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/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/buffer_head.h>
28 #include <linux/writeback.h>
29 #include <linux/pagevec.h>
30 #include <linux/mpage.h>
31 #include <linux/namei.h>
32 #include <linux/uio.h>
33 #include <linux/bio.h>
34 #include <linux/workqueue.h>
35 #include <linux/kernel.h>
36 #include <linux/printk.h>
37 #include <linux/slab.h>
38 #include <linux/aio.h>
39 #include <linux/bitops.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
51 struct ext4_inode_info
*ei
)
53 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
59 raw
->i_checksum_lo
= 0;
60 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
61 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
62 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
63 raw
->i_checksum_hi
= 0;
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
67 EXT4_INODE_SIZE(inode
->i_sb
));
69 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
70 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
71 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
72 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
77 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
78 struct ext4_inode_info
*ei
)
80 __u32 provided
, calculated
;
82 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
83 cpu_to_le32(EXT4_OS_LINUX
) ||
84 !ext4_has_metadata_csum(inode
->i_sb
))
87 provided
= le16_to_cpu(raw
->i_checksum_lo
);
88 calculated
= ext4_inode_csum(inode
, raw
, ei
);
89 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
90 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
91 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
95 return provided
== calculated
;
98 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
99 struct ext4_inode_info
*ei
)
103 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
104 cpu_to_le32(EXT4_OS_LINUX
) ||
105 !ext4_has_metadata_csum(inode
->i_sb
))
108 csum
= ext4_inode_csum(inode
, raw
, ei
);
109 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
110 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
111 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
112 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
115 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
118 trace_ext4_begin_ordered_truncate(inode
, new_size
);
120 * If jinode is zero, then we never opened the file for
121 * writing, so there's no need to call
122 * jbd2_journal_begin_ordered_truncate() since there's no
123 * outstanding writes we need to flush.
125 if (!EXT4_I(inode
)->jinode
)
127 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
128 EXT4_I(inode
)->jinode
,
132 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
133 unsigned int length
);
134 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
135 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
136 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
140 * Test whether an inode is a fast symlink.
142 int ext4_inode_is_fast_symlink(struct inode
*inode
)
144 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
145 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
147 if (ext4_has_inline_data(inode
))
150 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
170 jbd_debug(2, "restarting handle %p\n", handle
);
171 up_write(&EXT4_I(inode
)->i_data_sem
);
172 ret
= ext4_journal_restart(handle
, nblocks
);
173 down_write(&EXT4_I(inode
)->i_data_sem
);
174 ext4_discard_preallocations(inode
);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode
*inode
)
187 trace_ext4_evict_inode(inode
);
189 if (inode
->i_nlink
) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode
) &&
209 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
210 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
211 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
212 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
214 jbd2_complete_transaction(journal
, commit_tid
);
215 filemap_write_and_wait(&inode
->i_data
);
217 truncate_inode_pages_final(&inode
->i_data
);
219 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
223 if (is_bad_inode(inode
))
225 dquot_initialize(inode
);
227 if (ext4_should_order_data(inode
))
228 ext4_begin_ordered_truncate(inode
, 0);
229 truncate_inode_pages_final(&inode
->i_data
);
231 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode
->i_sb
);
238 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
239 ext4_blocks_for_truncate(inode
)+3);
240 if (IS_ERR(handle
)) {
241 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL
, inode
);
248 sb_end_intwrite(inode
->i_sb
);
253 ext4_handle_sync(handle
);
255 err
= ext4_mark_inode_dirty(handle
, inode
);
257 ext4_warning(inode
->i_sb
,
258 "couldn't mark inode dirty (err %d)", err
);
262 ext4_truncate(inode
);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle
, 3)) {
271 err
= ext4_journal_extend(handle
, 3);
273 err
= ext4_journal_restart(handle
, 3);
275 ext4_warning(inode
->i_sb
,
276 "couldn't extend journal (err %d)", err
);
278 ext4_journal_stop(handle
);
279 ext4_orphan_del(NULL
, inode
);
280 sb_end_intwrite(inode
->i_sb
);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle
, inode
);
294 EXT4_I(inode
)->i_dtime
= get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle
, inode
))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode
);
307 ext4_free_inode(handle
, inode
);
308 ext4_journal_stop(handle
);
309 sb_end_intwrite(inode
->i_sb
);
312 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
316 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
318 return &EXT4_I(inode
)->i_reserved_quota
;
323 * Called with i_data_sem down, which is important since we can call
324 * ext4_discard_preallocations() from here.
326 void ext4_da_update_reserve_space(struct inode
*inode
,
327 int used
, int quota_claim
)
329 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
330 struct ext4_inode_info
*ei
= EXT4_I(inode
);
332 spin_lock(&ei
->i_block_reservation_lock
);
333 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
334 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
335 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
336 "with only %d reserved data blocks",
337 __func__
, inode
->i_ino
, used
,
338 ei
->i_reserved_data_blocks
);
340 used
= ei
->i_reserved_data_blocks
;
343 /* Update per-inode reservations */
344 ei
->i_reserved_data_blocks
-= used
;
345 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
347 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
349 /* Update quota subsystem for data blocks */
351 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
354 * We did fallocate with an offset that is already delayed
355 * allocated. So on delayed allocated writeback we should
356 * not re-claim the quota for fallocated blocks.
358 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
362 * If we have done all the pending block allocations and if
363 * there aren't any writers on the inode, we can discard the
364 * inode's preallocations.
366 if ((ei
->i_reserved_data_blocks
== 0) &&
367 (atomic_read(&inode
->i_writecount
) == 0))
368 ext4_discard_preallocations(inode
);
371 static int __check_block_validity(struct inode
*inode
, const char *func
,
373 struct ext4_map_blocks
*map
)
375 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
377 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
378 "lblock %lu mapped to illegal pblock "
379 "(length %d)", (unsigned long) map
->m_lblk
,
386 #define check_block_validity(inode, map) \
387 __check_block_validity((inode), __func__, __LINE__, (map))
389 #ifdef ES_AGGRESSIVE_TEST
390 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
392 struct ext4_map_blocks
*es_map
,
393 struct ext4_map_blocks
*map
,
400 * There is a race window that the result is not the same.
401 * e.g. xfstests #223 when dioread_nolock enables. The reason
402 * is that we lookup a block mapping in extent status tree with
403 * out taking i_data_sem. So at the time the unwritten extent
404 * could be converted.
406 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
407 down_read(&EXT4_I(inode
)->i_data_sem
);
408 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
409 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
410 EXT4_GET_BLOCKS_KEEP_SIZE
);
412 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
413 EXT4_GET_BLOCKS_KEEP_SIZE
);
415 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
416 up_read((&EXT4_I(inode
)->i_data_sem
));
419 * We don't check m_len because extent will be collpased in status
420 * tree. So the m_len might not equal.
422 if (es_map
->m_lblk
!= map
->m_lblk
||
423 es_map
->m_flags
!= map
->m_flags
||
424 es_map
->m_pblk
!= map
->m_pblk
) {
425 printk("ES cache assertion failed for inode: %lu "
426 "es_cached ex [%d/%d/%llu/%x] != "
427 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
428 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
429 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
430 map
->m_len
, map
->m_pblk
, map
->m_flags
,
434 #endif /* ES_AGGRESSIVE_TEST */
437 * The ext4_map_blocks() function tries to look up the requested blocks,
438 * and returns if the blocks are already mapped.
440 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
441 * and store the allocated blocks in the result buffer head and mark it
444 * If file type is extents based, it will call ext4_ext_map_blocks(),
445 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
448 * On success, it returns the number of blocks being mapped or allocated.
449 * if create==0 and the blocks are pre-allocated and unwritten block,
450 * the result buffer head is unmapped. If the create ==1, it will make sure
451 * the buffer head is mapped.
453 * It returns 0 if plain look up failed (blocks have not been allocated), in
454 * that case, buffer head is unmapped
456 * It returns the error in case of allocation failure.
458 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
459 struct ext4_map_blocks
*map
, int flags
)
461 struct extent_status es
;
464 #ifdef ES_AGGRESSIVE_TEST
465 struct ext4_map_blocks orig_map
;
467 memcpy(&orig_map
, map
, sizeof(*map
));
471 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
472 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
473 (unsigned long) map
->m_lblk
);
476 * ext4_map_blocks returns an int, and m_len is an unsigned int
478 if (unlikely(map
->m_len
> INT_MAX
))
479 map
->m_len
= INT_MAX
;
481 /* We can handle the block number less than EXT_MAX_BLOCKS */
482 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
485 /* Lookup extent status tree firstly */
486 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
487 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
488 map
->m_pblk
= ext4_es_pblock(&es
) +
489 map
->m_lblk
- es
.es_lblk
;
490 map
->m_flags
|= ext4_es_is_written(&es
) ?
491 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
492 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
493 if (retval
> map
->m_len
)
496 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
501 #ifdef ES_AGGRESSIVE_TEST
502 ext4_map_blocks_es_recheck(handle
, inode
, map
,
509 * Try to see if we can get the block without requesting a new
512 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
513 down_read(&EXT4_I(inode
)->i_data_sem
);
514 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
515 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
516 EXT4_GET_BLOCKS_KEEP_SIZE
);
518 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
519 EXT4_GET_BLOCKS_KEEP_SIZE
);
524 if (unlikely(retval
!= map
->m_len
)) {
525 ext4_warning(inode
->i_sb
,
526 "ES len assertion failed for inode "
527 "%lu: retval %d != map->m_len %d",
528 inode
->i_ino
, retval
, map
->m_len
);
532 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
533 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
534 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
535 !(status
& EXTENT_STATUS_WRITTEN
) &&
536 ext4_find_delalloc_range(inode
, map
->m_lblk
,
537 map
->m_lblk
+ map
->m_len
- 1))
538 status
|= EXTENT_STATUS_DELAYED
;
539 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
540 map
->m_len
, map
->m_pblk
, status
);
544 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
545 up_read((&EXT4_I(inode
)->i_data_sem
));
548 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
549 ret
= check_block_validity(inode
, map
);
554 /* If it is only a block(s) look up */
555 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
559 * Returns if the blocks have already allocated
561 * Note that if blocks have been preallocated
562 * ext4_ext_get_block() returns the create = 0
563 * with buffer head unmapped.
565 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
567 * If we need to convert extent to unwritten
568 * we continue and do the actual work in
569 * ext4_ext_map_blocks()
571 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
575 * Here we clear m_flags because after allocating an new extent,
576 * it will be set again.
578 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
581 * New blocks allocate and/or writing to unwritten extent
582 * will possibly result in updating i_data, so we take
583 * the write lock of i_data_sem, and call get_block()
584 * with create == 1 flag.
586 down_write(&EXT4_I(inode
)->i_data_sem
);
589 * We need to check for EXT4 here because migrate
590 * could have changed the inode type in between
592 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
593 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
595 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
597 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
599 * We allocated new blocks which will result in
600 * i_data's format changing. Force the migrate
601 * to fail by clearing migrate flags
603 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
607 * Update reserved blocks/metadata blocks after successful
608 * block allocation which had been deferred till now. We don't
609 * support fallocate for non extent files. So we can update
610 * reserve space here.
613 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
614 ext4_da_update_reserve_space(inode
, retval
, 1);
620 if (unlikely(retval
!= map
->m_len
)) {
621 ext4_warning(inode
->i_sb
,
622 "ES len assertion failed for inode "
623 "%lu: retval %d != map->m_len %d",
624 inode
->i_ino
, retval
, map
->m_len
);
629 * If the extent has been zeroed out, we don't need to update
630 * extent status tree.
632 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
633 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
634 if (ext4_es_is_written(&es
))
637 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
638 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
639 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
640 !(status
& EXTENT_STATUS_WRITTEN
) &&
641 ext4_find_delalloc_range(inode
, map
->m_lblk
,
642 map
->m_lblk
+ map
->m_len
- 1))
643 status
|= EXTENT_STATUS_DELAYED
;
644 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
645 map
->m_pblk
, status
);
651 up_write((&EXT4_I(inode
)->i_data_sem
));
652 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
653 ret
= check_block_validity(inode
, map
);
660 static void ext4_end_io_unwritten(struct buffer_head
*bh
, int uptodate
)
662 struct inode
*inode
= bh
->b_assoc_map
->host
;
663 /* XXX: breaks on 32-bit > 16GB. Is that even supported? */
664 loff_t offset
= (loff_t
)(uintptr_t)bh
->b_private
<< inode
->i_blkbits
;
668 WARN_ON(!buffer_unwritten(bh
));
669 err
= ext4_convert_unwritten_extents(NULL
, inode
, offset
, bh
->b_size
);
672 /* Maximum number of blocks we map for direct IO at once. */
673 #define DIO_MAX_BLOCKS 4096
675 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
676 struct buffer_head
*bh
, int flags
)
678 handle_t
*handle
= ext4_journal_current_handle();
679 struct ext4_map_blocks map
;
680 int ret
= 0, started
= 0;
683 if (ext4_has_inline_data(inode
))
687 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
689 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
690 /* Direct IO write... */
691 if (map
.m_len
> DIO_MAX_BLOCKS
)
692 map
.m_len
= DIO_MAX_BLOCKS
;
693 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
694 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
696 if (IS_ERR(handle
)) {
697 ret
= PTR_ERR(handle
);
703 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
705 ext4_io_end_t
*io_end
= ext4_inode_aio(inode
);
707 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
708 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
709 if (IS_DAX(inode
) && buffer_unwritten(bh
) && !io_end
) {
710 bh
->b_assoc_map
= inode
->i_mapping
;
711 bh
->b_private
= (void *)(unsigned long)iblock
;
712 bh
->b_end_io
= ext4_end_io_unwritten
;
714 if (io_end
&& io_end
->flag
& EXT4_IO_END_UNWRITTEN
)
715 set_buffer_defer_completion(bh
);
716 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
720 ext4_journal_stop(handle
);
724 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
725 struct buffer_head
*bh
, int create
)
727 return _ext4_get_block(inode
, iblock
, bh
,
728 create
? EXT4_GET_BLOCKS_CREATE
: 0);
732 * `handle' can be NULL if create is zero
734 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
735 ext4_lblk_t block
, int create
)
737 struct ext4_map_blocks map
;
738 struct buffer_head
*bh
;
741 J_ASSERT(handle
!= NULL
|| create
== 0);
745 err
= ext4_map_blocks(handle
, inode
, &map
,
746 create
? EXT4_GET_BLOCKS_CREATE
: 0);
749 return create
? ERR_PTR(-ENOSPC
) : NULL
;
753 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
755 return ERR_PTR(-ENOMEM
);
756 if (map
.m_flags
& EXT4_MAP_NEW
) {
757 J_ASSERT(create
!= 0);
758 J_ASSERT(handle
!= NULL
);
761 * Now that we do not always journal data, we should
762 * keep in mind whether this should always journal the
763 * new buffer as metadata. For now, regular file
764 * writes use ext4_get_block instead, so it's not a
768 BUFFER_TRACE(bh
, "call get_create_access");
769 err
= ext4_journal_get_create_access(handle
, bh
);
774 if (!buffer_uptodate(bh
)) {
775 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
776 set_buffer_uptodate(bh
);
779 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
780 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
784 BUFFER_TRACE(bh
, "not a new buffer");
791 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
792 ext4_lblk_t block
, int create
)
794 struct buffer_head
*bh
;
796 bh
= ext4_getblk(handle
, inode
, block
, create
);
799 if (!bh
|| buffer_uptodate(bh
))
801 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
803 if (buffer_uptodate(bh
))
806 return ERR_PTR(-EIO
);
809 int ext4_walk_page_buffers(handle_t
*handle
,
810 struct buffer_head
*head
,
814 int (*fn
)(handle_t
*handle
,
815 struct buffer_head
*bh
))
817 struct buffer_head
*bh
;
818 unsigned block_start
, block_end
;
819 unsigned blocksize
= head
->b_size
;
821 struct buffer_head
*next
;
823 for (bh
= head
, block_start
= 0;
824 ret
== 0 && (bh
!= head
|| !block_start
);
825 block_start
= block_end
, bh
= next
) {
826 next
= bh
->b_this_page
;
827 block_end
= block_start
+ blocksize
;
828 if (block_end
<= from
|| block_start
>= to
) {
829 if (partial
&& !buffer_uptodate(bh
))
833 err
= (*fn
)(handle
, bh
);
841 * To preserve ordering, it is essential that the hole instantiation and
842 * the data write be encapsulated in a single transaction. We cannot
843 * close off a transaction and start a new one between the ext4_get_block()
844 * and the commit_write(). So doing the jbd2_journal_start at the start of
845 * prepare_write() is the right place.
847 * Also, this function can nest inside ext4_writepage(). In that case, we
848 * *know* that ext4_writepage() has generated enough buffer credits to do the
849 * whole page. So we won't block on the journal in that case, which is good,
850 * because the caller may be PF_MEMALLOC.
852 * By accident, ext4 can be reentered when a transaction is open via
853 * quota file writes. If we were to commit the transaction while thus
854 * reentered, there can be a deadlock - we would be holding a quota
855 * lock, and the commit would never complete if another thread had a
856 * transaction open and was blocking on the quota lock - a ranking
859 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
860 * will _not_ run commit under these circumstances because handle->h_ref
861 * is elevated. We'll still have enough credits for the tiny quotafile
864 int do_journal_get_write_access(handle_t
*handle
,
865 struct buffer_head
*bh
)
867 int dirty
= buffer_dirty(bh
);
870 if (!buffer_mapped(bh
) || buffer_freed(bh
))
873 * __block_write_begin() could have dirtied some buffers. Clean
874 * the dirty bit as jbd2_journal_get_write_access() could complain
875 * otherwise about fs integrity issues. Setting of the dirty bit
876 * by __block_write_begin() isn't a real problem here as we clear
877 * the bit before releasing a page lock and thus writeback cannot
878 * ever write the buffer.
881 clear_buffer_dirty(bh
);
882 BUFFER_TRACE(bh
, "get write access");
883 ret
= ext4_journal_get_write_access(handle
, bh
);
885 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
889 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
890 struct buffer_head
*bh_result
, int create
);
892 #ifdef CONFIG_EXT4_FS_ENCRYPTION
893 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
894 get_block_t
*get_block
)
896 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
897 unsigned to
= from
+ len
;
898 struct inode
*inode
= page
->mapping
->host
;
899 unsigned block_start
, block_end
;
902 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
904 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
905 bool decrypt
= false;
907 BUG_ON(!PageLocked(page
));
908 BUG_ON(from
> PAGE_CACHE_SIZE
);
909 BUG_ON(to
> PAGE_CACHE_SIZE
);
912 if (!page_has_buffers(page
))
913 create_empty_buffers(page
, blocksize
, 0);
914 head
= page_buffers(page
);
915 bbits
= ilog2(blocksize
);
916 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
918 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
919 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
920 block_end
= block_start
+ blocksize
;
921 if (block_end
<= from
|| block_start
>= to
) {
922 if (PageUptodate(page
)) {
923 if (!buffer_uptodate(bh
))
924 set_buffer_uptodate(bh
);
929 clear_buffer_new(bh
);
930 if (!buffer_mapped(bh
)) {
931 WARN_ON(bh
->b_size
!= blocksize
);
932 err
= get_block(inode
, block
, bh
, 1);
935 if (buffer_new(bh
)) {
936 unmap_underlying_metadata(bh
->b_bdev
,
938 if (PageUptodate(page
)) {
939 clear_buffer_new(bh
);
940 set_buffer_uptodate(bh
);
941 mark_buffer_dirty(bh
);
944 if (block_end
> to
|| block_start
< from
)
945 zero_user_segments(page
, to
, block_end
,
950 if (PageUptodate(page
)) {
951 if (!buffer_uptodate(bh
))
952 set_buffer_uptodate(bh
);
955 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
956 !buffer_unwritten(bh
) &&
957 (block_start
< from
|| block_end
> to
)) {
958 ll_rw_block(READ
, 1, &bh
);
960 decrypt
= ext4_encrypted_inode(inode
) &&
961 S_ISREG(inode
->i_mode
);
965 * If we issued read requests, let them complete.
967 while (wait_bh
> wait
) {
968 wait_on_buffer(*--wait_bh
);
969 if (!buffer_uptodate(*wait_bh
))
973 page_zero_new_buffers(page
, from
, to
);
975 err
= ext4_decrypt_one(inode
, page
);
980 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
981 loff_t pos
, unsigned len
, unsigned flags
,
982 struct page
**pagep
, void **fsdata
)
984 struct inode
*inode
= mapping
->host
;
985 int ret
, needed_blocks
;
992 trace_ext4_write_begin(inode
, pos
, len
, flags
);
994 * Reserve one block more for addition to orphan list in case
995 * we allocate blocks but write fails for some reason
997 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
998 index
= pos
>> PAGE_CACHE_SHIFT
;
999 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1002 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1003 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1012 * grab_cache_page_write_begin() can take a long time if the
1013 * system is thrashing due to memory pressure, or if the page
1014 * is being written back. So grab it first before we start
1015 * the transaction handle. This also allows us to allocate
1016 * the page (if needed) without using GFP_NOFS.
1019 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1025 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1026 if (IS_ERR(handle
)) {
1027 page_cache_release(page
);
1028 return PTR_ERR(handle
);
1032 if (page
->mapping
!= mapping
) {
1033 /* The page got truncated from under us */
1035 page_cache_release(page
);
1036 ext4_journal_stop(handle
);
1039 /* In case writeback began while the page was unlocked */
1040 wait_for_stable_page(page
);
1042 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1043 if (ext4_should_dioread_nolock(inode
))
1044 ret
= ext4_block_write_begin(page
, pos
, len
,
1045 ext4_get_block_write
);
1047 ret
= ext4_block_write_begin(page
, pos
, len
,
1050 if (ext4_should_dioread_nolock(inode
))
1051 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1053 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1055 if (!ret
&& ext4_should_journal_data(inode
)) {
1056 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1058 do_journal_get_write_access
);
1064 * __block_write_begin may have instantiated a few blocks
1065 * outside i_size. Trim these off again. Don't need
1066 * i_size_read because we hold i_mutex.
1068 * Add inode to orphan list in case we crash before
1071 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1072 ext4_orphan_add(handle
, inode
);
1074 ext4_journal_stop(handle
);
1075 if (pos
+ len
> inode
->i_size
) {
1076 ext4_truncate_failed_write(inode
);
1078 * If truncate failed early the inode might
1079 * still be on the orphan list; we need to
1080 * make sure the inode is removed from the
1081 * orphan list in that case.
1084 ext4_orphan_del(NULL
, inode
);
1087 if (ret
== -ENOSPC
&&
1088 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1090 page_cache_release(page
);
1097 /* For write_end() in data=journal mode */
1098 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1101 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1103 set_buffer_uptodate(bh
);
1104 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1105 clear_buffer_meta(bh
);
1106 clear_buffer_prio(bh
);
1111 * We need to pick up the new inode size which generic_commit_write gave us
1112 * `file' can be NULL - eg, when called from page_symlink().
1114 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1115 * buffers are managed internally.
1117 static int ext4_write_end(struct file
*file
,
1118 struct address_space
*mapping
,
1119 loff_t pos
, unsigned len
, unsigned copied
,
1120 struct page
*page
, void *fsdata
)
1122 handle_t
*handle
= ext4_journal_current_handle();
1123 struct inode
*inode
= mapping
->host
;
1124 loff_t old_size
= inode
->i_size
;
1126 int i_size_changed
= 0;
1128 trace_ext4_write_end(inode
, pos
, len
, copied
);
1129 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1130 ret
= ext4_jbd2_file_inode(handle
, inode
);
1133 page_cache_release(page
);
1138 if (ext4_has_inline_data(inode
)) {
1139 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1145 copied
= block_write_end(file
, mapping
, pos
,
1146 len
, copied
, page
, fsdata
);
1148 * it's important to update i_size while still holding page lock:
1149 * page writeout could otherwise come in and zero beyond i_size.
1151 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1153 page_cache_release(page
);
1156 pagecache_isize_extended(inode
, old_size
, pos
);
1158 * Don't mark the inode dirty under page lock. First, it unnecessarily
1159 * makes the holding time of page lock longer. Second, it forces lock
1160 * ordering of page lock and transaction start for journaling
1164 ext4_mark_inode_dirty(handle
, inode
);
1166 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1167 /* if we have allocated more blocks and copied
1168 * less. We will have blocks allocated outside
1169 * inode->i_size. So truncate them
1171 ext4_orphan_add(handle
, inode
);
1173 ret2
= ext4_journal_stop(handle
);
1177 if (pos
+ len
> inode
->i_size
) {
1178 ext4_truncate_failed_write(inode
);
1180 * If truncate failed early the inode might still be
1181 * on the orphan list; we need to make sure the inode
1182 * is removed from the orphan list in that case.
1185 ext4_orphan_del(NULL
, inode
);
1188 return ret
? ret
: copied
;
1191 static int ext4_journalled_write_end(struct file
*file
,
1192 struct address_space
*mapping
,
1193 loff_t pos
, unsigned len
, unsigned copied
,
1194 struct page
*page
, void *fsdata
)
1196 handle_t
*handle
= ext4_journal_current_handle();
1197 struct inode
*inode
= mapping
->host
;
1198 loff_t old_size
= inode
->i_size
;
1202 int size_changed
= 0;
1204 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1205 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1208 BUG_ON(!ext4_handle_valid(handle
));
1210 if (ext4_has_inline_data(inode
))
1211 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1215 if (!PageUptodate(page
))
1217 page_zero_new_buffers(page
, from
+copied
, to
);
1220 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1221 to
, &partial
, write_end_fn
);
1223 SetPageUptodate(page
);
1225 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1226 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1227 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1229 page_cache_release(page
);
1232 pagecache_isize_extended(inode
, old_size
, pos
);
1235 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1240 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1241 /* if we have allocated more blocks and copied
1242 * less. We will have blocks allocated outside
1243 * inode->i_size. So truncate them
1245 ext4_orphan_add(handle
, inode
);
1247 ret2
= ext4_journal_stop(handle
);
1250 if (pos
+ len
> inode
->i_size
) {
1251 ext4_truncate_failed_write(inode
);
1253 * If truncate failed early the inode might still be
1254 * on the orphan list; we need to make sure the inode
1255 * is removed from the orphan list in that case.
1258 ext4_orphan_del(NULL
, inode
);
1261 return ret
? ret
: copied
;
1265 * Reserve a single cluster located at lblock
1267 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1269 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1270 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1271 unsigned int md_needed
;
1275 * We will charge metadata quota at writeout time; this saves
1276 * us from metadata over-estimation, though we may go over by
1277 * a small amount in the end. Here we just reserve for data.
1279 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1284 * recalculate the amount of metadata blocks to reserve
1285 * in order to allocate nrblocks
1286 * worse case is one extent per block
1288 spin_lock(&ei
->i_block_reservation_lock
);
1290 * ext4_calc_metadata_amount() has side effects, which we have
1291 * to be prepared undo if we fail to claim space.
1294 trace_ext4_da_reserve_space(inode
, 0);
1296 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1297 spin_unlock(&ei
->i_block_reservation_lock
);
1298 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1301 ei
->i_reserved_data_blocks
++;
1302 spin_unlock(&ei
->i_block_reservation_lock
);
1304 return 0; /* success */
1307 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1309 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1310 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1313 return; /* Nothing to release, exit */
1315 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1317 trace_ext4_da_release_space(inode
, to_free
);
1318 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1320 * if there aren't enough reserved blocks, then the
1321 * counter is messed up somewhere. Since this
1322 * function is called from invalidate page, it's
1323 * harmless to return without any action.
1325 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1326 "ino %lu, to_free %d with only %d reserved "
1327 "data blocks", inode
->i_ino
, to_free
,
1328 ei
->i_reserved_data_blocks
);
1330 to_free
= ei
->i_reserved_data_blocks
;
1332 ei
->i_reserved_data_blocks
-= to_free
;
1334 /* update fs dirty data blocks counter */
1335 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1337 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1339 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1342 static void ext4_da_page_release_reservation(struct page
*page
,
1343 unsigned int offset
,
1344 unsigned int length
)
1347 struct buffer_head
*head
, *bh
;
1348 unsigned int curr_off
= 0;
1349 struct inode
*inode
= page
->mapping
->host
;
1350 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1351 unsigned int stop
= offset
+ length
;
1355 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1357 head
= page_buffers(page
);
1360 unsigned int next_off
= curr_off
+ bh
->b_size
;
1362 if (next_off
> stop
)
1365 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1367 clear_buffer_delay(bh
);
1369 curr_off
= next_off
;
1370 } while ((bh
= bh
->b_this_page
) != head
);
1373 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1374 ext4_es_remove_extent(inode
, lblk
, to_release
);
1377 /* If we have released all the blocks belonging to a cluster, then we
1378 * need to release the reserved space for that cluster. */
1379 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1380 while (num_clusters
> 0) {
1381 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1382 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1383 if (sbi
->s_cluster_ratio
== 1 ||
1384 !ext4_find_delalloc_cluster(inode
, lblk
))
1385 ext4_da_release_space(inode
, 1);
1392 * Delayed allocation stuff
1395 struct mpage_da_data
{
1396 struct inode
*inode
;
1397 struct writeback_control
*wbc
;
1399 pgoff_t first_page
; /* The first page to write */
1400 pgoff_t next_page
; /* Current page to examine */
1401 pgoff_t last_page
; /* Last page to examine */
1403 * Extent to map - this can be after first_page because that can be
1404 * fully mapped. We somewhat abuse m_flags to store whether the extent
1405 * is delalloc or unwritten.
1407 struct ext4_map_blocks map
;
1408 struct ext4_io_submit io_submit
; /* IO submission data */
1411 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1416 struct pagevec pvec
;
1417 struct inode
*inode
= mpd
->inode
;
1418 struct address_space
*mapping
= inode
->i_mapping
;
1420 /* This is necessary when next_page == 0. */
1421 if (mpd
->first_page
>= mpd
->next_page
)
1424 index
= mpd
->first_page
;
1425 end
= mpd
->next_page
- 1;
1427 ext4_lblk_t start
, last
;
1428 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1429 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1430 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1433 pagevec_init(&pvec
, 0);
1434 while (index
<= end
) {
1435 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1438 for (i
= 0; i
< nr_pages
; i
++) {
1439 struct page
*page
= pvec
.pages
[i
];
1440 if (page
->index
> end
)
1442 BUG_ON(!PageLocked(page
));
1443 BUG_ON(PageWriteback(page
));
1445 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1446 ClearPageUptodate(page
);
1450 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1451 pagevec_release(&pvec
);
1455 static void ext4_print_free_blocks(struct inode
*inode
)
1457 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1458 struct super_block
*sb
= inode
->i_sb
;
1459 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1461 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1462 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1463 ext4_count_free_clusters(sb
)));
1464 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1465 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1466 (long long) EXT4_C2B(EXT4_SB(sb
),
1467 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1468 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1469 (long long) EXT4_C2B(EXT4_SB(sb
),
1470 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1471 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1472 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1473 ei
->i_reserved_data_blocks
);
1477 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1479 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1483 * This function is grabs code from the very beginning of
1484 * ext4_map_blocks, but assumes that the caller is from delayed write
1485 * time. This function looks up the requested blocks and sets the
1486 * buffer delay bit under the protection of i_data_sem.
1488 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1489 struct ext4_map_blocks
*map
,
1490 struct buffer_head
*bh
)
1492 struct extent_status es
;
1494 sector_t invalid_block
= ~((sector_t
) 0xffff);
1495 #ifdef ES_AGGRESSIVE_TEST
1496 struct ext4_map_blocks orig_map
;
1498 memcpy(&orig_map
, map
, sizeof(*map
));
1501 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1505 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1506 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1507 (unsigned long) map
->m_lblk
);
1509 /* Lookup extent status tree firstly */
1510 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1511 if (ext4_es_is_hole(&es
)) {
1513 down_read(&EXT4_I(inode
)->i_data_sem
);
1518 * Delayed extent could be allocated by fallocate.
1519 * So we need to check it.
1521 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1522 map_bh(bh
, inode
->i_sb
, invalid_block
);
1524 set_buffer_delay(bh
);
1528 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1529 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1530 if (retval
> map
->m_len
)
1531 retval
= map
->m_len
;
1532 map
->m_len
= retval
;
1533 if (ext4_es_is_written(&es
))
1534 map
->m_flags
|= EXT4_MAP_MAPPED
;
1535 else if (ext4_es_is_unwritten(&es
))
1536 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1540 #ifdef ES_AGGRESSIVE_TEST
1541 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1547 * Try to see if we can get the block without requesting a new
1548 * file system block.
1550 down_read(&EXT4_I(inode
)->i_data_sem
);
1551 if (ext4_has_inline_data(inode
))
1553 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1554 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1556 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1562 * XXX: __block_prepare_write() unmaps passed block,
1566 * If the block was allocated from previously allocated cluster,
1567 * then we don't need to reserve it again. However we still need
1568 * to reserve metadata for every block we're going to write.
1570 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
<= 1 ||
1571 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1572 ret
= ext4_da_reserve_space(inode
, iblock
);
1574 /* not enough space to reserve */
1580 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1581 ~0, EXTENT_STATUS_DELAYED
);
1587 map_bh(bh
, inode
->i_sb
, invalid_block
);
1589 set_buffer_delay(bh
);
1590 } else if (retval
> 0) {
1592 unsigned int status
;
1594 if (unlikely(retval
!= map
->m_len
)) {
1595 ext4_warning(inode
->i_sb
,
1596 "ES len assertion failed for inode "
1597 "%lu: retval %d != map->m_len %d",
1598 inode
->i_ino
, retval
, map
->m_len
);
1602 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1603 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1604 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1605 map
->m_pblk
, status
);
1611 up_read((&EXT4_I(inode
)->i_data_sem
));
1617 * This is a special get_block_t callback which is used by
1618 * ext4_da_write_begin(). It will either return mapped block or
1619 * reserve space for a single block.
1621 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1622 * We also have b_blocknr = -1 and b_bdev initialized properly
1624 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1625 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1626 * initialized properly.
1628 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1629 struct buffer_head
*bh
, int create
)
1631 struct ext4_map_blocks map
;
1634 BUG_ON(create
== 0);
1635 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1637 map
.m_lblk
= iblock
;
1641 * first, we need to know whether the block is allocated already
1642 * preallocated blocks are unmapped but should treated
1643 * the same as allocated blocks.
1645 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1649 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1650 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1652 if (buffer_unwritten(bh
)) {
1653 /* A delayed write to unwritten bh should be marked
1654 * new and mapped. Mapped ensures that we don't do
1655 * get_block multiple times when we write to the same
1656 * offset and new ensures that we do proper zero out
1657 * for partial write.
1660 set_buffer_mapped(bh
);
1665 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1671 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1677 static int __ext4_journalled_writepage(struct page
*page
,
1680 struct address_space
*mapping
= page
->mapping
;
1681 struct inode
*inode
= mapping
->host
;
1682 struct buffer_head
*page_bufs
= NULL
;
1683 handle_t
*handle
= NULL
;
1684 int ret
= 0, err
= 0;
1685 int inline_data
= ext4_has_inline_data(inode
);
1686 struct buffer_head
*inode_bh
= NULL
;
1688 ClearPageChecked(page
);
1691 BUG_ON(page
->index
!= 0);
1692 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1693 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1694 if (inode_bh
== NULL
)
1697 page_bufs
= page_buffers(page
);
1702 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1705 /* As soon as we unlock the page, it can go away, but we have
1706 * references to buffers so we are safe */
1709 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1710 ext4_writepage_trans_blocks(inode
));
1711 if (IS_ERR(handle
)) {
1712 ret
= PTR_ERR(handle
);
1716 BUG_ON(!ext4_handle_valid(handle
));
1719 BUFFER_TRACE(inode_bh
, "get write access");
1720 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1722 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1725 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1726 do_journal_get_write_access
);
1728 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1733 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1734 err
= ext4_journal_stop(handle
);
1738 if (!ext4_has_inline_data(inode
))
1739 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1741 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1748 * Note that we don't need to start a transaction unless we're journaling data
1749 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1750 * need to file the inode to the transaction's list in ordered mode because if
1751 * we are writing back data added by write(), the inode is already there and if
1752 * we are writing back data modified via mmap(), no one guarantees in which
1753 * transaction the data will hit the disk. In case we are journaling data, we
1754 * cannot start transaction directly because transaction start ranks above page
1755 * lock so we have to do some magic.
1757 * This function can get called via...
1758 * - ext4_writepages after taking page lock (have journal handle)
1759 * - journal_submit_inode_data_buffers (no journal handle)
1760 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1761 * - grab_page_cache when doing write_begin (have journal handle)
1763 * We don't do any block allocation in this function. If we have page with
1764 * multiple blocks we need to write those buffer_heads that are mapped. This
1765 * is important for mmaped based write. So if we do with blocksize 1K
1766 * truncate(f, 1024);
1767 * a = mmap(f, 0, 4096);
1769 * truncate(f, 4096);
1770 * we have in the page first buffer_head mapped via page_mkwrite call back
1771 * but other buffer_heads would be unmapped but dirty (dirty done via the
1772 * do_wp_page). So writepage should write the first block. If we modify
1773 * the mmap area beyond 1024 we will again get a page_fault and the
1774 * page_mkwrite callback will do the block allocation and mark the
1775 * buffer_heads mapped.
1777 * We redirty the page if we have any buffer_heads that is either delay or
1778 * unwritten in the page.
1780 * We can get recursively called as show below.
1782 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1785 * But since we don't do any block allocation we should not deadlock.
1786 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1788 static int ext4_writepage(struct page
*page
,
1789 struct writeback_control
*wbc
)
1794 struct buffer_head
*page_bufs
= NULL
;
1795 struct inode
*inode
= page
->mapping
->host
;
1796 struct ext4_io_submit io_submit
;
1797 bool keep_towrite
= false;
1799 trace_ext4_writepage(page
);
1800 size
= i_size_read(inode
);
1801 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1802 len
= size
& ~PAGE_CACHE_MASK
;
1804 len
= PAGE_CACHE_SIZE
;
1806 page_bufs
= page_buffers(page
);
1808 * We cannot do block allocation or other extent handling in this
1809 * function. If there are buffers needing that, we have to redirty
1810 * the page. But we may reach here when we do a journal commit via
1811 * journal_submit_inode_data_buffers() and in that case we must write
1812 * allocated buffers to achieve data=ordered mode guarantees.
1814 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1815 ext4_bh_delay_or_unwritten
)) {
1816 redirty_page_for_writepage(wbc
, page
);
1817 if (current
->flags
& PF_MEMALLOC
) {
1819 * For memory cleaning there's no point in writing only
1820 * some buffers. So just bail out. Warn if we came here
1821 * from direct reclaim.
1823 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1828 keep_towrite
= true;
1831 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1833 * It's mmapped pagecache. Add buffers and journal it. There
1834 * doesn't seem much point in redirtying the page here.
1836 return __ext4_journalled_writepage(page
, len
);
1838 ext4_io_submit_init(&io_submit
, wbc
);
1839 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1840 if (!io_submit
.io_end
) {
1841 redirty_page_for_writepage(wbc
, page
);
1845 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
1846 ext4_io_submit(&io_submit
);
1847 /* Drop io_end reference we got from init */
1848 ext4_put_io_end_defer(io_submit
.io_end
);
1852 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1855 loff_t size
= i_size_read(mpd
->inode
);
1858 BUG_ON(page
->index
!= mpd
->first_page
);
1859 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1860 len
= size
& ~PAGE_CACHE_MASK
;
1862 len
= PAGE_CACHE_SIZE
;
1863 clear_page_dirty_for_io(page
);
1864 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
1866 mpd
->wbc
->nr_to_write
--;
1872 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1875 * mballoc gives us at most this number of blocks...
1876 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1877 * The rest of mballoc seems to handle chunks up to full group size.
1879 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1882 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1884 * @mpd - extent of blocks
1885 * @lblk - logical number of the block in the file
1886 * @bh - buffer head we want to add to the extent
1888 * The function is used to collect contig. blocks in the same state. If the
1889 * buffer doesn't require mapping for writeback and we haven't started the
1890 * extent of buffers to map yet, the function returns 'true' immediately - the
1891 * caller can write the buffer right away. Otherwise the function returns true
1892 * if the block has been added to the extent, false if the block couldn't be
1895 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
1896 struct buffer_head
*bh
)
1898 struct ext4_map_blocks
*map
= &mpd
->map
;
1900 /* Buffer that doesn't need mapping for writeback? */
1901 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
1902 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
1903 /* So far no extent to map => we write the buffer right away */
1904 if (map
->m_len
== 0)
1909 /* First block in the extent? */
1910 if (map
->m_len
== 0) {
1913 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
1917 /* Don't go larger than mballoc is willing to allocate */
1918 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
1921 /* Can we merge the block to our big extent? */
1922 if (lblk
== map
->m_lblk
+ map
->m_len
&&
1923 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
1931 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1933 * @mpd - extent of blocks for mapping
1934 * @head - the first buffer in the page
1935 * @bh - buffer we should start processing from
1936 * @lblk - logical number of the block in the file corresponding to @bh
1938 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1939 * the page for IO if all buffers in this page were mapped and there's no
1940 * accumulated extent of buffers to map or add buffers in the page to the
1941 * extent of buffers to map. The function returns 1 if the caller can continue
1942 * by processing the next page, 0 if it should stop adding buffers to the
1943 * extent to map because we cannot extend it anymore. It can also return value
1944 * < 0 in case of error during IO submission.
1946 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
1947 struct buffer_head
*head
,
1948 struct buffer_head
*bh
,
1951 struct inode
*inode
= mpd
->inode
;
1953 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
1954 >> inode
->i_blkbits
;
1957 BUG_ON(buffer_locked(bh
));
1959 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
1960 /* Found extent to map? */
1963 /* Everything mapped so far and we hit EOF */
1966 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
1967 /* So far everything mapped? Submit the page for IO. */
1968 if (mpd
->map
.m_len
== 0) {
1969 err
= mpage_submit_page(mpd
, head
->b_page
);
1973 return lblk
< blocks
;
1977 * mpage_map_buffers - update buffers corresponding to changed extent and
1978 * submit fully mapped pages for IO
1980 * @mpd - description of extent to map, on return next extent to map
1982 * Scan buffers corresponding to changed extent (we expect corresponding pages
1983 * to be already locked) and update buffer state according to new extent state.
1984 * We map delalloc buffers to their physical location, clear unwritten bits,
1985 * and mark buffers as uninit when we perform writes to unwritten extents
1986 * and do extent conversion after IO is finished. If the last page is not fully
1987 * mapped, we update @map to the next extent in the last page that needs
1988 * mapping. Otherwise we submit the page for IO.
1990 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
1992 struct pagevec pvec
;
1994 struct inode
*inode
= mpd
->inode
;
1995 struct buffer_head
*head
, *bh
;
1996 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
2002 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2003 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2004 lblk
= start
<< bpp_bits
;
2005 pblock
= mpd
->map
.m_pblk
;
2007 pagevec_init(&pvec
, 0);
2008 while (start
<= end
) {
2009 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2013 for (i
= 0; i
< nr_pages
; i
++) {
2014 struct page
*page
= pvec
.pages
[i
];
2016 if (page
->index
> end
)
2018 /* Up to 'end' pages must be contiguous */
2019 BUG_ON(page
->index
!= start
);
2020 bh
= head
= page_buffers(page
);
2022 if (lblk
< mpd
->map
.m_lblk
)
2024 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2026 * Buffer after end of mapped extent.
2027 * Find next buffer in the page to map.
2030 mpd
->map
.m_flags
= 0;
2032 * FIXME: If dioread_nolock supports
2033 * blocksize < pagesize, we need to make
2034 * sure we add size mapped so far to
2035 * io_end->size as the following call
2036 * can submit the page for IO.
2038 err
= mpage_process_page_bufs(mpd
, head
,
2040 pagevec_release(&pvec
);
2045 if (buffer_delay(bh
)) {
2046 clear_buffer_delay(bh
);
2047 bh
->b_blocknr
= pblock
++;
2049 clear_buffer_unwritten(bh
);
2050 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2053 * FIXME: This is going to break if dioread_nolock
2054 * supports blocksize < pagesize as we will try to
2055 * convert potentially unmapped parts of inode.
2057 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
2058 /* Page fully mapped - let IO run! */
2059 err
= mpage_submit_page(mpd
, page
);
2061 pagevec_release(&pvec
);
2066 pagevec_release(&pvec
);
2068 /* Extent fully mapped and matches with page boundary. We are done. */
2070 mpd
->map
.m_flags
= 0;
2074 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2076 struct inode
*inode
= mpd
->inode
;
2077 struct ext4_map_blocks
*map
= &mpd
->map
;
2078 int get_blocks_flags
;
2079 int err
, dioread_nolock
;
2081 trace_ext4_da_write_pages_extent(inode
, map
);
2083 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2084 * to convert an unwritten extent to be initialized (in the case
2085 * where we have written into one or more preallocated blocks). It is
2086 * possible that we're going to need more metadata blocks than
2087 * previously reserved. However we must not fail because we're in
2088 * writeback and there is nothing we can do about it so it might result
2089 * in data loss. So use reserved blocks to allocate metadata if
2092 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2093 * the blocks in question are delalloc blocks. This indicates
2094 * that the blocks and quotas has already been checked when
2095 * the data was copied into the page cache.
2097 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2098 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
2099 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2101 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2102 if (map
->m_flags
& (1 << BH_Delay
))
2103 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2105 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2108 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2109 if (!mpd
->io_submit
.io_end
->handle
&&
2110 ext4_handle_valid(handle
)) {
2111 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2112 handle
->h_rsv_handle
= NULL
;
2114 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2117 BUG_ON(map
->m_len
== 0);
2118 if (map
->m_flags
& EXT4_MAP_NEW
) {
2119 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2122 for (i
= 0; i
< map
->m_len
; i
++)
2123 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2129 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2130 * mpd->len and submit pages underlying it for IO
2132 * @handle - handle for journal operations
2133 * @mpd - extent to map
2134 * @give_up_on_write - we set this to true iff there is a fatal error and there
2135 * is no hope of writing the data. The caller should discard
2136 * dirty pages to avoid infinite loops.
2138 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2139 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2140 * them to initialized or split the described range from larger unwritten
2141 * extent. Note that we need not map all the described range since allocation
2142 * can return less blocks or the range is covered by more unwritten extents. We
2143 * cannot map more because we are limited by reserved transaction credits. On
2144 * the other hand we always make sure that the last touched page is fully
2145 * mapped so that it can be written out (and thus forward progress is
2146 * guaranteed). After mapping we submit all mapped pages for IO.
2148 static int mpage_map_and_submit_extent(handle_t
*handle
,
2149 struct mpage_da_data
*mpd
,
2150 bool *give_up_on_write
)
2152 struct inode
*inode
= mpd
->inode
;
2153 struct ext4_map_blocks
*map
= &mpd
->map
;
2158 mpd
->io_submit
.io_end
->offset
=
2159 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2161 err
= mpage_map_one_extent(handle
, mpd
);
2163 struct super_block
*sb
= inode
->i_sb
;
2165 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2166 goto invalidate_dirty_pages
;
2168 * Let the uper layers retry transient errors.
2169 * In the case of ENOSPC, if ext4_count_free_blocks()
2170 * is non-zero, a commit should free up blocks.
2172 if ((err
== -ENOMEM
) ||
2173 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2175 goto update_disksize
;
2178 ext4_msg(sb
, KERN_CRIT
,
2179 "Delayed block allocation failed for "
2180 "inode %lu at logical offset %llu with"
2181 " max blocks %u with error %d",
2183 (unsigned long long)map
->m_lblk
,
2184 (unsigned)map
->m_len
, -err
);
2185 ext4_msg(sb
, KERN_CRIT
,
2186 "This should not happen!! Data will "
2189 ext4_print_free_blocks(inode
);
2190 invalidate_dirty_pages
:
2191 *give_up_on_write
= true;
2196 * Update buffer state, submit mapped pages, and get us new
2199 err
= mpage_map_and_submit_buffers(mpd
);
2201 goto update_disksize
;
2202 } while (map
->m_len
);
2206 * Update on-disk size after IO is submitted. Races with
2207 * truncate are avoided by checking i_size under i_data_sem.
2209 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2210 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2214 down_write(&EXT4_I(inode
)->i_data_sem
);
2215 i_size
= i_size_read(inode
);
2216 if (disksize
> i_size
)
2218 if (disksize
> EXT4_I(inode
)->i_disksize
)
2219 EXT4_I(inode
)->i_disksize
= disksize
;
2220 err2
= ext4_mark_inode_dirty(handle
, inode
);
2221 up_write(&EXT4_I(inode
)->i_data_sem
);
2223 ext4_error(inode
->i_sb
,
2224 "Failed to mark inode %lu dirty",
2233 * Calculate the total number of credits to reserve for one writepages
2234 * iteration. This is called from ext4_writepages(). We map an extent of
2235 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2236 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2237 * bpp - 1 blocks in bpp different extents.
2239 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2241 int bpp
= ext4_journal_blocks_per_page(inode
);
2243 return ext4_meta_trans_blocks(inode
,
2244 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2248 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2249 * and underlying extent to map
2251 * @mpd - where to look for pages
2253 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2254 * IO immediately. When we find a page which isn't mapped we start accumulating
2255 * extent of buffers underlying these pages that needs mapping (formed by
2256 * either delayed or unwritten buffers). We also lock the pages containing
2257 * these buffers. The extent found is returned in @mpd structure (starting at
2258 * mpd->lblk with length mpd->len blocks).
2260 * Note that this function can attach bios to one io_end structure which are
2261 * neither logically nor physically contiguous. Although it may seem as an
2262 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2263 * case as we need to track IO to all buffers underlying a page in one io_end.
2265 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2267 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2268 struct pagevec pvec
;
2269 unsigned int nr_pages
;
2270 long left
= mpd
->wbc
->nr_to_write
;
2271 pgoff_t index
= mpd
->first_page
;
2272 pgoff_t end
= mpd
->last_page
;
2275 int blkbits
= mpd
->inode
->i_blkbits
;
2277 struct buffer_head
*head
;
2279 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2280 tag
= PAGECACHE_TAG_TOWRITE
;
2282 tag
= PAGECACHE_TAG_DIRTY
;
2284 pagevec_init(&pvec
, 0);
2286 mpd
->next_page
= index
;
2287 while (index
<= end
) {
2288 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2289 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2293 for (i
= 0; i
< nr_pages
; i
++) {
2294 struct page
*page
= pvec
.pages
[i
];
2297 * At this point, the page may be truncated or
2298 * invalidated (changing page->mapping to NULL), or
2299 * even swizzled back from swapper_space to tmpfs file
2300 * mapping. However, page->index will not change
2301 * because we have a reference on the page.
2303 if (page
->index
> end
)
2307 * Accumulated enough dirty pages? This doesn't apply
2308 * to WB_SYNC_ALL mode. For integrity sync we have to
2309 * keep going because someone may be concurrently
2310 * dirtying pages, and we might have synced a lot of
2311 * newly appeared dirty pages, but have not synced all
2312 * of the old dirty pages.
2314 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2317 /* If we can't merge this page, we are done. */
2318 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2323 * If the page is no longer dirty, or its mapping no
2324 * longer corresponds to inode we are writing (which
2325 * means it has been truncated or invalidated), or the
2326 * page is already under writeback and we are not doing
2327 * a data integrity writeback, skip the page
2329 if (!PageDirty(page
) ||
2330 (PageWriteback(page
) &&
2331 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2332 unlikely(page
->mapping
!= mapping
)) {
2337 wait_on_page_writeback(page
);
2338 BUG_ON(PageWriteback(page
));
2340 if (mpd
->map
.m_len
== 0)
2341 mpd
->first_page
= page
->index
;
2342 mpd
->next_page
= page
->index
+ 1;
2343 /* Add all dirty buffers to mpd */
2344 lblk
= ((ext4_lblk_t
)page
->index
) <<
2345 (PAGE_CACHE_SHIFT
- blkbits
);
2346 head
= page_buffers(page
);
2347 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2353 pagevec_release(&pvec
);
2358 pagevec_release(&pvec
);
2362 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2365 struct address_space
*mapping
= data
;
2366 int ret
= ext4_writepage(page
, wbc
);
2367 mapping_set_error(mapping
, ret
);
2371 static int ext4_writepages(struct address_space
*mapping
,
2372 struct writeback_control
*wbc
)
2374 pgoff_t writeback_index
= 0;
2375 long nr_to_write
= wbc
->nr_to_write
;
2376 int range_whole
= 0;
2378 handle_t
*handle
= NULL
;
2379 struct mpage_da_data mpd
;
2380 struct inode
*inode
= mapping
->host
;
2381 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2382 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2384 struct blk_plug plug
;
2385 bool give_up_on_write
= false;
2387 trace_ext4_writepages(inode
, wbc
);
2390 * No pages to write? This is mainly a kludge to avoid starting
2391 * a transaction for special inodes like journal inode on last iput()
2392 * because that could violate lock ordering on umount
2394 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2395 goto out_writepages
;
2397 if (ext4_should_journal_data(inode
)) {
2398 struct blk_plug plug
;
2400 blk_start_plug(&plug
);
2401 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2402 blk_finish_plug(&plug
);
2403 goto out_writepages
;
2407 * If the filesystem has aborted, it is read-only, so return
2408 * right away instead of dumping stack traces later on that
2409 * will obscure the real source of the problem. We test
2410 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2411 * the latter could be true if the filesystem is mounted
2412 * read-only, and in that case, ext4_writepages should
2413 * *never* be called, so if that ever happens, we would want
2416 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2418 goto out_writepages
;
2421 if (ext4_should_dioread_nolock(inode
)) {
2423 * We may need to convert up to one extent per block in
2424 * the page and we may dirty the inode.
2426 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2430 * If we have inline data and arrive here, it means that
2431 * we will soon create the block for the 1st page, so
2432 * we'd better clear the inline data here.
2434 if (ext4_has_inline_data(inode
)) {
2435 /* Just inode will be modified... */
2436 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2437 if (IS_ERR(handle
)) {
2438 ret
= PTR_ERR(handle
);
2439 goto out_writepages
;
2441 BUG_ON(ext4_test_inode_state(inode
,
2442 EXT4_STATE_MAY_INLINE_DATA
));
2443 ext4_destroy_inline_data(handle
, inode
);
2444 ext4_journal_stop(handle
);
2447 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2450 if (wbc
->range_cyclic
) {
2451 writeback_index
= mapping
->writeback_index
;
2452 if (writeback_index
)
2454 mpd
.first_page
= writeback_index
;
2457 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2458 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2463 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2465 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2466 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2468 blk_start_plug(&plug
);
2469 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2470 /* For each extent of pages we use new io_end */
2471 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2472 if (!mpd
.io_submit
.io_end
) {
2478 * We have two constraints: We find one extent to map and we
2479 * must always write out whole page (makes a difference when
2480 * blocksize < pagesize) so that we don't block on IO when we
2481 * try to write out the rest of the page. Journalled mode is
2482 * not supported by delalloc.
2484 BUG_ON(ext4_should_journal_data(inode
));
2485 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2487 /* start a new transaction */
2488 handle
= ext4_journal_start_with_reserve(inode
,
2489 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2490 if (IS_ERR(handle
)) {
2491 ret
= PTR_ERR(handle
);
2492 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2493 "%ld pages, ino %lu; err %d", __func__
,
2494 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2495 /* Release allocated io_end */
2496 ext4_put_io_end(mpd
.io_submit
.io_end
);
2500 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2501 ret
= mpage_prepare_extent_to_map(&mpd
);
2504 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2508 * We scanned the whole range (or exhausted
2509 * nr_to_write), submitted what was mapped and
2510 * didn't find anything needing mapping. We are
2516 ext4_journal_stop(handle
);
2517 /* Submit prepared bio */
2518 ext4_io_submit(&mpd
.io_submit
);
2519 /* Unlock pages we didn't use */
2520 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2521 /* Drop our io_end reference we got from init */
2522 ext4_put_io_end(mpd
.io_submit
.io_end
);
2524 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2526 * Commit the transaction which would
2527 * free blocks released in the transaction
2530 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2534 /* Fatal error - ENOMEM, EIO... */
2538 blk_finish_plug(&plug
);
2539 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2541 mpd
.last_page
= writeback_index
- 1;
2547 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2549 * Set the writeback_index so that range_cyclic
2550 * mode will write it back later
2552 mapping
->writeback_index
= mpd
.first_page
;
2555 trace_ext4_writepages_result(inode
, wbc
, ret
,
2556 nr_to_write
- wbc
->nr_to_write
);
2560 static int ext4_nonda_switch(struct super_block
*sb
)
2562 s64 free_clusters
, dirty_clusters
;
2563 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2566 * switch to non delalloc mode if we are running low
2567 * on free block. The free block accounting via percpu
2568 * counters can get slightly wrong with percpu_counter_batch getting
2569 * accumulated on each CPU without updating global counters
2570 * Delalloc need an accurate free block accounting. So switch
2571 * to non delalloc when we are near to error range.
2574 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2576 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2578 * Start pushing delalloc when 1/2 of free blocks are dirty.
2580 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2581 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2583 if (2 * free_clusters
< 3 * dirty_clusters
||
2584 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2586 * free block count is less than 150% of dirty blocks
2587 * or free blocks is less than watermark
2594 /* We always reserve for an inode update; the superblock could be there too */
2595 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2597 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
2598 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
)))
2601 if (pos
+ len
<= 0x7fffffffULL
)
2604 /* We might need to update the superblock to set LARGE_FILE */
2608 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2609 loff_t pos
, unsigned len
, unsigned flags
,
2610 struct page
**pagep
, void **fsdata
)
2612 int ret
, retries
= 0;
2615 struct inode
*inode
= mapping
->host
;
2618 index
= pos
>> PAGE_CACHE_SHIFT
;
2620 if (ext4_nonda_switch(inode
->i_sb
)) {
2621 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2622 return ext4_write_begin(file
, mapping
, pos
,
2623 len
, flags
, pagep
, fsdata
);
2625 *fsdata
= (void *)0;
2626 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2628 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2629 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2639 * grab_cache_page_write_begin() can take a long time if the
2640 * system is thrashing due to memory pressure, or if the page
2641 * is being written back. So grab it first before we start
2642 * the transaction handle. This also allows us to allocate
2643 * the page (if needed) without using GFP_NOFS.
2646 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2652 * With delayed allocation, we don't log the i_disksize update
2653 * if there is delayed block allocation. But we still need
2654 * to journalling the i_disksize update if writes to the end
2655 * of file which has an already mapped buffer.
2658 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2659 ext4_da_write_credits(inode
, pos
, len
));
2660 if (IS_ERR(handle
)) {
2661 page_cache_release(page
);
2662 return PTR_ERR(handle
);
2666 if (page
->mapping
!= mapping
) {
2667 /* The page got truncated from under us */
2669 page_cache_release(page
);
2670 ext4_journal_stop(handle
);
2673 /* In case writeback began while the page was unlocked */
2674 wait_for_stable_page(page
);
2676 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2677 ret
= ext4_block_write_begin(page
, pos
, len
,
2678 ext4_da_get_block_prep
);
2680 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2684 ext4_journal_stop(handle
);
2686 * block_write_begin may have instantiated a few blocks
2687 * outside i_size. Trim these off again. Don't need
2688 * i_size_read because we hold i_mutex.
2690 if (pos
+ len
> inode
->i_size
)
2691 ext4_truncate_failed_write(inode
);
2693 if (ret
== -ENOSPC
&&
2694 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2697 page_cache_release(page
);
2706 * Check if we should update i_disksize
2707 * when write to the end of file but not require block allocation
2709 static int ext4_da_should_update_i_disksize(struct page
*page
,
2710 unsigned long offset
)
2712 struct buffer_head
*bh
;
2713 struct inode
*inode
= page
->mapping
->host
;
2717 bh
= page_buffers(page
);
2718 idx
= offset
>> inode
->i_blkbits
;
2720 for (i
= 0; i
< idx
; i
++)
2721 bh
= bh
->b_this_page
;
2723 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2728 static int ext4_da_write_end(struct file
*file
,
2729 struct address_space
*mapping
,
2730 loff_t pos
, unsigned len
, unsigned copied
,
2731 struct page
*page
, void *fsdata
)
2733 struct inode
*inode
= mapping
->host
;
2735 handle_t
*handle
= ext4_journal_current_handle();
2737 unsigned long start
, end
;
2738 int write_mode
= (int)(unsigned long)fsdata
;
2740 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2741 return ext4_write_end(file
, mapping
, pos
,
2742 len
, copied
, page
, fsdata
);
2744 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2745 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2746 end
= start
+ copied
- 1;
2749 * generic_write_end() will run mark_inode_dirty() if i_size
2750 * changes. So let's piggyback the i_disksize mark_inode_dirty
2753 new_i_size
= pos
+ copied
;
2754 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2755 if (ext4_has_inline_data(inode
) ||
2756 ext4_da_should_update_i_disksize(page
, end
)) {
2757 ext4_update_i_disksize(inode
, new_i_size
);
2758 /* We need to mark inode dirty even if
2759 * new_i_size is less that inode->i_size
2760 * bu greater than i_disksize.(hint delalloc)
2762 ext4_mark_inode_dirty(handle
, inode
);
2766 if (write_mode
!= CONVERT_INLINE_DATA
&&
2767 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2768 ext4_has_inline_data(inode
))
2769 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2772 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2778 ret2
= ext4_journal_stop(handle
);
2782 return ret
? ret
: copied
;
2785 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2786 unsigned int length
)
2789 * Drop reserved blocks
2791 BUG_ON(!PageLocked(page
));
2792 if (!page_has_buffers(page
))
2795 ext4_da_page_release_reservation(page
, offset
, length
);
2798 ext4_invalidatepage(page
, offset
, length
);
2804 * Force all delayed allocation blocks to be allocated for a given inode.
2806 int ext4_alloc_da_blocks(struct inode
*inode
)
2808 trace_ext4_alloc_da_blocks(inode
);
2810 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
2814 * We do something simple for now. The filemap_flush() will
2815 * also start triggering a write of the data blocks, which is
2816 * not strictly speaking necessary (and for users of
2817 * laptop_mode, not even desirable). However, to do otherwise
2818 * would require replicating code paths in:
2820 * ext4_writepages() ->
2821 * write_cache_pages() ---> (via passed in callback function)
2822 * __mpage_da_writepage() -->
2823 * mpage_add_bh_to_extent()
2824 * mpage_da_map_blocks()
2826 * The problem is that write_cache_pages(), located in
2827 * mm/page-writeback.c, marks pages clean in preparation for
2828 * doing I/O, which is not desirable if we're not planning on
2831 * We could call write_cache_pages(), and then redirty all of
2832 * the pages by calling redirty_page_for_writepage() but that
2833 * would be ugly in the extreme. So instead we would need to
2834 * replicate parts of the code in the above functions,
2835 * simplifying them because we wouldn't actually intend to
2836 * write out the pages, but rather only collect contiguous
2837 * logical block extents, call the multi-block allocator, and
2838 * then update the buffer heads with the block allocations.
2840 * For now, though, we'll cheat by calling filemap_flush(),
2841 * which will map the blocks, and start the I/O, but not
2842 * actually wait for the I/O to complete.
2844 return filemap_flush(inode
->i_mapping
);
2848 * bmap() is special. It gets used by applications such as lilo and by
2849 * the swapper to find the on-disk block of a specific piece of data.
2851 * Naturally, this is dangerous if the block concerned is still in the
2852 * journal. If somebody makes a swapfile on an ext4 data-journaling
2853 * filesystem and enables swap, then they may get a nasty shock when the
2854 * data getting swapped to that swapfile suddenly gets overwritten by
2855 * the original zero's written out previously to the journal and
2856 * awaiting writeback in the kernel's buffer cache.
2858 * So, if we see any bmap calls here on a modified, data-journaled file,
2859 * take extra steps to flush any blocks which might be in the cache.
2861 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2863 struct inode
*inode
= mapping
->host
;
2868 * We can get here for an inline file via the FIBMAP ioctl
2870 if (ext4_has_inline_data(inode
))
2873 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2874 test_opt(inode
->i_sb
, DELALLOC
)) {
2876 * With delalloc we want to sync the file
2877 * so that we can make sure we allocate
2880 filemap_write_and_wait(mapping
);
2883 if (EXT4_JOURNAL(inode
) &&
2884 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2886 * This is a REALLY heavyweight approach, but the use of
2887 * bmap on dirty files is expected to be extremely rare:
2888 * only if we run lilo or swapon on a freshly made file
2889 * do we expect this to happen.
2891 * (bmap requires CAP_SYS_RAWIO so this does not
2892 * represent an unprivileged user DOS attack --- we'd be
2893 * in trouble if mortal users could trigger this path at
2896 * NB. EXT4_STATE_JDATA is not set on files other than
2897 * regular files. If somebody wants to bmap a directory
2898 * or symlink and gets confused because the buffer
2899 * hasn't yet been flushed to disk, they deserve
2900 * everything they get.
2903 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2904 journal
= EXT4_JOURNAL(inode
);
2905 jbd2_journal_lock_updates(journal
);
2906 err
= jbd2_journal_flush(journal
);
2907 jbd2_journal_unlock_updates(journal
);
2913 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2916 static int ext4_readpage(struct file
*file
, struct page
*page
)
2919 struct inode
*inode
= page
->mapping
->host
;
2921 trace_ext4_readpage(page
);
2923 if (ext4_has_inline_data(inode
))
2924 ret
= ext4_readpage_inline(inode
, page
);
2927 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
2933 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2934 struct list_head
*pages
, unsigned nr_pages
)
2936 struct inode
*inode
= mapping
->host
;
2938 /* If the file has inline data, no need to do readpages. */
2939 if (ext4_has_inline_data(inode
))
2942 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
2945 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
2946 unsigned int length
)
2948 trace_ext4_invalidatepage(page
, offset
, length
);
2950 /* No journalling happens on data buffers when this function is used */
2951 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2953 block_invalidatepage(page
, offset
, length
);
2956 static int __ext4_journalled_invalidatepage(struct page
*page
,
2957 unsigned int offset
,
2958 unsigned int length
)
2960 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2962 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
2965 * If it's a full truncate we just forget about the pending dirtying
2967 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
2968 ClearPageChecked(page
);
2970 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
2973 /* Wrapper for aops... */
2974 static void ext4_journalled_invalidatepage(struct page
*page
,
2975 unsigned int offset
,
2976 unsigned int length
)
2978 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
2981 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2983 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2985 trace_ext4_releasepage(page
);
2987 /* Page has dirty journalled data -> cannot release */
2988 if (PageChecked(page
))
2991 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2993 return try_to_free_buffers(page
);
2997 * ext4_get_block used when preparing for a DIO write or buffer write.
2998 * We allocate an uinitialized extent if blocks haven't been allocated.
2999 * The extent will be converted to initialized after the IO is complete.
3001 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3002 struct buffer_head
*bh_result
, int create
)
3004 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3005 inode
->i_ino
, create
);
3006 return _ext4_get_block(inode
, iblock
, bh_result
,
3007 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3010 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3011 struct buffer_head
*bh_result
, int create
)
3013 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3014 inode
->i_ino
, create
);
3015 return _ext4_get_block(inode
, iblock
, bh_result
,
3016 EXT4_GET_BLOCKS_NO_LOCK
);
3019 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3020 ssize_t size
, void *private)
3022 ext4_io_end_t
*io_end
= iocb
->private;
3024 /* if not async direct IO just return */
3028 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3029 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3030 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3033 iocb
->private = NULL
;
3034 io_end
->offset
= offset
;
3035 io_end
->size
= size
;
3036 ext4_put_io_end(io_end
);
3040 * For ext4 extent files, ext4 will do direct-io write to holes,
3041 * preallocated extents, and those write extend the file, no need to
3042 * fall back to buffered IO.
3044 * For holes, we fallocate those blocks, mark them as unwritten
3045 * If those blocks were preallocated, we mark sure they are split, but
3046 * still keep the range to write as unwritten.
3048 * The unwritten extents will be converted to written when DIO is completed.
3049 * For async direct IO, since the IO may still pending when return, we
3050 * set up an end_io call back function, which will do the conversion
3051 * when async direct IO completed.
3053 * If the O_DIRECT write will extend the file then add this inode to the
3054 * orphan list. So recovery will truncate it back to the original size
3055 * if the machine crashes during the write.
3058 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3059 struct iov_iter
*iter
, loff_t offset
)
3061 struct file
*file
= iocb
->ki_filp
;
3062 struct inode
*inode
= file
->f_mapping
->host
;
3064 size_t count
= iov_iter_count(iter
);
3066 get_block_t
*get_block_func
= NULL
;
3068 loff_t final_size
= offset
+ count
;
3069 ext4_io_end_t
*io_end
= NULL
;
3071 /* Use the old path for reads and writes beyond i_size. */
3072 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3073 return ext4_ind_direct_IO(rw
, iocb
, iter
, offset
);
3075 BUG_ON(iocb
->private == NULL
);
3078 * Make all waiters for direct IO properly wait also for extent
3079 * conversion. This also disallows race between truncate() and
3080 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3083 atomic_inc(&inode
->i_dio_count
);
3085 /* If we do a overwrite dio, i_mutex locking can be released */
3086 overwrite
= *((int *)iocb
->private);
3089 down_read(&EXT4_I(inode
)->i_data_sem
);
3090 mutex_unlock(&inode
->i_mutex
);
3094 * We could direct write to holes and fallocate.
3096 * Allocated blocks to fill the hole are marked as
3097 * unwritten to prevent parallel buffered read to expose
3098 * the stale data before DIO complete the data IO.
3100 * As to previously fallocated extents, ext4 get_block will
3101 * just simply mark the buffer mapped but still keep the
3102 * extents unwritten.
3104 * For non AIO case, we will convert those unwritten extents
3105 * to written after return back from blockdev_direct_IO.
3107 * For async DIO, the conversion needs to be deferred when the
3108 * IO is completed. The ext4 end_io callback function will be
3109 * called to take care of the conversion work. Here for async
3110 * case, we allocate an io_end structure to hook to the iocb.
3112 iocb
->private = NULL
;
3113 ext4_inode_aio_set(inode
, NULL
);
3114 if (!is_sync_kiocb(iocb
)) {
3115 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3121 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3123 iocb
->private = ext4_get_io_end(io_end
);
3125 * we save the io structure for current async direct
3126 * IO, so that later ext4_map_blocks() could flag the
3127 * io structure whether there is a unwritten extents
3128 * needs to be converted when IO is completed.
3130 ext4_inode_aio_set(inode
, io_end
);
3134 get_block_func
= ext4_get_block_write_nolock
;
3136 get_block_func
= ext4_get_block_write
;
3137 dio_flags
= DIO_LOCKING
;
3139 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3140 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3143 ret
= dax_do_io(rw
, iocb
, inode
, iter
, offset
, get_block_func
,
3144 ext4_end_io_dio
, dio_flags
);
3146 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3147 inode
->i_sb
->s_bdev
, iter
, offset
,
3149 ext4_end_io_dio
, NULL
, dio_flags
);
3152 * Put our reference to io_end. This can free the io_end structure e.g.
3153 * in sync IO case or in case of error. It can even perform extent
3154 * conversion if all bios we submitted finished before we got here.
3155 * Note that in that case iocb->private can be already set to NULL
3159 ext4_inode_aio_set(inode
, NULL
);
3160 ext4_put_io_end(io_end
);
3162 * When no IO was submitted ext4_end_io_dio() was not
3163 * called so we have to put iocb's reference.
3165 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3166 WARN_ON(iocb
->private != io_end
);
3167 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3168 ext4_put_io_end(io_end
);
3169 iocb
->private = NULL
;
3172 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3173 EXT4_STATE_DIO_UNWRITTEN
)) {
3176 * for non AIO case, since the IO is already
3177 * completed, we could do the conversion right here
3179 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3183 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3188 inode_dio_done(inode
);
3189 /* take i_mutex locking again if we do a ovewrite dio */
3191 up_read(&EXT4_I(inode
)->i_data_sem
);
3192 mutex_lock(&inode
->i_mutex
);
3198 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3199 struct iov_iter
*iter
, loff_t offset
)
3201 struct file
*file
= iocb
->ki_filp
;
3202 struct inode
*inode
= file
->f_mapping
->host
;
3203 size_t count
= iov_iter_count(iter
);
3206 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3207 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3212 * If we are doing data journalling we don't support O_DIRECT
3214 if (ext4_should_journal_data(inode
))
3217 /* Let buffer I/O handle the inline data case. */
3218 if (ext4_has_inline_data(inode
))
3221 trace_ext4_direct_IO_enter(inode
, offset
, count
, rw
);
3222 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3223 ret
= ext4_ext_direct_IO(rw
, iocb
, iter
, offset
);
3225 ret
= ext4_ind_direct_IO(rw
, iocb
, iter
, offset
);
3226 trace_ext4_direct_IO_exit(inode
, offset
, count
, rw
, ret
);
3231 * Pages can be marked dirty completely asynchronously from ext4's journalling
3232 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3233 * much here because ->set_page_dirty is called under VFS locks. The page is
3234 * not necessarily locked.
3236 * We cannot just dirty the page and leave attached buffers clean, because the
3237 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3238 * or jbddirty because all the journalling code will explode.
3240 * So what we do is to mark the page "pending dirty" and next time writepage
3241 * is called, propagate that into the buffers appropriately.
3243 static int ext4_journalled_set_page_dirty(struct page
*page
)
3245 SetPageChecked(page
);
3246 return __set_page_dirty_nobuffers(page
);
3249 static const struct address_space_operations ext4_aops
= {
3250 .readpage
= ext4_readpage
,
3251 .readpages
= ext4_readpages
,
3252 .writepage
= ext4_writepage
,
3253 .writepages
= ext4_writepages
,
3254 .write_begin
= ext4_write_begin
,
3255 .write_end
= ext4_write_end
,
3257 .invalidatepage
= ext4_invalidatepage
,
3258 .releasepage
= ext4_releasepage
,
3259 .direct_IO
= ext4_direct_IO
,
3260 .migratepage
= buffer_migrate_page
,
3261 .is_partially_uptodate
= block_is_partially_uptodate
,
3262 .error_remove_page
= generic_error_remove_page
,
3265 static const struct address_space_operations ext4_journalled_aops
= {
3266 .readpage
= ext4_readpage
,
3267 .readpages
= ext4_readpages
,
3268 .writepage
= ext4_writepage
,
3269 .writepages
= ext4_writepages
,
3270 .write_begin
= ext4_write_begin
,
3271 .write_end
= ext4_journalled_write_end
,
3272 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3274 .invalidatepage
= ext4_journalled_invalidatepage
,
3275 .releasepage
= ext4_releasepage
,
3276 .direct_IO
= ext4_direct_IO
,
3277 .is_partially_uptodate
= block_is_partially_uptodate
,
3278 .error_remove_page
= generic_error_remove_page
,
3281 static const struct address_space_operations ext4_da_aops
= {
3282 .readpage
= ext4_readpage
,
3283 .readpages
= ext4_readpages
,
3284 .writepage
= ext4_writepage
,
3285 .writepages
= ext4_writepages
,
3286 .write_begin
= ext4_da_write_begin
,
3287 .write_end
= ext4_da_write_end
,
3289 .invalidatepage
= ext4_da_invalidatepage
,
3290 .releasepage
= ext4_releasepage
,
3291 .direct_IO
= ext4_direct_IO
,
3292 .migratepage
= buffer_migrate_page
,
3293 .is_partially_uptodate
= block_is_partially_uptodate
,
3294 .error_remove_page
= generic_error_remove_page
,
3297 void ext4_set_aops(struct inode
*inode
)
3299 switch (ext4_inode_journal_mode(inode
)) {
3300 case EXT4_INODE_ORDERED_DATA_MODE
:
3301 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3303 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3304 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3306 case EXT4_INODE_JOURNAL_DATA_MODE
:
3307 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3312 if (test_opt(inode
->i_sb
, DELALLOC
))
3313 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3315 inode
->i_mapping
->a_ops
= &ext4_aops
;
3318 static int __ext4_block_zero_page_range(handle_t
*handle
,
3319 struct address_space
*mapping
, loff_t from
, loff_t length
)
3321 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3322 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3323 unsigned blocksize
, pos
;
3325 struct inode
*inode
= mapping
->host
;
3326 struct buffer_head
*bh
;
3330 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3331 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3335 blocksize
= inode
->i_sb
->s_blocksize
;
3337 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3339 if (!page_has_buffers(page
))
3340 create_empty_buffers(page
, blocksize
, 0);
3342 /* Find the buffer that contains "offset" */
3343 bh
= page_buffers(page
);
3345 while (offset
>= pos
) {
3346 bh
= bh
->b_this_page
;
3350 if (buffer_freed(bh
)) {
3351 BUFFER_TRACE(bh
, "freed: skip");
3354 if (!buffer_mapped(bh
)) {
3355 BUFFER_TRACE(bh
, "unmapped");
3356 ext4_get_block(inode
, iblock
, bh
, 0);
3357 /* unmapped? It's a hole - nothing to do */
3358 if (!buffer_mapped(bh
)) {
3359 BUFFER_TRACE(bh
, "still unmapped");
3364 /* Ok, it's mapped. Make sure it's up-to-date */
3365 if (PageUptodate(page
))
3366 set_buffer_uptodate(bh
);
3368 if (!buffer_uptodate(bh
)) {
3370 ll_rw_block(READ
, 1, &bh
);
3372 /* Uhhuh. Read error. Complain and punt. */
3373 if (!buffer_uptodate(bh
))
3375 if (S_ISREG(inode
->i_mode
) &&
3376 ext4_encrypted_inode(inode
)) {
3377 /* We expect the key to be set. */
3378 BUG_ON(!ext4_has_encryption_key(inode
));
3379 BUG_ON(blocksize
!= PAGE_CACHE_SIZE
);
3380 WARN_ON_ONCE(ext4_decrypt_one(inode
, page
));
3383 if (ext4_should_journal_data(inode
)) {
3384 BUFFER_TRACE(bh
, "get write access");
3385 err
= ext4_journal_get_write_access(handle
, bh
);
3389 zero_user(page
, offset
, length
);
3390 BUFFER_TRACE(bh
, "zeroed end of block");
3392 if (ext4_should_journal_data(inode
)) {
3393 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3396 mark_buffer_dirty(bh
);
3397 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3398 err
= ext4_jbd2_file_inode(handle
, inode
);
3403 page_cache_release(page
);
3408 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3409 * starting from file offset 'from'. The range to be zero'd must
3410 * be contained with in one block. If the specified range exceeds
3411 * the end of the block it will be shortened to end of the block
3412 * that cooresponds to 'from'
3414 static int ext4_block_zero_page_range(handle_t
*handle
,
3415 struct address_space
*mapping
, loff_t from
, loff_t length
)
3417 struct inode
*inode
= mapping
->host
;
3418 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3419 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3420 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3423 * correct length if it does not fall between
3424 * 'from' and the end of the block
3426 if (length
> max
|| length
< 0)
3430 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3431 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3435 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3436 * up to the end of the block which corresponds to `from'.
3437 * This required during truncate. We need to physically zero the tail end
3438 * of that block so it doesn't yield old data if the file is later grown.
3440 static int ext4_block_truncate_page(handle_t
*handle
,
3441 struct address_space
*mapping
, loff_t from
)
3443 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3446 struct inode
*inode
= mapping
->host
;
3448 blocksize
= inode
->i_sb
->s_blocksize
;
3449 length
= blocksize
- (offset
& (blocksize
- 1));
3451 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3454 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3455 loff_t lstart
, loff_t length
)
3457 struct super_block
*sb
= inode
->i_sb
;
3458 struct address_space
*mapping
= inode
->i_mapping
;
3459 unsigned partial_start
, partial_end
;
3460 ext4_fsblk_t start
, end
;
3461 loff_t byte_end
= (lstart
+ length
- 1);
3464 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3465 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3467 start
= lstart
>> sb
->s_blocksize_bits
;
3468 end
= byte_end
>> sb
->s_blocksize_bits
;
3470 /* Handle partial zero within the single block */
3472 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3473 err
= ext4_block_zero_page_range(handle
, mapping
,
3477 /* Handle partial zero out on the start of the range */
3478 if (partial_start
) {
3479 err
= ext4_block_zero_page_range(handle
, mapping
,
3480 lstart
, sb
->s_blocksize
);
3484 /* Handle partial zero out on the end of the range */
3485 if (partial_end
!= sb
->s_blocksize
- 1)
3486 err
= ext4_block_zero_page_range(handle
, mapping
,
3487 byte_end
- partial_end
,
3492 int ext4_can_truncate(struct inode
*inode
)
3494 if (S_ISREG(inode
->i_mode
))
3496 if (S_ISDIR(inode
->i_mode
))
3498 if (S_ISLNK(inode
->i_mode
))
3499 return !ext4_inode_is_fast_symlink(inode
);
3504 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3505 * associated with the given offset and length
3507 * @inode: File inode
3508 * @offset: The offset where the hole will begin
3509 * @len: The length of the hole
3511 * Returns: 0 on success or negative on failure
3514 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3516 struct super_block
*sb
= inode
->i_sb
;
3517 ext4_lblk_t first_block
, stop_block
;
3518 struct address_space
*mapping
= inode
->i_mapping
;
3519 loff_t first_block_offset
, last_block_offset
;
3521 unsigned int credits
;
3524 if (!S_ISREG(inode
->i_mode
))
3527 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3530 * Write out all dirty pages to avoid race conditions
3531 * Then release them.
3533 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3534 ret
= filemap_write_and_wait_range(mapping
, offset
,
3535 offset
+ length
- 1);
3540 mutex_lock(&inode
->i_mutex
);
3542 /* No need to punch hole beyond i_size */
3543 if (offset
>= inode
->i_size
)
3547 * If the hole extends beyond i_size, set the hole
3548 * to end after the page that contains i_size
3550 if (offset
+ length
> inode
->i_size
) {
3551 length
= inode
->i_size
+
3552 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3556 if (offset
& (sb
->s_blocksize
- 1) ||
3557 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3559 * Attach jinode to inode for jbd2 if we do any zeroing of
3562 ret
= ext4_inode_attach_jinode(inode
);
3568 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3569 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3571 /* Now release the pages and zero block aligned part of pages*/
3572 if (last_block_offset
> first_block_offset
)
3573 truncate_pagecache_range(inode
, first_block_offset
,
3576 /* Wait all existing dio workers, newcomers will block on i_mutex */
3577 ext4_inode_block_unlocked_dio(inode
);
3578 inode_dio_wait(inode
);
3580 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3581 credits
= ext4_writepage_trans_blocks(inode
);
3583 credits
= ext4_blocks_for_truncate(inode
);
3584 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3585 if (IS_ERR(handle
)) {
3586 ret
= PTR_ERR(handle
);
3587 ext4_std_error(sb
, ret
);
3591 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3596 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3597 EXT4_BLOCK_SIZE_BITS(sb
);
3598 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3600 /* If there are no blocks to remove, return now */
3601 if (first_block
>= stop_block
)
3604 down_write(&EXT4_I(inode
)->i_data_sem
);
3605 ext4_discard_preallocations(inode
);
3607 ret
= ext4_es_remove_extent(inode
, first_block
,
3608 stop_block
- first_block
);
3610 up_write(&EXT4_I(inode
)->i_data_sem
);
3614 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3615 ret
= ext4_ext_remove_space(inode
, first_block
,
3618 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3621 up_write(&EXT4_I(inode
)->i_data_sem
);
3623 ext4_handle_sync(handle
);
3625 /* Now release the pages again to reduce race window */
3626 if (last_block_offset
> first_block_offset
)
3627 truncate_pagecache_range(inode
, first_block_offset
,
3630 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3631 ext4_mark_inode_dirty(handle
, inode
);
3633 ext4_journal_stop(handle
);
3635 ext4_inode_resume_unlocked_dio(inode
);
3637 mutex_unlock(&inode
->i_mutex
);
3641 int ext4_inode_attach_jinode(struct inode
*inode
)
3643 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3644 struct jbd2_inode
*jinode
;
3646 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3649 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3650 spin_lock(&inode
->i_lock
);
3653 spin_unlock(&inode
->i_lock
);
3656 ei
->jinode
= jinode
;
3657 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3660 spin_unlock(&inode
->i_lock
);
3661 if (unlikely(jinode
!= NULL
))
3662 jbd2_free_inode(jinode
);
3669 * We block out ext4_get_block() block instantiations across the entire
3670 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3671 * simultaneously on behalf of the same inode.
3673 * As we work through the truncate and commit bits of it to the journal there
3674 * is one core, guiding principle: the file's tree must always be consistent on
3675 * disk. We must be able to restart the truncate after a crash.
3677 * The file's tree may be transiently inconsistent in memory (although it
3678 * probably isn't), but whenever we close off and commit a journal transaction,
3679 * the contents of (the filesystem + the journal) must be consistent and
3680 * restartable. It's pretty simple, really: bottom up, right to left (although
3681 * left-to-right works OK too).
3683 * Note that at recovery time, journal replay occurs *before* the restart of
3684 * truncate against the orphan inode list.
3686 * The committed inode has the new, desired i_size (which is the same as
3687 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3688 * that this inode's truncate did not complete and it will again call
3689 * ext4_truncate() to have another go. So there will be instantiated blocks
3690 * to the right of the truncation point in a crashed ext4 filesystem. But
3691 * that's fine - as long as they are linked from the inode, the post-crash
3692 * ext4_truncate() run will find them and release them.
3694 void ext4_truncate(struct inode
*inode
)
3696 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3697 unsigned int credits
;
3699 struct address_space
*mapping
= inode
->i_mapping
;
3702 * There is a possibility that we're either freeing the inode
3703 * or it's a completely new inode. In those cases we might not
3704 * have i_mutex locked because it's not necessary.
3706 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3707 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3708 trace_ext4_truncate_enter(inode
);
3710 if (!ext4_can_truncate(inode
))
3713 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3715 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3716 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3718 if (ext4_has_inline_data(inode
)) {
3721 ext4_inline_data_truncate(inode
, &has_inline
);
3726 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3727 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3728 if (ext4_inode_attach_jinode(inode
) < 0)
3732 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3733 credits
= ext4_writepage_trans_blocks(inode
);
3735 credits
= ext4_blocks_for_truncate(inode
);
3737 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3738 if (IS_ERR(handle
)) {
3739 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3743 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3744 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3747 * We add the inode to the orphan list, so that if this
3748 * truncate spans multiple transactions, and we crash, we will
3749 * resume the truncate when the filesystem recovers. It also
3750 * marks the inode dirty, to catch the new size.
3752 * Implication: the file must always be in a sane, consistent
3753 * truncatable state while each transaction commits.
3755 if (ext4_orphan_add(handle
, inode
))
3758 down_write(&EXT4_I(inode
)->i_data_sem
);
3760 ext4_discard_preallocations(inode
);
3762 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3763 ext4_ext_truncate(handle
, inode
);
3765 ext4_ind_truncate(handle
, inode
);
3767 up_write(&ei
->i_data_sem
);
3770 ext4_handle_sync(handle
);
3774 * If this was a simple ftruncate() and the file will remain alive,
3775 * then we need to clear up the orphan record which we created above.
3776 * However, if this was a real unlink then we were called by
3777 * ext4_evict_inode(), and we allow that function to clean up the
3778 * orphan info for us.
3781 ext4_orphan_del(handle
, inode
);
3783 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3784 ext4_mark_inode_dirty(handle
, inode
);
3785 ext4_journal_stop(handle
);
3787 trace_ext4_truncate_exit(inode
);
3791 * ext4_get_inode_loc returns with an extra refcount against the inode's
3792 * underlying buffer_head on success. If 'in_mem' is true, we have all
3793 * data in memory that is needed to recreate the on-disk version of this
3796 static int __ext4_get_inode_loc(struct inode
*inode
,
3797 struct ext4_iloc
*iloc
, int in_mem
)
3799 struct ext4_group_desc
*gdp
;
3800 struct buffer_head
*bh
;
3801 struct super_block
*sb
= inode
->i_sb
;
3803 int inodes_per_block
, inode_offset
;
3806 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3809 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3810 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3815 * Figure out the offset within the block group inode table
3817 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3818 inode_offset
= ((inode
->i_ino
- 1) %
3819 EXT4_INODES_PER_GROUP(sb
));
3820 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3821 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3823 bh
= sb_getblk(sb
, block
);
3826 if (!buffer_uptodate(bh
)) {
3830 * If the buffer has the write error flag, we have failed
3831 * to write out another inode in the same block. In this
3832 * case, we don't have to read the block because we may
3833 * read the old inode data successfully.
3835 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3836 set_buffer_uptodate(bh
);
3838 if (buffer_uptodate(bh
)) {
3839 /* someone brought it uptodate while we waited */
3845 * If we have all information of the inode in memory and this
3846 * is the only valid inode in the block, we need not read the
3850 struct buffer_head
*bitmap_bh
;
3853 start
= inode_offset
& ~(inodes_per_block
- 1);
3855 /* Is the inode bitmap in cache? */
3856 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3857 if (unlikely(!bitmap_bh
))
3861 * If the inode bitmap isn't in cache then the
3862 * optimisation may end up performing two reads instead
3863 * of one, so skip it.
3865 if (!buffer_uptodate(bitmap_bh
)) {
3869 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3870 if (i
== inode_offset
)
3872 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3876 if (i
== start
+ inodes_per_block
) {
3877 /* all other inodes are free, so skip I/O */
3878 memset(bh
->b_data
, 0, bh
->b_size
);
3879 set_buffer_uptodate(bh
);
3887 * If we need to do any I/O, try to pre-readahead extra
3888 * blocks from the inode table.
3890 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3891 ext4_fsblk_t b
, end
, table
;
3893 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
3895 table
= ext4_inode_table(sb
, gdp
);
3896 /* s_inode_readahead_blks is always a power of 2 */
3897 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
3901 num
= EXT4_INODES_PER_GROUP(sb
);
3902 if (ext4_has_group_desc_csum(sb
))
3903 num
-= ext4_itable_unused_count(sb
, gdp
);
3904 table
+= num
/ inodes_per_block
;
3908 sb_breadahead(sb
, b
++);
3912 * There are other valid inodes in the buffer, this inode
3913 * has in-inode xattrs, or we don't have this inode in memory.
3914 * Read the block from disk.
3916 trace_ext4_load_inode(inode
);
3918 bh
->b_end_io
= end_buffer_read_sync
;
3919 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3921 if (!buffer_uptodate(bh
)) {
3922 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3923 "unable to read itable block");
3933 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3935 /* We have all inode data except xattrs in memory here. */
3936 return __ext4_get_inode_loc(inode
, iloc
,
3937 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3940 void ext4_set_inode_flags(struct inode
*inode
)
3942 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3943 unsigned int new_fl
= 0;
3945 if (flags
& EXT4_SYNC_FL
)
3947 if (flags
& EXT4_APPEND_FL
)
3949 if (flags
& EXT4_IMMUTABLE_FL
)
3950 new_fl
|= S_IMMUTABLE
;
3951 if (flags
& EXT4_NOATIME_FL
)
3952 new_fl
|= S_NOATIME
;
3953 if (flags
& EXT4_DIRSYNC_FL
)
3954 new_fl
|= S_DIRSYNC
;
3955 if (test_opt(inode
->i_sb
, DAX
))
3957 inode_set_flags(inode
, new_fl
,
3958 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
3961 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3962 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3964 unsigned int vfs_fl
;
3965 unsigned long old_fl
, new_fl
;
3968 vfs_fl
= ei
->vfs_inode
.i_flags
;
3969 old_fl
= ei
->i_flags
;
3970 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3971 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3973 if (vfs_fl
& S_SYNC
)
3974 new_fl
|= EXT4_SYNC_FL
;
3975 if (vfs_fl
& S_APPEND
)
3976 new_fl
|= EXT4_APPEND_FL
;
3977 if (vfs_fl
& S_IMMUTABLE
)
3978 new_fl
|= EXT4_IMMUTABLE_FL
;
3979 if (vfs_fl
& S_NOATIME
)
3980 new_fl
|= EXT4_NOATIME_FL
;
3981 if (vfs_fl
& S_DIRSYNC
)
3982 new_fl
|= EXT4_DIRSYNC_FL
;
3983 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3986 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3987 struct ext4_inode_info
*ei
)
3990 struct inode
*inode
= &(ei
->vfs_inode
);
3991 struct super_block
*sb
= inode
->i_sb
;
3993 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3994 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3995 /* we are using combined 48 bit field */
3996 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3997 le32_to_cpu(raw_inode
->i_blocks_lo
);
3998 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3999 /* i_blocks represent file system block size */
4000 return i_blocks
<< (inode
->i_blkbits
- 9);
4005 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4009 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4010 struct ext4_inode
*raw_inode
,
4011 struct ext4_inode_info
*ei
)
4013 __le32
*magic
= (void *)raw_inode
+
4014 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4015 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4016 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4017 ext4_find_inline_data_nolock(inode
);
4019 EXT4_I(inode
)->i_inline_off
= 0;
4022 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4024 struct ext4_iloc iloc
;
4025 struct ext4_inode
*raw_inode
;
4026 struct ext4_inode_info
*ei
;
4027 struct inode
*inode
;
4028 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4034 inode
= iget_locked(sb
, ino
);
4036 return ERR_PTR(-ENOMEM
);
4037 if (!(inode
->i_state
& I_NEW
))
4043 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4046 raw_inode
= ext4_raw_inode(&iloc
);
4048 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4049 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4050 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4051 EXT4_INODE_SIZE(inode
->i_sb
)) {
4052 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4053 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4054 EXT4_INODE_SIZE(inode
->i_sb
));
4059 ei
->i_extra_isize
= 0;
4061 /* Precompute checksum seed for inode metadata */
4062 if (ext4_has_metadata_csum(sb
)) {
4063 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4065 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4066 __le32 gen
= raw_inode
->i_generation
;
4067 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4069 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4073 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4074 EXT4_ERROR_INODE(inode
, "checksum invalid");
4079 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4080 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4081 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4082 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4083 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4084 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4086 i_uid_write(inode
, i_uid
);
4087 i_gid_write(inode
, i_gid
);
4088 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4090 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4091 ei
->i_inline_off
= 0;
4092 ei
->i_dir_start_lookup
= 0;
4093 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4094 /* We now have enough fields to check if the inode was active or not.
4095 * This is needed because nfsd might try to access dead inodes
4096 * the test is that same one that e2fsck uses
4097 * NeilBrown 1999oct15
4099 if (inode
->i_nlink
== 0) {
4100 if ((inode
->i_mode
== 0 ||
4101 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4102 ino
!= EXT4_BOOT_LOADER_INO
) {
4103 /* this inode is deleted */
4107 /* The only unlinked inodes we let through here have
4108 * valid i_mode and are being read by the orphan
4109 * recovery code: that's fine, we're about to complete
4110 * the process of deleting those.
4111 * OR it is the EXT4_BOOT_LOADER_INO which is
4112 * not initialized on a new filesystem. */
4114 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4115 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4116 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4117 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4119 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4120 inode
->i_size
= ext4_isize(raw_inode
);
4121 ei
->i_disksize
= inode
->i_size
;
4123 ei
->i_reserved_quota
= 0;
4125 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4126 ei
->i_block_group
= iloc
.block_group
;
4127 ei
->i_last_alloc_group
= ~0;
4129 * NOTE! The in-memory inode i_data array is in little-endian order
4130 * even on big-endian machines: we do NOT byteswap the block numbers!
4132 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4133 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4134 INIT_LIST_HEAD(&ei
->i_orphan
);
4137 * Set transaction id's of transactions that have to be committed
4138 * to finish f[data]sync. We set them to currently running transaction
4139 * as we cannot be sure that the inode or some of its metadata isn't
4140 * part of the transaction - the inode could have been reclaimed and
4141 * now it is reread from disk.
4144 transaction_t
*transaction
;
4147 read_lock(&journal
->j_state_lock
);
4148 if (journal
->j_running_transaction
)
4149 transaction
= journal
->j_running_transaction
;
4151 transaction
= journal
->j_committing_transaction
;
4153 tid
= transaction
->t_tid
;
4155 tid
= journal
->j_commit_sequence
;
4156 read_unlock(&journal
->j_state_lock
);
4157 ei
->i_sync_tid
= tid
;
4158 ei
->i_datasync_tid
= tid
;
4161 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4162 if (ei
->i_extra_isize
== 0) {
4163 /* The extra space is currently unused. Use it. */
4164 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4165 EXT4_GOOD_OLD_INODE_SIZE
;
4167 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4171 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4172 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4173 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4174 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4176 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4177 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4178 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4179 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4181 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4186 if (ei
->i_file_acl
&&
4187 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4188 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4192 } else if (!ext4_has_inline_data(inode
)) {
4193 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4194 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4195 (S_ISLNK(inode
->i_mode
) &&
4196 !ext4_inode_is_fast_symlink(inode
))))
4197 /* Validate extent which is part of inode */
4198 ret
= ext4_ext_check_inode(inode
);
4199 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4200 (S_ISLNK(inode
->i_mode
) &&
4201 !ext4_inode_is_fast_symlink(inode
))) {
4202 /* Validate block references which are part of inode */
4203 ret
= ext4_ind_check_inode(inode
);
4209 if (S_ISREG(inode
->i_mode
)) {
4210 inode
->i_op
= &ext4_file_inode_operations
;
4211 if (test_opt(inode
->i_sb
, DAX
))
4212 inode
->i_fop
= &ext4_dax_file_operations
;
4214 inode
->i_fop
= &ext4_file_operations
;
4215 ext4_set_aops(inode
);
4216 } else if (S_ISDIR(inode
->i_mode
)) {
4217 inode
->i_op
= &ext4_dir_inode_operations
;
4218 inode
->i_fop
= &ext4_dir_operations
;
4219 } else if (S_ISLNK(inode
->i_mode
)) {
4220 if (ext4_inode_is_fast_symlink(inode
) &&
4221 !ext4_encrypted_inode(inode
)) {
4222 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4223 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4224 sizeof(ei
->i_data
) - 1);
4226 inode
->i_op
= &ext4_symlink_inode_operations
;
4227 ext4_set_aops(inode
);
4229 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4230 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4231 inode
->i_op
= &ext4_special_inode_operations
;
4232 if (raw_inode
->i_block
[0])
4233 init_special_inode(inode
, inode
->i_mode
,
4234 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4236 init_special_inode(inode
, inode
->i_mode
,
4237 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4238 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4239 make_bad_inode(inode
);
4242 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4246 ext4_set_inode_flags(inode
);
4247 unlock_new_inode(inode
);
4253 return ERR_PTR(ret
);
4256 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4258 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4259 return ERR_PTR(-EIO
);
4260 return ext4_iget(sb
, ino
);
4263 static int ext4_inode_blocks_set(handle_t
*handle
,
4264 struct ext4_inode
*raw_inode
,
4265 struct ext4_inode_info
*ei
)
4267 struct inode
*inode
= &(ei
->vfs_inode
);
4268 u64 i_blocks
= inode
->i_blocks
;
4269 struct super_block
*sb
= inode
->i_sb
;
4271 if (i_blocks
<= ~0U) {
4273 * i_blocks can be represented in a 32 bit variable
4274 * as multiple of 512 bytes
4276 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4277 raw_inode
->i_blocks_high
= 0;
4278 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4281 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4284 if (i_blocks
<= 0xffffffffffffULL
) {
4286 * i_blocks can be represented in a 48 bit variable
4287 * as multiple of 512 bytes
4289 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4290 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4291 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4293 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4294 /* i_block is stored in file system block size */
4295 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4296 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4297 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4302 struct other_inode
{
4303 unsigned long orig_ino
;
4304 struct ext4_inode
*raw_inode
;
4307 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4310 struct other_inode
*oi
= (struct other_inode
*) data
;
4312 if ((inode
->i_ino
!= ino
) ||
4313 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4314 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4315 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4317 spin_lock(&inode
->i_lock
);
4318 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4319 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4320 (inode
->i_state
& I_DIRTY_TIME
)) {
4321 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4323 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4324 spin_unlock(&inode
->i_lock
);
4326 spin_lock(&ei
->i_raw_lock
);
4327 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4328 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4329 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4330 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4331 spin_unlock(&ei
->i_raw_lock
);
4332 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4335 spin_unlock(&inode
->i_lock
);
4340 * Opportunistically update the other time fields for other inodes in
4341 * the same inode table block.
4343 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4344 unsigned long orig_ino
, char *buf
)
4346 struct other_inode oi
;
4348 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4349 int inode_size
= EXT4_INODE_SIZE(sb
);
4351 oi
.orig_ino
= orig_ino
;
4352 ino
= orig_ino
& ~(inodes_per_block
- 1);
4353 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4354 if (ino
== orig_ino
)
4356 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4357 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4362 * Post the struct inode info into an on-disk inode location in the
4363 * buffer-cache. This gobbles the caller's reference to the
4364 * buffer_head in the inode location struct.
4366 * The caller must have write access to iloc->bh.
4368 static int ext4_do_update_inode(handle_t
*handle
,
4369 struct inode
*inode
,
4370 struct ext4_iloc
*iloc
)
4372 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4373 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4374 struct buffer_head
*bh
= iloc
->bh
;
4375 struct super_block
*sb
= inode
->i_sb
;
4376 int err
= 0, rc
, block
;
4377 int need_datasync
= 0, set_large_file
= 0;
4381 spin_lock(&ei
->i_raw_lock
);
4383 /* For fields not tracked in the in-memory inode,
4384 * initialise them to zero for new inodes. */
4385 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4386 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4388 ext4_get_inode_flags(ei
);
4389 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4390 i_uid
= i_uid_read(inode
);
4391 i_gid
= i_gid_read(inode
);
4392 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4393 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4394 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4396 * Fix up interoperability with old kernels. Otherwise, old inodes get
4397 * re-used with the upper 16 bits of the uid/gid intact
4400 raw_inode
->i_uid_high
=
4401 cpu_to_le16(high_16_bits(i_uid
));
4402 raw_inode
->i_gid_high
=
4403 cpu_to_le16(high_16_bits(i_gid
));
4405 raw_inode
->i_uid_high
= 0;
4406 raw_inode
->i_gid_high
= 0;
4409 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4410 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4411 raw_inode
->i_uid_high
= 0;
4412 raw_inode
->i_gid_high
= 0;
4414 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4416 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4417 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4418 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4419 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4421 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4423 spin_unlock(&ei
->i_raw_lock
);
4426 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4427 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4428 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4429 raw_inode
->i_file_acl_high
=
4430 cpu_to_le16(ei
->i_file_acl
>> 32);
4431 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4432 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4433 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4436 if (ei
->i_disksize
> 0x7fffffffULL
) {
4437 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4438 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4439 EXT4_SB(sb
)->s_es
->s_rev_level
==
4440 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4443 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4444 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4445 if (old_valid_dev(inode
->i_rdev
)) {
4446 raw_inode
->i_block
[0] =
4447 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4448 raw_inode
->i_block
[1] = 0;
4450 raw_inode
->i_block
[0] = 0;
4451 raw_inode
->i_block
[1] =
4452 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4453 raw_inode
->i_block
[2] = 0;
4455 } else if (!ext4_has_inline_data(inode
)) {
4456 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4457 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4460 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4461 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4462 if (ei
->i_extra_isize
) {
4463 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4464 raw_inode
->i_version_hi
=
4465 cpu_to_le32(inode
->i_version
>> 32);
4466 raw_inode
->i_extra_isize
=
4467 cpu_to_le16(ei
->i_extra_isize
);
4470 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4471 spin_unlock(&ei
->i_raw_lock
);
4472 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4473 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4476 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4477 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4480 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4481 if (set_large_file
) {
4482 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4483 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4486 ext4_update_dynamic_rev(sb
);
4487 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4488 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4489 ext4_handle_sync(handle
);
4490 err
= ext4_handle_dirty_super(handle
, sb
);
4492 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4495 ext4_std_error(inode
->i_sb
, err
);
4500 * ext4_write_inode()
4502 * We are called from a few places:
4504 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4505 * Here, there will be no transaction running. We wait for any running
4506 * transaction to commit.
4508 * - Within flush work (sys_sync(), kupdate and such).
4509 * We wait on commit, if told to.
4511 * - Within iput_final() -> write_inode_now()
4512 * We wait on commit, if told to.
4514 * In all cases it is actually safe for us to return without doing anything,
4515 * because the inode has been copied into a raw inode buffer in
4516 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4519 * Note that we are absolutely dependent upon all inode dirtiers doing the
4520 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4521 * which we are interested.
4523 * It would be a bug for them to not do this. The code:
4525 * mark_inode_dirty(inode)
4527 * inode->i_size = expr;
4529 * is in error because write_inode() could occur while `stuff()' is running,
4530 * and the new i_size will be lost. Plus the inode will no longer be on the
4531 * superblock's dirty inode list.
4533 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4537 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4540 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4541 if (ext4_journal_current_handle()) {
4542 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4548 * No need to force transaction in WB_SYNC_NONE mode. Also
4549 * ext4_sync_fs() will force the commit after everything is
4552 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4555 err
= ext4_force_commit(inode
->i_sb
);
4557 struct ext4_iloc iloc
;
4559 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4563 * sync(2) will flush the whole buffer cache. No need to do
4564 * it here separately for each inode.
4566 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4567 sync_dirty_buffer(iloc
.bh
);
4568 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4569 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4570 "IO error syncing inode");
4579 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4580 * buffers that are attached to a page stradding i_size and are undergoing
4581 * commit. In that case we have to wait for commit to finish and try again.
4583 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4587 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4588 tid_t commit_tid
= 0;
4591 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4593 * All buffers in the last page remain valid? Then there's nothing to
4594 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4597 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4600 page
= find_lock_page(inode
->i_mapping
,
4601 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4604 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4605 PAGE_CACHE_SIZE
- offset
);
4607 page_cache_release(page
);
4611 read_lock(&journal
->j_state_lock
);
4612 if (journal
->j_committing_transaction
)
4613 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4614 read_unlock(&journal
->j_state_lock
);
4616 jbd2_log_wait_commit(journal
, commit_tid
);
4623 * Called from notify_change.
4625 * We want to trap VFS attempts to truncate the file as soon as
4626 * possible. In particular, we want to make sure that when the VFS
4627 * shrinks i_size, we put the inode on the orphan list and modify
4628 * i_disksize immediately, so that during the subsequent flushing of
4629 * dirty pages and freeing of disk blocks, we can guarantee that any
4630 * commit will leave the blocks being flushed in an unused state on
4631 * disk. (On recovery, the inode will get truncated and the blocks will
4632 * be freed, so we have a strong guarantee that no future commit will
4633 * leave these blocks visible to the user.)
4635 * Another thing we have to assure is that if we are in ordered mode
4636 * and inode is still attached to the committing transaction, we must
4637 * we start writeout of all the dirty pages which are being truncated.
4638 * This way we are sure that all the data written in the previous
4639 * transaction are already on disk (truncate waits for pages under
4642 * Called with inode->i_mutex down.
4644 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4646 struct inode
*inode
= dentry
->d_inode
;
4649 const unsigned int ia_valid
= attr
->ia_valid
;
4651 error
= inode_change_ok(inode
, attr
);
4655 if (is_quota_modification(inode
, attr
))
4656 dquot_initialize(inode
);
4657 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4658 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4661 /* (user+group)*(old+new) structure, inode write (sb,
4662 * inode block, ? - but truncate inode update has it) */
4663 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4664 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4665 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4666 if (IS_ERR(handle
)) {
4667 error
= PTR_ERR(handle
);
4670 error
= dquot_transfer(inode
, attr
);
4672 ext4_journal_stop(handle
);
4675 /* Update corresponding info in inode so that everything is in
4676 * one transaction */
4677 if (attr
->ia_valid
& ATTR_UID
)
4678 inode
->i_uid
= attr
->ia_uid
;
4679 if (attr
->ia_valid
& ATTR_GID
)
4680 inode
->i_gid
= attr
->ia_gid
;
4681 error
= ext4_mark_inode_dirty(handle
, inode
);
4682 ext4_journal_stop(handle
);
4685 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4688 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4689 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4691 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4695 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4696 inode_inc_iversion(inode
);
4698 if (S_ISREG(inode
->i_mode
) &&
4699 (attr
->ia_size
< inode
->i_size
)) {
4700 if (ext4_should_order_data(inode
)) {
4701 error
= ext4_begin_ordered_truncate(inode
,
4706 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4707 if (IS_ERR(handle
)) {
4708 error
= PTR_ERR(handle
);
4711 if (ext4_handle_valid(handle
)) {
4712 error
= ext4_orphan_add(handle
, inode
);
4715 down_write(&EXT4_I(inode
)->i_data_sem
);
4716 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4717 rc
= ext4_mark_inode_dirty(handle
, inode
);
4721 * We have to update i_size under i_data_sem together
4722 * with i_disksize to avoid races with writeback code
4723 * running ext4_wb_update_i_disksize().
4726 i_size_write(inode
, attr
->ia_size
);
4727 up_write(&EXT4_I(inode
)->i_data_sem
);
4728 ext4_journal_stop(handle
);
4730 ext4_orphan_del(NULL
, inode
);
4734 loff_t oldsize
= inode
->i_size
;
4736 i_size_write(inode
, attr
->ia_size
);
4737 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
4741 * Blocks are going to be removed from the inode. Wait
4742 * for dio in flight. Temporarily disable
4743 * dioread_nolock to prevent livelock.
4746 if (!ext4_should_journal_data(inode
)) {
4747 ext4_inode_block_unlocked_dio(inode
);
4748 inode_dio_wait(inode
);
4749 ext4_inode_resume_unlocked_dio(inode
);
4751 ext4_wait_for_tail_page_commit(inode
);
4754 * Truncate pagecache after we've waited for commit
4755 * in data=journal mode to make pages freeable.
4757 truncate_pagecache(inode
, inode
->i_size
);
4760 * We want to call ext4_truncate() even if attr->ia_size ==
4761 * inode->i_size for cases like truncation of fallocated space
4763 if (attr
->ia_valid
& ATTR_SIZE
)
4764 ext4_truncate(inode
);
4767 setattr_copy(inode
, attr
);
4768 mark_inode_dirty(inode
);
4772 * If the call to ext4_truncate failed to get a transaction handle at
4773 * all, we need to clean up the in-core orphan list manually.
4775 if (orphan
&& inode
->i_nlink
)
4776 ext4_orphan_del(NULL
, inode
);
4778 if (!rc
&& (ia_valid
& ATTR_MODE
))
4779 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
4782 ext4_std_error(inode
->i_sb
, error
);
4788 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4791 struct inode
*inode
;
4792 unsigned long long delalloc_blocks
;
4794 inode
= dentry
->d_inode
;
4795 generic_fillattr(inode
, stat
);
4798 * If there is inline data in the inode, the inode will normally not
4799 * have data blocks allocated (it may have an external xattr block).
4800 * Report at least one sector for such files, so tools like tar, rsync,
4801 * others doen't incorrectly think the file is completely sparse.
4803 if (unlikely(ext4_has_inline_data(inode
)))
4804 stat
->blocks
+= (stat
->size
+ 511) >> 9;
4807 * We can't update i_blocks if the block allocation is delayed
4808 * otherwise in the case of system crash before the real block
4809 * allocation is done, we will have i_blocks inconsistent with
4810 * on-disk file blocks.
4811 * We always keep i_blocks updated together with real
4812 * allocation. But to not confuse with user, stat
4813 * will return the blocks that include the delayed allocation
4814 * blocks for this file.
4816 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4817 EXT4_I(inode
)->i_reserved_data_blocks
);
4818 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
4822 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
4825 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4826 return ext4_ind_trans_blocks(inode
, lblocks
);
4827 return ext4_ext_index_trans_blocks(inode
, pextents
);
4831 * Account for index blocks, block groups bitmaps and block group
4832 * descriptor blocks if modify datablocks and index blocks
4833 * worse case, the indexs blocks spread over different block groups
4835 * If datablocks are discontiguous, they are possible to spread over
4836 * different block groups too. If they are contiguous, with flexbg,
4837 * they could still across block group boundary.
4839 * Also account for superblock, inode, quota and xattr blocks
4841 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
4844 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4850 * How many index blocks need to touch to map @lblocks logical blocks
4851 * to @pextents physical extents?
4853 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
4858 * Now let's see how many group bitmaps and group descriptors need
4861 groups
= idxblocks
+ pextents
;
4863 if (groups
> ngroups
)
4865 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4866 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4868 /* bitmaps and block group descriptor blocks */
4869 ret
+= groups
+ gdpblocks
;
4871 /* Blocks for super block, inode, quota and xattr blocks */
4872 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4878 * Calculate the total number of credits to reserve to fit
4879 * the modification of a single pages into a single transaction,
4880 * which may include multiple chunks of block allocations.
4882 * This could be called via ext4_write_begin()
4884 * We need to consider the worse case, when
4885 * one new block per extent.
4887 int ext4_writepage_trans_blocks(struct inode
*inode
)
4889 int bpp
= ext4_journal_blocks_per_page(inode
);
4892 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
4894 /* Account for data blocks for journalled mode */
4895 if (ext4_should_journal_data(inode
))
4901 * Calculate the journal credits for a chunk of data modification.
4903 * This is called from DIO, fallocate or whoever calling
4904 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4906 * journal buffers for data blocks are not included here, as DIO
4907 * and fallocate do no need to journal data buffers.
4909 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4911 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4915 * The caller must have previously called ext4_reserve_inode_write().
4916 * Give this, we know that the caller already has write access to iloc->bh.
4918 int ext4_mark_iloc_dirty(handle_t
*handle
,
4919 struct inode
*inode
, struct ext4_iloc
*iloc
)
4923 if (IS_I_VERSION(inode
))
4924 inode_inc_iversion(inode
);
4926 /* the do_update_inode consumes one bh->b_count */
4929 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4930 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4936 * On success, We end up with an outstanding reference count against
4937 * iloc->bh. This _must_ be cleaned up later.
4941 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4942 struct ext4_iloc
*iloc
)
4946 err
= ext4_get_inode_loc(inode
, iloc
);
4948 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4949 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4955 ext4_std_error(inode
->i_sb
, err
);
4960 * Expand an inode by new_extra_isize bytes.
4961 * Returns 0 on success or negative error number on failure.
4963 static int ext4_expand_extra_isize(struct inode
*inode
,
4964 unsigned int new_extra_isize
,
4965 struct ext4_iloc iloc
,
4968 struct ext4_inode
*raw_inode
;
4969 struct ext4_xattr_ibody_header
*header
;
4971 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4974 raw_inode
= ext4_raw_inode(&iloc
);
4976 header
= IHDR(inode
, raw_inode
);
4978 /* No extended attributes present */
4979 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4980 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4981 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4983 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4987 /* try to expand with EAs present */
4988 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4993 * What we do here is to mark the in-core inode as clean with respect to inode
4994 * dirtiness (it may still be data-dirty).
4995 * This means that the in-core inode may be reaped by prune_icache
4996 * without having to perform any I/O. This is a very good thing,
4997 * because *any* task may call prune_icache - even ones which
4998 * have a transaction open against a different journal.
5000 * Is this cheating? Not really. Sure, we haven't written the
5001 * inode out, but prune_icache isn't a user-visible syncing function.
5002 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5003 * we start and wait on commits.
5005 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5007 struct ext4_iloc iloc
;
5008 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5009 static unsigned int mnt_count
;
5013 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5014 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5015 if (ext4_handle_valid(handle
) &&
5016 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5017 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5019 * We need extra buffer credits since we may write into EA block
5020 * with this same handle. If journal_extend fails, then it will
5021 * only result in a minor loss of functionality for that inode.
5022 * If this is felt to be critical, then e2fsck should be run to
5023 * force a large enough s_min_extra_isize.
5025 if ((jbd2_journal_extend(handle
,
5026 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5027 ret
= ext4_expand_extra_isize(inode
,
5028 sbi
->s_want_extra_isize
,
5031 ext4_set_inode_state(inode
,
5032 EXT4_STATE_NO_EXPAND
);
5034 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5035 ext4_warning(inode
->i_sb
,
5036 "Unable to expand inode %lu. Delete"
5037 " some EAs or run e2fsck.",
5040 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5046 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5051 * ext4_dirty_inode() is called from __mark_inode_dirty()
5053 * We're really interested in the case where a file is being extended.
5054 * i_size has been changed by generic_commit_write() and we thus need
5055 * to include the updated inode in the current transaction.
5057 * Also, dquot_alloc_block() will always dirty the inode when blocks
5058 * are allocated to the file.
5060 * If the inode is marked synchronous, we don't honour that here - doing
5061 * so would cause a commit on atime updates, which we don't bother doing.
5062 * We handle synchronous inodes at the highest possible level.
5064 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5065 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5066 * to copy into the on-disk inode structure are the timestamp files.
5068 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5072 if (flags
== I_DIRTY_TIME
)
5074 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5078 ext4_mark_inode_dirty(handle
, inode
);
5080 ext4_journal_stop(handle
);
5087 * Bind an inode's backing buffer_head into this transaction, to prevent
5088 * it from being flushed to disk early. Unlike
5089 * ext4_reserve_inode_write, this leaves behind no bh reference and
5090 * returns no iloc structure, so the caller needs to repeat the iloc
5091 * lookup to mark the inode dirty later.
5093 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5095 struct ext4_iloc iloc
;
5099 err
= ext4_get_inode_loc(inode
, &iloc
);
5101 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5102 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5104 err
= ext4_handle_dirty_metadata(handle
,
5110 ext4_std_error(inode
->i_sb
, err
);
5115 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5122 * We have to be very careful here: changing a data block's
5123 * journaling status dynamically is dangerous. If we write a
5124 * data block to the journal, change the status and then delete
5125 * that block, we risk forgetting to revoke the old log record
5126 * from the journal and so a subsequent replay can corrupt data.
5127 * So, first we make sure that the journal is empty and that
5128 * nobody is changing anything.
5131 journal
= EXT4_JOURNAL(inode
);
5134 if (is_journal_aborted(journal
))
5136 /* We have to allocate physical blocks for delalloc blocks
5137 * before flushing journal. otherwise delalloc blocks can not
5138 * be allocated any more. even more truncate on delalloc blocks
5139 * could trigger BUG by flushing delalloc blocks in journal.
5140 * There is no delalloc block in non-journal data mode.
5142 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5143 err
= ext4_alloc_da_blocks(inode
);
5148 /* Wait for all existing dio workers */
5149 ext4_inode_block_unlocked_dio(inode
);
5150 inode_dio_wait(inode
);
5152 jbd2_journal_lock_updates(journal
);
5155 * OK, there are no updates running now, and all cached data is
5156 * synced to disk. We are now in a completely consistent state
5157 * which doesn't have anything in the journal, and we know that
5158 * no filesystem updates are running, so it is safe to modify
5159 * the inode's in-core data-journaling state flag now.
5163 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5165 err
= jbd2_journal_flush(journal
);
5167 jbd2_journal_unlock_updates(journal
);
5168 ext4_inode_resume_unlocked_dio(inode
);
5171 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5173 ext4_set_aops(inode
);
5175 jbd2_journal_unlock_updates(journal
);
5176 ext4_inode_resume_unlocked_dio(inode
);
5178 /* Finally we can mark the inode as dirty. */
5180 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5182 return PTR_ERR(handle
);
5184 err
= ext4_mark_inode_dirty(handle
, inode
);
5185 ext4_handle_sync(handle
);
5186 ext4_journal_stop(handle
);
5187 ext4_std_error(inode
->i_sb
, err
);
5192 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5194 return !buffer_mapped(bh
);
5197 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5199 struct page
*page
= vmf
->page
;
5203 struct file
*file
= vma
->vm_file
;
5204 struct inode
*inode
= file_inode(file
);
5205 struct address_space
*mapping
= inode
->i_mapping
;
5207 get_block_t
*get_block
;
5210 sb_start_pagefault(inode
->i_sb
);
5211 file_update_time(vma
->vm_file
);
5212 /* Delalloc case is easy... */
5213 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5214 !ext4_should_journal_data(inode
) &&
5215 !ext4_nonda_switch(inode
->i_sb
)) {
5217 ret
= __block_page_mkwrite(vma
, vmf
,
5218 ext4_da_get_block_prep
);
5219 } while (ret
== -ENOSPC
&&
5220 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5225 size
= i_size_read(inode
);
5226 /* Page got truncated from under us? */
5227 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5229 ret
= VM_FAULT_NOPAGE
;
5233 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5234 len
= size
& ~PAGE_CACHE_MASK
;
5236 len
= PAGE_CACHE_SIZE
;
5238 * Return if we have all the buffers mapped. This avoids the need to do
5239 * journal_start/journal_stop which can block and take a long time
5241 if (page_has_buffers(page
)) {
5242 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5244 ext4_bh_unmapped
)) {
5245 /* Wait so that we don't change page under IO */
5246 wait_for_stable_page(page
);
5247 ret
= VM_FAULT_LOCKED
;
5252 /* OK, we need to fill the hole... */
5253 if (ext4_should_dioread_nolock(inode
))
5254 get_block
= ext4_get_block_write
;
5256 get_block
= ext4_get_block
;
5258 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5259 ext4_writepage_trans_blocks(inode
));
5260 if (IS_ERR(handle
)) {
5261 ret
= VM_FAULT_SIGBUS
;
5264 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5265 if (!ret
&& ext4_should_journal_data(inode
)) {
5266 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5267 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5269 ret
= VM_FAULT_SIGBUS
;
5270 ext4_journal_stop(handle
);
5273 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5275 ext4_journal_stop(handle
);
5276 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5279 ret
= block_page_mkwrite_return(ret
);
5281 sb_end_pagefault(inode
->i_sb
);