273 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3274 test_opt(inode
->i_sb
, DELALLOC
)) {
3276 * With delalloc we want to sync the file
3277 * so that we can make sure we allocate
3280 filemap_write_and_wait(mapping
);
3283 if (EXT4_JOURNAL(inode
) &&
3284 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3286 * This is a REALLY heavyweight approach, but the use of
3287 * bmap on dirty files is expected to be extremely rare:
3288 * only if we run lilo or swapon on a freshly made file
3289 * do we expect this to happen.
3291 * (bmap requires CAP_SYS_RAWIO so this does not
3292 * represent an unprivileged user DOS attack --- we'd be
3293 * in trouble if mortal users could trigger this path at
3296 * NB. EXT4_STATE_JDATA is not set on files other than
3297 * regular files. If somebody wants to bmap a directory
3298 * or symlink and gets confused because the buffer
3299 * hasn't yet been flushed to disk, they deserve
3300 * everything they get.
3303 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3304 journal
= EXT4_JOURNAL(inode
);
3305 jbd2_journal_lock_updates(journal
);
3306 err
= jbd2_journal_flush(journal
);
3307 jbd2_journal_unlock_updates(journal
);
3313 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3316 static int ext4_readpage(struct file
*file
, struct page
*page
)
3319 struct inode
*inode
= page
->mapping
->host
;
3321 trace_ext4_readpage(page
);
3323 if (ext4_has_inline_data(inode
))
3324 ret
= ext4_readpage_inline(inode
, page
);
3327 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3333 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3334 struct list_head
*pages
, unsigned nr_pages
)
3336 struct inode
*inode
= mapping
->host
;
3338 /* If the file has inline data, no need to do readpages. */
3339 if (ext4_has_inline_data(inode
))
3342 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3345 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3346 unsigned int length
)
3348 trace_ext4_invalidatepage(page
, offset
, length
);
3350 /* No journalling happens on data buffers when this function is used */
3351 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3353 block_invalidatepage(page
, offset
, length
);
3356 static int __ext4_journalled_invalidatepage(struct page
*page
,
3357 unsigned int offset
,
3358 unsigned int length
)
3360 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3362 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3365 * If it's a full truncate we just forget about the pending dirtying
3367 if (offset
== 0 && length
== PAGE_SIZE
)
3368 ClearPageChecked(page
);
3370 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3373 /* Wrapper for aops... */
3374 static void ext4_journalled_invalidatepage(struct page
*page
,
3375 unsigned int offset
,
3376 unsigned int length
)
3378 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3381 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3383 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3385 trace_ext4_releasepage(page
);
3387 /* Page has dirty journalled data -> cannot release */
3388 if (PageChecked(page
))
3391 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3393 return try_to_free_buffers(page
);
3396 #ifdef CONFIG_FS_DAX
3397 static bool ext4_inode_datasync_dirty(struct inode
*inode
)
3399 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
3402 return !jbd2_transaction_committed(journal
,
3403 EXT4_I(inode
)->i_datasync_tid
);
3404 /* Any metadata buffers to write? */
3405 if (!list_empty(&inode
->i_mapping
->private_list
))
3407 return inode
->i_state
& I_DIRTY_DATASYNC
;
3410 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3411 unsigned flags
, struct iomap
*iomap
)
3413 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3414 unsigned int blkbits
= inode
->i_blkbits
;
3415 unsigned long first_block
= offset
>> blkbits
;
3416 unsigned long last_block
= (offset
+ length
- 1) >> blkbits
;
3417 struct ext4_map_blocks map
;
3420 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3423 map
.m_lblk
= first_block
;
3424 map
.m_len
= last_block
- first_block
+ 1;
3426 if (!(flags
& IOMAP_WRITE
)) {
3427 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3433 /* Trim mapping request to maximum we can map at once for DIO */
3434 if (map
.m_len
> DIO_MAX_BLOCKS
)
3435 map
.m_len
= DIO_MAX_BLOCKS
;
3436 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3439 * Either we allocate blocks and then we don't get unwritten
3440 * extent so we have reserved enough credits, or the blocks
3441 * are already allocated and unwritten and in that case
3442 * extent conversion fits in the credits as well.
3444 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3447 return PTR_ERR(handle
);
3449 ret
= ext4_map_blocks(handle
, inode
, &map
,
3450 EXT4_GET_BLOCKS_CREATE_ZERO
);
3452 ext4_journal_stop(handle
);
3453 if (ret
== -ENOSPC
&&
3454 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3460 * If we added blocks beyond i_size, we need to make sure they
3461 * will get truncated if we crash before updating i_size in
3462 * ext4_iomap_end(). For faults we don't need to do that (and
3463 * even cannot because for orphan list operations inode_lock is
3464 * required) - if we happen to instantiate block beyond i_size,
3465 * it is because we race with truncate which has already added
3466 * the inode to the orphan list.
3468 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3469 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3472 err
= ext4_orphan_add(handle
, inode
);
3474 ext4_journal_stop(handle
);
3478 ext4_journal_stop(handle
);
3482 if ((flags
& IOMAP_WRITE
) && ext4_inode_datasync_dirty(inode
))
3483 iomap
->flags
|= IOMAP_F_DIRTY
;
3484 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3485 iomap
->dax_dev
= sbi
->s_daxdev
;
3486 iomap
->offset
= first_block
<< blkbits
;
3489 iomap
->type
= IOMAP_HOLE
;
3490 iomap
->blkno
= IOMAP_NULL_BLOCK
;
3491 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3493 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3494 iomap
->type
= IOMAP_MAPPED
;
3495 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3496 iomap
->type
= IOMAP_UNWRITTEN
;
3501 iomap
->blkno
= (sector_t
)map
.m_pblk
<< (blkbits
- 9);
3502 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3505 if (map
.m_flags
& EXT4_MAP_NEW
)
3506 iomap
->flags
|= IOMAP_F_NEW
;
3510 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3511 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3515 int blkbits
= inode
->i_blkbits
;
3516 bool truncate
= false;
3518 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3521 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3522 if (IS_ERR(handle
)) {
3523 ret
= PTR_ERR(handle
);
3526 if (ext4_update_inode_size(inode
, offset
+ written
))
3527 ext4_mark_inode_dirty(handle
, inode
);
3529 * We may need to truncate allocated but not written blocks beyond EOF.
3531 if (iomap
->offset
+ iomap
->length
>
3532 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3533 ext4_lblk_t written_blk
, end_blk
;
3535 written_blk
= (offset
+ written
) >> blkbits
;
3536 end_blk
= (offset
+ length
) >> blkbits
;
3537 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3541 * Remove inode from orphan list if we were extending a inode and
3542 * everything went fine.
3544 if (!truncate
&& inode
->i_nlink
&&
3545 !list_empty(&EXT4_I(inode
)->i_orphan
))
3546 ext4_orphan_del(handle
, inode
);
3547 ext4_journal_stop(handle
);
3549 ext4_truncate_failed_write(inode
);
3552 * If truncate failed early the inode might still be on the
3553 * orphan list; we need to make sure the inode is removed from
3554 * the orphan list in that case.
3557 ext4_orphan_del(NULL
, inode
);
3562 const struct iomap_ops ext4_iomap_ops
= {
3563 .iomap_begin
= ext4_iomap_begin
,
3564 .iomap_end
= ext4_iomap_end
,
3569 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3570 ssize_t size
, void *private)
3572 ext4_io_end_t
*io_end
= private;
3574 /* if not async direct IO just return */
3578 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3579 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3580 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3583 * Error during AIO DIO. We cannot convert unwritten extents as the
3584 * data was not written. Just clear the unwritten flag and drop io_end.
3587 ext4_clear_io_unwritten_flag(io_end
);
3590 io_end
->offset
= offset
;
3591 io_end
->size
= size
;
3592 ext4_put_io_end(io_end
);
3598 * Handling of direct IO writes.
3600 * For ext4 extent files, ext4 will do direct-io write even to holes,
3601 * preallocated extents, and those write extend the file, no need to
3602 * fall back to buffered IO.
3604 * For holes, we fallocate those blocks, mark them as unwritten
3605 * If those blocks were preallocated, we mark sure they are split, but
3606 * still keep the range to write as unwritten.
3608 * The unwritten extents will be converted to written when DIO is completed.
3609 * For async direct IO, since the IO may still pending when return, we
3610 * set up an end_io call back function, which will do the conversion
3611 * when async direct IO completed.
3613 * If the O_DIRECT write will extend the file then add this inode to the
3614 * orphan list. So recovery will truncate it back to the original size
3615 * if the machine crashes during the write.
3618 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3620 struct file
*file
= iocb
->ki_filp
;
3621 struct inode
*inode
= file
->f_mapping
->host
;
3622 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3624 loff_t offset
= iocb
->ki_pos
;
3625 size_t count
= iov_iter_count(iter
);
3627 get_block_t
*get_block_func
= NULL
;
3629 loff_t final_size
= offset
+ count
;
3633 if (final_size
> inode
->i_size
) {
3634 /* Credits for sb + inode write */
3635 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3636 if (IS_ERR(handle
)) {
3637 ret
= PTR_ERR(handle
);
3640 ret
= ext4_orphan_add(handle
, inode
);
3642 ext4_journal_stop(handle
);
3646 ei
->i_disksize
= inode
->i_size
;
3647 ext4_journal_stop(handle
);
3650 BUG_ON(iocb
->private == NULL
);
3653 * Make all waiters for direct IO properly wait also for extent
3654 * conversion. This also disallows race between truncate() and
3655 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3657 inode_dio_begin(inode
);
3659 /* If we do a overwrite dio, i_mutex locking can be released */
3660 overwrite
= *((int *)iocb
->private);
3663 inode_unlock(inode
);
3666 * For extent mapped files we could direct write to holes and fallocate.
3668 * Allocated blocks to fill the hole are marked as unwritten to prevent
3669 * parallel buffered read to expose the stale data before DIO complete
3672 * As to previously fallocated extents, ext4 get_block will just simply
3673 * mark the buffer mapped but still keep the extents unwritten.
3675 * For non AIO case, we will convert those unwritten extents to written
3676 * after return back from blockdev_direct_IO. That way we save us from
3677 * allocating io_end structure and also the overhead of offloading
3678 * the extent convertion to a workqueue.
3680 * For async DIO, the conversion needs to be deferred when the
3681 * IO is completed. The ext4 end_io callback function will be
3682 * called to take care of the conversion work. Here for async
3683 * case, we allocate an io_end structure to hook to the iocb.
3685 iocb
->private = NULL
;
3687 get_block_func
= ext4_dio_get_block_overwrite
;
3688 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3689 round_down(offset
, i_blocksize(inode
)) >= inode
->i_size
) {
3690 get_block_func
= ext4_dio_get_block
;
3691 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3692 } else if (is_sync_kiocb(iocb
)) {
3693 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3694 dio_flags
= DIO_LOCKING
;
3696 get_block_func
= ext4_dio_get_block_unwritten_async
;
3697 dio_flags
= DIO_LOCKING
;
3699 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3700 get_block_func
, ext4_end_io_dio
, NULL
,
3703 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3704 EXT4_STATE_DIO_UNWRITTEN
)) {
3707 * for non AIO case, since the IO is already
3708 * completed, we could do the conversion right here
3710 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3714 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3717 inode_dio_end(inode
);
3718 /* take i_mutex locking again if we do a ovewrite dio */
3722 if (ret
< 0 && final_size
> inode
->i_size
)
3723 ext4_truncate_failed_write(inode
);
3725 /* Handle extending of i_size after direct IO write */
3729 /* Credits for sb + inode write */
3730 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3731 if (IS_ERR(handle
)) {
3732 /* This is really bad luck. We've written the data
3733 * but cannot extend i_size. Bail out and pretend
3734 * the write failed... */
3735 ret
= PTR_ERR(handle
);
3737 ext4_orphan_del(NULL
, inode
);
3742 ext4_orphan_del(handle
, inode
);
3744 loff_t end
= offset
+ ret
;
3745 if (end
> inode
->i_size
) {
3746 ei
->i_disksize
= end
;
3747 i_size_write(inode
, end
);
3749 * We're going to return a positive `ret'
3750 * here due to non-zero-length I/O, so there's
3751 * no way of reporting error returns from
3752 * ext4_mark_inode_dirty() to userspace. So
3755 ext4_mark_inode_dirty(handle
, inode
);
3758 err
= ext4_journal_stop(handle
);
3766 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3768 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3769 struct inode
*inode
= mapping
->host
;
3770 size_t count
= iov_iter_count(iter
);
3774 * Shared inode_lock is enough for us - it protects against concurrent
3775 * writes & truncates and since we take care of writing back page cache,
3776 * we are protected against page writeback as well.
3778 inode_lock_shared(inode
);
3779 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3780 iocb
->ki_pos
+ count
- 1);
3783 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3784 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3786 inode_unlock_shared(inode
);
3790 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3792 struct file
*file
= iocb
->ki_filp
;
3793 struct inode
*inode
= file
->f_mapping
->host
;
3794 size_t count
= iov_iter_count(iter
);
3795 loff_t offset
= iocb
->ki_pos
;
3798 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3799 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3804 * If we are doing data journalling we don't support O_DIRECT
3806 if (ext4_should_journal_data(inode
))
3809 /* Let buffer I/O handle the inline data case. */
3810 if (ext4_has_inline_data(inode
))
3813 /* DAX uses iomap path now */
3814 if (WARN_ON_ONCE(IS_DAX(inode
)))
3817 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3818 if (iov_iter_rw(iter
) == READ
)
3819 ret
= ext4_direct_IO_read(iocb
, iter
);
3821 ret
= ext4_direct_IO_write(iocb
, iter
);
3822 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3827 * Pages can be marked dirty completely asynchronously from ext4's journalling
3828 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3829 * much here because ->set_page_dirty is called under VFS locks. The page is
3830 * not necessarily locked.
3832 * We cannot just dirty the page and leave attached buffers clean, because the
3833 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3834 * or jbddirty because all the journalling code will explode.
3836 * So what we do is to mark the page "pending dirty" and next time writepage
3837 * is called, propagate that into the buffers appropriately.
3839 static int ext4_journalled_set_page_dirty(struct page
*page
)
3841 SetPageChecked(page
);
3842 return __set_page_dirty_nobuffers(page
);
3845 static int ext4_set_page_dirty(struct page
*page
)
3847 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3848 WARN_ON_ONCE(!page_has_buffers(page
));
3849 return __set_page_dirty_buffers(page
);
3852 static const struct address_space_operations ext4_aops
= {
3853 .readpage
= ext4_readpage
,
3854 .readpages
= ext4_readpages
,
3855 .writepage
= ext4_writepage
,
3856 .writepages
= ext4_writepages
,
3857 .write_begin
= ext4_write_begin
,
3858 .write_end
= ext4_write_end
,
3859 .set_page_dirty
= ext4_set_page_dirty
,
3861 .invalidatepage
= ext4_invalidatepage
,
3862 .releasepage
= ext4_releasepage
,
3863 .direct_IO
= ext4_direct_IO
,
3864 .migratepage
= buffer_migrate_page
,
3865 .is_partially_uptodate
= block_is_partially_uptodate
,
3866 .error_remove_page
= generic_error_remove_page
,
3869 static const struct address_space_operations ext4_journalled_aops
= {
3870 .readpage
= ext4_readpage
,
3871 .readpages
= ext4_readpages
,
3872 .writepage
= ext4_writepage
,
3873 .writepages
= ext4_writepages
,
3874 .write_begin
= ext4_write_begin
,
3875 .write_end
= ext4_journalled_write_end
,
3876 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3878 .invalidatepage
= ext4_journalled_invalidatepage
,
3879 .releasepage
= ext4_releasepage
,
3880 .direct_IO
= ext4_direct_IO
,
3881 .is_partially_uptodate
= block_is_partially_uptodate
,
3882 .error_remove_page
= generic_error_remove_page
,
3885 static const struct address_space_operations ext4_da_aops
= {
3886 .readpage
= ext4_readpage
,
3887 .readpages
= ext4_readpages
,
3888 .writepage
= ext4_writepage
,
3889 .writepages
= ext4_writepages
,
3890 .write_begin
= ext4_da_write_begin
,
3891 .write_end
= ext4_da_write_end
,
3892 .set_page_dirty
= ext4_set_page_dirty
,
3894 .invalidatepage
= ext4_da_invalidatepage
,
3895 .releasepage
= ext4_releasepage
,
3896 .direct_IO
= ext4_direct_IO
,
3897 .migratepage
= buffer_migrate_page
,
3898 .is_partially_uptodate
= block_is_partially_uptodate
,
3899 .error_remove_page
= generic_error_remove_page
,
3902 void ext4_set_aops(struct inode
*inode
)
3904 switch (ext4_inode_journal_mode(inode
)) {
3905 case EXT4_INODE_ORDERED_DATA_MODE
:
3906 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3908 case EXT4_INODE_JOURNAL_DATA_MODE
:
3909 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3914 if (test_opt(inode
->i_sb
, DELALLOC
))
3915 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3917 inode
->i_mapping
->a_ops
= &ext4_aops
;
3920 static int __ext4_block_zero_page_range(handle_t
*handle
,
3921 struct address_space
*mapping
, loff_t from
, loff_t length
)
3923 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3924 unsigned offset
= from
& (PAGE_SIZE
-1);
3925 unsigned blocksize
, pos
;
3927 struct inode
*inode
= mapping
->host
;
3928 struct buffer_head
*bh
;
3932 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3933 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3937 blocksize
= inode
->i_sb
->s_blocksize
;
3939 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3941 if (!page_has_buffers(page
))
3942 create_empty_buffers(page
, blocksize
, 0);
3944 /* Find the buffer that contains "offset" */
3945 bh
= page_buffers(page
);
3947 while (offset
>= pos
) {
3948 bh
= bh
->b_this_page
;
3952 if (buffer_freed(bh
)) {
3953 BUFFER_TRACE(bh
, "freed: skip");
3956 if (!buffer_mapped(bh
)) {
3957 BUFFER_TRACE(bh
, "unmapped");
3958 ext4_get_block(inode
, iblock
, bh
, 0);
3959 /* unmapped? It's a hole - nothing to do */
3960 if (!buffer_mapped(bh
)) {
3961 BUFFER_TRACE(bh
, "still unmapped");
3966 /* Ok, it's mapped. Make sure it's up-to-date */
3967 if (PageUptodate(page
))
3968 set_buffer_uptodate(bh
);
3970 if (!buffer_uptodate(bh
)) {
3972 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3974 /* Uhhuh. Read error. Complain and punt. */
3975 if (!buffer_uptodate(bh
))
3977 if (S_ISREG(inode
->i_mode
) &&
3978 ext4_encrypted_inode(inode
)) {
3979 /* We expect the key to be set. */
3980 BUG_ON(!fscrypt_has_encryption_key(inode
));
3981 BUG_ON(blocksize
!= PAGE_SIZE
);
3982 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
3983 page
, PAGE_SIZE
, 0, page
->index
));
3986 if (ext4_should_journal_data(inode
)) {
3987 BUFFER_TRACE(bh
, "get write access");
3988 err
= ext4_journal_get_write_access(handle
, bh
);
3992 zero_user(page
, offset
, length
);
3993 BUFFER_TRACE(bh
, "zeroed end of block");
3995 if (ext4_should_journal_data(inode
)) {
3996 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3999 mark_buffer_dirty(bh
);
4000 if (ext4_should_order_data(inode
))
4001 err
= ext4_jbd2_inode_add_write(handle
, inode
);
4011 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4012 * starting from file offset 'from'. The range to be zero'd must
4013 * be contained with in one block. If the specified range exceeds
4014 * the end of the block it will be shortened to end of the block
4015 * that cooresponds to 'from'
4017 static int ext4_block_zero_page_range(handle_t
*handle
,
4018 struct address_space
*mapping
, loff_t from
, loff_t length
)
4020 struct inode
*inode
= mapping
->host
;
4021 unsigned offset
= from
& (PAGE_SIZE
-1);
4022 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4023 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
4026 * correct length if it does not fall between
4027 * 'from' and the end of the block
4029 if (length
> max
|| length
< 0)
4032 if (IS_DAX(inode
)) {
4033 return iomap_zero_range(inode
, from
, length
, NULL
,
4036 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4040 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4041 * up to the end of the block which corresponds to `from'.
4042 * This required during truncate. We need to physically zero the tail end
4043 * of that block so it doesn't yield old data if the file is later grown.
4045 static int ext4_block_truncate_page(handle_t
*handle
,
4046 struct address_space
*mapping
, loff_t from
)
4048 unsigned offset
= from
& (PAGE_SIZE
-1);
4051 struct inode
*inode
= mapping
->host
;
4053 /* If we are processing an encrypted inode during orphan list handling */
4054 if (ext4_encrypted_inode(inode
) && !fscrypt_has_encryption_key(inode
))
4057 blocksize
= inode
->i_sb
->s_blocksize
;
4058 length
= blocksize
- (offset
& (blocksize
- 1));
4060 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4063 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
4064 loff_t lstart
, loff_t length
)
4066 struct super_block
*sb
= inode
->i_sb
;
4067 struct address_space
*mapping
= inode
->i_mapping
;
4068 unsigned partial_start
, partial_end
;
4069 ext4_fsblk_t start
, end
;
4070 loff_t byte_end
= (lstart
+ length
- 1);
4073 partial_start
= lstart
& (sb
->s_blocksize
- 1);
4074 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
4076 start
= lstart
>> sb
->s_blocksize_bits
;
4077 end
= byte_end
>> sb
->s_blocksize_bits
;
4079 /* Handle partial zero within the single block */
4081 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
4082 err
= ext4_block_zero_page_range(handle
, mapping
,
4086 /* Handle partial zero out on the start of the range */
4087 if (partial_start
) {
4088 err
= ext4_block_zero_page_range(handle
, mapping
,
4089 lstart
, sb
->s_blocksize
);
4093 /* Handle partial zero out on the end of the range */
4094 if (partial_end
!= sb
->s_blocksize
- 1)
4095 err
= ext4_block_zero_page_range(handle
, mapping
,
4096 byte_end
- partial_end
,
4101 int ext4_can_truncate(struct inode
*inode
)
4103 if (S_ISREG(inode
->i_mode
))
4105 if (S_ISDIR(inode
->i_mode
))
4107 if (S_ISLNK(inode
->i_mode
))
4108 return !ext4_inode_is_fast_symlink(inode
);
4113 * We have to make sure i_disksize gets properly updated before we truncate
4114 * page cache due to hole punching or zero range. Otherwise i_disksize update
4115 * can get lost as it may have been postponed to submission of writeback but
4116 * that will never happen after we truncate page cache.
4118 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
4122 loff_t size
= i_size_read(inode
);
4124 WARN_ON(!inode_is_locked(inode
));
4125 if (offset
> size
|| offset
+ len
< size
)
4128 if (EXT4_I(inode
)->i_disksize
>= size
)
4131 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4133 return PTR_ERR(handle
);
4134 ext4_update_i_disksize(inode
, size
);
4135 ext4_mark_inode_dirty(handle
, inode
);
4136 ext4_journal_stop(handle
);
4142 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4143 * associated with the given offset and length
4145 * @inode: File inode
4146 * @offset: The offset where the hole will begin
4147 * @len: The length of the hole
4149 * Returns: 0 on success or negative on failure
4152 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4154 struct super_block
*sb
= inode
->i_sb
;
4155 ext4_lblk_t first_block
, stop_block
;
4156 struct address_space
*mapping
= inode
->i_mapping
;
4157 loff_t first_block_offset
, last_block_offset
;
4159 unsigned int credits
;
4162 if (!S_ISREG(inode
->i_mode
))
4165 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4168 * Write out all dirty pages to avoid race conditions
4169 * Then release them.
4171 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4172 ret
= filemap_write_and_wait_range(mapping
, offset
,
4173 offset
+ length
- 1);
4180 /* No need to punch hole beyond i_size */
4181 if (offset
>= inode
->i_size
)
4185 * If the hole extends beyond i_size, set the hole
4186 * to end after the page that contains i_size
4188 if (offset
+ length
> inode
->i_size
) {
4189 length
= inode
->i_size
+
4190 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4194 if (offset
& (sb
->s_blocksize
- 1) ||
4195 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4197 * Attach jinode to inode for jbd2 if we do any zeroing of
4200 ret
= ext4_inode_attach_jinode(inode
);
4206 /* Wait all existing dio workers, newcomers will block on i_mutex */
4207 ext4_inode_block_unlocked_dio(inode
);
4208 inode_dio_wait(inode
);
4211 * Prevent page faults from reinstantiating pages we have released from
4214 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4215 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4216 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4218 /* Now release the pages and zero block aligned part of pages*/
4219 if (last_block_offset
> first_block_offset
) {
4220 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4223 truncate_pagecache_range(inode
, first_block_offset
,
4227 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4228 credits
= ext4_writepage_trans_blocks(inode
);
4230 credits
= ext4_blocks_for_truncate(inode
);
4231 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4232 if (IS_ERR(handle
)) {
4233 ret
= PTR_ERR(handle
);
4234 ext4_std_error(sb
, ret
);
4238 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4243 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4244 EXT4_BLOCK_SIZE_BITS(sb
);
4245 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4247 /* If there are no blocks to remove, return now */
4248 if (first_block
>= stop_block
)
4251 down_write(&EXT4_I(inode
)->i_data_sem
);
4252 ext4_discard_preallocations(inode
);
4254 ret
= ext4_es_remove_extent(inode
, first_block
,
4255 stop_block
- first_block
);
4257 up_write(&EXT4_I(inode
)->i_data_sem
);
4261 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4262 ret
= ext4_ext_remove_space(inode
, first_block
,
4265 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4268 up_write(&EXT4_I(inode
)->i_data_sem
);
4270 ext4_handle_sync(handle
);
4272 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4273 ext4_mark_inode_dirty(handle
, inode
);
4275 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4277 ext4_journal_stop(handle
);
4279 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4280 ext4_inode_resume_unlocked_dio(inode
);
4282 inode_unlock(inode
);
4286 int ext4_inode_attach_jinode(struct inode
*inode
)
4288 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4289 struct jbd2_inode
*jinode
;
4291 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4294 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4295 spin_lock(&inode
->i_lock
);
4298 spin_unlock(&inode
->i_lock
);
4301 ei
->jinode
= jinode
;
4302 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4305 spin_unlock(&inode
->i_lock
);
4306 if (unlikely(jinode
!= NULL
))
4307 jbd2_free_inode(jinode
);
4314 * We block out ext4_get_block() block instantiations across the entire
4315 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4316 * simultaneously on behalf of the same inode.
4318 * As we work through the truncate and commit bits of it to the journal there
4319 * is one core, guiding principle: the file's tree must always be consistent on
4320 * disk. We must be able to restart the truncate after a crash.
4322 * The file's tree may be transiently inconsistent in memory (although it
4323 * probably isn't), but whenever we close off and commit a journal transaction,
4324 * the contents of (the filesystem + the journal) must be consistent and
4325 * restartable. It's pretty simple, really: bottom up, right to left (although
4326 * left-to-right works OK too).
4328 * Note that at recovery time, journal replay occurs *before* the restart of
4329 * truncate against the orphan inode list.
4331 * The committed inode has the new, desired i_size (which is the same as
4332 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4333 * that this inode's truncate did not complete and it will again call
4334 * ext4_truncate() to have another go. So there will be instantiated blocks
4335 * to the right of the truncation point in a crashed ext4 filesystem. But
4336 * that's fine - as long as they are linked from the inode, the post-crash
4337 * ext4_truncate() run will find them and release them.
4339 int ext4_truncate(struct inode
*inode
)
4341 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4342 unsigned int credits
;
4345 struct address_space
*mapping
= inode
->i_mapping
;
4348 * There is a possibility that we're either freeing the inode
4349 * or it's a completely new inode. In those cases we might not
4350 * have i_mutex locked because it's not necessary.
4352 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4353 WARN_ON(!inode_is_locked(inode
));
4354 trace_ext4_truncate_enter(inode
);
4356 if (!ext4_can_truncate(inode
))
4359 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4361 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4362 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4364 if (ext4_has_inline_data(inode
)) {
4367 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4374 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4375 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4376 if (ext4_inode_attach_jinode(inode
) < 0)
4380 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4381 credits
= ext4_writepage_trans_blocks(inode
);
4383 credits
= ext4_blocks_for_truncate(inode
);
4385 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4387 return PTR_ERR(handle
);
4389 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4390 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4393 * We add the inode to the orphan list, so that if this
4394 * truncate spans multiple transactions, and we crash, we will
4395 * resume the truncate when the filesystem recovers. It also
4396 * marks the inode dirty, to catch the new size.
4398 * Implication: the file must always be in a sane, consistent
4399 * truncatable state while each transaction commits.
4401 err
= ext4_orphan_add(handle
, inode
);
4405 down_write(&EXT4_I(inode
)->i_data_sem
);
4407 ext4_discard_preallocations(inode
);
4409 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4410 err
= ext4_ext_truncate(handle
, inode
);
4412 ext4_ind_truncate(handle
, inode
);
4414 up_write(&ei
->i_data_sem
);
4419 ext4_handle_sync(handle
);
4423 * If this was a simple ftruncate() and the file will remain alive,
4424 * then we need to clear up the orphan record which we created above.
4425 * However, if this was a real unlink then we were called by
4426 * ext4_evict_inode(), and we allow that function to clean up the
4427 * orphan info for us.
4430 ext4_orphan_del(handle
, inode
);
4432 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4433 ext4_mark_inode_dirty(handle
, inode
);
4434 ext4_journal_stop(handle
);
4436 trace_ext4_truncate_exit(inode
);
4441 * ext4_get_inode_loc returns with an extra refcount against the inode's
4442 * underlying buffer_head on success. If 'in_mem' is true, we have all
4443 * data in memory that is needed to recreate the on-disk version of this
4446 static int __ext4_get_inode_loc(struct inode
*inode
,
4447 struct ext4_iloc
*iloc
, int in_mem
)
4449 struct ext4_group_desc
*gdp
;
4450 struct buffer_head
*bh
;
4451 struct super_block
*sb
= inode
->i_sb
;
4453 int inodes_per_block
, inode_offset
;
4456 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4457 return -EFSCORRUPTED
;
4459 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4460 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4465 * Figure out the offset within the block group inode table
4467 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4468 inode_offset
= ((inode
->i_ino
- 1) %
4469 EXT4_INODES_PER_GROUP(sb
));
4470 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4471 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4473 bh
= sb_getblk(sb
, block
);
4476 if (!buffer_uptodate(bh
)) {
4480 * If the buffer has the write error flag, we have failed
4481 * to write out another inode in the same block. In this
4482 * case, we don't have to read the block because we may
4483 * read the old inode data successfully.
4485 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4486 set_buffer_uptodate(bh
);
4488 if (buffer_uptodate(bh
)) {
4489 /* someone brought it uptodate while we waited */
4495 * If we have all information of the inode in memory and this
4496 * is the only valid inode in the block, we need not read the
4500 struct buffer_head
*bitmap_bh
;
4503 start
= inode_offset
& ~(inodes_per_block
- 1);
4505 /* Is the inode bitmap in cache? */
4506 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4507 if (unlikely(!bitmap_bh
))
4511 * If the inode bitmap isn't in cache then the
4512 * optimisation may end up performing two reads instead
4513 * of one, so skip it.
4515 if (!buffer_uptodate(bitmap_bh
)) {
4519 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4520 if (i
== inode_offset
)
4522 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4526 if (i
== start
+ inodes_per_block
) {
4527 /* all other inodes are free, so skip I/O */
4528 memset(bh
->b_data
, 0, bh
->b_size
);
4529 set_buffer_uptodate(bh
);
4537 * If we need to do any I/O, try to pre-readahead extra
4538 * blocks from the inode table.
4540 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4541 ext4_fsblk_t b
, end
, table
;
4543 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4545 table
= ext4_inode_table(sb
, gdp
);
4546 /* s_inode_readahead_blks is always a power of 2 */
4547 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4551 num
= EXT4_INODES_PER_GROUP(sb
);
4552 if (ext4_has_group_desc_csum(sb
))
4553 num
-= ext4_itable_unused_count(sb
, gdp
);
4554 table
+= num
/ inodes_per_block
;
4558 sb_breadahead(sb
, b
++);
4562 * There are other valid inodes in the buffer, this inode
4563 * has in-inode xattrs, or we don't have this inode in memory.
4564 * Read the block from disk.
4566 trace_ext4_load_inode(inode
);
4568 bh
->b_end_io
= end_buffer_read_sync
;
4569 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4571 if (!buffer_uptodate(bh
)) {
4572 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4573 "unable to read itable block");
4583 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4585 /* We have all inode data except xattrs in memory here. */
4586 return __ext4_get_inode_loc(inode
, iloc
,
4587 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4590 void ext4_set_inode_flags(struct inode
*inode
)
4592 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4593 unsigned int new_fl
= 0;
4595 if (flags
& EXT4_SYNC_FL
)
4597 if (flags
& EXT4_APPEND_FL
)
4599 if (flags
& EXT4_IMMUTABLE_FL
)
4600 new_fl
|= S_IMMUTABLE
;
4601 if (flags
& EXT4_NOATIME_FL
)
4602 new_fl
|= S_NOATIME
;
4603 if (flags
& EXT4_DIRSYNC_FL
)
4604 new_fl
|= S_DIRSYNC
;
4605 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
) &&
4606 !ext4_should_journal_data(inode
) && !ext4_has_inline_data(inode
) &&
4607 !ext4_encrypted_inode(inode
))
4609 inode_set_flags(inode
, new_fl
,
4610 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4613 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4614 struct ext4_inode_info
*ei
)
4617 struct inode
*inode
= &(ei
->vfs_inode
);
4618 struct super_block
*sb
= inode
->i_sb
;
4620 if (ext4_has_feature_huge_file(sb
)) {
4621 /* we are using combined 48 bit field */
4622 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4623 le32_to_cpu(raw_inode
->i_blocks_lo
);
4624 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4625 /* i_blocks represent file system block size */
4626 return i_blocks
<< (inode
->i_blkbits
- 9);
4631 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4635 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4636 struct ext4_inode
*raw_inode
,
4637 struct ext4_inode_info
*ei
)
4639 __le32
*magic
= (void *)raw_inode
+
4640 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4641 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4642 EXT4_INODE_SIZE(inode
->i_sb
) &&
4643 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4644 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4645 ext4_find_inline_data_nolock(inode
);
4647 EXT4_I(inode
)->i_inline_off
= 0;
4650 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4652 if (!ext4_has_feature_project(inode
->i_sb
))
4654 *projid
= EXT4_I(inode
)->i_projid
;
4658 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4660 struct ext4_iloc iloc
;
4661 struct ext4_inode
*raw_inode
;
4662 struct ext4_inode_info
*ei
;
4663 struct inode
*inode
;
4664 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4672 inode
= iget_locked(sb
, ino
);
4674 return ERR_PTR(-ENOMEM
);
4675 if (!(inode
->i_state
& I_NEW
))
4681 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4684 raw_inode
= ext4_raw_inode(&iloc
);
4686 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4687 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4688 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4689 EXT4_INODE_SIZE(inode
->i_sb
) ||
4690 (ei
->i_extra_isize
& 3)) {
4691 EXT4_ERROR_INODE(inode
,
4692 "bad extra_isize %u (inode size %u)",
4694 EXT4_INODE_SIZE(inode
->i_sb
));
4695 ret
= -EFSCORRUPTED
;
4699 ei
->i_extra_isize
= 0;
4701 /* Precompute checksum seed for inode metadata */
4702 if (ext4_has_metadata_csum(sb
)) {
4703 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4705 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4706 __le32 gen
= raw_inode
->i_generation
;
4707 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4709 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4713 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4714 EXT4_ERROR_INODE(inode
, "checksum invalid");
4719 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4720 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4721 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4722 if (ext4_has_feature_project(sb
) &&
4723 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4724 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4725 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4727 i_projid
= EXT4_DEF_PROJID
;
4729 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4730 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4731 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4733 i_uid_write(inode
, i_uid
);
4734 i_gid_write(inode
, i_gid
);
4735 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4736 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4738 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4739 ei
->i_inline_off
= 0;
4740 ei
->i_dir_start_lookup
= 0;
4741 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4742 /* We now have enough fields to check if the inode was active or not.
4743 * This is needed because nfsd might try to access dead inodes
4744 * the test is that same one that e2fsck uses
4745 * NeilBrown 1999oct15
4747 if (inode
->i_nlink
== 0) {
4748 if ((inode
->i_mode
== 0 ||
4749 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4750 ino
!= EXT4_BOOT_LOADER_INO
) {
4751 /* this inode is deleted */
4755 /* The only unlinked inodes we let through here have
4756 * valid i_mode and are being read by the orphan
4757 * recovery code: that's fine, we're about to complete
4758 * the process of deleting those.
4759 * OR it is the EXT4_BOOT_LOADER_INO which is
4760 * not initialized on a new filesystem. */
4762 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4763 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4764 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4765 if (ext4_has_feature_64bit(sb
))
4767 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4768 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4769 if ((size
= i_size_read(inode
)) < 0) {
4770 EXT4_ERROR_INODE(inode
, "bad i_size value: %lld", size
);
4771 ret
= -EFSCORRUPTED
;
4774 ei
->i_disksize
= inode
->i_size
;
4776 ei
->i_reserved_quota
= 0;
4778 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4779 ei
->i_block_group
= iloc
.block_group
;
4780 ei
->i_last_alloc_group
= ~0;
4782 * NOTE! The in-memory inode i_data array is in little-endian order
4783 * even on big-endian machines: we do NOT byteswap the block numbers!
4785 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4786 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4787 INIT_LIST_HEAD(&ei
->i_orphan
);
4790 * Set transaction id's of transactions that have to be committed
4791 * to finish f[data]sync. We set them to currently running transaction
4792 * as we cannot be sure that the inode or some of its metadata isn't
4793 * part of the transaction - the inode could have been reclaimed and
4794 * now it is reread from disk.
4797 transaction_t
*transaction
;
4800 read_lock(&journal
->j_state_lock
);
4801 if (journal
->j_running_transaction
)
4802 transaction
= journal
->j_running_transaction
;
4804 transaction
= journal
->j_committing_transaction
;
4806 tid
= transaction
->t_tid
;
4808 tid
= journal
->j_commit_sequence
;
4809 read_unlock(&journal
->j_state_lock
);
4810 ei
->i_sync_tid
= tid
;
4811 ei
->i_datasync_tid
= tid
;
4814 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4815 if (ei
->i_extra_isize
== 0) {
4816 /* The extra space is currently unused. Use it. */
4817 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4818 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4819 EXT4_GOOD_OLD_INODE_SIZE
;
4821 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4825 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4826 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4827 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4828 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4830 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4831 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4832 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4833 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4835 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4840 if (ei
->i_file_acl
&&
4841 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4842 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4844 ret
= -EFSCORRUPTED
;
4846 } else if (!ext4_has_inline_data(inode
)) {
4847 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4848 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4849 (S_ISLNK(inode
->i_mode
) &&
4850 !ext4_inode_is_fast_symlink(inode
))))
4851 /* Validate extent which is part of inode */
4852 ret
= ext4_ext_check_inode(inode
);
4853 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4854 (S_ISLNK(inode
->i_mode
) &&
4855 !ext4_inode_is_fast_symlink(inode
))) {
4856 /* Validate block references which are part of inode */
4857 ret
= ext4_ind_check_inode(inode
);
4863 if (S_ISREG(inode
->i_mode
)) {
4864 inode
->i_op
= &ext4_file_inode_operations
;
4865 inode
->i_fop
= &ext4_file_operations
;
4866 ext4_set_aops(inode
);
4867 } else if (S_ISDIR(inode
->i_mode
)) {
4868 inode
->i_op
= &ext4_dir_inode_operations
;
4869 inode
->i_fop
= &ext4_dir_operations
;
4870 } else if (S_ISLNK(inode
->i_mode
)) {
4871 if (ext4_encrypted_inode(inode
)) {
4872 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4873 ext4_set_aops(inode
);
4874 } else if (ext4_inode_is_fast_symlink(inode
)) {
4875 inode
->i_link
= (char *)ei
->i_data
;
4876 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4877 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4878 sizeof(ei
->i_data
) - 1);
4880 inode
->i_op
= &ext4_symlink_inode_operations
;
4881 ext4_set_aops(inode
);
4883 inode_nohighmem(inode
);
4884 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4885 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4886 inode
->i_op
= &ext4_special_inode_operations
;
4887 if (raw_inode
->i_block
[0])
4888 init_special_inode(inode
, inode
->i_mode
,
4889 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4891 init_special_inode(inode
, inode
->i_mode
,
4892 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4893 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4894 make_bad_inode(inode
);
4896 ret
= -EFSCORRUPTED
;
4897 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4901 ext4_set_inode_flags(inode
);
4903 unlock_new_inode(inode
);
4909 return ERR_PTR(ret
);
4912 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4914 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4915 return ERR_PTR(-EFSCORRUPTED
);
4916 return ext4_iget(sb
, ino
);
4919 static int ext4_inode_blocks_set(handle_t
*handle
,
4920 struct ext4_inode
*raw_inode
,
4921 struct ext4_inode_info
*ei
)
4923 struct inode
*inode
= &(ei
->vfs_inode
);
4924 u64 i_blocks
= inode
->i_blocks
;
4925 struct super_block
*sb
= inode
->i_sb
;
4927 if (i_blocks
<= ~0U) {
4929 * i_blocks can be represented in a 32 bit variable
4930 * as multiple of 512 bytes
4932 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4933 raw_inode
->i_blocks_high
= 0;
4934 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4937 if (!ext4_has_feature_huge_file(sb
))
4940 if (i_blocks
<= 0xffffffffffffULL
) {
4942 * i_blocks can be represented in a 48 bit variable
4943 * as multiple of 512 bytes
4945 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4946 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4947 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4949 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4950 /* i_block is stored in file system block size */
4951 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4952 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4953 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4958 struct other_inode
{
4959 unsigned long orig_ino
;
4960 struct ext4_inode
*raw_inode
;
4963 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4966 struct other_inode
*oi
= (struct other_inode
*) data
;
4968 if ((inode
->i_ino
!= ino
) ||
4969 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4970 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4971 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4973 spin_lock(&inode
->i_lock
);
4974 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4975 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4976 (inode
->i_state
& I_DIRTY_TIME
)) {
4977 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4979 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4980 spin_unlock(&inode
->i_lock
);
4982 spin_lock(&ei
->i_raw_lock
);
4983 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4984 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4985 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4986 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4987 spin_unlock(&ei
->i_raw_lock
);
4988 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4991 spin_unlock(&inode
->i_lock
);
4996 * Opportunistically update the other time fields for other inodes in
4997 * the same inode table block.
4999 static void ext4_update_other_inodes_time(struct super_block
*sb
,
5000 unsigned long orig_ino
, char *buf
)
5002 struct other_inode oi
;
5004 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
5005 int inode_size
= EXT4_INODE_SIZE(sb
);
5007 oi
.orig_ino
= orig_ino
;
5009 * Calculate the first inode in the inode table block. Inode
5010 * numbers are one-based. That is, the first inode in a block
5011 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5013 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
5014 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
5015 if (ino
== orig_ino
)
5017 oi
.raw_inode
= (struct ext4_inode
*) buf
;
5018 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
5023 * Post the struct inode info into an on-disk inode location in the
5024 * buffer-cache. This gobbles the caller's reference to the
5025 * buffer_head in the inode location struct.
5027 * The caller must have write access to iloc->bh.
5029 static int ext4_do_update_inode(handle_t
*handle
,
5030 struct inode
*inode
,
5031 struct ext4_iloc
*iloc
)
5033 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5034 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5035 struct buffer_head
*bh
= iloc
->bh
;
5036 struct super_block
*sb
= inode
->i_sb
;
5037 int err
= 0, rc
, block
;
5038 int need_datasync
= 0, set_large_file
= 0;
5043 spin_lock(&ei
->i_raw_lock
);
5045 /* For fields not tracked in the in-memory inode,
5046 * initialise them to zero for new inodes. */
5047 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5048 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5050 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5051 i_uid
= i_uid_read(inode
);
5052 i_gid
= i_gid_read(inode
);
5053 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
5054 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5055 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
5056 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
5058 * Fix up interoperability with old kernels. Otherwise, old inodes get
5059 * re-used with the upper 16 bits of the uid/gid intact
5061 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
5062 raw_inode
->i_uid_high
= 0;
5063 raw_inode
->i_gid_high
= 0;
5065 raw_inode
->i_uid_high
=
5066 cpu_to_le16(high_16_bits(i_uid
));
5067 raw_inode
->i_gid_high
=
5068 cpu_to_le16(high_16_bits(i_gid
));
5071 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
5072 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
5073 raw_inode
->i_uid_high
= 0;
5074 raw_inode
->i_gid_high
= 0;
5076 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5078 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5079 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5080 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5081 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5083 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5085 spin_unlock(&ei
->i_raw_lock
);
5088 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5089 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5090 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5091 raw_inode
->i_file_acl_high
=
5092 cpu_to_le16(ei
->i_file_acl
>> 32);
5093 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5094 if (ei
->i_disksize
!= ext4_isize(inode
->i_sb
, raw_inode
)) {
5095 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5098 if (ei
->i_disksize
> 0x7fffffffULL
) {
5099 if (!ext4_has_feature_large_file(sb
) ||
5100 EXT4_SB(sb
)->s_es
->s_rev_level
==
5101 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5104 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5105 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5106 if (old_valid_dev(inode
->i_rdev
)) {
5107 raw_inode
->i_block
[0] =
5108 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5109 raw_inode
->i_block
[1] = 0;
5111 raw_inode
->i_block
[0] = 0;
5112 raw_inode
->i_block
[1] =
5113 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5114 raw_inode
->i_block
[2] = 0;
5116 } else if (!ext4_has_inline_data(inode
)) {
5117 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5118 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5121 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5122 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5123 if (ei
->i_extra_isize
) {
5124 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5125 raw_inode
->i_version_hi
=
5126 cpu_to_le32(inode
->i_version
>> 32);
5127 raw_inode
->i_extra_isize
=
5128 cpu_to_le16(ei
->i_extra_isize
);
5132 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5133 i_projid
!= EXT4_DEF_PROJID
);
5135 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5136 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5137 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5139 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5140 spin_unlock(&ei
->i_raw_lock
);
5141 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
5142 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5145 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5146 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5149 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5150 if (set_large_file
) {
5151 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5152 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5155 ext4_update_dynamic_rev(sb
);
5156 ext4_set_feature_large_file(sb
);
5157 ext4_handle_sync(handle
);
5158 err
= ext4_handle_dirty_super(handle
, sb
);
5160 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5163 ext4_std_error(inode
->i_sb
, err
);
5168 * ext4_write_inode()
5170 * We are called from a few places:
5172 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5173 * Here, there will be no transaction running. We wait for any running
5174 * transaction to commit.
5176 * - Within flush work (sys_sync(), kupdate and such).
5177 * We wait on commit, if told to.
5179 * - Within iput_final() -> write_inode_now()
5180 * We wait on commit, if told to.
5182 * In all cases it is actually safe for us to return without doing anything,
5183 * because the inode has been copied into a raw inode buffer in
5184 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5187 * Note that we are absolutely dependent upon all inode dirtiers doing the
5188 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5189 * which we are interested.
5191 * It would be a bug for them to not do this. The code:
5193 * mark_inode_dirty(inode)
5195 * inode->i_size = expr;
5197 * is in error because write_inode() could occur while `stuff()' is running,
5198 * and the new i_size will be lost. Plus the inode will no longer be on the
5199 * superblock's dirty inode list.
5201 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5205 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
5208 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5209 if (ext4_journal_current_handle()) {
5210 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5216 * No need to force transaction in WB_SYNC_NONE mode. Also
5217 * ext4_sync_fs() will force the commit after everything is
5220 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5223 err
= ext4_force_commit(inode
->i_sb
);
5225 struct ext4_iloc iloc
;
5227 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5231 * sync(2) will flush the whole buffer cache. No need to do
5232 * it here separately for each inode.
5234 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5235 sync_dirty_buffer(iloc
.bh
);
5236 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5237 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5238 "IO error syncing inode");
5247 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5248 * buffers that are attached to a page stradding i_size and are undergoing
5249 * commit. In that case we have to wait for commit to finish and try again.
5251 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5255 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5256 tid_t commit_tid
= 0;
5259 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5261 * All buffers in the last page remain valid? Then there's nothing to
5262 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5265 if (offset
> PAGE_SIZE
- i_blocksize(inode
))
5268 page
= find_lock_page(inode
->i_mapping
,
5269 inode
->i_size
>> PAGE_SHIFT
);
5272 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5273 PAGE_SIZE
- offset
);
5279 read_lock(&journal
->j_state_lock
);
5280 if (journal
->j_committing_transaction
)
5281 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5282 read_unlock(&journal
->j_state_lock
);
5284 jbd2_log_wait_commit(journal
, commit_tid
);
5291 * Called from notify_change.
5293 * We want to trap VFS attempts to truncate the file as soon as
5294 * possible. In particular, we want to make sure that when the VFS
5295 * shrinks i_size, we put the inode on the orphan list and modify
5296 * i_disksize immediately, so that during the subsequent flushing of
5297 * dirty pages and freeing of disk blocks, we can guarantee that any
5298 * commit will leave the blocks being flushed in an unused state on
5299 * disk. (On recovery, the inode will get truncated and the blocks will
5300 * be freed, so we have a strong guarantee that no future commit will
5301 * leave these blocks visible to the user.)
5303 * Another thing we have to assure is that if we are in ordered mode
5304 * and inode is still attached to the committing transaction, we must
5305 * we start writeout of all the dirty pages which are being truncated.
5306 * This way we are sure that all the data written in the previous
5307 * transaction are already on disk (truncate waits for pages under
5310 * Called with inode->i_mutex down.
5312 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5314 struct inode
*inode
= d_inode(dentry
);
5317 const unsigned int ia_valid
= attr
->ia_valid
;
5319 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5322 error
= setattr_prepare(dentry
, attr
);
5326 if (is_quota_modification(inode
, attr
)) {
5327 error
= dquot_initialize(inode
);
5331 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5332 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5335 /* (user+group)*(old+new) structure, inode write (sb,
5336 * inode block, ? - but truncate inode update has it) */
5337 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5338 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5339 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5340 if (IS_ERR(handle
)) {
5341 error
= PTR_ERR(handle
);
5345 /* dquot_transfer() calls back ext4_get_inode_usage() which
5346 * counts xattr inode references.
5348 down_read(&EXT4_I(inode
)->xattr_sem
);
5349 error
= dquot_transfer(inode
, attr
);
5350 up_read(&EXT4_I(inode
)->xattr_sem
);
5353 ext4_journal_stop(handle
);
5356 /* Update corresponding info in inode so that everything is in
5357 * one transaction */
5358 if (attr
->ia_valid
& ATTR_UID
)
5359 inode
->i_uid
= attr
->ia_uid
;
5360 if (attr
->ia_valid
& ATTR_GID
)
5361 inode
->i_gid
= attr
->ia_gid
;
5362 error
= ext4_mark_inode_dirty(handle
, inode
);
5363 ext4_journal_stop(handle
);
5366 if (attr
->ia_valid
& ATTR_SIZE
) {
5368 loff_t oldsize
= inode
->i_size
;
5369 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5371 if (ext4_encrypted_inode(inode
)) {
5372 error
= fscrypt_get_encryption_info(inode
);
5375 if (!fscrypt_has_encryption_key(inode
))
5379 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5380 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5382 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5385 if (!S_ISREG(inode
->i_mode
))
5388 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5389 inode_inc_iversion(inode
);
5391 if (ext4_should_order_data(inode
) &&
5392 (attr
->ia_size
< inode
->i_size
)) {
5393 error
= ext4_begin_ordered_truncate(inode
,
5398 if (attr
->ia_size
!= inode
->i_size
) {
5399 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5400 if (IS_ERR(handle
)) {
5401 error
= PTR_ERR(handle
);
5404 if (ext4_handle_valid(handle
) && shrink
) {
5405 error
= ext4_orphan_add(handle
, inode
);
5409 * Update c/mtime on truncate up, ext4_truncate() will
5410 * update c/mtime in shrink case below
5413 inode
->i_mtime
= current_time(inode
);
5414 inode
->i_ctime
= inode
->i_mtime
;
5416 down_write(&EXT4_I(inode
)->i_data_sem
);
5417 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5418 rc
= ext4_mark_inode_dirty(handle
, inode
);
5422 * We have to update i_size under i_data_sem together
5423 * with i_disksize to avoid races with writeback code
5424 * running ext4_wb_update_i_disksize().
5427 i_size_write(inode
, attr
->ia_size
);
5428 up_write(&EXT4_I(inode
)->i_data_sem
);
5429 ext4_journal_stop(handle
);
5432 ext4_orphan_del(NULL
, inode
);
5437 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5440 * Blocks are going to be removed from the inode. Wait
5441 * for dio in flight. Temporarily disable
5442 * dioread_nolock to prevent livelock.
5445 if (!ext4_should_journal_data(inode
)) {
5446 ext4_inode_block_unlocked_dio(inode
);
5447 inode_dio_wait(inode
);
5448 ext4_inode_resume_unlocked_dio(inode
);
5450 ext4_wait_for_tail_page_commit(inode
);
5452 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5454 * Truncate pagecache after we've waited for commit
5455 * in data=journal mode to make pages freeable.
5457 truncate_pagecache(inode
, inode
->i_size
);
5459 rc
= ext4_truncate(inode
);
5463 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5467 setattr_copy(inode
, attr
);
5468 mark_inode_dirty(inode
);
5472 * If the call to ext4_truncate failed to get a transaction handle at
5473 * all, we need to clean up the in-core orphan list manually.
5475 if (orphan
&& inode
->i_nlink
)
5476 ext4_orphan_del(NULL
, inode
);
5478 if (!error
&& (ia_valid
& ATTR_MODE
))
5479 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5482 ext4_std_error(inode
->i_sb
, error
);
5488 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5489 u32 request_mask
, unsigned int query_flags
)
5491 struct inode
*inode
= d_inode(path
->dentry
);
5492 struct ext4_inode
*raw_inode
;
5493 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5496 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5497 stat
->result_mask
|= STATX_BTIME
;
5498 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5499 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5502 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5503 if (flags
& EXT4_APPEND_FL
)
5504 stat
->attributes
|= STATX_ATTR_APPEND
;
5505 if (flags
& EXT4_COMPR_FL
)
5506 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5507 if (flags
& EXT4_ENCRYPT_FL
)
5508 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5509 if (flags
& EXT4_IMMUTABLE_FL
)
5510 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5511 if (flags
& EXT4_NODUMP_FL
)
5512 stat
->attributes
|= STATX_ATTR_NODUMP
;
5514 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5515 STATX_ATTR_COMPRESSED
|
5516 STATX_ATTR_ENCRYPTED
|
5517 STATX_ATTR_IMMUTABLE
|
5520 generic_fillattr(inode
, stat
);
5524 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5525 u32 request_mask
, unsigned int query_flags
)
5527 struct inode
*inode
= d_inode(path
->dentry
);
5528 u64 delalloc_blocks
;
5530 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5533 * If there is inline data in the inode, the inode will normally not
5534 * have data blocks allocated (it may have an external xattr block).
5535 * Report at least one sector for such files, so tools like tar, rsync,
5536 * others don't incorrectly think the file is completely sparse.
5538 if (unlikely(ext4_has_inline_data(inode
)))
5539 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5542 * We can't update i_blocks if the block allocation is delayed
5543 * otherwise in the case of system crash before the real block
5544 * allocation is done, we will have i_blocks inconsistent with
5545 * on-disk file blocks.
5546 * We always keep i_blocks updated together with real
5547 * allocation. But to not confuse with user, stat
5548 * will return the blocks that include the delayed allocation
5549 * blocks for this file.
5551 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5552 EXT4_I(inode
)->i_reserved_data_blocks
);
5553 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5557 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5560 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5561 return ext4_ind_trans_blocks(inode
, lblocks
);
5562 return ext4_ext_index_trans_blocks(inode
, pextents
);
5566 * Account for index blocks, block groups bitmaps and block group
5567 * descriptor blocks if modify datablocks and index blocks
5568 * worse case, the indexs blocks spread over different block groups
5570 * If datablocks are discontiguous, they are possible to spread over
5571 * different block groups too. If they are contiguous, with flexbg,
5572 * they could still across block group boundary.
5574 * Also account for superblock, inode, quota and xattr blocks
5576 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5579 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5585 * How many index blocks need to touch to map @lblocks logical blocks
5586 * to @pextents physical extents?
5588 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5593 * Now let's see how many group bitmaps and group descriptors need
5596 groups
= idxblocks
+ pextents
;
5598 if (groups
> ngroups
)
5600 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5601 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5603 /* bitmaps and block group descriptor blocks */
5604 ret
+= groups
+ gdpblocks
;
5606 /* Blocks for super block, inode, quota and xattr blocks */
5607 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5613 * Calculate the total number of credits to reserve to fit
5614 * the modification of a single pages into a single transaction,
5615 * which may include multiple chunks of block allocations.
5617 * This could be called via ext4_write_begin()
5619 * We need to consider the worse case, when
5620 * one new block per extent.
5622 int ext4_writepage_trans_blocks(struct inode
*inode
)
5624 int bpp
= ext4_journal_blocks_per_page(inode
);
5627 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5629 /* Account for data blocks for journalled mode */
5630 if (ext4_should_journal_data(inode
))
5636 * Calculate the journal credits for a chunk of data modification.
5638 * This is called from DIO, fallocate or whoever calling
5639 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5641 * journal buffers for data blocks are not included here, as DIO
5642 * and fallocate do no need to journal data buffers.
5644 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5646 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5650 * The caller must have previously called ext4_reserve_inode_write().
5651 * Give this, we know that the caller already has write access to iloc->bh.
5653 int ext4_mark_iloc_dirty(handle_t
*handle
,
5654 struct inode
*inode
, struct ext4_iloc
*iloc
)
5658 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5661 if (IS_I_VERSION(inode
))
5662 inode_inc_iversion(inode
);
5664 /* the do_update_inode consumes one bh->b_count */
5667 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5668 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5674 * On success, We end up with an outstanding reference count against
5675 * iloc->bh. This _must_ be cleaned up later.
5679 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5680 struct ext4_iloc
*iloc
)
5684 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5687 err
= ext4_get_inode_loc(inode
, iloc
);
5689 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5690 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5696 ext4_std_error(inode
->i_sb
, err
);
5700 static int __ext4_expand_extra_isize(struct inode
*inode
,
5701 unsigned int new_extra_isize
,
5702 struct ext4_iloc
*iloc
,
5703 handle_t
*handle
, int *no_expand
)
5705 struct ext4_inode
*raw_inode
;
5706 struct ext4_xattr_ibody_header
*header
;
5709 raw_inode
= ext4_raw_inode(iloc
);
5711 header
= IHDR(inode
, raw_inode
);
5713 /* No extended attributes present */
5714 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5715 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5716 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5717 EXT4_I(inode
)->i_extra_isize
, 0,
5718 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5719 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5723 /* try to expand with EAs present */
5724 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5728 * Inode size expansion failed; don't try again
5737 * Expand an inode by new_extra_isize bytes.
5738 * Returns 0 on success or negative error number on failure.
5740 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5741 unsigned int new_extra_isize
,
5742 struct ext4_iloc iloc
,
5748 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5752 * In nojournal mode, we can immediately attempt to expand
5753 * the inode. When journaled, we first need to obtain extra
5754 * buffer credits since we may write into the EA block
5755 * with this same handle. If journal_extend fails, then it will
5756 * only result in a minor loss of functionality for that inode.
5757 * If this is felt to be critical, then e2fsck should be run to
5758 * force a large enough s_min_extra_isize.
5760 if (ext4_handle_valid(handle
) &&
5761 jbd2_journal_extend(handle
,
5762 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) != 0)
5765 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5768 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5769 handle
, &no_expand
);
5770 ext4_write_unlock_xattr(inode
, &no_expand
);
5775 int ext4_expand_extra_isize(struct inode
*inode
,
5776 unsigned int new_extra_isize
,
5777 struct ext4_iloc
*iloc
)
5783 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5788 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5789 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5790 if (IS_ERR(handle
)) {
5791 error
= PTR_ERR(handle
);
5796 ext4_write_lock_xattr(inode
, &no_expand
);
5798 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5799 error
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5805 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
5806 handle
, &no_expand
);
5808 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
5812 ext4_write_unlock_xattr(inode
, &no_expand
);
5814 ext4_journal_stop(handle
);
5819 * What we do here is to mark the in-core inode as clean with respect to inode
5820 * dirtiness (it may still be data-dirty).
5821 * This means that the in-core inode may be reaped by prune_icache
5822 * without having to perform any I/O. This is a very good thing,
5823 * because *any* task may call prune_icache - even ones which
5824 * have a transaction open against a different journal.
5826 * Is this cheating? Not really. Sure, we haven't written the
5827 * inode out, but prune_icache isn't a user-visible syncing function.
5828 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5829 * we start and wait on commits.
5831 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5833 struct ext4_iloc iloc
;
5834 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5838 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5839 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5843 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
5844 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
5847 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5851 * ext4_dirty_inode() is called from __mark_inode_dirty()
5853 * We're really interested in the case where a file is being extended.
5854 * i_size has been changed by generic_commit_write() and we thus need
5855 * to include the updated inode in the current transaction.
5857 * Also, dquot_alloc_block() will always dirty the inode when blocks
5858 * are allocated to the file.
5860 * If the inode is marked synchronous, we don't honour that here - doing
5861 * so would cause a commit on atime updates, which we don't bother doing.
5862 * We handle synchronous inodes at the highest possible level.
5864 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5865 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5866 * to copy into the on-disk inode structure are the timestamp files.
5868 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5872 if (flags
== I_DIRTY_TIME
)
5874 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5878 ext4_mark_inode_dirty(handle
, inode
);
5880 ext4_journal_stop(handle
);
5887 * Bind an inode's backing buffer_head into this transaction, to prevent
5888 * it from being flushed to disk early. Unlike
5889 * ext4_reserve_inode_write, this leaves behind no bh reference and
5890 * returns no iloc structure, so the caller needs to repeat the iloc
5891 * lookup to mark the inode dirty later.
5893 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5895 struct ext4_iloc iloc
;
5899 err
= ext4_get_inode_loc(inode
, &iloc
);
5901 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5902 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5904 err
= ext4_handle_dirty_metadata(handle
,
5910 ext4_std_error(inode
->i_sb
, err
);
5915 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5920 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5923 * We have to be very careful here: changing a data block's
5924 * journaling status dynamically is dangerous. If we write a
5925 * data block to the journal, change the status and then delete
5926 * that block, we risk forgetting to revoke the old log record
5927 * from the journal and so a subsequent replay can corrupt data.
5928 * So, first we make sure that the journal is empty and that
5929 * nobody is changing anything.
5932 journal
= EXT4_JOURNAL(inode
);
5935 if (is_journal_aborted(journal
))
5938 /* Wait for all existing dio workers */
5939 ext4_inode_block_unlocked_dio(inode
);
5940 inode_dio_wait(inode
);
5943 * Before flushing the journal and switching inode's aops, we have
5944 * to flush all dirty data the inode has. There can be outstanding
5945 * delayed allocations, there can be unwritten extents created by
5946 * fallocate or buffered writes in dioread_nolock mode covered by
5947 * dirty data which can be converted only after flushing the dirty
5948 * data (and journalled aops don't know how to handle these cases).
5951 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5952 err
= filemap_write_and_wait(inode
->i_mapping
);
5954 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5955 ext4_inode_resume_unlocked_dio(inode
);
5960 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5961 jbd2_journal_lock_updates(journal
);
5964 * OK, there are no updates running now, and all cached data is
5965 * synced to disk. We are now in a completely consistent state
5966 * which doesn't have anything in the journal, and we know that
5967 * no filesystem updates are running, so it is safe to modify
5968 * the inode's in-core data-journaling state flag now.
5972 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5974 err
= jbd2_journal_flush(journal
);
5976 jbd2_journal_unlock_updates(journal
);
5977 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5978 ext4_inode_resume_unlocked_dio(inode
);
5981 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5983 ext4_set_aops(inode
);
5985 * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated.
5986 * E.g. S_DAX may get cleared / set.
5988 ext4_set_inode_flags(inode
);
5990 jbd2_journal_unlock_updates(journal
);
5991 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
5994 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5995 ext4_inode_resume_unlocked_dio(inode
);
5997 /* Finally we can mark the inode as dirty. */
5999 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
6001 return PTR_ERR(handle
);
6003 err
= ext4_mark_inode_dirty(handle
, inode
);
6004 ext4_handle_sync(handle
);
6005 ext4_journal_stop(handle
);
6006 ext4_std_error(inode
->i_sb
, err
);
6011 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
6013 return !buffer_mapped(bh
);
6016 int ext4_page_mkwrite(struct vm_fault
*vmf
)
6018 struct vm_area_struct
*vma
= vmf
->vma
;
6019 struct page
*page
= vmf
->page
;
6023 struct file
*file
= vma
->vm_file
;
6024 struct inode
*inode
= file_inode(file
);
6025 struct address_space
*mapping
= inode
->i_mapping
;
6027 get_block_t
*get_block
;
6030 sb_start_pagefault(inode
->i_sb
);
6031 file_update_time(vma
->vm_file
);
6033 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6035 ret
= ext4_convert_inline_data(inode
);
6039 /* Delalloc case is easy... */
6040 if (test_opt(inode
->i_sb
, DELALLOC
) &&
6041 !ext4_should_journal_data(inode
) &&
6042 !ext4_nonda_switch(inode
->i_sb
)) {
6044 ret
= block_page_mkwrite(vma
, vmf
,
6045 ext4_da_get_block_prep
);
6046 } while (ret
== -ENOSPC
&&
6047 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
6052 size
= i_size_read(inode
);
6053 /* Page got truncated from under us? */
6054 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6056 ret
= VM_FAULT_NOPAGE
;
6060 if (page
->index
== size
>> PAGE_SHIFT
)
6061 len
= size
& ~PAGE_MASK
;
6065 * Return if we have all the buffers mapped. This avoids the need to do
6066 * journal_start/journal_stop which can block and take a long time
6068 if (page_has_buffers(page
)) {
6069 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
6071 ext4_bh_unmapped
)) {
6072 /* Wait so that we don't change page under IO */
6073 wait_for_stable_page(page
);
6074 ret
= VM_FAULT_LOCKED
;
6079 /* OK, we need to fill the hole... */
6080 if (ext4_should_dioread_nolock(inode
))
6081 get_block
= ext4_get_block_unwritten
;
6083 get_block
= ext4_get_block
;
6085 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6086 ext4_writepage_trans_blocks(inode
));
6087 if (IS_ERR(handle
)) {
6088 ret
= VM_FAULT_SIGBUS
;
6091 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
6092 if (!ret
&& ext4_should_journal_data(inode
)) {
6093 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
6094 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
6096 ret
= VM_FAULT_SIGBUS
;
6097 ext4_journal_stop(handle
);
6100 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6102 ext4_journal_stop(handle
);
6103 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6106 ret
= block_page_mkwrite_return(ret
);
6108 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6109 sb_end_pagefault(inode
->i_sb
);
6113 int ext4_filemap_fault(struct vm_fault
*vmf
)
6115 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
6118 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6119 err
= filemap_fault(vmf
);
6120 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6126 * Find the first extent at or after @lblk in an inode that is not a hole.
6127 * Search for @map_len blocks at most. The extent is returned in @result.
6129 * The function returns 1 if we found an extent. The function returns 0 in
6130 * case there is no extent at or after @lblk and in that case also sets
6131 * @result->es_len to 0. In case of error, the error code is returned.
6133 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
6134 unsigned int map_len
, struct extent_status
*result
)
6136 struct ext4_map_blocks map
;
6137 struct extent_status es
= {};
6141 map
.m_len
= map_len
;
6144 * For non-extent based files this loop may iterate several times since
6145 * we do not determine full hole size.
6147 while (map
.m_len
> 0) {
6148 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
6151 /* There's extent covering m_lblk? Just return it. */
6155 ext4_es_store_pblock(result
, map
.m_pblk
);
6156 result
->es_lblk
= map
.m_lblk
;
6157 result
->es_len
= map
.m_len
;
6158 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
6159 status
= EXTENT_STATUS_UNWRITTEN
;
6161 status
= EXTENT_STATUS_WRITTEN
;
6162 ext4_es_store_status(result
, status
);
6165 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
6166 map
.m_lblk
+ map
.m_len
- 1,
6168 /* Is delalloc data before next block in extent tree? */
6169 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
6170 ext4_lblk_t offset
= 0;
6172 if (es
.es_lblk
< lblk
)
6173 offset
= lblk
- es
.es_lblk
;
6174 result
->es_lblk
= es
.es_lblk
+ offset
;
6175 ext4_es_store_pblock(result
,
6176 ext4_es_pblock(&es
) + offset
);
6177 result
->es_len
= es
.es_len
- offset
;
6178 ext4_es_store_status(result
, ext4_es_status(&es
));
6182 /* There's a hole at m_lblk, advance us after it */
6183 map
.m_lblk
+= map
.m_len
;
6184 map_len
-= map
.m_len
;
6185 map
.m_len
= map_len
;